Despite significant advancements in dynamic neural rendering, existing methods fail to address the unique challenges posed by UAV-captured scenarios, particularly those involving monocular camera setups, top-down perspective, and multiple small, moving humans, which are not adequately represented in existing datasets. In this work, we introduce UAV4D, a framework for enabling photorealistic rendering for dynamic real-world scenes captured by UAVs. Specifically, we address the challenge of reconstructing dynamic scenes with multiple moving pedestrians from monocular video data without the need for additional sensors. We use a combination of a 3D foundation model and a human mesh reconstruction model to reconstruct both the scene background and humans. We propose a novel approach to resolve the scene scale ambiguity and place both humans and the scene in world coordinates by identifying human-scene contact points. Additionally, we exploit the SMPL model and background mesh to initialize Gaussian splats, enabling holistic scene rendering. We evaluated our method on three complex UAV-captured datasets: VisDrone, Manipal-UAV, and Okutama-Action, each with distinct characteristics and 10-50 humans. Our results demonstrate the benefits of our approach over existing methods in novel view synthesis, achieving a 1.5 dB PSNR improvement and superior visual sharpness.

Audio comprehension-including speech, non-speech sounds, and music-is essential for achieving human-level intelligence. Consequently, AI agents must demonstrate holistic audio understanding to qualify as generally intelligent. However, evaluating auditory intelligence comprehensively remains challenging. To address this gap, we introduce MMAU-Pro, the most comprehensive and rigorously curated benchmark for assessing audio intelligence in AI systems. MMAU-Pro contains 5,305 instances, where each instance has one or more audios paired with human expert-generated question-answer pairs, spanning speech, sound, music, and their combinations. Unlike existing benchmarks, MMAU-Pro evaluates auditory intelligence across 49 unique skills and multiple complex dimensions, including long-form audio comprehension, spatial audio reasoning, multi-audio understanding, among others. All questions are meticulously designed to require deliberate multi-hop reasoning, including both multiple-choice and open-ended response formats. Importantly, audio data is sourced directly ``from the wild" rather than from existing datasets with known distributions. We evaluate 22 leading open-source and proprietary multimodal AI models, revealing significant limitations: even state-of-the-art models such as Gemini 2.5 Flash and Audio Flamingo 3 achieve only 59.2% and 51.7% accuracy, respectively, approaching random performance in multiple categories. Our extensive analysis highlights specific shortcomings and provides novel insights, offering actionable perspectives for the community to enhance future AI systems' progression toward audio general intelligence. The benchmark and code is available at https://sonalkum.github.io/mmau-pro.

We address the critical gap between the computational demands of vision-language models and the possible ultra-low-bit weight precision (bitwidth

Retrieving music using natural language descriptions has improved with contrastive audio-text models such as CLAP, but current systems remain limited to coarse semantic queries.When descriptions specify fine-grained musical attributes such as tempo, key, chord progression, or rhythmic structure, existing models often fail to retrieve the correct audio.We show that this limitation stems from the contrastive learning objective itself: despite being trained on long captions, CLAP-based models effectively utilize only the first few tokens, discarding much of the information encoded in detailed prompts.Then, we propose FIGMA (FIne-Grained Music RetrievAl), a multi-view contrastive architecture that addresses this limitation by jointly optimizing global audio-text alignment and frame-level, token-wise alignment.This design enables FIGMA to capture both high-level semantic context and finegrained musical attributes within a unified representation space.Moreover, we formalize the task of Fine-Grained Music Retrieval and construct Fine-Grained Music Caption dataset (FGMCaps), a large-scale dataset of 380K music-caption pairs for training along with a 10K test set, both annotated with tempo, key, chord progression, beat count, as well as genre and mood.Extensive experiments demonstrate that FIGMA consistently outperforms existing CLAP-based music retrieval models across multiple music retrieval benchmarks, including out-of-domain evaluations, with relative improvements of up to 73.3%.

Most existing traffic video datasets including Waymo are structured, focusing predominantly on Western traffic, which hinders global applicability. Specifically, most Asian scenarios are far more complex, involving numerous objects with distinct motions and behaviors. Addressing this gap, we present a new dataset, DAVE, designed for evaluating perception methods with high representation of Vulnerable Road Users (VRUs: e.g. pedestrians, animals, motorbikes, and bicycles) in complex and unpredictable environments. DAVE is a manually annotated dataset encompassing 16 diverse actor categories (spanning animals, humans, vehicles, etc.) and 16 action types (complex and rare cases like cut-ins, zigzag movement, U-turn, etc.), which require high reasoning ability. DAVE densely annotates over 13 million bounding boxes (bboxes) actors with identification, and more than 1.6 million boxes are annotated with both actor identification and action/behavior details. The videos within DAVE are collected based on a broad spectrum of factors, such as weather conditions, the time of day, road scenarios, and traffic density. DAVE can benchmark video tasks like Tracking, Detection, Spatiotemporal Action Localization, Language-Visual Moment retrieval, and Multi-label Video Action Recognition. Given the critical importance of accurately identifying VRUs to prevent accidents and ensure road safety, in DAVE, vulnerable road users constitute 41.13% of instances, compared to 23.71% in Waymo. DAVE provides an invaluable resource for the development of more sensitive and accurate visual perception algorithms in the complex real world. Our experiments show that existing methods suffer degradation in performance when evaluated on DAVE, highlighting its benefit for future video recognition research.

Monocular 3D foundation models offer an extensible solution for perception tasks, making them attractive for broader 3D vision applications. In this paper, we propose MoRe, a training-free Monocular Geometry Refinement method designed to improve cross-view consistency and achieve scale alignment. To induce inter-frame relationships, our method employs feature matching between frames to establish correspondences. Rather than applying simple least squares optimization on these matched points, we formulate a graph-based optimization framework that performs local planar approximation using the estimated 3D points and surface normals estimated by monocular foundation models. This formulation addresses the scale ambiguity inherent in monocular geometric priors while preserving the underlying 3D structure. We further demonstrate that MoRe not only enhances 3D reconstruction but also improves novel view synthesis, particularly in sparse-view rendering scenarios.

Manan Suri, Puneet Mathur, Nedim Lipka, Franck Dernoncourt, Ryan A. Rossi, Dinesh Manocha. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

We present ImPoster, a novel algorithm for generating a target image of a ‘source’ subject performing a ‘driving’ action. The inputs to our algorithm are a single pair of a source image with the subject that we wish to edit and a driving image with a subject of an arbitrary class performing the driving action, along with the text descriptions of the two images. Our approach is completely unsupervised and does not require any access to additional annotations like keypoints or pose. Our approach builds on a pretrained text-to-image latent diffusion model and learns the characteristics of the source and the driving image by finetuning the diffusion model for a small number of iterations. At inference time, ImPoster performs step-wise text prompting i.e. it denoises by first moving in the direction of the image manifold corresponding to the driving image followed by the direction of the image manifold corresponding to the text description of the desired target image. We propose a novel diffusion guidance formulation, image frequency guidance, to steer the generation towards the manifold of the source subject and the driving action at every step of the inference denoising. Our frequency guidance formulations are derived from the frequency domain properties of images. We extensively evaluate ImPoster on a diverse set of source-driving image pairs to demonstrate improvements over baselines. To the best of our knowledge, ImPoster is the first approach towards achieving both subject-driven as well as action-driven image personalization.

With the widespread deployment of Multimodal Large Language Models (MLLMs) for visual-reasoning tasks, improving their safety has become crucial. Recent research indicates that despite training-time safety alignment, these models remain vulnerable to jailbreak attacks. In this work, we first highlight an important safety gap to describe that alignment achieved solely through safety training may be insufficient against jailbreak attacks. To address this vulnerability, we propose Immune, an inference-time defense framework that leverages a safety reward model through controlled decoding to defend against jailbreak attacks. Additionally, we provide a mathematical characterization of Immune, offering insights on why it improves safety against jailbreaks. Extensive evaluations on diverse jailbreak benchmarks using recent MLLMs reveal that Immune effectively enhances model safety while preserving the model’s original capabilities. For instance, against text-based jailbreak attacks on LLaVA-1.6, Immune reduces the attack success rate by 57.82% and 16.78% compared to the base MLLM and state-of-the-art defense strategy, respectively.

We introduce the first learning-based dense matching algorithm, termed Equirectangular Projection-Oriented Dense Kernelized Feature Matching (EDM), specifically designed for omnidirectional images. Equirectangular projection (ERP) images, with their large fields of view, are particularly suited for dense matching techniques that aim to establish comprehensive correspondences across images. However, ERP images are subject to significant distortions, which we address by leveraging the spherical camera model and geodesic flow refinement in the dense matching method. To further mitigate these distortions, we propose spherical positional embeddings based on 3D Cartesian coordinates of the feature grid. Additionally, our method incorporates bidirectional transformations between spherical and Cartesian coordinate systems during refinement, utilizing a unit sphere to improve matching performance. We demonstrate that our proposed method achieves notable performance enhancements, with improvements of +26.72 and +42.62 in AUC@5° on the Matterport3D and Stanford2D3D datasets. Project Page: https://jdk9405.github.io/EDM

Ramaneswaran Selvakumar, Ashish Seth, Nishit Anand, Utkarsh Tyagi, Sonal Kumar, Sreyan Ghosh, Dinesh Manocha. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Ashish Seth, Utkarsh Tyagi, Ramaneswaran Selvakumar, Nishit Anand, Sonal Kumar, Sreyan Ghosh, Ramani Duraiswami, Chirag Agarwal, Dinesh Manocha. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Flowcharts are a critical tool for visualizing decision-making processes.However, their non-linear structure and complex visual-textual relationships make it challenging to interpret them using LLMs, as vision-language models frequently hallucinate nonexistent connections and decision paths when analyzing these diagrams.This leads to compromised reliability for automated flowchart processing in critical domains such as logistics, health, and engineering.We introduce the task of Fine-grained Flowchart Attribution, which traces specific components grounding a flowchart referring LLM response.Flowchart Attribution ensures the verifiability of LLM predictions and improves explainability by linking generated responses to the flowchart's structure.We propose FlowPathAgent, a neurosymbolic agent that performs fine-grained post hoc attribution through graph-based reasoning.It first segments the flowchart, then converts it into a structured symbolic graph, and then employs an agentic approach to dynamically interact with the graph, to generate attribution paths.Additionally, we present FlowExplainBench, a novel benchmark for evaluating flowchart attributions across diverse styles, domains, and question types.Experimental results show that FlowPathAgent mitigates visual hallucinations in LLM answers over flowchart QA, outperforming strong baselines by 10-14% on our proposed FlowExplainBench dataset.

Open-vocabulary audio-language models, like CLAP [1], offer a promising approach for zero-shot audio classification (ZSAC) by enabling classification with any arbitrary set of categories specified with natural language prompts. In this paper, we propose a simple but effective method to improve ZSAC with CLAP. Specifically, we shift from the conventional method of using prompts with abstract category labels (e.g., Sound of an organ) to prompts that describe sounds using their inherent descriptive features in a diverse context (e.g., The organ’s deep and resonant tones filled the cathedral.). To achieve this, we first propose ReCLAP, a CLAP model trained with rewritten audio captions for improved understanding of sounds in the wild. These rewritten captions describe each sound event in the original caption using their unique discriminative characteristics. ReCLAP outperforms all baselines on both multi-modal audio-text retrieval and ZSAC. Next, to improve zero-shot audio classification with ReCLAP, we propose prompt augmentation. In contrast to the traditional method of employing hand-written template prompts, we generate custom prompts for each unique label in the dataset. These custom prompts first describe the sound event in the label and then employ them in diverse scenes. Our proposed method improves ReCLAP’s performance on ZSAC by 1%-18% and outperforms 1 all baselines by 1% - 55%1.

Modern perception models, particularly those designed for multisensory egocentric tasks, have achieved remarkable performance but often come with substantial computational costs. These high demands pose challenges for real-world deployment, especially in resource-constrained environments. In this paper, we introduce EgoAdapt, a framework that adaptively performs cross-modal distillation and policy learning to enable efficient inference across different egocentric perception tasks, including egocentric action recognition, active speaker localization, and behavior anticipation. Our proposed policy module is adaptable to task-specific action spaces, making it broadly applicable. Experimental results on three challenging egocentric datasets EPIC-Kitchens, EasyCom, and Aria Everyday Activities demonstrate that our method significantly enhances efficiency, reducing GMACs by up to 89.09%, parameters up to 82.02%, and energy up to 9.6x, while still on-par and in many cases outperforming, the performance of corresponding state-of-the-art models.

Recent advancements in reasoning optimization have greatly enhanced the performance of large language models (LLMs). However, existing work fails to address the complexities of audio-visual scenarios, underscoring the need for further research. In this paper, we introduce AURELIA, a novel actor-critic based audio-visual (AV) reasoning framework that distills structured, step-by-step reasoning into AVLLMs at test time, improving their ability to process complex multi-modal inputs without additional training or fine-tuning. To further advance AVLLM reasoning skills, we present AVReasonBench, a challenging benchmark comprising 4500 audio-visual questions, each paired with detailed step-by-step reasoning. Our benchmark spans six distinct tasks, including AV-GeoIQ, which evaluates AV reasoning combined with geographical and cultural knowledge. Evaluating 18 AVLLMs on AVReasonBench reveals significant limitations in their multi-modal reasoning capabilities. Using AURELIA, we achieve up to a 100% relative improvement, demonstrating its effectiveness. This performance gain highlights the potential of reasoning-enhanced data generation for advancing AVLLMs in real-world applications. Our code and data will be publicly released at: https: //github.com/schowdhury671/aurelia.

With the rapid advancement of Multi-modal Large Language Models (MLLMs), several diagnostic benchmarks have recently been developed to assess these models' multi-modal reasoning proficiency. However, these benchmarks are restricted to assessing primarily the visual aspect and do not examine the holistic audio-visual (AV) understanding. Moreover, currently, there are no benchmarks that investigate the capabilities of AVLLMs to calibrate their responses when presented with perturbed inputs. To this end, we introduce Audio-Visual Trustworthiness assessment Benchmark (AVTrustBench), comprising 600K samples spanning over 9 meticulously crafted tasks, evaluating the capabilities of AVLLMs across three distinct dimensions: Adversarial attack, Compositional reasoning, and Modality-specific dependency. Using our benchmark we extensively evaluate 13 state-of-the-art AVLLMs. The findings reveal that the majority of existing models fall significantly short of achieving human-like comprehension, offering valuable insights for future research directions. To alleviate the limitations in the existing approaches, we further propose a robust, model-agnostic calibrated audio-visual preference optimization based training strategy CAVPref, obtaining a gain up to 30.19% across all 9 tasks. We will publicly release our code and benchmark to facilitate future research in this direction.

Neural networks that map between low dimensional spaces are ubiquitous in computer graphics and scientific computing; however, in their naive implementation, they are unable to learn high frequency information. We present a comprehensive analysis comparing the two most common techniques for mitigating this spectral bias: Fourier feature encodings (FFE) and multigrid parametric encodings (MPE). FFEs are seen as the standard for low dimensional mappings, but MPEs often outperform them and learn representations with higher resolution and finer detail. FFE's roots in the Fourier transform, make it susceptible to aliasing if pushed too far, while MPEs, which use a learned grid structure, have no such limitation. To understand the difference in performance, we use the neural tangent kernel (NTK) to evaluate these encodings through the lens of an analogous kernel regression. By finding a lower bound on the smallest eigenvalue of the NTK, we prove that MPEs improve a network's performance through the structure of their grid and not their learnable embedding. This mechanism is fundamentally different from FFEs, which rely solely on their embedding space to improve performance. Results are empirically validated on a 2D image regression task using images taken from 100 synonym sets of ImageNet and 3D implicit surface regression on objects from the Stanford graphics dataset. Using peak signal-to-noise ratio (PSNR) and multiscale structural similarity (MS-SSIM) to evaluate how well fine details are learned, we show that the MPE increases the minimum eigenvalue by 8 orders of magnitude over the baseline and 2 orders of magnitude over the FFE. The increase in spectrum corresponds to a 15 dB (PSNR) / 0.65 (MS-SSIM) increase over baseline and a 12 dB (PSNR) / 0.33 (MS-SSIM) increase over the FFE.

Alignment of Large Language models (LLMs) is crucial for safe and trustworthy deployment in applications. Reinforcement learning from human feedback (RLHF) has emerged as an effective technique to align LLMs to human preferences, and broader utilities, but it requires updating billions of model parameters which is computationally expensive. Controlled Decoding, by contrast, provides a mechanism for aligning a model at inference time without retraining. However, single-agent decoding approaches often struggle to adapt to diverse tasks due to the complexity and variability inherent in these tasks. To strengthen the test-time performance w.r.t the target task, we propose a mixture of agents-based decoding strategies leveraging the existing off-the-shelf aligned LLM policies. Treating each prior policy as an agent in the spirit of mixture of agent collaboration, we develop a decoding method that allows for inference-time alignment through a token-level selection strategy among multiple agents. For each token, the most suitable LLM is dynamically chosen from a pool of models based on a long-term utility metric. This policy-switching mechanism ensures optimal model selection at each step, enabling efficient collaboration and alignment among LLMs during decoding. Theoretical analysis of our proposed algorithm establishes optimal performance with respect to the target task represented via a target reward, for the given off-the-shelf models. We conduct comprehensive empirical evaluations with open-source aligned models on diverse tasks and preferences, which demonstrates the merits of this approach over single-agent decoding baselines. Notably, COLLAB surpasses the current SoTA decoding strategy, achieving an improvement of {up to 1.56x} in average reward and $71.89\%$ in GPT-4 based win-tie rate.

Large Vision-Language Models (LVLMs) often produce responses that misalign with factual information, a phenomenon known as hallucinations. While hallucinations are well-studied, the exact causes behind them remain underexplored. In this paper, we first investigate the root causes of hallucinations in LVLMs. Our findings reveal that existing mitigation techniques primarily reduce hallucinations for visual recognition prompts—those that require simple descriptions of visual elements—but fail for cognitive prompts that demand deliberate reasoning. We identify the core issue as a lack of true visual perception in LVLMs: although they can accurately recognize visual elements, they struggle to fully interpret these elements in the context of the input prompt and effectively link this recognition to their internal knowledge, which is critical for reasoning. To address this gap, we introduce Visual Description Grounded Decoding (VDGD), a simple, robust, and training-free method designed to enhance visual perception and improve reasoning capabilities in LVLMs. VDGD works by first generating a detailed description of the image and appending it as a prefix to the instruction. During response generation, tokens are sampled based on their KL divergence to the description, favoring candidates with lower divergence. Experimental results on multiple visual reasoning benchmarks and LVLMs demonstrate that VDGD consistently outperforms existing baselines 2% - 33%. Finally, we introduce VaLLu, a benchmark designed for comprehensive evaluation of the cognitive capabilities of LVLMs.

We present Synthio, a novel approach for augmenting small-scale audio classification datasets with synthetic data. Our goal is to improve audio classification accuracy with limited labeled data. Traditional data augmentation techniques, which apply artificial transformations (e.g., adding random noise or masking segments), struggle to create data that captures the true diversity present in real-world audios. To address this shortcoming, we propose to augment the dataset with synthetic audio generated from text-to-audio (T2A) diffusion models. However, synthesizing effective augmentations is challenging because not only should the generated data be acoustically consistent with the underlying small-scale dataset, but they should also have sufficient compositional diversity. To overcome the first challenge, we align the generations of the T2A model with the small-scale dataset using preference optimization. This ensures that the acoustic characteristics of the generated data remain consistent with the small-scale dataset. To address the second challenge, we propose a novel caption generation technique that leverages the reasoning capabilities of Large Language Models to (1) generate diverse and meaningful audio captions and (2) iteratively refine their quality. The generated captions are then used to prompt the aligned T2A model. We extensively evaluate Synthio on ten datasets and four simulated limited-data settings. Results indicate our method consistently outperforms all baselines by 0.1%-39% using a T2A model trained only on weakly-captioned AudioSet.

Large Language Models (LLMs) have become an indispensable part of natural language processing tasks. However, autoregressive sampling has become an efficiency bottleneck. Multi-Draft Speculative Decoding (MDSD) is a recent approach where, when generating each token, a small draft model generates multiple drafts, and the target LLM verifies them in parallel, ensuring that the final output conforms to the target model distribution. The two main design choices in MDSD are the draft sampling method and the verification algorithm. For a fixed draft sampling method, the optimal acceptance rate is a solution to an optimal transport problem, but the complexity of this problem makes it difficult to solve for the optimal acceptance rate and measure the gap between existing verification algorithms and the theoretical upper bound. This paper discusses the dual of the optimal transport problem, providing a way to efficiently compute the optimal acceptance rate. For the first time, we measure the theoretical upper bound of MDSD efficiency for vocabulary sizes in the thousands and quantify the gap between existing verification algorithms and this bound. We also compare different draft sampling methods based on their optimal acceptance rates. Our results show that the draft sampling method strongly influences the optimal acceptance rate, with sampling without replacement outperforming sampling with replacement. Additionally, existing verification algorithms do not reach the theoretical upper bound for both without replacement and with replacement sampling. Our findings suggest that carefully designed draft sampling methods can potentially improve the optimal acceptance rate and enable the development of verification algorithms that closely match the theoretical upper bound.

The ability to comprehend audio—which includes speech, non-speech sounds, and music—is crucial for AI agents to interact effectively with the world. We present MMAU, a novel benchmark designed to evaluate multimodal audio understanding models on tasks requiring expert-level knowledge and complex reasoning. MMAU comprises 10k carefully curated audio clips paired with human-annotated natural language questions and answers spanning speech, environmental sounds, and music. It includes information extraction and reasoning questions, requiring models to demonstrate 27 distinct skills across unique and challenging tasks. Unlike existing benchmarks, MMAU emphasizes advanced perception and reasoning with domain-specific knowledge, challenging models to tackle tasks akin to those faced by experts. We assess 18 open-source and proprietary (Large) Audio-Language Models, demonstrating the significant challenges posed by MMAU. Notably, even the most advanced Gemini 2.0 Flash achieves only 59.93% accuracy, and the state-of-the-art open-source Qwen2-Audio achieves only 52.50%, highlighting considerable room for improvement. We believe MMAU will drive the audio and multimodal research community to develop more advanced audio understanding models capable of solving complex audio tasks.

Aligning large language models with humans is challenging due to the inherently multifaceted nature of preference feedback. While existing approaches typically frame this as a multi-objective optimization problem, they often overlook how humans actually make decisions. Research on bounded rationality suggests that human decision making follows satisficing strategies- optimizing primary objectives while ensuring others meet acceptable thresholds. To bridge this gap and operationalize the notion of satisficing alignment, we propose SITAlign: an inference-time framework that addresses the multifaceted nature of alignment by maximizing a primary objective while satisfying threshold-based constraints on secondary criteria. We provide theoretical insights by deriving sub-optimality bounds of our satisficing-based inference alignment approach. We empirically validate SITAlign’s performance through extensive experimentation on multiple benchmarks. For instance, on the PKU-SafeRLHF dataset with the primary objective of maximizing helpfulness while ensuring a threshold on harmlessness, SITAlign outperforms the state-of-the-art multi-objective decoding strategy by a margin of 22.3% in terms of GPT-4 win-tie rate for helpfulness reward while adhering to the threshold on harmlessness.

Understanding and reasoning over non-speech sounds and music are crucial for both humans and AI agents to interact effectively with their environments. In this paper, we introduce Audio Flamingo 2 (AF2), an Audio-Language Model (ALM) with advanced audio understanding and reasoning capabilities. AF2 leverages (i) a custom CLAP model, (ii) synthetic Audio QA data for fine-grained audio reasoning, and (iii) a multi-stage curriculum learning strategy. AF2 achieves state-of-the-art performance with only a 3B parameter small language model, surpassing large open-source and proprietary models across over 20 benchmarks. Next, for the first time, we extend audio understanding to long audio segments (30 secs to 5 mins) and propose LongAudio, a large and novel dataset for training ALMs on long audio captioning and question-answering tasks. Fine-tuning AF2 on LongAudio leads to exceptional performance on our proposed LongAudioBench, an expert-annotated benchmark for evaluating ALMs on long audio understanding capabilities. We conduct extensive ablation studies to confirm the efficacy of our approach.

We propose a new method for improving zero-shot ObjectNav that aims to utilize potentially available environmental percepts for navigational assistance. Our approach takes into account that the ground agent may have limited and sometimes obstructed view. Our formulation encourages Generative Communication (GC) between an assistive overhead agent with a global view containing the target object and the ground agent with an obfuscated view; both equipped with Vision-Language Models (VLMs) for vision-to-language translation. In this assisted setup, the embodied agents communicate environmental information before the ground agent executes actions towards a target. Despite the overhead agent having a global view with the target, we note a drop in performance (−13% in OSR and −13% in SPL) of a fully cooperative assistance scheme over an unassisted baseline. In contrast, a selective assistance scheme where the ground agent retains its independent exploratory behaviour shows a 10% OSR and 7.65% SPL improvement. To explain navigation performance, we analyze the GC for unique traits, quantifying the presence of hallucination and cooperation. Specifically, we identify the novel linguistic trait of preemptive hallucination in our embodied setting, where the overhead agent assumes that the ground agent has executed an action in the dialogue when it is yet to move, and note its strong correlation with navigation performance. We conduct real-world experiments and present some qualitative examples where we mitigate hallucinations via prompt finetuning to improve ObjectNav performance.

We present a novel autonomous robot navigation algorithm for outdoor environments that is capable of handling diverse terrain traversability conditions. Our approach, VLM-GroNav, uses vision-language models (VLMs) and integrates them with physical grounding that is used to assess intrinsic terrain properties such as deformability and slipperiness. We use proprioceptive-based sensing, which provides direct measurements of these physical properties, and enhances the overall semantic understanding of the terrains. Our formulation uses in-context learning to ground the VLM's semantic understanding with proprioceptive data to allow dynamic updates of traversability estimates based on the robot's real-time physical interactions with the environment. We use the updated traversability estimations to inform both the local and global planners for real-time trajectory replanning. We validate our method on a legged robot (Ghost Vision 60) and a wheeled robot (Clearpath Husky), in diverse real-world outdoor environments with different deformable and slippery terrains. In practice, we observe significant improvements over state-of-the-art methods by up to 50% increase in navigation success rate.

Navigating large-scale outdoor environments requires complex reasoning in terms of geometric structures, environmental semantics, and terrain characteristics, which are typically captured by onboard sensors such as LiDAR and cameras. While current mobile robots can navigate such environments using pre-defined, high-precision maps based on hand-crafted rules catered for the specific environment, they lack commonsense reasoning capabilities, especially the traversability analysis, that most humans possess when navigating unknown outdoor spaces. To address this gap, we introduce the Global Navigation Dataset (GND), a large-scale dataset that integrates multi-modal sensory data, including 3D LiDAR point clouds and RGB and 360° images, as well as multi-category traversability maps (pedestrian walkways, vehicle roadways, stairs, off-road terrain, and obstacles) from ten university campuses. These environments encompass a variety of parks, urban settings, elevation changes, and campus layouts of different scales. The dataset covers approximately$2.7 ~\text{km}^{2}$and includes at least 350 buildings in total. We also present a set of novel applications of GND to showcase its utility to enable global robot navigation, such as map-based global navigation, mapless navigation, and global place recognition. GND's website can be found at https://cs.gmu.edu/xiao/Research/GND/.

We present BehAV, a novel approach for autonomous robot navigation in outdoor scenes guided by human instructions and leveraging Vision Language Models (VLMs). Our method interprets human commands using a Large Language Model (LLM), and categorizes the instructions into navigation and behavioral guidelines. Navigation guidelines consist of directional commands (e.g., “move forward until“) and associated landmarks (e.g., “the building with blue windows”), while behavioral guidelines encompass regulatory actions (e.g., “stay on“) and their corresponding objects (e.g., “pavements“). We use VLMs for their zero-shot scene understanding capabilities to estimate landmark locations from RGB images for robot navigation. Further, we introduce a novel scene representation that utilizes VLMs to ground behavioral rules into a behavioral cost map. This cost map encodes the presence of behavioral objects within the scene and assigns costs based on their regulatory actions. The behavioral cost map is integrated with a LiDAR-based occupancy map for navigation. To navigate outdoor scenes while adhering to the instructed behaviors, we present an unconstrained Model Predictive Control (MPC)based planner that prioritizes both reaching landmarks and following behavioral guidelines. We evaluate the performance of BehAV on a quadruped robot across diverse real-world scenarios, demonstrating a 22.49 % improvement in alignment with human-teleoperated actions, as measured by Fréchet distance, and achieving a 40 % higher navigation success rate compared to state-of-the-art methods.

We present and tackle the problem of Embodied Question Answering (EQA) with Situational Queries (S-EQA) in a household environment. Unlike prior EQA work tackling simple queries that directly reference target objects and properties ("What is the color of the car?"), situational queries (such as "Is the house ready for sleeptime?") are challenging as they require the agent to correctly identify multiple object-states (Doors: Closed, Lights: Off, etc.) and reach a consensus on their states for an answer. Towards this objective, we first introduce a novel Prompt-Generate-Evaluate (PGE) scheme that wraps around an LLM’s output to generate unique situational queries and corresponding consensus object information. PGE is used to generate 2K datapoints in the VirtualHome simulator, which is then annotated for ground truth answers via a large scale user-study conducted on M-Turk. With a high rate of answerability (97.26%) on this study, we establish that LLMs are good at generating situational data. However, in evaluating the data using an LLM, we observe a low correlation of 46.2% with the ground truth human annotations; indicating that while LLMs are good at generating situational data, they struggle to answer them according to consensus. When asked for reasoning, we observe the LLM often goes against commonsense in justifying its answer. Finally, we utilize PGE to generate situational data in a real-world environment, exposing LLM hallucination in generating reliable object-states when a structured scene graph is unavailable. To the best of our knowledge, this is the first work to introduce EQA in the context of situational queries and also the first to present a generative approach for query creation. We aim to foster research on improving the real-world usability of embodied agents through this work.

We present a novel method, AutoSpatial, an efficient approach with structured spatial grounding to enhance VLMs’ spatial reasoning. By combining minimal manual supervision with large-scale Visual Question-Answering (VQA) pairs auto-labeling, our approach tackles the challenge of VLMs’ limited spatial understanding in social navigation tasks. By applying a hierarchical two-round VQA strategy during training, AutoSpatial achieves both global and detailed understanding of scenarios, demonstrating more accurate spatial perception, movement prediction, Chain of Thought (CoT) reasoning, final action, and explanation compared to other SOTA approaches. These five components are essential for comprehensive social navigation reasoning. Our approach was evaluated using both expert systems (GPT-4o, Gemini 2.0 Flash, and Claude 3.5 Sonnet) that provided cross-validation scores and human evaluators who assigned relative rankings to compare model performances across four key aspects. Augmented by the enhanced spatial reasoning capabilities, AutoSpatial demonstrates substantial improvements by averaged cross-validation score from expert systems in: perception & prediction (up to 10.71%), reasoning (up to 16.26%), action (up to 20.50%), and explanation (up to 18.73%) compared to baseline models trained only on manually annotated data.

We introduce ET-Former, a novel end-to-end algorithm for semantic scene completion using a single monocular camera. Our approach generates a semantic occupancy map from single RGB observation while simultaneously providing uncertainty estimates for semantic predictions. By designing a triplane-based deformable attention mechanism, our approach improves geometric understanding of the scene than other SOTA approaches and reduces noise in semantic predictions. Additionally, through the use of a Conditional Variational AutoEncoder (CVAE), we estimate the uncertainties of these predictions. The generated semantic and uncertainty maps will help formulate navigation strategies that facilitate safe and permissible decision making in the future. Evaluated on the Semantic-KITTI dataset, ET-Former achieves the highest Intersection over Union (IoU) and mean IoU (mIoU) scores while maintaining the lowest GPU memory usage, surpassing state-of-the-art (SOTA) methods. It improves the SOTA scores of IoU from 44.71 to 51.49 and mIoU from 15.04 to 16.30 on SeamnticKITTI test, with a notably low training memory consumption of 10.9 GB, achieving at least a 25% reduction compared to previous methods. Project page: https://github.com/amazon-science/ET-Former.

In this paper, we present a new approach to improve the neural rendering fidelity of in-the-wild unmanned aerial vehicle (UAV)-based scenes. Our formulation is designed for dynamic scenes, consisting of small moving objects or human actions in particular. We propose an extension of K-Planes Neural Radiance Field (NeRF), wherein our algorithm stores a set of tiered high dimensional feature vectors. The tiered feature vectors are generated to effectively model conceptual information about a scene as well as to be processed by an image decoder that transforms output feature maps into RGB images. Our technique leverages the information among both static and dynamic objects within a scene and is able to capture salient scene attributes of high altitude videos. We evaluate its performance on challenging datasets, including Okutama Action and UG2, and observe considerable improvement in accuracy over state of the art neural rendering methods.

Large language models (LLMs) showcase many desirable traits for intelligent and helpful robots. However, they are also known to hallucinate predictions. This issue is exacerbated in robotics where LLM hallucinations may result in robots confidently executing plans that are contrary to user goals or relying more frequently on human assistance. In this work, we present LBAP, a novel approach for utilizing off-the-shelf LLMs, alongside Bayesian inference for uncertainty Alignment in robotic Planners that minimizes hallucinations and human intervention. Our key finding is that we can use Bayesian inference to more accurately calibrate a robots confidence measure through accounting for both scene grounding and world knowledge. This process allows us to mitigate hallucinations and better align the LLM’s confidence measure with the probability of success. Through experiments in both simulation and the real world on tasks with a variety of ambiguities, we show that LBAP significantly increases success rate and decreases the amount of human intervention required relative to prior art. For example, in our real-world testing paradigm, LBAP decreases the human help rate of previous methods by over 33% at a success rate of 70%.

Reinforcement learning (RL) is a promising approach for robotic navigation, allowing robots to learn through trial and error. However, real-world robotic tasks often suffer from sparse rewards, leading to inefficient exploration and suboptimal policies due to sample inefficiency of RL. In this work, we introduce Confidence-Controlled Exploration (CCE), a novel method that improves sample efficiency in RL-based robotic navigation without modifying the reward function. Unlike existing approaches, such as entropy regularization and reward shaping, which can introduce instability by altering rewards, CCE dynamically adjusts trajectory length based on policy entropy. Specifically, it shortens trajectories when uncertainty is high to enhance exploration and extends them when confidence is high to prioritize exploitation. CCE is a principled and practical solution inspired by a theoretical connection between policy entropy and gradient estimation. It integrates seamlessly with on-policy and off-policy RL methods and requires minimal modifications. We validate CCE across REINFORCE, PPO, and SAC in both simulated and real-world navigation tasks. CCE outperforms fixed-trajectory and entropy-regularized baselines, achieving an 18% higher success rate, 20-38% shorter paths, and 9.32% lower elevation costs under a fixed training sample budget. Finally, we deploy CCE on a Clearpath Husky robot, demonstrating its effectiveness in complex outdoor environments.

As mobile robots become increasingly common in human-centric environments, social navigation—adhering to unwritten social norms rather than merely avoiding pedestrians—has drawn growing attention. Existing methods, from hand-crafted techniques to learning-based approaches, often overlook the nuanced context and scene understanding that humans naturally exhibit. Inspired by studies indicating the critical role of language in cognition and reasoning, we propose a new approach to bridge robot perception and socially aware actions through human-like language reasoning. We introduce Social robot Navigation via Explainable Interactions (SNEI), a human-annotated vision-language dataset comprising over 40K Visual Question Answering (VQA) pairs across 2K unique social scenarios, drawn from diverse, unstructured public spaces. SNEI contains perception, prediction, chain-of-thought reasoning, action, and explanation, thereby allowing robots to interpret social contexts in human language. We fine-tune a Vision-Language Model, Social-LLaVA, on SNEI to demonstrate the potential of language-guided reasoning for high-level navigation tasks. Experimental evaluations—both quantitative and qualitative—demonstrate that Social-LLaVA can outperform state-of-the-art models.†.

We present CROSS-GAiT, a novel algorithm for quadruped robots that uses Cross Attention to fuse terrain representations derived from visual and time-series inputs; including linear accelerations, angular velocities, and joint efforts. These fused representations are used to continuously adjust two critical gait parameters (step height and hip splay), enabling adaptive gaits that respond dynamically to varying terrain conditions. To generate terrain representations, we process visual inputs through a masked Vision Transformer (ViT) encoder and time-series data through a dilated causal convolutional encoder. The Cross Attention mechanism then selects and integrates the most relevant features from each modality, combining terrain characteristics with robot dynamics for informed gait adaptation. This fused representation allows CROSS-GAiT to continuously adjust gait parameters in response to unpredictable terrain conditions in real-time. We train CROSS-GAiT on a diverse set of terrains including asphalt, concrete, brick pavements, grass, dense vegetation, pebbles, gravel, and sand and validate its generalization ability on unseen environments. Our hardware implementation on the Ghost Robotics Vision 60 demonstrates superior performance in challenging terrains, such as high-density vegetation, unstable surfaces, sandbanks, and deformable substrates. We observe at least a 7.04% reduction in IMU energy density and a 27.3% reduction in total joint effort, which directly correlates with increased stability and reduced energy usage when compared to state-of-the-art methods. Furthermore, CROSS-GAiT demonstrates at least a 64.5% increase in success rate and a 4.91% reduction in time to reach the goal in four complex scenarios. Additionally, the learned representations perform 4.48% better than the state-of-the-art on a terrain classification task.

In this work, we highlight vulnerabilities in robotic systems integrating large language models (LLMs) and vision-language models (VLMs) due to input modality sensitivities. While LLM/VLM-controlled robots show impressive performance across various tasks, their reliability under slight input variations remains underexplored yet critical. These models are highly sensitive to instruction or perceptual input changes, which can trigger misalignment issues, leading to execution failures with severe real-world consequences. To study this issue, we analyze the misalignment-induced vulnerabilities within LLM/VLM-controlled robotic systems and present a mathematical formulation for failure modes arising from variations in input modalities. We propose empirical perturbation strategies to expose these vulnerabilities and validate their effectiveness through experiments on multiple robot manipulation tasks. Our results show that simple input perturbations reduce task execution success rates by 22.2% and 14.6% in two representative LLM/VLM-controlled robotic systems. These findings underscore the importance of input modality robustness and motivate further research to ensure the safe and reliable deployment of advanced LLM/VLM-controlled robotic systems.

Soumya Suvra Ghosal, Soumyabrata Pal, Koyel Mukherjee, Dinesh Manocha. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Sonal Kumar, Sreyan Ghosh, Utkarsh Tyagi, Anton Jeran Ratnarajah, Chandra Kiran Reddy Evuru, Ramani Duraiswami, Dinesh Manocha. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Ashish Seth, Ramaneswaran Selvakumar, Sonal Kumar, Sreyan Ghosh, Dinesh Manocha. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Manan Suri, Puneet Mathur, Franck Dernoncourt, Kanika Goswami, Ryan A. Rossi, Dinesh Manocha. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Large multimodal models (LMMs) have shown remarkable progress in audiovisual understanding, yet they struggle with real-world scenarios that require complex reasoning across extensive video collections. Existing benchmarks for video question answering remain limited in scope, typically involving one clip per query, which falls short of representing the challenges of large-scale, audiovisual retrieval and reasoning encountered in practical applications. To bridge this gap, we introduce a novel task named AVHaystacksQA, where the goal is to identify salient segments across different videos in response to a query and link them together to generate the most informative answer. To this end, we present AVHaystacks, an audio-visual benchmark comprising 3100 annotated QA pairs designed to assess the capabilities of LMMs in multi-video retrieval and temporal grounding task. Additionally, we propose a model-agnostic, multi-agent framework MAGNET to address this challenge, achieving up to 89% and 65% relative improvements over baseline methods on BLEU@4 and GPT evaluation scores in QA task on our proposed AVHaystacks. To enable robust evaluation of multi-video retrieval and temporal grounding for optimal response generation, we introduce two new metrics, STEM, which captures alignment errors between a ground truth and a predicted step sequence and MTGS, to facilitate balanced and interpretable evaluation of segment-level grounding performance.

Recent trends in test-time scaling for reasoning models (e.g., OpenAI o1, DeepSeek R1) have led to a popular belief that extending thinking traces using prompts like “Wait” or “Let me rethink” can improve performance. This raises a natural question: Does thinking more at test-time truly lead to better reasoning? To answer this question, we perform a detailed empirical study across models and benchmarks, which reveals a consistent pattern of initial performance improvements from additional thinking followed by a decline, due to "overthinking". To understand this non-monotonic trend, we consider a simple probabilistic model, which reveals that additional thinking increases output variance—creating an illusion of improved reasoning while ultimately undermining precision. Thus, observed gains from "more thinking" are not true indicators of improved reasoning, but artifacts stemming from the connection between model uncertainty and evaluation metric. This suggests that test-time scaling through extended thinking is not an effective way to utilize the inference thinking budget. Recognizing these limitations, we introduce an alternative test-time scaling approach, parallel thinking, inspired by Best-of-N sampling. Our method generates multiple independent reasoning paths within the same inference budget and selects the most consistent response via majority vote, achieving up to 20% higher accuracy compared to extended thinking. This provides a simple yet effective mechanism for test-time scaling of reasoning models.

We present Audio Flamingo 3 (AF3), a fully open state-of-the-art (SOTA) large audio-language model that advances reasoning and understanding across speech, sound, and music. AF3 introduces: (i) AF-Whisper, a unified audio encoder trained using a novel strategy for joint representation learning across all 3 modalities of speech, sound, and music; (ii) flexible, on-demand thinking, allowing the model to do chain-of-thought-type reasoning before answering; (iii) multi-turn, multi-audio chat; (iv) long audio understanding and reasoning (including speech) up to 10 minutes; and (v) voice-to-voice interaction. To enable these capabilities, we propose several large-scale training datasets curated using novel strategies, including AudioSkills-XL, LongAudio-XL, AF-Think, and AF-Chat, and train AF3 with a novel five-stage curriculum-based training strategy. Trained on only open-source audio data, AF3 achieves new SOTA results on over 20+ (long) audio understanding and reasoning benchmarks, surpassing both open-weight and closed-source models trained on much larger datasets.

The increasing use of 360$^\circ$ images across various domains has emphasized the need for robust depth estimation techniques tailored for omnidirectional images. However, obtaining large-scale labeled datasets for 360$^\circ$ depth estimation remains a significant challenge. In this paper, we propose RPG360, a training-free robust 360$^\circ$ monocular depth estimation method that leverages perspective foundation models and graph optimization. Our approach converts 360$^\circ$ images into six- face cubemap representations, where a perspective foundation model is employed to estimate depth and surface normals. To address depth scale inconsistencies across different faces of the cubemap, we introduce a novel depth scale alignment technique using graph-based optimization, which parameterizes the predicted depth and normal maps while incorporating an additional per-face scale parameter. This optimization ensures depth scale consistency across the six-face cubemap while preserving 3D structural integrity. Furthermore, as foundation models exhibit inherent robustness in zero-shot settings, our method achieves superior performance across diverse datasets, including Matterport3D, Stanford2D3D, and 360Loc. We also demonstrate the versatility of our depth estimation approach by validating its benefits in downstream tasks such as feature matching 3.2 ∼ 5.4% and Structure from Motion 0.2 ∼ 9.7% in AUC@5$^\circ$.

Vision Language models (VLMs) have achieved remarkable success in video understanding tasks. Yet, a key question remains: Do they comprehend visual information or merely learn superficial mappings between visual and textual patterns? Understanding visual cues, particularly those related to physics and common sense, is crucial for AI systems interacting with the physical world. However, existing VLM evaluations primarily rely on positive-control tests using real-world videos that resemble training distributions. While VLMs perform well on such benchmarks, it is unclear whether they grasp underlying visual and contextual signals or simply exploit visual-language correlations. To fill this gap, we propose incorporating negative-control tests, i.e., videos depicting physically impossible or logically inconsistent scenarios, and evaluating whether models can recognize these violations. True visual understanding should evince comparable performance across both positive and negative tests. Since such content is rare in the real world, we introduce VideoHallu, a synthetic video dataset featuring physics- and commonsense-violating scenes generated using state-of-the-art tools such as Veo2, Sora, and Kling. The dataset includes expert-annotated question-answer pairs spanning four categories of physical and commonsense violations, designed to be straightforward for human reasoning. We evaluate several leading VLMs, including Qwen-2.5-VL, Video-R1, and VideoChat-R1. Despite their strong performance on real-world benchmarks (e.g., MVBench, MMVU), these models hallucinate or fail to detect physical or logical violations, revealing fundamental weaknesses in visual understanding. Finally, we explore reinforcement learning-based post-training on our negative dataset: fine-tuning improves performance on VideoHallu without degrading results on standard benchmarks, indicating enhanced visual reasoning in VLMs. Our data is available at https://github.com/zli12321/VideoHallu.git.

We study the correlations between redirected walking (RDW) rotation gains and patterns in users' posture and gaze data during locomotion in virtual reality (VR). To do this, we conducted a psychophysical experiment to measure users' sensitivity to RDW rotation gains and collect gaze and posture data during the experiment. Using multilevel modeling, we studied how different factors of the VR system and user affected their physiological signals. In particular, we studied the effects of redirection gain, trial duration, trial number (i.e., time spent in VR), and participant gender on postural sway, gaze velocity (a proxy for gaze stability), and saccade and blink rate. Our results showed that, in general, physiological signals were significantly positively correlated with the strength of redirection gain, the duration of trials, and the trial number. Gaze velocity was negatively correlated with trial duration. Additionally, we measured users' sensitivity to rotation gains in well-lit (photopic) and dimly-lit (mesopic) virtual lighting conditions. Results showed that there were no significant differences in RDW detection thresholds between the photopic and mesopic luminance conditions.

Sreyan Ghosh, Utkarsh Tyagi, Sonal Kumar, Chandra Kiran Evuru, Ramaneswaran S, S Sakshi, Dinesh Manocha. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Predicting price variations of financial instruments for risk modeling and stock trading is challenging due to the stochastic nature of the stock market. While recent advancements in the Financial AI realm have expanded the scope of data and methods they use, such as textual and audio cues from financial earnings calls, limitations exist. Most datasets are small, and show domain distribution shifts due to the nature of their source, suggesting the exploration for data augmentation for robust augmentation strategies such as Mixup. To tackle such challenges in the financial domain, we propose SH-Mix: Saliency-guided Hierarchical Mixup augmentation technique for multimodal financial prediction tasks. SH-Mix combines multi-level embedding mixup strategies based on the contribution of each modality and context subsequences. Through extensive quantitative and qualitative experiments on financial earnings and conference call datasets consisting of text and speech, we show that SH-Mix outperforms state-of-the-art methods by 3-7%. Additionally, we show that SH-Mix is generalizable across different modalities and models.

We propose DOC-RAG - Domain-distributed Co-occurrence Retrieval Augmentation for ASR language model personalization aiming to improve the automatic speech recognition of rare word patterns in unseen domains. Our approach involves contrastively training a document retrieval module to rank external knowledge domains based on their semantic similarity with respect to the input query. We further use n-gram co-occurrence distribution to recognize rare word patterns associated with specific domains. We aggregate the next word probability distribution based on the relative importance of different domains. Extensive experiments on three user-specific speech-to-text tasks for meetings, TED talks, and financial earnings calls show that DOC-RAG significantly outperforms strong baselines with an 8-15% improvement in terms of perplexity and a 4-7% reduction in terms of Word Error Rates in various settings.

We present a novel task of document-level script event prediction, which aims to predict the next event given a candidate list of narrative events in long-form documents. To enable this, we introduce DocSEP, a challenging dataset in two new domains - contractual documents and Wikipedia articles, where timeline events may be paragraphs apart and may require multi-hop temporal and causal reasoning. We benchmark existing baselines and present a novel architecture called DocScript to learn sequential ordering between events at the document scale. Our experimental results on the DocSEP dataset demonstrate that learning longer-range dependencies between events is a key challenge and show that contemporary LLMs such as ChatGPT and FlanT5 struggle to solve this task, indicating their lack of reasoning abilities for understanding causal relationships and temporal sequences within long texts.

Advancements in neural signed distance fields (SDFs) have enabled modeling 3D surface geometry from a set of 2D images of real-world scenes. Baking neural SDFs can extract explicit mesh with appearance baked into texture maps as neural features. The baked meshes still have a large memory footprint and require a powerful GPU for real-time rendering. Neural optimization of such large meshes with differentiable rendering pose significant challenges. We propose a method to produce optimized meshes for large unbounded scenes with low triangle budget and high fidelity of geometry and appearance. We achieve this by combining advancements in baking neural SDFs with classical mesh simplification techniques and proposing a joint appearance-geometry refinement step. The visual quality is comparable to or better than state-of-the-art neural meshing and baking methods with high geometric accuracy despite significant reduction in triangle count, making the produced meshes efficient for storage, transmission, and rendering on mobile hardware. We validate the effectiveness of the proposed method on large unbounded scenes from mip-NeRF 360, Tanks & Temples, and Deep Blending datasets, achieving at-par rendering quality with 73 x reduced triangles and 11 x reduction in memory footprint.

Music is a universal language that can communicate emotions and feelings. It forms an essential part of the whole spectrum of creative media, ranging from movies to social media posts. Machine learning models that can synthesize music are predominantly conditioned on textual descriptions of it. Inspired by how musicians compose music not just from a movie script, but also through visualizations, we propose MELFuSION, a model that can effectively use cues from a textual description and the corresponding image to synthesize music. MELFuSION is a text-to-music diffusion model with a novel “visual synapse ”, which effectively infuses the semantics from the visual modality into the generated music. To facilitate research in this area, we introduce a new dataset MeLBench, and propose a new evaluation metric IMSM. Our exhaustive experimental evaluation suggests that adding visual information to the music synthesis pipeline significantly improves the quality of generated music, measured both objectively and subjectively, with a relative gain of up to 67.98% on the FAD score. We hope that our work will gather attention to this pragmatic, yet relatively under-explored research area.

We introduce “HALLUSIONBENCH11“Hallusion” is a portmanteau of “hallucination” and “illusion.”,” a comprehensive benchmark designed for the evaluation of image-context rea-soning. This benchmark presents significant challenges to advanced large visual-language models (LVLMs), such as GPT-4V(ision), Gemini Pro Vision, Claude 3, and LLaVA-1.5, by emphasizing nuanced understanding and interpre-tation of visual data. The benchmark comprises 346 images paired with 1129 questions, all meticulously crafted by human experts. We introduce a novel structure for these visual questions designed to establish control groups. This structure enables us to conduct a quantitative analysis of the models' response tendencies, logical consistency, and various failure modes. In our evaluation on Hallusion-bench, we benchmarked 15 different models, highlighting a 31.42% question-pair accuracy achieved by the state-of-the-art GPT-4V. Notably, all other evaluated models achieve accuracy below 16%. Moreover, our analysis not only high-lights the observed failure modes, including language hal-lucination and visual illusion but also deepens an under-standing of these pitfalls. Our comprehensive case studies within Hallusionbench shed light on the challenges of hallucination and illusion in LVLMs. Based on these in-sights, we suggest potential pathways for their future im-provement. The benchmark and codebase can be accessed at https://github.com/tianyi-labIHallusionBench.

Accurate estimation of Room Impulse Response (RIR), which captures an environment's acoustic properties, is important for speech processing and AR/VR applications. We propose AV-RIR, a novel multi-modal multi-task learning approach to accurately estimate the RIR from a given reverberant speech signal and the visual cues of its corresponding environment. AV-RIR builds on a novel neural codec-based architecture that effectively captures environment geometry and materials properties and solves speech dereverberation as an auxiliary task by using multi-task learning. We also propose Geo-Mat features that augment material information into visual cues and CRIP that improves late reverberation components in the estimated RIR via image-to-RIR retrieval by 86%. Empirical results show that AV-RIR quantitatively outperforms previous audio-only and visual-only approaches by achieving 36% - 63% improvement across various acoustic metrics in RIR estimation. Additionally, it also achieves higher preference scores in human evaluation. As an auxiliary benefit, dereverbed speech from AV-RIR shows competitive performance with the state-of-the-art in various spoken language processing tasks and outperforms reverberation time error score in the real-world AVSpeech dataset. Qualitative examples of both synthesized reverberant speech and enhanced speech are available online11https://anton-jeran.github.io/AVRIR/.

Image-text contrastive models such as CLIP learn transferable and robust representations for zero-shot transfer to a variety of downstream tasks. However, to obtain strong downstream performances, prompts need to be carefully curated, which can be a tedious engineering task. To address the issue of manual prompt engineering, prompt-tuning is used where a set of contextual vectors are learned by leveraging information from the training data. Despite their effectiveness, existing prompt-tuning frameworks often lack interpretability, thus limiting their ability to understand the compositional nature of images. In this work, we first identify that incorporating compositional attributes (e.g., a “green” tree frog) in the design of manual prompts can significantly enhance image-text alignment scores. Building upon this observation, we propose a novel and interpretable prompt-tuning method named IntCoOp, which learns to jointly align attribute-level inductive biases and class embeddings during prompt-tuning. To assess the effectiveness of our approach, we evaluate IntCoOp across two representative tasks in a few-shot learning setup: generalization to novel classes, and unseen domain shifts. Through extensive experiments across 10 downstream datasets on CLIP, we find that introducing attribute-level inductive biases leads to superior performance against state-of-art prompt tuning frameworks. Notably, in a 16-shot setup, IntCoOp improves CoOp by 7.35% in average performance across 10 diverse datasets.

Sreyan Ghosh, Sonal Kumar, Ashish Seth, Chandra Kiran Reddy Evuru, Utkarsh Tyagi, S Sakshi, Oriol Nieto, Ramani Duraiswami, Dinesh Manocha. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

In this paper, we present EH-MAM (Easyto-Hard adaptive Masked Acoustic Modeling), a novel self-supervised learning approach for speech representation learning.In contrast to the prior methods that use random masking schemes for Masked Acoustic Modeling (MAM), we introduce a novel selective and adaptive masking strategy.Specifically, during SSL training, we progressively introduce harder regions to the model for reconstruction.Our approach automatically selects hard regions and is built on the observation that the reconstruction loss of individual frames in MAM can provide natural signals to judge the difficulty of solving the MAM pre-text task for that frame.To identify these hard regions, we employ a teacher model that first predicts the frame-wise losses and then decides which frames to mask.By learning to create challenging problems, such as identifying harder frames and solving them simultaneously, the model is able to learn more effective representations and thereby acquire a more comprehensive understanding of the speech.Quantitatively, EH-MAM outperforms several state-ofthe-art baselines across various low-resource speech recognition and SUPERB benchmarks by 5%-10%.Additionally, we conduct a thorough analysis to show that the regions masked by EH-MAM effectively capture useful context across speech frames 1 . * Equal contribution, † Equal Ideation 1 Code:

Manan Suri, Puneet Mathur, Franck Dernoncourt, Rajiv Jain, Vlad I Morariu, Ramit Sawhney, Preslav Nakov, Dinesh Manocha. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

We present RECAP (REtrieval-Augmented Audio CAPtioning), a novel and effective audio captioning system that generates captions conditioned on an input audio and other captions similar to the audio retrieved from a datastore. Additionally, our proposed method can transfer to any domain without the need for any additional fine-tuning. To generate a caption for an audio sample, we leverage an audio-text model CLAP [1] to retrieve captions similar to it from a replaceable datastore, which are then used to construct a prompt. Next, we feed this prompt to a GPT-2 decoder and introduce cross-attention layers between the CLAP encoder and GPT-2 to condition the audio for caption generation. Experiments on two benchmark datasets, Clotho and AudioCaps, show that RECAP achieves competitive performance in in-domain settings and significant improvements in out-of-domain settings. Additionally, due to its capability to exploit a large text-captions-only datastore in a training-free fashion, RECAP shows unique capabilities of captioning novel audio events never seen during training and compositional audios with multiple events. To promote research in this space, we also release 150,000+ new weakly labeled captions for AudioSet, AudioCaps, and Clotho1.

Continued self-supervised (SSL) pre-training for adapting existing SSL models to the target domain has shown to be extremely effective for low-resource Automatic Speech Recognition (ASR). This paper proposes Stable Distillation, a simple and novel approach for SSL-based continued pre-training that boosts ASR performance in the target domain where both labeled and unlabeled data are limited. Stable Distillation employs self-distillation as regularization for continued pre-training, alleviating the over-fitting issue, a common problem continued pre-training faces when the source and target domains differ. Specifically, first, we perform vanilla continued pre-training on an initial SSL pre-trained model on the target domain ASR dataset and call it the teacher. Next, we take the same initial pre-trained model as a student to perform continued pre-training while enforcing its hidden representations to be close to that of the teacher (via MSE loss). This student is then used for downstream ASR fine-tuning on the target dataset. In practice, Stable Distillation outperforms all our baselines by 0.8 - 7 WER when evaluated in various experimental settings1.

Continued pre-training (CP) offers multiple advantages, like target domain adaptation and the potential to exploit the continuous stream of unlabeled data available online. However, continued pre-training on out-of-domain distributions often leads to catastrophic forgetting of previously acquired knowledge, leading to sub-optimal ASR performance. This paper presents FusDom, a simple and novel methodology for SSL-based continued pre-training. FusDom learns speech representations that are robust and adaptive yet not forgetful of concepts seen in the past. Instead of solving the SSL pre-text task on the output representations of a single model, FusDom leverages two identical pre-trained SSL models, a teacher and a student, with a modified pre-training head to solve the CP SSL pre-text task. This head employs a cross-attention mechanism between the representations of both models while only the student receives gradient updates and the teacher does not. Finally, the student is fine-tuned for ASR. In practice, FusDom outperforms all our baselines across settings significantly, with WER improvements in the range of 0.2 WER - 7.3 WER in the target domain, while retaining the performance in the earlier domain1.

We present a novel unified bilevel optimization-based framework, \textsf{PARL}, formulated to address the recently highlighted critical issue of policy alignment in reinforcement learning using utility or preference-based feedback. We identify a major gap within current algorithmic designs for solving policy alignment due to a lack of precise characterization of the dependence of the alignment objective on the data generated by policy trajectories. This shortfall contributes to the sub-optimal performance observed in contemporary algorithms. Our framework addressed these concerns by explicitly parameterizing the distribution of the upper alignment objective (reward design) by the lower optimal variable (optimal policy for the designed reward). Interestingly, from an optimization perspective, our formulation leads to a new class of stochastic bilevel problems where the stochasticity at the upper objective depends upon the lower-level variable. {True to our best knowledge, this work presents the first formulation of the RLHF as a bilevel optimization problem which generalizes the existing RLHF formulations and addresses the existing distribution shift issues in RLHF formulations.} To demonstrate the efficacy of our formulation in resolving alignment issues in RL, we devised an algorithm named \textsf{A-PARL} to solve PARL problem, establishing sample complexity bounds of order $\mathcal{O}(1/T)$. Our empirical results substantiate that the proposed \textsf{PARL} can address the alignment concerns in RL by showing significant improvements (up to 63\% in terms of required samples) for policy alignment in large-scale environments of the Deepmind control suite and Meta world tasks.

A fundamental characteristic of audio is its compositional nature. Audio-language models (ALMs) trained using a contrastive approach (e.g., CLAP) that learns a shared representation between audio and language modalities have improved performance in many downstream applications, including zero-shot audio classification, audio retrieval, etc. However, the ability of these models to effectively perform compositional reasoning remains largely unexplored and necessitates additional research. In this paper, we propose CompA, a collection of two expert-annotated benchmarks with a majority of real-world audio samples, to evaluate compositional reasoning in ALMs. Our proposed CompA-order evaluates how well an ALM understands the order or occurrence of acoustic events in audio, and CompA-attribute evaluates attribute-binding of acoustic events. An instance from either benchmark consists of two audio-caption pairs, where both audios have the same acoustic events but with different compositions. An ALM is evaluated on how well it matches the right audio to the right caption. Using this benchmark, we first show that current ALMs perform only marginally better than random chance, thereby struggling with compositional reasoning. Next, we propose CompA-CLAP, where we fine-tune CLAP using a novel learning method to improve its compositional reasoning abilities. To train CompA-CLAP, we first propose improvements to contrastive training with composition-aware hard negatives, allowing for more focused training. Next, we propose a novel modular contrastive loss that helps the model learn fine-grained compositional understanding and overcomes the acute scarcity of openly available compositional audios. CompA-CLAP significantly improves over all our baseline models on the CompA benchmark, indicating its superior compositional reasoning capabilities.

Our study addresses the challenge of distinguishing human-written text from Large Language Model (LLM) outputs. We provide evidence that this differentiation is consistently feasible, except when human and machine text distributions are indistinguishable across their entire support. Employing information theory, we show that while detecting machine-generated text becomes harder as it nears human quality, it remains possible with adequate text data. We introduce guidelines on the required text data quantity, either through sample size or sequence length, for reliable AI text detection, through derivations of sample complexity bounds. This research paves the way for advanced detection methods. Our comprehensive empirical tests, conducted across various datasets (Xsum, Squad, IMDb, and Kaggle FakeNews) and with several state-of-the-art text generators (GPT-2, GPT-3.5-Turbo, Llama, Llama-2-13B-Chat-HF, Llama-2-70B-Chat-HF), assess the viability of enhanced detection methods against detectors like RoBERTa-Large/Base-Detector and GPTZero, with increasing sample sizes and sequence lengths. Our findings align with OpenAI’s empirical data related to sequence length, marking the first theoretical substantiation for these observations.

Reinforcement Learning from Human Feedback (RLHF) aligns language models to human preferences by employing a singular reward model derived from preference data. However, the single reward model overlooks the rich diversity of human preferences inherent in data collected from multiple users. In this work, we first derive an impossibility result of alignment with single reward RLHF, thereby highlighting its insufficiency in representing diverse human preferences. Next, we propose to learn a mixture of reward models via an expectation-maximization algorithm and solve a MaxMin alignment objective inspired by the Egalitarian principle in social choice theory to better honor diverse human preferences. We present comprehensive experimental results on small-scale (GPT-2) and large-scale language (with Tulu2-7B)) and show the efficacy of the proposed approach in the presence of diversity among human preferences. We remark that our findings in this work are not only limited to language models but also extend to reinforcement learning in general.

Instruction Tuning (IT), the process of training large language models (LLMs) using instruction-response pairs, has emerged as the predominant method for transforming base pre-trained LLMs into open-domain conversational agents. While IT has achieved notable success and widespread adoption, its limitations and shortcomings remain underexplored. In this paper, through rigorous experiments and an in-depth analysis of the changes LLMs undergo through IT, we reveal various limitations of IT. In particular, we show that (1) IT fails to enhance knowledge or skills in LLMs. LoRA fine-tuning is limited to learning response initiation and style tokens, and full-parameter fine-tuning leads to knowledge degradation. (2) Copying response patterns from IT datasets derived from knowledgeable sources leads to a decline in response quality. (3) Full-parameter fine-tuning increases hallucination by inaccurately borrowing tokens from conceptually similar instances in the IT dataset for generating responses. (4) Popular methods to improve IT do not lead to performance improvements over a simple LoRA fine-tuned model. Our findings reveal that responses generated solely from pre-trained knowledge consistently outperform responses by models that learn any form of new knowledge from IT on open-source datasets. We hope the insights and challenges revealed in this paper inspire future work in related directions.

In the context of average-reward reinforcement learning, the requirement for oracle knowledge of the mixing time, a measure of the duration a Markov chain under a fixed policy needs to achieve its stationary distribution, poses a significant challenge for the global convergence of policy gradient methods. This requirement is particularly problematic due to the difficulty and expense of estimating mixing time in environments with large state spaces, leading to the necessity of impractically long trajectories for effective gradient estimation in practical applications. To address this limitation, we consider the Multi-level Actor-Critic (MAC) framework, which incorporates a Multi-level Monte-Carlo (MLMC) gradient estimator. With our approach, we effectively alleviate the dependency on mixing time knowledge, a first for average-reward MDPs global convergence. Furthermore, our approach exhibits the tightest available dependence of $\mathcal{O}(\sqrt{\tau_{mix}})$ known from prior work. With a 2D grid world goal-reaching navigation experiment, we demonstrate that MAC outperforms the existing state-of-the-art policy gradient-based method for average reward settings.

In this paper, we present a probabilistic and unconstrained model predictive control formulation for robot navigation under uncertainty. We present (1) a closed-form approximation of the probability of collision that naturally models the propagation of uncertainty over the planning horizon and is computationally cheap to evaluate, and (2) a collision-cost formulation which provably preserves forward invariance (i.e., keeps the robot away from obstacles) when combined with the probability formulation. Notably, our formulation avoids hard constraints by construction, which in turn avoids abrupt transitions in robot behavior around the constraint boundaries ensuring graceful navigation. Further, we present proof for the forward invariance and the stability of the approach. We compare the efficacy of our method with the baseline [1], which the proposed approach builds on. We demonstrate that the approach results in confident and safe robot navigation in tight spaces by smoothly slowing down the robot in low survivability environments (e.g., tight corridors), but also allows it to move away from obstacles safely when needed.

Robotic sketching in real-world scenarios poses a challenging problem with diverse applications in art, robotics, and digital design. We present a novel approach that bridges the gap between digital and robotic sketching, leveraging behavior cloning and reinforcement learning techniques. This paper introduces an approach aimed at bringing the gap between simulated and real-world robotic sketching closer together through the integration of behavior cloning and reinforcement learning techniques. Our approach trains painting policies that operate effectively in both virtual environments and real-world robotic sketching systems. We have implemented a robotic sketching system featuring an UltraArm robot equipped with a RealSense D415 camera, closely emulating the MyPaint virtual environment. Our system can perceive its environment and adapt painting policies to natural painting media. Our results highlight the effectiveness of our agent in terms of acquiring policies for high-dimensional continuous action spaces, enabling the seamless transfer of brush manipulation techniques from simulation to practical robotic sketching. Furthermore, we demonstrate our robotic sketching system’s capability to generate complex images and strokes using various configurations. https://sites.google.com/view/sketchingrobot

In this paper, we present an approach for coverage path planning for a team of an energy-constrained Unmanned Aerial Vehicle (UAV) and an Unmanned Ground Vehicle (UGV). Both the UAV and the UGV have predefined areas that they have to cover. The goal is to perform complete coverage by both robots while minimizing the coverage time. The UGV can also serve as a mobile recharging station. The UAV and UGV need to occasionally rendezvous for recharging. We propose a heuristic method to address this NP-Hard planning problem. Our approach involves initially determining coverage paths without factoring in energy constraints. Subsequently, we cluster segments of these paths and employ graph matching to assign UAV clusters to UGV clusters for efficient recharging management. We perform numerical analysis on real-world coverage applications and show that compared with a greedy approach our method reduces rendezvous overhead on average by 11.33%. We demonstrate proof-of-concept with a team of a VOXL m500 drone and a Clearpath Jackal ground vehicle, providing a complete system from the offline algorithm to the field execution.

We present a novel learning-based trajectory generation algorithm for outdoor robot navigation. Our goal is to compute collision-free paths that also satisfy the environment-specific traversability constraints. Our approach is designed for global planning using limited onboard robot perception in mapless environments while ensuring comprehensive coverage of all traversable directions. Our formulation uses a Conditional Variational Autoencoder (CVAE) generative model that is enhanced with traversability constraints and an optimization formulation used for the coverage. We highlight the benefits of our approach over state-of-the-art trajectory generation approaches and demonstrate its performance in challenging and large outdoor environments, including around buildings, across intersections, along trails, and off-road terrain, using a Clearpath Husky and a Boston Dynamics Spot robot. In practice, our approach results in a 6% improvement in coverage of traversable areas and an 89% reduction in trajectory portions residing in non-traversable regions. Our video is here: https://youtu.be/3eJ2soAzXnU

Tremendous variations coupled with large degrees of freedom in UAV-based imaging conditions lead to a significant lack of data in adequately learning UAV-based perception models. Using various synthetic renderers in conjunction with perception models is prevalent to create synthetic data to augment the learning in the ground-based imaging domain. However, severe challenges in the austere UAV-based domain require distinctive solutions to image synthesis for data augmentation. In this work, we leverage recent advancements in neural rendering to improve static and dynamic novel-view UAV-based image synthesis, especially from high altitudes, capturing salient scene attributes. Finally, we demonstrate a considerable performance boost is achieved when a state-of-the-art detection model is optimized primarily on hybrid sets of real and synthetic data instead of the real or synthetic data separately.

We present MIM (Multi-Layer Intensity Map), a novel 3D object representation for robot perception and autonomous navigation. MIMs consist of multiple stacked layers of 2D grid maps each derived from reflected point cloud intensities corresponding to a certain height interval. The different layers of MIMs can be used to simultaneously estimate obstacles’ height, solidity/density, and opacity. We demonstrate that MIMs’ can help accurately differentiate obstacles that are safe to navigate through (e.g. beaded/string curtains, pliable tall grass), from ones that must be avoided (e.g. transparent surfaces such as glass walls, bushes, trees, etc.) in indoor and outdoor environments. Further, to handle narrow passages, and navigate through non-solid obstacles in dense environments, we propose an approach to adaptively inflate or enlarge the obstacles detected on MIMs based on their solidity, and the robot’s preferred velocity direction. We demonstrate these improved navigation capabilities in real-world narrow, dense environments using a real Turtlebot and Boston Dynamics Spot robots. We observe significant increases in success rates to more than 50%, up to a 9.5% decrease in normalized trajectory length, and up to a 22.6% increase in the F-score compared to current navigation methods using other sensor modalities.

We present VAPOR, a novel method for autonomous legged robot navigation in unstructured, densely vegetated outdoor environments using offline Reinforcement Learning (RL). Our method trains a novel RL policy using an actor-critic network and arbitrary data collected in real outdoor vegetation. Our policy uses height and intensity-based cost maps derived from 3D LiDAR point clouds, a goal cost map, and processed proprioception data as state inputs, and learns the physical and geometric properties of the surrounding obstacles such as height, density, and solidity/stiffness. The fully-trained policy’s critic network is then used to evaluate the quality of dynamically feasible velocities generated from a novel contextaware planner. Our planner adapts the robot’s velocity space based on the presence of entrapment including vegetation, and narrow passages in dense environments. We demonstrate our method’s capabilities on a Spot robot in complex real-world outdoor scenes, including dense vegetation. We observe that VAPOR’s actions improve success rates by up to 40%, decrease the average current consumption by up to 2.9%, and decrease the normalized trajectory length by up to 11.2% compared to existing end-to-end offline RL and other outdoor navigation methods.

We present a new learning approach, Soft Conditional Prompt Learning (SCP), which leverages the strengths of prompt learning for aerial video action recognition. Our approach is designed to predict the action of each agent by helping the models focus on the descriptions or instructions associated with actions in the input videos for aerial/robot visual perception. Our formulation supports various prompts, including learnable prompts, auxiliary visual information, and large vision models to improve the recognition performance. We present a soft conditional prompt method that learns to dynamically generate prompts from a pool of prompt experts under different video inputs. By sharing the same objective with the task, our proposed SCP can optimize prompts that guide the model’s predictions while explicitly learning input-invariant (prompt experts pool) and input-specific (data-dependent) prompt knowledge. In practice, we observe a 3.17 − 10.2% accuracy improvement on the aerial video datasets (Okutama [1], NECDrone [2]), which consist of scenes with single-agent and multi-agent actions. We further evaluate our approach on ground camera videos to verify the effectiveness and generalization and achieve a 1.0 − 3.6% improvement on SSV2 [3]. We integrate our method into the ROS2 as well.

We present a novel end-to-end algorithm (PoCo) for the indoor RGB-D place recognition task, aimed at identifying the most likely match for a given query frame within a reference database. The task presents inherent challenges attributed to the constrained field of view and limited range of perception sensors. We propose a new network architecture, which generalizes the recent Context of Clusters (CoCs) to extract global descriptors directly from the noisy point clouds through end-to-end learning. Moreover, we develop the architecture by integrating both color and geometric modalities into the point features to enhance the global descriptor representation. We conducted evaluations on public datasets ScanNet-PR and ARKit with 807 and 5047 scenarios, respectively. PoCo achieves SOTA performance: on ScanNet-PR, we achieve R@1 of 64.63%, a 5.7% improvement from the best-published result CGis (61.12%); on Arkit, we achieve R@1 of 45.12%, a 13.3% improvement from the best-published result CGis (39.82%). In addition, PoCo shows higher efficiency than CGis in inference time (1.75X-faster), and we demonstrate the effectiveness of PoCo in recognizing places within a real-world laboratory environment. Video: https://youtu.be/D8dObAeMiCw;

We present a novel end-to-end diffusion-based trajectory generation method, DTG, for mapless global navigation in challenging outdoor scenarios with occlusions and unstructured off-road features like grass, buildings, bushes, etc. Given a distant goal, our approach computes a trajectory that satisfies the following goals: (1) minimize the travel distance to the goal; (2) maximize the traversability by choosing paths that do not lie in undesirable areas. Specifically, we present a novel Conditional RNN(CRNN) for diffusion models to efficiently generate trajectories. Furthermore, we propose an adaptive training method that ensures that the diffusion model generates more traversable trajectories. We evaluate our methods in various outdoor scenes and compare the performance with other global navigation algorithms on a Husky robot. In practice, we observe at least a 15% improvement in traveling distance and around a 7% improvement in traversability. Video and Code: https://github.com/jingGM/DTG.git.

Decentralized navigation methods rely primarily on local observations, lacking the global awareness needed to coordinate effectively within a multi-agent system. Exchanging relevant messages between agents can promote cooperation and improve navigation efficiency. We present a Reinforcement Learning (RL)-based decentralized navigation approach that learns ‘when,’ ‘what,’ and ‘with whom’ to communicate for safe and cooperative navigation. Our method leverages a visual transformer and self-attention mechanism to encode the local occupancy map and the state information of neighbors into fixed-length encodings, allowing it to handle an arbitrary number of neighbors for collision-free navigation. In addition, the network encodes the agent’s state information and observations of neighboring agents into a concise message vector by learning what information is crucial to communicate, which is shared with neighboring agents upon request. Moreover, to avoid indiscriminate broadcasting, the network learns when and with whom to communicate and request message vectors. Subsequently, the messages communicated alongside the local information are used to guide navigation decisions. We evaluate our method against state-of-the-art baselines in complex scenarios, including narrow corridors and environments with multiple agents. We observe considerable improvements in terms of navigation performance, showing up to ∼ 2× improvement in navigation success rates and a reduction of up to ∼ 20% in path length.

We present VLPG-Nav, a visual language navigation method for guiding robots to specified objects within household scenes. Unlike existing methods primarily focused on navigating the robot toward objects, our approach considers the additional challenge of centering the object within the robot’s camera view. Our method builds a visual language pose graph (VLPG) that functions as a spatial map of VL embeddings. Given an open-vocabulary object query, we plan a viewpoint for object navigation using the VLPG. Despite navigating to the viewpoint, real-world challenges such as object occlusion, displacement, and the robot’s localization errors can prevent visibility. We build an object localization probability map that leverages the robot’s current observations and prior VLPG. When the object is not visible, the probability map is updated, and an alternate viewpoint is computed. In addition, we propose an object-centering formulation that locally adjusts the robot’s pose to center the object in the camera view. We evaluate the effectiveness of our approach through simulations and real-world experiments, evaluating its ability to successfully view and center the object within the camera’s field of view. VLPG-Nav demonstrates improved performance in locating the object, navigating around occlusions, and centering the object within the robot’s camera view, outperforming selected baselines in the evaluation settings.

We present AMCO, a novel navigation method for quadruped robots that adaptively combines vision-based and proprioception-based perception capabilities. Our approach uses three cost maps: general knowledge map; traversability history map; and current proprioception map; which are derived from a robot’s vision and proprioception data, and couples them to obtain a coupled traversability cost map for navigation. The general knowledge map encodes terrains semantically segmented from visual sensing, and represents a terrain’s typically expected traversability. The traversability history map encodes the robot’s recent proprioceptive measurements on a terrain and its semantic segmentation as a cost map. Further, the robot’s present proprioceptive measurement is encoded as a cost map in the current proprioception map. As the general knowledge map and traversability history map rely on semantic segmentation, we evaluate the reliability of the visual sensory data by estimating the brightness and motion blur of input RGB images and accordingly combine the three cost maps to obtain the coupled traversability cost map used for navigation. Leveraging this adaptive coupling, the robot can depend on the most reliable input modality available. Finally, we present a novel planner that selects appropriate gaits and velocities for traversing challenging outdoor environments using the coupled traversability cost map. We demonstrate AMCO’s navigation performance in different real-world outdoor environments and observe 10.8%-34.9% reduction w.r.t. two stability metrics, and up to 50% improvement in terms of success rate compared to current navigation methods.

We present AGL-NET, a novel learning-based method for global localization using LiDAR point clouds and satellite maps. AGL-Net tackles two critical challenges: bridging the representation gap between image and points modalities for robust feature matching, and handling inherent scale discrepancies between global view and local view. To address these challenges, AGL-Net leverages a unified network architecture with a novel two-stage matching design. The first stage extracts informative neural features directly from raw sensor data and performs initial feature matching. The second stage refines this matching process by extracting informative skeleton features and incorporating a novel scale alignment step to rectify scale variations between LiDAR and map data. Furthermore, a novel scale and skeleton loss function guides the network toward learning scale-invariant feature representations, eliminating the need for pre-processing satellite maps. This significantly improves real-world applicability in scenarios with unknown map scales. To facilitate rigorous performance evaluation, we introduce a meticulously designed dataset within the CARLA simulator specifically tailored for metric localization training and assessment.

We present CoNVOI, a novel method for autonomous robot navigation in real-world indoor and outdoor environments using Vision Language Models (VLMs). We employ VLMs in two ways: first, we leverage their zero-shot image classification capability to identify the context or scenario (e.g., indoor corridor, outdoor terrain, crosswalk, etc) of the robot’s surroundings, and formulate context-based navigation behaviors as simple text prompts (e.g. "stay on the pavement"). Second, we utilize their state-of-the-art semantic understanding and logical reasoning capabilities to compute a suitable trajectory given the identified context. To this end, we propose a novel multi-modal visual marking approach to annotate the obstacle-free regions in the RGB image used as input to the VLM with numbers, by correlating it with a local occupancy map of the environment. The marked numbers ground image locations in the real-world, direct the VLM’s attention solely to navigable locations, and elucidate the spatial relationships between them and terrains depicted in the image to the VLM. Next, we query the VLM to select numbers on the marked image that satisfy the context-based behavior text prompt, and construct a reference path using the selected numbers. Finally, we propose a method to extrapolate the reference trajectory when the robot’s environmental context has not changed to prevent unnecessary VLM queries. We use the reference trajectory to guide a motion planner, and demonstrate that it leads to human-like behaviors (e.g. not cutting through a group of people, using crosswalks, etc.) in various real-world indoor and outdoor scenarios. We perform several ablations and navigation comparisons and demonstrate that CoNVOI’s trajectories are most similar to human teleoperated ground truth in terms of Fréchet distance (9.7-58.2% closer), lowest path errors (up to 88.13% lower), and up to 86.09% lower % of unacceptable paths.

Navigating robots through unstructured terrains is challenging, primarily due to the dynamic environmental changes. While humans adeptly navigate such terrains by using context from their observations, creating a similar context-aware navigation system for robots is difficult. The essence of the issue lies in the acquisition and interpretation of context information, a task complicated by the inherent ambiguity of human language. In this work, we introduce LANCAR, which addresses this issue by combining a context translator with reinforcement learning (RL) agents for context-aware locomotion. LANCAR allows robots to comprehend context information through Large Language Models (LLMs) sourced from human observers and convert this information into actionable context embeddings. These embeddings, combined with the robot’s sensor data, provide a complete input for the RL agent’s policy network. We provide an extensive evaluation of LANCAR under different levels of context ambiguity and compare with alternative methods. The experimental results showcase the superior generalizability and adaptability across different terrains. Notably, LANCAR shows at least a 7.4% increase in episodic reward over the best alternatives, highlighting its potential to enhance robotic navigation in unstructured environments. More details and experiment videos could be found in this link.

Aligning foundation models is essential for their safe and trustworthy deployment. However, traditional fine-tuning methods are computationally intensive and require updating billions of model parameters. A promising alternative, alignment via decoding, adjusts the response distribution directly without model updates to maximize a target reward $r$, thus providing a lightweight and adaptable framework for alignment. However, principled decoding methods rely on oracle access to an optimal Q-function ($Q^*$), which is often unavailable in practice. Hence, prior SoTA methods either approximate this $Q^*$ using $Q^{\pi_{\text{sft}}}$ (derived from the reference $\texttt{SFT}$ model) or rely on short-term rewards, resulting in sub-optimal decoding performance. In this work, we propose $\texttt{Transfer Q}^*$, which implicitly estimates the optimal value function for a target reward $r$ through a baseline model $\rho_{\texttt{BL}}$ aligned with a baseline reward $r_{\texttt{BL}}$ (which can be different from the target reward $r$). Theoretical analyses of $\texttt{Transfer Q}^*$ provide a rigorous characterization of its optimality, deriving an upper bound on the sub-optimality gap and identifying a hyperparameter to control the deviation from the pre-trained reference $\texttt{SFT}$ model based on user needs. Our approach significantly reduces the sub-optimality gap observed in prior SoTA methods and demonstrates superior empirical performance across key metrics such as coherence, diversity, and quality in extensive tests on several synthetic and real datasets.

Reading monolithic instructional documents in VR is often challenging, especially when tasks are collaborative. Here we present DocuBits, a novel method for transforming monolithic documents into small, interactive instructional elements. Our approach allows users to:(i) create instructional elements (ii) position them within VR and (iii) use them to monitor and share progress in a multi-user VR learning environment. We describe our design methodology as well as two user studies evaluating how both individual users and pairs of users interact with DocuBits compared to monolithic documents while performing a chemistry lab task. Our analysis shows that, for both studies, DocuBits had substantially higher usability, while decreasing perceived workload $(p \lt 0.001)$. Our collaborative study showed that participants perceived higher social presence, collaborator awareness as well as immersion and presence $(p \lt 0.001)$. We discuss our insights for using text-based instructions to support enhanced collaboration in VR environments.

We present an end-to-end binaural audio rendering approach (Listen2Scene) for virtual reality (VR) and augmented reality (AR) applications. We propose a novel neural-network-based binaural sound propagation method to generate acoustic effects for indoor 3D models of real environments. Any clean audio or dry audio can be convolved with the generated acoustic effects to render audio corresponding to the real environment. We propose a graph neural network that uses both the material and the topology information of the 3D scenes and generates a scene latent vector. Moreover, we use a conditional generative adversarial network (CGAN) to generate acoustic effects from the scene latent vector. Our network can handle holes or other artifacts in the reconstructed 3D mesh model. We present an efficient cost function for the generator network to incorporate spatial audio effects. Given the source and the listener position, our learning-based binaural sound propagation approach can generate an acoustic effect in 0.1 milliseconds on an NVIDIA GeForce RTX 2080 Ti GPU. We have evaluated the accuracy of our approach with binaural acoustic effects generated using an interactive geometric sound propagation algorithm and captured real acoustic effects /real-world recordings. We also performed a perceptual evaluation and observed that the audio rendered by our approach is more plausible than audio rendered using prior learning-based and geometric-based sound propagation algorithms. We quantitatively evaluated the accuracy of our approach using statistical acoustic parameters, and energy decay curves. The demo videos, code and dataset are available online1.

We present a new algorithm for the selection of informative frames in video action recognition. Our approach is designed for aerial videos captured using a moving camera where human actors occupy a small spatial resolution of video frames. Our algorithm utilizes the motion bias within aerial videos, which enables the selection of motion-salient frames. We introduce the concept of patch mutual information (PMI) score to quantify the motion bias between adjacent frames, by measuring the similarity of patches. We use this score to assess the amount of discriminative motion information contained in one frame relative to another. We present an adaptive frame selection strategy using shifted leaky ReLu and cumulative distribution function, which ensures that the sampled frames comprehensively cover all the essential segments with high motion salience. Our approach can be integrated with any action recognition model to enhance its accuracy. In practice, our method achieves a relative improvement of 2.2 - 13.8% in top-1 accuracy on UAV-Human, 6.8% on NEC Drone, and 9.0% on Diving48 datasets. The code is available at https://github.com/Ricky-Xian/PMI-Sampler.

We present a novel approach for action recognition in UAV videos. Our formulation is designed to handle occlusion and viewpoint changes caused by the movement of a UAV. We use the concept of mutual information to compute and align the regions corresponding to human action or motion in the temporal domain. This enables our recognition model to learn from the key features associated with the motion. We also propose a novel frame sampling method that uses joint mutual information to acquire the most informative frame sequence in UAV videos. We have integrated our approach with X3D and evaluated the performance on multiple datasets. In practice, we achieve 18.9% improvement in Top-1 accuracy over current state-of-the-art methods on UAV-Human [30], 7.3% improvement on Drone-Action [41], and 7.16% improvement on NEC Drones [7]. The code is available at https://github.com/Ricky-Xian/MITFAS

Although distance learning presents a number of interesting educational advantages as compared to in-person instruction, it is not without its downsides. We first assess the educational challenges presented by distance learning as a whole and identify 4 main challenges that distance learning currently presents as compared to in-person instruction: the lack of social interaction, reduced student engagement and focus, reduced comprehension and information retention, and the lack of flexible and customizable instructor resources. After assessing each of these challenges in-depth, we examine how AR/VR technologies might serve to address each challenge along with their current shortcomings, and finally outline the further research that is required to fully understand the potential of AR/VR technologies as they apply to distance learning.

In this paper, we present a novel multi-modal attention guidance method designed to address the challenges of turn-taking dynamics in meetings and enhance group conversations within virtual reality (VR) environments. Recognizing the difficulties posed by a confined field of view and the absence of detailed gesture tracking in VR, our proposed method aims to mitigate the challenges of noticing new speakers attempting to join the conversation. This approach tailors attention guidance, providing a nuanced experience for highly engaged participants while offering subtler cues for those less engaged, thereby enriching the overall meeting dynamics. Through group interview studies, we gathered insights to guide our design, resulting in a prototype that employs light as a diegetic guidance mechanism, complemented by spatial audio. The combination creates an intuitive and immersive meeting environment, effectively directing users' attention to new speakers. An evaluation study, comparing our method to state-of-the-art attention guidance approaches, demonstrated significantly faster response times (p < 0.001), heightened perceived conversation satisfaction (p < 0.001), and preference (p < 0.001) for our method. Our findings contribute to the understanding of design implications for VR social attention guidance, opening avenues for future research and development.

We provide the first perceptual quantification of user's sensitivity to radial optic flow artifacts and demonstrate a promising approach for masking this optic flow artifact via blink suppression. Near-eye HMOs allow users to feel immersed in virtual environments by providing visual cues, like motion parallax and stereoscopy, that mimic how we view the physical world. However, these systems exhibit a variety of perceptual artifacts that can limit their usability and the user's sense of presence in VR. One well-known artifact is the vergence-accommodation conflict (VAC). Varifocal displays can mitigate VAC, but bring with them other artifacts such as a change in virtual image size (radial optic flow) when the focal plane changes. We conducted a set of psychophysical studies to measure users' ability to perceive this radial flow artifact before, during, and after self-initiated blinks. Our results showed that visual sensitivity was reduced by a factor of 10 at the start and for ~70 ms after a blink was detected. Pre- and post-blink sensitivity was, on average, ~O.15% image size change during normal viewing and increased to ~1.5- 2.0% during blinks. Our results imply that a rapid (under 70 ms) radial optic flow distortion can go unnoticed during a blink. Furthermore, our results provide empirical data that can be used to inform engineering requirements for both hardware design and software-based graphical correction algorithms for future varifocal near-eye displays. Our project website is available at https://gamma.umd.edu/ROF/.

Model-based approaches to reinforcement learning (MBRL) exhibit favorable performance in practice, but their theoretical guarantees in large spaces are mostly restricted to the setting when transition model is Gaussian or Lipschitz, and demands a posterior estimate whose representational complexity grows unbounded with time. In this work, we develop a novel MBRL method (i) which relaxes the assumptions on the target transition model to belong to a generic family of mixture models; (ii) is applicable to large-scale training by incorporating a compression step such that the posterior estimate consists of a Bayesian coreset of only statistically significant past state-action pairs; and (iii) exhibits a sublinear Bayesian regret. To achieve these results, we adopt an approach based upon Stein's method, which, under a smoothness condition on the constructed posterior and target, allows distributional distance to be evaluated in closed form as the kernelized Stein discrepancy (KSD). The aforementioned compression step is then computed in terms of greedily retaining only those samples which are more than a certain KSD away from the previous model estimate. Experimentally, we observe that this approach is competitive with several state-of-the-art RL methodologies, and can achieve up-to 50 percent reduction in wall clock time in some continuous control environments.

Professional document editing tools require a certain level of expertise to perform complex edit operations. To make editing tools accessible to increasingly novice users, we investigate intelligent document assistant systems that can make or suggest edits based on a user's natural language request. Such a system should be able to understand the user's ambiguous requests and contextualize them to the visual cues and textual content found in a document image to edit localized unstructured text and structured layouts. To this end, we propose a new task of language-guided localized document editing, where the user provides a document and an open vocabulary editing request, and the intelligent system produces a command that can be used to automate edits in real-world document editing software. In support of this task, we curate the DocEdit dataset, a collection of approximately 28K instances of user edit requests over PDF and design templates along with their corresponding ground truth software executable commands. To our knowledge, this is the first dataset that provides a diverse mix of edit operations with direct and indirect references to the embedded text and visual objects such as paragraphs, lists, tables, etc. We also propose DocEditor, a Transformer-based localization-aware multimodal (textual, spatial, and visual) model that performs the new task. The model attends to both document objects and related text contents which may be referred to in a user edit request, generating a multimodal embedding that is used to predict an edit command and associated bounding box localizing it. Our proposed model empirically outperforms other baseline deep learning approaches by 15-18%, providing a strong starting point for future work.

Sreyan Ghosh, Utkarsh Tyagi, Manan Suri, Sonal Kumar, Ramaneswaran S, Dinesh Manocha. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

We present a new pipeline for acquiring a textured mesh in the wild with a single smartphone which offers access to images, depth maps, and valid poses. Our method first introduces an RGBD-aided structure from motion, which can yield filtered depth maps and refines camera poses guided by corresponding depth. Then, we adopt the neural implicit surface reconstruction method, which allows for high-quality mesh and develops a new training process for applying a regularization provided by classical multi-view stereo methods. Moreover, we apply a differentiable rendering to fine-tune incomplete texture maps and generate textures which are perceptually closer to the original scene. Our pipeline can be applied to any common objects in the real world without the need for either in-the-lab environments or accurate mask images. We demonstrate results of captured objects with complex shapes and validate our method numerically against existing 3D reconstruction and texture mapping methods.

The choice of input text prompt plays a critical role in the performance of Vision-Language Pretrained (VLP) models such as CLIP.We present APoLLo , a unified multi-modal approach that combines Adapter and Prompt learning for Vision-Language models.Our method is designed to substantially improve the generalization capabilities of VLP models when they are fine-tuned in a few-shot setting.We introduce trainable cross-attention-based adapter layers in conjunction with vision and language encoders to strengthen the alignment between the two modalities.We enforce consistency between the respective encoder branches (receiving augmented inputs) to prevent overfitting in downstream tasks.Our method is evaluated on three representative tasks: generalization to novel classes, cross-dataset evaluation, and unseen domain shifts.In practice, APoLLo achieves a relative gain up to 6.03% over MaPLe (SOTA) on novel classes for 10 diverse image recognition datasets.

Sreyan Ghosh, Chandra Kiran Reddy Evuru, Sonal Kumar, S Ramaneswaran, S Sakshi, Utkarsh Tyagi, Dinesh Manocha. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023.

The tremendous growth of social media users interacting in online conversations has led to significant growth in hate speech affecting people from various demographics.Most of the prior works focus on detecting explicit hate speech, which is overt and leverages hateful phrases, with very little work focusing on detecting hate speech that is implicit or denotes hatred through indirect or coded language.In this paper, we present CoSyn, a context synergized neural network that explicitly incorporates user-and conversational-context for detecting implicit hate speech in online conversations.CoSyn introduces novel ways to encode these external contexts and employs a novel context interaction mechanism that clearly captures the interplay between them, making independent assessments of the amounts of information to be retrieved from these noisy contexts.Additionally, it carries out all these operations in the hyperbolic space to account for the scalefree dynamics of social media.We demonstrate the effectiveness of CoSyn on 6 hate speech datasets and show that CoSyn outperforms all our baselines in detecting implicit hate speech with absolute improvements in the range of 1.24% -57.8%.We make our code available 1 .

We present Multiscale Audio Spectrogram Transformer (MAST) for audio classification, which brings the concept of multiscale feature hierarchies to the Audio Spectrogram Transformer (AST) [1]. Given an input audio spectrogram, we first patchify and project it into an initial temporal resolution and embedding dimension, post which the multiple stages in MAST progressively expand the embedding dimension while reducing the temporal resolution of the input. We use a pyramid structure that allows early layers of MAST operating at a high temporal resolution but low embedding space to model simple low-level acoustic information and deeper temporally coarse layers to model high-level acoustic information with high-dimensional embeddings. We also extend our approach to present a new Self-Supervised Learning (SSL) method called SS-MAST, which calculates a symmetric contrastive loss between latent representations from a student and a teacher encoder, leveraging patch-drop, a novel audio augmentation approach that we introduce. In practice, MAST significantly outperforms AST by an average accuracy of 3.4% across 8 speech and non-speech tasks from the LAPE Benchmark [2], achieving state-of-the-art results on keyword spotting in Speech Commands. Additionally, our proposed SS-MAST achieves an absolute average improvement of 2.6% over the previously proposed SSAST [3]1.

We propose a novel approach for blind room impulse response (RIR) estimation systems in the context of a downstream application scenario, far-field automatic speech recognition (ASR). We first draw the connection between improved RIR estimation and improved ASR performance, as a means of evaluating neural RIR estimators. We then propose a generative adversarial network (GAN) based architecture that encodes RIR features from reverberant speech and constructs an RIR from the encoded features, and uses a novel energy decay relief loss to optimize for capturing energy-based properties of the input reverberant speech. We show that our model outperforms the state-of-the-art baselines on acoustic benchmarks (by 17% on the energy decay relief and 22% on an early-reflection energy metric), as well as in an ASR evaluation task (by 6.9% in word error rate).

We present a new Self-Supervised Learning (SSL) approach to pre-train encoders on unlabeled audio data that reduces the need for large amounts of labeled data for audio and speech classification. Our primary aim is to learn au-dio representations that can generalize across a large vari-ety of speech and non-speech tasks in a low-resource un-labeled audio pre-training setting. Inspired by the recent success of clustering and contrasting learning paradigms for SSL-based speech representation learning, we propose SLICER (Symmetrical Learning of Instance and Cluster-level Efficient Representations) which brings together the best of both clustering and contrasting learning paradigms. We use a symmetric loss between latent representations from student and teacher encoders and simultaneously solve in-stance and cluster-level contrastive learning tasks. We obtain cluster representations online by just projecting the input spectrogram into an output subspace with dimensions equal to the number of clusters. In addition, we propose a novel mel-spectrogram augmentation procedure k-mix, based on mixup [1], which does not require labels and aids unsupervised representation learning for audio. Overall, SLICER achieves state-of-the-art results on the LAPE Benchmark [2], significantly outperforming all other prior approaches, some-times pre-trained on 10× larger unsupervised data than our setting. Code https: https://github.com/Sreyan88/audio-ssl.

We present AdVerb, a novel audio-visual dereverberation framework that uses visual cues in addition to the reverberant sound to estimate clean audio. Although audioonly dereverberation is a well-studied problem, our approach incorporates the complementary visual modality to perform audio dereverberation. Given an image of the environment where the reverberated sound signal has been recorded, AdVerb employs a novel geometry-aware cross-modal transformer architecture that captures scene geometry and audio-visual cross-modal relationship to generate a complex ideal ratio mask, which, when applied to the reverberant audio predicts the clean sound. The effectiveness of our method is demonstrated through extensive quantitative and qualitative evaluations. Our approach significantly outperforms traditional audio-only and audio-visual baselines on three downstream tasks: speech enhancement, speech recognition, and speaker verification, with relative improvements in the range of 18% - 82% on the LibriSpeech test-clean set. We also achieve highly satisfactory RT60 error scores on the AVSpeech dataset.

We present CrossLoc3D, a novel 3D place recognition method that solves a large-scale point matching problem in a cross-source setting. Cross-source point cloud data corresponds to point sets captured by depth sensors with different accuracies or from different distances and perspectives. We address the challenges in terms of developing 3D place recognition methods that account for the representation gap between points captured by different sources. Our method handles cross-source data by utilizing multi-grained features and selecting convolution kernel sizes that correspond to most prominent features. Inspired by the diffusion models, our method uses a novel iterative refinement process that gradually shifts the embedding spaces from different sources to a single canonical space for better metric learning. In addition, we present CS-Campus3D, the first 3D aerial-ground cross-source dataset consisting of point cloud data from both aerial and ground LiDAR scans. The point clouds in CS-Campus3D have representation gaps and other features like different views, point densities, and noise patterns. We show that our CrossLoc3D algorithm can achieve an improvement of 4.74% - 15.37% in terms of the top 1 average recall on our CS-Campus3D benchmark and achieves performance comparable to state-of-the-art 3D place recognition method on the Oxford RobotCar. The code and CS-Campus3D benchmark will be available at github.com/rayguan97/crossloc3d.

We propose a novel method, LoLep, which regresses Locally-Learned planes from a single RGB image to represent scenes accurately, thus generating better novel views. Without the depth information, regressing appropriate plane locations is a challenging problem. To solve this issue, we pre-partition the disparity space into bins and design a disparity sampler to regress local offsets for multiple planes in each bin. However, only using such a sampler makes the network not convergent; we further propose two optimizing strategies that combine with different disparity distributions of datasets and propose an occlusion-aware reprojection loss as a simple yet effective geometric supervision technique. We also introduce a self-attention mechanism to improve occlusion inference and present a Block-Sampling Self-Attention (BS-SA) module to address the problem of applying self-attention to large feature maps. We demonstrate the effectiveness of our approach and generate state-of-the-art results on different datasets. Compared to MINE, our approach has an LPIPS reduction of 4.8%∼9.0% and an RV reduction of 74.9% ~ 83.5%. We also evaluate the performance on real-world images and demonstrate the benefits.

Directed Exploration is a crucial challenge in reinforcement learning (RL), especially when rewards are sparse. Information-directed sampling (IDS), which optimizes the information ratio, seeks to do so by augmenting regret with information gain. However, estimating information gain is computationally intractable or relies on restrictive assumptions which prohibit its use in many practical instances. In this work, we posit an alternative exploration incentive in terms of the integral probability metric (IPM) between a current estimate of the transition model and the unknown optimal, which under suitable conditions, can be computed in closed form with the kernelized Stein discrepancy (KSD). Based on KSD, we develop a novel algorithm STEERING: STEin information dirEcted exploration for model-based Reinforcement LearnING. To enable its derivation, we develop fundamentally new variants of KSD for discrete conditional distributions. We further establish that STEERING archives sublinear Bayesian regret, improving upon prior learning rates of information-augmented MBRL, IDS included. Experimentally, we show that the proposed algorithm is computationally affordable and outperforms several prior approaches.

Many existing reinforcement learning (RL) methods employ stochastic gradient iteration on the back end, whose stability hinges upon a hypothesis that the data-generating process mixes exponentially fast with a rate parameter that appears in the step-size selection. Unfortunately, this assumption is violated for large state spaces or settings with sparse rewards, and the mixing time is unknown, making the step size inoperable. In this work, we propose an RL methodology attuned to the mixing time by employing a multi-level Monte Carlo estimator for the critic, the actor, and the average reward embedded within an actor-critic (AC) algorithm. This method, which we call Multi-level Actor-Critic (MAC), is developed specifically for infinite-horizon average-reward settings and neither relies on oracle knowledge of the mixing time in its parameter selection nor assumes its exponential decay; it is therefore readily applicable to applications with slower mixing times. Nonetheless, it achieves a convergence rate comparable to SOTA actor-critic algorithms. We experimentally show that these alleviated restrictions on the technical conditions required for stability translate to superior performance in practice for RL problems with sparse rewards.

We present a novel reinforcement learning based algorithm for multi-robot task allocation problem in ware-house environments. We formulate it as a Markov Decision Process and solve via a novel deep multi-agent reinforcement learning method (called RTAW) with attention inspired policy architecture. Hence, our proposed policy network uses global embeddings that are independent of the number of robots/tasks. We utilize proximal policy optimization algorithm for training and use a carefully designed reward to obtain a converged policy. The converged policy ensures cooperation among different robots to minimize total travel delay (TTD) which ultimately improves the makespan for a sufficiently large task-list. In our extensive experiments, we compare the performance of our RTAW algorithm to state of the art methods such as myopic pickup distance minimization (greedy) and regret based baselines on different navigation schemes. We show an improvement of upto 14% (25–1000 seconds) in TTD on scenarios with hundreds or thousands of tasks for different challenging warehouse layouts and task generation schemes. We also demonstrate the scalability of our approach by showing performance with up to 1000 robots in simulations.

In this paper, we present a novel Heavy-Tailed Stochastic Policy Gradient (HT-PSG) algorithm to deal with the challenges of sparse rewards in continuous control problems. Sparse rewards are common in continuous control robotics tasks such as manipulation and navigation and make the learning problem hard due to the non-trivial estimation of value functions over the state space. This demands either reward shaping or expert demonstrations for the sparse reward environment. However, obtaining high-quality demonstrations is quite expensive and sometimes even impossible. We propose a heavy-tailed policy parametrization along with a modified momentum-based policy gradient tracking scheme (HT-SPG) to induce a stable exploratory behavior in the algorithm. The proposed algorithm does not require access to expert demonstrations. We test the performance of HT-SPG on various benchmark tasks of continuous control with sparse rewards such as 1D Mario, Pathological Mountain Car, Sparse Pendulum in OpenAI Gym, and Sparse MuJoCo environments (Hopper-v2, Half-Cheetah, Walker-2D). We show consistent performance improvement across all tasks in terms of high average cumulative reward without requiring access to expert demonstrations. We further demonstrate that a navigation policy trained using HT-SPG can be easily transferred into a Clearpath Husky robot to perform real-world navigation tasks.

We present a new traffic dataset, Meteor, which captures traffic patterns and multi-agent driving behaviors in unstructured scenarios. Meteor consists of more than 1000 one-minute videos, over 2 million annotated frames with bounding boxes and GPS trajectories for 16 unique agent categories, and more than 13 million bounding boxes for traffic agents. Meteor is a dataset for rare and interesting, multi-agent driving behaviors that are grouped into traffic violations, atypical interactions, and diverse scenarios. Every video in Meteor is tagged using a diverse range of factors corresponding to weather, time of the day, road conditions, and traffic density. We use Meteor to benchmark perception methods for object detection and multi-agent behavior prediction. Our key finding is that state-of-the-art models for object detection and behavior prediction, which otherwise succeed on existing datasets such as Waymo, fail on the Meteor dataset. Meteor is a step towards developing more sophisticated perception models for dense, heterogeneous, and unstructured scenarios.

We present a lightweight, decentralized algorithm for navigating multiple nonholonomic agents through challenging environments with narrow passages. Our key idea is to allow agents to yield to each other in large open areas instead of narrow passages, to increase the success rate of conventional decentralized algorithms. At pre-processing time, our method computes a medial axis for the freespace. A reference trajectory is then computed and projected onto the medial axis for each agent. During run time, when an agent senses other agents moving in the opposite direction, our algorithm uses the medial axis to estimate a Point of Impact (POI) as well as the available area around the POI. If the area around the POI is not large enough for yielding behaviors to be successful, we shift the POI to nearby large areas by modulating the agent's reference trajectory and traveling speed. We evaluate our method on a row of 4 environments with up to 15 robots, and we find our method incurs a marginal computational overhead of 10–30 ms on average, achieving real-time performance. Afterward, our planned reference trajectories can be tracked using local navigation algorithms to achieve up to a 100% higher success rate over local navigation algorithms alone.

We present a learning algorithm, DifFAR, for human activity recognition in videos. Our approach is designed for UAV videos, which are mainly acquired from obliquely placed dynamic cameras that contain a human actor along with background motion. Typically, the human actors occupy less than one-tenth of the spatial resolution. DifFAR simultaneously harnesses the benefits of frequency domain representations, a classical analysis tool in signal processing, and data driven neural networks. We build a differentiable static-dynamic frequency mask prior to model the salient static and dynamic pixels in the video, crucial for the underlying task of action recognition. We use this differentiable mask prior to enable the neural network to intrinsically learn disentangled feature representations via an identity loss function. Our formulation empowers the network to inherently compute disentangled salient features within its layers. Further, we propose a cost-function encapsulating temporal relevance and spatial content to sample the most important frame within uniformly spaced video segments. We conduct extensive experiments on the UAV Human dataset and the NEC Drone dataset and demonstrate relative improvements of 5.72% - 13.00% over the state-of-the-art and 14.28% - 38.05% over the corresponding baseline model.

Human action recognition is a challenging problem, particularly when there is high variability in factors such as subject appearance, backgrounds and viewpoint. While deep neural networks (DNNs) have been shown to perform well on action recognition tasks, they typically require large amounts of high-quality labeled data to achieve robust performance across a variety of conditions. Synthetic data has shown promise as a way to avoid the substantial costs and potential ethical concerns associated with collecting and labeling enormous amounts of data in the real-world. However, synthetic data may differ from real data in important ways. This phenomenon, known as domain shift, can limit the utility of synthetic data in robotics applications. To mitigate the effects of domain shift, substantial effort is being dedicated to the development of domain adaptation (DA) techniques. Yet, much remains to be understood about how best to develop these techniques. In this paper, we introduce a new dataset called Robot Control Gestures (RoCoG-v2). The dataset is composed of both real and synthetic videos from seven gesture classes, and is intended to support the study of synthetic-to-real domain shift for video-based action recognition. Our work expands upon existing datasets by focusing the action classes on gestures for human-robot teaming, as well as by enabling investigation of domain shift in both ground and aerial views. We present baseline results using state-of-the-art action recognition and domain adaptation algorithms and offer initial insight on tackling the synthetic-to-real and ground-to-air domain shifts. Instructions on accessing the dataset can be found at https://github.com/reddyav1/RoCoG-v2.

We propose a novel approach for aerial video action recognition. Our method is designed for videos captured using UAVs and can run on edge or mobile devices. We present a learning-based approach that uses customized auto zoom to automatically identify the human target and scale it appropriately. This makes it easier to extract the key features and reduces the computational overhead. We also present an efficient temporal reasoning algorithm to capture the action information along the spatial and temporal domains within a controllable computational cost. Our approach has been implemented and evaluated both on the desktop with high-end GPUs and on the low power Robotics RB5 Platform for robots and drones. In practice, we achieve$6.1-7.4 \%$improvement over SOTA in Top-1 accuracy on the RoCoG-v2 dataset, 8.3-10.4% improvement on the UAV-Human dataset and 3.2% improvement on the Drone Action dataset.

We present an algorithm for safe robot navigation in complex dynamic environments using a variant of model predictive equilibrium point control. We use an optimization formulation to navigate robots gracefully in dynamic environments by optimizing over a trajectory cost function at each timestep. We present a novel trajectory cost formulation that significantly reduces conservative and deadlocking behaviors and generates smooth trajectories. In particular, we propose a new collision probability function that effectively captures the risk associated with a given configuration and the time to avoid collisions based on the velocity direction. Moreover, we propose a terminal state cost based on the expected time-to-goal and time-to-collision values that helps in avoiding trajectories that could result in deadlock. We evaluate our cost formulation in multiple simulated scenarios, including narrow corridors with dynamic obstacles, and observe significantly improved navigation behavior and reduced deadlocks as compared to prior methods.

We propose a novel method for autonomous legged robot navigation in densely vegetated environments with a variety of pliable/traversable and non-pliable/untraversable vegetation. We present a novel few-shot learning classifier that can be trained on a few hundred RGB images to differentiate flora that can be navigated through, from the ones that must be circumvented. Using the vegetation classification and 2D lidar scans, our method constructs a vegetation-aware traversability cost map that accurately represents the pliable and non-pliable obstacles with lower, and higher traversability costs, respectively. Our cost map construction accounts for misclassifications of the vegetation and further lowers the risk of collisions, freezing and entrapment in vegetation during navigation. Furthermore, we propose holonomic recovery behaviors for the robot for scenarios where it freezes, or gets physically entrapped in dense, pliable vegetation. We demonstrate our method on a Boston Dynamics Spot robot in real-world unstructured environments with sparse and dense tall grass, bushes, trees, etc. We observe an increase of 25-90% in success rates, 10-90% decrease in freezing rate, and up to 65% decrease in the false positive rate compared to existing methods.

Radio applications are increasingly being used in urban environments for cellular radio systems and safety applications that use vehicle-vehicle, and vehicle-to-infrastructure. We present a novel ray tracing-based radio propagation algorithm that can handle large urban scenes with hundreds or thousands of dynamic objects and receivers. Our approach is based on the use of coherence-based techniques that exploit spatial and temporal coherence for efficient wireless propagation and radio network planning. Our formulation also utilizes channel coherence which is used to determine the effectiveness of the propagation model within a certain time in dynamically generated paths; and spatial consistency which is used to estimate the similarity and accuracy of changes in a dynamic environment with varying propagation models and blocking obstacles. We highlight the performance of our simulator in large urban traffic scenes with an area of $2 \ast 2 km ^{2}$ and more than 10,000 users and devices. We evaluate the accuracy by comparing the results with discrete model simulations performed using WinProp. In practice, our approach scales linearly with the area of the urban environment and the number of dynamic obstacles or receivers.

Biomedical Named Entity Recognition (BioNER) is the fundamental task of identifying named entities from biomedical text. However, BioNER suffers from severe data scarcity and lacks high-quality labeled data due to the highly specialized and expert knowledge required for annotation. Though data augmentation has shown to be highly effective for low-resource NER in general, existing data augmentation techniques fail to produce factual and diverse augmentations for BioNER. In this paper, we present BioAug, a novel data augmentation framework for low-resource BioNER. BioAug, built on BART, is trained to solve a novel text reconstruction task based on selective masking and knowledge augmentation. Post training, we perform conditional generation and generate diverse augmentations conditioning BioAug on selectively corrupted text similar to the training stage. We demonstrate the effectiveness of BioAug on 5 benchmark BioNER datasets and show that BioAug outperforms all our baselines by a significant margin (1.5%-21.5% absolute improvement) and is able to generate augmentations that are both more factual and diverse. Code: https://github.com/Sreyan88/BioAug.

We present a novel method, SALAD, for the challenging vision task of adapting a pre-trained "source" domain network to a "target" domain, with a small budget for annotation in the "target" domain and a shift in the label space. Further, the task assumes that the source data is not available for adaptation, due to privacy concerns or otherwise. We postulate that such systems need to jointly optimize the dual task of (i) selecting fixed number of samples from the target domain for annotation and (ii) transfer of knowledge from the pre-trained network to the target domain. To do this, SALAD consists of a novel Guided Attention Transfer Network (GATN) and an active learning function, HAL. The GATN enables feature distillation from pre-trained network to the target network, complemented with the target samples mined by HALusing transfer-ability and uncertainty criteria. SALAD has three key benefits: (i) it is task-agnostic, and can be applied across various visual tasks such as classification, segmentation and detection; (ii) it can handle shifts in output label space from the pre-trained source network to the target domain; (iii) it does not require access to source data for adaptation. We conduct extensive experiments across 3 visual tasks, viz. digits classification (MNIST, SVHN, VISDA), synthetic (GTA5) to real (CityScapes) image segmentation, and document layout detection (PubLayNet to DSSE). We show that our source-free approach, SALAD, results in an improvement of 0.5%−31.3% (across datasets and tasks) over prior adaptation methods that assume access to large amounts of annotated source data for adaptation. Code is available here.

Digital documents often contain images and scanned text. Parsing such visually-rich documents is a core task for work-flow automation, but it remains challenging since most documents do not encode explicit layout information, e.g., how characters and words are grouped into boxes and ordered into larger semantic entities. Current state-of-the-art layout extraction methods are challenged by such documents as they rely on word sequences to have correct reading order and do not exploit their hierarchical structure. We propose LayerDoc, an approach that uses visual features, textual semantics, and spatial coordinates along with constraint inference to extract the hierarchical layout structure of documents in a bottom-up layer-wise fashion. LayerDoc recursively groups smaller regions into larger semantic elements in 2D to infer complex nested hierarchies. Experiments show that our approach outperforms competitive baselines by 10-15% on three diverse datasets of forms and mobile app screen layouts for the tasks of spatial region classification, higher-order group identification, layout hierarchy extraction, reading order detection, and word grouping.

We present a novel method for placing a 3D human animation into a 3D scene while maintaining any human-scene interactions in the animation. We use the notion of computing the most important meshes in the animation for the interaction with the scene, which we call "keyframes." These keyframes allow us to better optimize the placement of the animation into the scene such that interactions in the animations (standing, laying, sitting, etc.) match the affordances ofthe scene (e.g., standing on the floor or laying in a bed). We compare our method, which we call PAAK, with prior approaches, including POSA, PROX ground truth, and a motion synthesis method, and highlight the benefits of our method with a perceptual study. Human raters preferred our PAAK method over the PROX ground truth data 64.6% of the time. Additionally, in direct comparisons, the raters preferred PAAK over competing methods including 61.5% compared to POSA. Our project website is available at https://gamma.umd.edu/paak/.

In developing virtual acoustic environments, it is important to understand the relationship between the computation cost and the perceptual significance of the resultant numerical error. In this article, we propose a quality criterion that evaluates the error significance of path-tracing-based sound propagation simulators. We present an analytical formula that estimates the error signal power spectrum. With this spectrum estimation, we can use a modified Zwicker's loudness model to calculate the relative loudness of the error signal masked by the ideal output. Our experimental results show that the proposed criterion can explain the human perception of simulation error in a variety of cases.

We present PACE, a novel method for modifying motion-captured virtual agents to interact with and move throughout dense, cluttered 3D scenes. Our approach changes a given motion sequence of a virtual agent as needed to adjust to the obstacles and objects in the environment. We first take the individual frames of the motion sequence most important for modeling interactions with the scene and pair them with the relevant scene geometry, obstacles, and semantics such that interactions in the agents motion match the affordances of the scene (e.g., standing on a floor or sitting in a chair). We then optimize the motion of the human by directly altering the high-DOF pose at each frame in the motion to better account for the unique geometric constraints of the scene. Our formulation uses novel loss functions that maintain a realistic flow and natural-looking motion. We compare our method with prior motion generating techniques and highlight the benefits of our method with a perceptual study and physical plausibility metrics. Human raters preferred our method over the prior approaches. Specifically, they preferred our method 57.1% of the time versus the state-of-the-art method using existing motions, and 81.0% of the time versus a state-of-the-art motion synthesis method. Additionally, our method performs significantly higher on established physical plausibility and interaction metrics. Specifically, we outperform competing methods by over 1.2% in terms of the non-collision metric and by over 18% in terms of the contact metric. We have integrated our interactive system with Microsoft HoloLens and demonstrate its benefits in real-world indoor scenes. Our project website is available at https://gamma.mnd.edu/pace/.

Accurately detecting and tracking pedestrians in 3D space is challenging due to large variations in rotations, poses and scales. The situation becomes even worse for dense crowds with severe occlusions. However, existing benchmarks either only provide 2D annotations, or have limited 3D annotations with low-density pedestrian distribution, making it difficult to build a reliable pedestrian perception system especially in crowded scenes. To better evaluate pedestrian perception algorithms in crowded scenarios, we introduce a large-scale multimodal dataset, STCrowd. Specifically, in STCrowd, there are a total of 219 K pedestrian instances and 20 persons per frame on average, with various levels of occlusion. We provide synchronized LiDAR point clouds and camera images as well as their corresponding 3D labels and joint IDs. STCrowd can be used for various tasks, including LiDAR-only, image-only, and sensor-fusion based pedestrian detection and tracking. We provide baselines for most of the tasks. In addition, considering the property of sparse global distribution and density-varying local distribution of pedestrians, we further propose a novel method, Density-aware Hierarchical heatmap Aggregation (DHA), to enhance pedestrian perception in crowded scenes. Extensive experiments show that our new method achieves state-of-the-art performance for pedestrian detection on various datasets. https://github.com/4DVLab/STCrowd.git.

We present 3MASSIV, a multilingual, multimodal and multi-aspect, expertly-annotated dataset of diverse short videos extracted from short-video social media platform - Moj. 3MASSIV comprises of 50k short videos (20 seconds average duration) and 100K unlabeled videos in 11 different languages and captures popular short video trends like pranks, fails, romance, comedy expressed via unique audio-visual formats like self-shot videos, reaction videos, lip-synching, self-sung songs, etc. 3MASSIV presents an opportunity for multimodal and multilingual semantic understanding on these unique videos by annotating them for concepts, affective states, media types, and audio language. We present a thorough analysis of 3MASSIV and highlight the variety and unique aspects of our dataset compared to other contemporary popular datasets with strong baselines. We also show how the social media content in 3MASSIV is dynamic and temporal in nature, which can be used for semantic understanding tasks and cross-lingual analysis.

We present DocInfer -a novel, end-to-end Document-level Natural Language Inference model that builds a hierarchical document graph enriched through inter-sentence relations (topical, entity-based, concept-based), performs paragraph pruning using the novel SubGraph Pooling layer, followed by optimal evidence selection based on REINFORCE algorithm to identify the most important context sentences for a given hypothesis.Our evidence selection mechanism allows it to transcend the input length limitation of modern BERT-like Transformer models while presenting the entire evidence together for inferential reasoning.We show this is an important property needed to reason on large documents where the evidence may be fragmented and located arbitrarily far from each other.Extensive experiments on popular corpora -DocNLI, ContractNLI, and ConTRoL datasets, and our new proposed dataset called CaseHoldNLI on the task of legal judicial reasoning, demonstrate significant performance gains of 8-12% over SOTA methods.Our ablation studies validate the impact of our model.Performance improvement of ∼ 3 -6% on annotation-scarce downstream tasks of fact verification, multiple-choice QA, and contract clause retrieval demonstrates the usefulness of DocInfer beyond primary NLI tasks.

We present a neural-network-based fast diffuse room impulse response generator (FAST-RIR) for generating room impulse responses (RIRs) for a given acoustic environment. Our FAST-RIR takes rectangular room dimensions, listener and speaker positions, and reverberation time (T60) as inputs and generates specular and diffuse reflections for a given acoustic environment. Our FAST-RIR is capable of generating RIRs for a given input T60with an average error of 0.02s. We evaluate our generated RIRs in automatic speech recognition (ASR) applications using Google Speech API, Microsoft Speech API, and Kaldi tools. We show that our proposed FAST-RIR with batch size 1 is 400 times faster than a state-of-the-art diffuse acoustic simulator (DAS) on a CPU and gives similar performance to DAS in ASR experiments. Our FAST-RIR is 12 times faster than an existing GPU-based RIR generator (gpuRIR). We show that our FAST-RIR outperforms gpuRIR by 2.5% in an AMI far-field ASR benchmark.

We present a robust learning algorithm to detect and handle collisions in 3D deforming meshes. We first train a neural network to detect collisions and then use a numerical optimization algorithm to resolve penetrations guided by the network. Our learned collision handler can resolve collisions for unseen, high-dimensional meshes with thousands of vertices. To obtain stable network performance in such large and unseen spaces, we apply active learning by progressively inserting new collision data based on the network inferences. We automatically label these new data using an analytical collision detector and progressively fine-tune our detection networks. We evaluate our method for collision handling of complex, 3D meshes coming from several datasets with different shapes and topologies, including datasets corresponding to dressed and undressed human poses, cloth simulations, and human hand poses acquired using multi-view capture systems.

We present a novel approach for risk-aware planning with human agents in multi-agent traffic scenarios. Our approach takes into account the wide range of human driver behaviors on the road, from aggressive maneuvers like speeding and overtaking, to conservative traits like driving slowly and conforming to the right-most lane. In our approach, we learn a mapping from a data-driven human driver behavior model called the CMetric to a driver's entropic risk preference. We then use the derived risk preference within a game-theoretic risk-sensitive planner to model risk-aware interactions among human drivers and an autonomous vehicle in various traffic scenarios. We demonstrate our method in a merging scenario, where our results show that the final trajectories obtained from the risk-aware planner generate desirable emergent behaviors. Particularly, our planner recognizes aggressive human drivers and yields to them while maintaining a greater distance from them. In a user study, participants were able to distinguish between aggressive and conservative simulated drivers based on trajectories generated from our risk-sensitive planner. We also observe that aggressive human driving results in more frequent lane-changing in the planner. Finally, we compare the performance of our modified risk-aware planner with existing methods and show that modeling human driver behavior leads to safer navigation.

Monocular depth estimation in the wild inherently predicts depth up to an unknown scale. To resolve scale ambiguity issue, we present a learning algorithm that leverages monocular simultaneous localization and mapping (SLAM) with proprioceptive sensors. Such monocular SLAM systems can provide metrically scaled camera poses. Given these metric poses and monocular sequences, we propose a self-supervised learning method for the pre-trained supervised monocular depth networks to enable metrically scaled depth estimation. Our approach is based on a teacher-student formulation which guides our network to predict high-quality depths. We demonstrate that our approach is useful for various applications such as mobile robot navigation and is applicable to diverse environments. Our full system shows improvements over recent self-supervised depth estimation and completion methods on EuRoC, OpenLORIS, and ScanNet datasets.

We present a novel self-supervised algorithm named MotionHint for monocular visual odometry (VO) that takes motion constraints into account. A key aspect of our approach is to use an appropriate motion model that can help existing self-supervised monocular VO (SSM-VO) algorithms to overcome issues related to the local minima within their self-supervised loss functions. The motion model is expressed with a neural network named PPnet. It is trained to coarsely predict the next pose of the camera and the uncertainty of this prediction. Our self-supervised approach combines the original loss and the motion loss, which is the weighted difference between the prediction and the generated ego-motion. Taking two existing SSM-VO systems as our baseline, we evaluate our MotionHint algorithm on the standard KITTI benchmark. Experimental results show that our MotionHint algorithm can be easily applied to existing open-sourced state-of-the-art SSM-VO systems to greatly improve the performance by reducing the resulting ATE by up to 28.73%.

We present a novel method for reliable robot navigation in uneven outdoor terrains. Our approach employs a fully-trained Deep Reinforcement Learning (DRL) network that uses elevation maps of the environment, robot pose, and goal as inputs to compute an attention mask of the environment. The attention mask is used to identify reduced stability regions in the elevation map and is computed using channel and spatial attention modules and a novel reward function. We continuously compute and update a navigation cost-map that encodes the elevation information or the level-of-flatness of the terrain using the attention mask. We then generate locally least-cost waypoints on the cost-map and compute the final dynamically feasible trajectory using another DRL-based method. Our approach guarantees safe, locally least-cost paths and dynamically feasible robot velocities in uneven terrains. We observe an increase of 35.18% in terms of success rate and, a decrease of 26.14% in the cumulative elevation gradient of the robot's trajectory compared to prior navigation methods in high-elevation regions. We evaluate our method on a Husky robot in real-world uneven terrains (∼$4m$of elevation gain) and demonstrate its benefits.

We present a novel reinforcement learning (RL) based task allocation and decentralized navigation algorithm for mobile robots in warehouse environments. Our approach is designed for scenarios in which multiple robots are used to perform various pick up and delivery tasks. We consider the problem of joint decentralized task allocation and navigation and present a two level approach to solve it. At the higher level, we solve the task allocation by formulating it in terms of Markov Decision Processes and choosing the appropriate rewards to minimize the Total Travel Delay (TTD). At the lower level, we use a decentralized navigation scheme based on ORCA that enables each robot to perform these tasks in an independent manner, and avoid collisions with other robots and dynamic obstacles. We combine these lower and upper levels by defining rewards for the higher level as the feedback from the lower level navigation algorithm. We perform extensive evaluation in complex warehouse layouts with large number of agents and highlight the benefits over state-of-the-art algorithms based on myopic pickup distance minimization and regret-based task selection. We observe improvement up to 14% in terms of task completion time and up-to 40% improvement in terms of computing collision-free trajectories for the robots.

We present a decentralized path-planning algorithm for navigating multiple differential-drive robots in dense environments. In contrast to prior decentralized methods, we propose a novel congestion metric-based replanning that couples local and global planning techniques to efficiently navigate in scenarios with multiple corridors. To handle dense scenes with narrow passages, our approach computes the initial path for each agent to its assigned goal using a lattice planner. Based on neighbors' information, each agent performs online replanning using a congestion metric that tends to reduce the collisions and improves the navigation performance. Furthermore, we use the Voronoi cells of each agent to plan the local motion as well as a corridor selection strategy to limit the congestion in narrow passages. We evaluate the performance of our approach in complex scenes with tens of agents and narrow passages. We show that our Coupled Global-Local approach and Replanning (CGLR) improves the performance and efficiency over prior decentralized methods. In addition, our approach results in a higher success rate in terms of collision-free navigation to the goals, showing improvement in the range of 3-70% over prior decentralized solutions in certain scenarios.

We present a centralized algorithm for labeled, disk-shaped Multi-Robot Path Planning (MPP) in a continuous planar workspace with polygonal boundaries. Our method automatically transform the continuous problem into a discrete, graph-based variant termed the pebble motion problem, which can be solved efficiently. To construct the underlying pebble graph, we identify inscribed circles in the workspace via a medial axis transform and organize robots into layers within each inscribed circle. We show that our layered pebble-graph enables collision-free motions, allowing all graph-restricted MPP instances to be feasible. MPP instances with continuous start and goal positions can then be solved via local navigations that route robots from and to graph vertices. We tested our method on several environments with high robot-packing densities (up to 61.6% of the workspace). For environments with narrow passages, such density violates the well-separated assumptions made by state-of-the-art MPP planners, while our method achieves an average success rate of 83%.

We present TerraPN, a novel method that learns the surface properties (traction, bumpiness, deformability, etc.) of complex outdoor terrains directly from robot-terrain interactions through self-supervised learning, and uses it for autonomous robot navigation. Our method uses RGB images of terrain surfaces and the robot's velocities as inputs, and the IMU vibrations and odometry errors experienced by the robot as labels for self-supervision. Our method computes a surface cost map that differentiates smooth, high-traction surfaces (low navigation costs) from bumpy, slippery, deformable surfaces (high navigation costs). We compute the cost map by non-uniformly sampling patches from the input RGB image by detecting boundaries between surfaces resulting in low inference times (47.27% lower) compared to uniform sampling and existing segmentation methods. We present a novel navigation algorithm that accounts for a surface's cost, computes cost-based acceleration limits for the robot, and dynamically feasible, collision-free trajectories. TerraPN's surface cost prediction can be trained in ∼ 25 minutes for five different surfaces, compared to several hours for previous learning-based segmentation methods. In terms of navigation, our method outperforms previous works in terms of success rates (up to 35.84% higher), vibration cost of the trajectories (up to 21.52% lower), and slowing the robot on bumpy, deformable surfaces (up to 46.76% slower) in different scenarios.

Communication between robots and the server is a major problem for cloud robotic systems. In this paper, we address the problem caused by data loss during such communications and propose an efficient buffering algorithm, called AFR, to solve the problem. We model the problem into an optimization problem to maximize the received Quantity of Information (QoI). Our AFR algorithm is formally proved to achieve near-optimal QoI, which has a lower bound that is a constant multiple of the unrealizable optimal QoI. We implement our AFR algorithm in ROS without changing the API for the applications. Our experiments on two cloud robot applications show that our AFR algorithm can efficiently and effectively reduce the impact of data loss. For the remote mapping application, the RMSE caused by data loss can be reduced by about 20%. For the remote tracking application, the probability of tracking failure caused by data loss can be reduced from about 40 %-60 % to under 10%. Meanwhile, our AFR algorithm introduces time overhead of under 10 microseconds.

Content personalization is one of the foundations of today’s digital marketing. Often the same image needs to be adapted for different design schemes for content that is created for different occasions, geographic locations or other aspects of the target population. We present a novel reinforcement learning (RL) based method for automatically stylizing images to complement the design scheme of media, e.g., interactive websites, apps, or posters. Our approach considers attributes related to the design of the media and adapts the style of the input image to match the context. We do so using a preferential reward system in the RL framework that learns a reward function using human feedback. We conducted several user studies to evaluate our approach and demonstrate that we are able to effectively adapt image styles to different design schemes. In user studies, images stylized through our approach were the most preferred variation across a majority of our experiments. Additionally, we also release a dataset consisting of perceptual associations of web context with the associated image style.

Virtual reality (VR) offers the promise of an infinite office and remote collaboration, however, existing interactions in VR do not strongly support one of the most essential tasks for most knowledge workers, reading. This paper presents VRDoc, a set of gaze-based interaction methods designed to improve the reading experience in VR. We introduce three key components: Gaze Select-and-Snap for document selection, Gaze MagGlass for enhanced text legibility, and Gaze Scroll for ease of document traversal. We implemented each of these tools using a commodity VR headset with eye-tracking. In a series of user studies with 13 participants, we show that VRDoc makes VR reading both more efficient (p < 0.01) and less demanding (p < 0.01), and when given a choice, users preferred to use our tools over the current VR reading methods.

We present a novel data-driven algorithm to learn the perceived emotions of individuals based on their walking motion or gaits. Given an RGB video of an individual walking, we extract their walking gait as a sequence of 3D poses. Our goal is to exploit the gait features to learn and model the emotional state of the individual into one of four categorical emotions: happy, sad, angry, or neutral. Our perceived emotion identification approach uses deep features learned using long short-term memory networks (LSTMs) on datasets with labeled emotive gaits. We combine these features with gait-based affective features consisting of posture and movement measures. Our algorithm identifies both the categorical emotions from the gaits and the corresponding values for the dimensional emotion components - valence and arousal. We also introduce and benchmark a dataset called Emotion Walk (EWalk), consisting of videos of gaits of individuals annotated with emotions. We show that our algorithm mapping the combined feature space to the perceived emotional state provides an accuracy of 80.07% on the EWalk dataset, outperforming the current baselines by an absolute 13–24%.

We propose a method to detect individualized highlights for users on given target videos based on their preferred highlight clips marked on previous videos they have watched. Our method explicitly leverages the contents of both the preferred clips and the target videos using pre-trained features for the objects and the human activities. We design a multi-head attention mechanism to adaptively weigh the preferred clips based on their object- and human-activity-based contents, and fuse them using these weights into a single feature representation for each user. We compute similarities between these per-user feature representations and the per-frame features computed from the desired target videos to estimate the user-specific highlight clips from the target videos. We test our method on a large-scale highlight detection dataset containing the annotated highlights of individual users. Compared to current baselines, we observe an absolute improvement of 2-4% in the mean average precision of the detected highlights. We also perform extensive ablation experiments on the number of preferred highlight clips associated with each user as well as on the object- and human-activity-based feature representations to validate that our method is indeed both content-based and user-specific.

Risk prediction and price movement classification are essential tasks in financial markets. Monetary policy calls (MPC) provide important insights into the actions taken by a country's central bank on economic goals related to inflation, employment, prices, and interest rates. Analyzing visual, vocal, and textual cues from MPC calls can help analysts and policymakers evaluate the economic risks and make sound investment decisions. To aid the analysis of MPC calls, we curate the Monopoly dataset, a collection of public conference call videos along with their corresponding audio recordings and text transcripts released by six international banks between 2009 and 2022. Our dataset is the first attempt to explore the benefits of visual cues in addition to audio and textual signals for financial prediction tasks. We introduce MPCNet, a competitive baseline architecture that takes advantage of the cross-modal transformer blocks and modality-specific attention fusion to forecast the financial risk and price movement associated with the MPC calls. Empirical results prove that the task is challenging, with the proposed architecture performing 5-18% better than strong Transformer-based baselines. We release the MPC dataset and benchmark models to motivate future research in this new challenging domain.

We propose a mesh-based neural network (MESH2IR) to generate acoustic impulse responses (IRs) for indoor 3D scenes represented using a mesh. The IRs are used to create a high-quality sound experience in interactive applications and audio processing. Our method can handle input triangular meshes with arbitrary topologies (2K - 3M triangles). We present a novel training technique to train MESH2IR using energy decay relief and highlight its benefits. We also show that training MESH2IR on IRs preprocessed using our proposed technique significantly improves the accuracy of IR generation. We reduce the non-linearity in the mesh space by transforming 3D scene meshes to latent space using a graph convolution network. Our MESH2IR is more than 200 times faster than a geometric acoustic algorithm on a CPU and can generate more than 10,000 IRs per second on an NVIDIA GeForce RTX 2080 Ti GPU for a given furnished indoor 3D scene. The acoustic metrics are used to characterize the acoustic environment. We show that the acoustic metrics of the IRs predicted from our MESH2IR match the ground truth with less than 10% error. We also highlight the benefits of MESH2IR on audio and speech processing applications such as speech dereverberation and speech separation. To the best of our knowledge, ours is the first neural-network-based approach to predict IRs from a given 3D scene mesh in real-time.

Puneet Mathur, Vlad Morariu, Verena Kaynig-Fittkau, Jiuxiang Gu, Franck Dernoncourt, Quan Tran, Ani Nenkova, Dinesh Manocha, Rajiv Jain. Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2022.

We present a novel metric to analyze the similarity between the physical environment and the virtual environment for natural walking in virtual reality. Our approach is general and can be applied to any pair of physical and virtual environments. We use geometric techniques based on conforming constrained Delaunay triangulations and visibility polygons to compute the Environment Navigation Incompatibility (ENI) metric that can be used to measure the complexity of performing simultaneous navigation. We demonstrate applications of ENI for highlighting regions of incompatibility for a pair of environments, guiding the design of the virtual environments to make them more compatible with a fixed physical environment, and evaluating the performance of different redirected walking controllers. We validate the ENI metric using simulations and two user studies. Results of our simulations and user studies show that in the environment pair that our metric identified as more navigable, users were able to walk for longer before colliding with objects in the physical environment. Overall, ENI is the first general metric that can automatically identify regions of high and low compatibility in physical and virtual environments. Our project website is available at https://gamma.umd.edu/eni/.

5G applications have become increasingly popular in recent years as the spread of fifth-generation (5G) network deployment has grown. For vehicular networks, mmWave band signals have been well studied and used for communication and sensing. In this work, we propose a new dynamic ray tracing algorithm that exploits spatial and temporal coherence. We evaluate the performance by comparing the results on typical vehicular communication scenarios with GEMV2, which uses a combination of deterministic and stochastic models, and WinProp, which utilizes the deterministic model for simulations with given environment information. We also compare the performance of our algorithm on complex, urban models and observe a reduction in computation time by 36% compared to GEMV2and by 30% compared to WinProp, while maintaining similar prediction accuracy.

We present a novel architecture for 3D object detection, M3DETR, which combines different point cloud representations (raw, voxels, bird-eye view) with different feature scales based on multi-scale feature pyramids. M3DETR is the first approach that unifies multiple point cloud representations, feature scales, as well as models mutual relationships between point clouds simultaneously using transformers. We perform extensive ablation experiments that highlight the benefits of fusing representation and scale, and modeling the relationships. Our method achieves state-of-the-art performance on the KITTI 3D object detection dataset and Waymo Open Dataset. Results show that M3DETR improves the baseline significantly by 1.48% mAP for all classes on Waymo Open Dataset. In particular, our approach ranks 1ston the well-known KITTI 3D Detection Benchmark for both car and cyclist classes, and ranks 1ston Waymo Open Dataset with single frame point cloud input. Our code is available at: https://github.com/rayguan97/M3DETR.

Abstract We present a novel mesh‐based learning approach (N‐Cloth) for plausible 3D cloth deformation prediction. Our approach is general and can handle cloth or obstacles represented by triangle meshes with arbitrary topologies. We use graph convolution to transform the cloth and object meshes into a latent space to reduce the non‐linearity in the mesh space. Our network can predict the target 3D cloth mesh deformation based on the initial state of the cloth mesh template and the target obstacle mesh. Our approach can handle complex cloth meshes with up to 100 K triangles and scenes with various objects corresponding to SMPL humans, non‐SMPL humans or rigid bodies. In practice, our approach can be used to generate plausible cloth simulation at 30 – 45 fps on an NVIDIA GeForce RTX 3090 GPU. We highlight its benefits over prior learning‐based methods and physically‐based cloth simulators.

With robotics rapidly advancing, more effective human–robot interaction is increasingly needed to realize the full potential of robots for society. While spoken language must be part of the solution, our ability to provide spoken language interaction capabilities is still very limited. In this article, based on the report of an interdisciplinary workshop convened by the National Science Foundation, we identify key scientific and engineering advances needed to enable effective spoken language interaction with robotics. We make 25 recommendations, involving eight general themes: putting human needs first, better modeling the social and interactive aspects of language, improving robustness, creating new methods for rapid adaptation, better integrating speech and language with other communication modalities, giving speech and language components access to rich representations of the robot’s current knowledge and state, making all components operate in real time, and improving research infrastructure and resources. Research and development that prioritizes these topics will, we believe, provide a solid foundation for the creation of speech-capable robots that are easy and effective for humans to work with.

Antagonistic crowd behaviors are often observed in cases of serious conflict. Antagonistic emotions, which is the typical psychological state of agents in different roles (i.e., cops, activists, and civilians) in crowd violence scenes, and the way they spread through contagion in a crowd are important causes of crowd antagonistic behaviors. Moreover, games, which refers to the interaction between opposing groups adopting different strategies to obtain higher benefits and less casualties, determine the level of crowd violence. We present an antagonistic crowd simulation model (ACSEE), which is integrated with antagonistic emotional contagion and evolutionary game theories. Our approach models the antagonistic emotions between agents in different roles using two components: mental emotion and external emotion. We combine enhanced susceptible-infectious-susceptible (SIS) and game approaches to evaluate the role of antagonistic emotional contagion in crowd violence. Our evolutionary game theoretic approach incorporates antagonistic emotional contagion through deterrent force, which is modelled by a mixture of emotional forces and physical forces defeating the opponents. Antagonistic emotional contagion and evolutionary game theories influence each other to determine antagonistic crowd behaviors. We evaluate our approach on real-world scenarios consisting of different kinds of agents. We also compare the simulated crowd behaviors with real-world crowd videos and use our approach to predict the trends of crowd movements in violence incidents. We investigate the impact of various factors (number of agents, emotion, strategy, etc.) on the outcome of crowd violence. We present results from user studies suggesting that our model can simulate antagonistic crowd behaviors similar to those seen in real-world scenarios.

Studies have shown that autonomous vehicles (AVs) behave conservatively in a traffic environment composed of human drivers and do not adapt to local conditions and socio-cultural norms. It is known that socially aware AVs can be designed if there exists a mechanism to understand the behaviors of human drivers. We present an approach that leverages machine learning to predict, the behaviors of human drivers. This is similar to how humans implicitly interpret the behaviors of drivers on the road, by only observing the trajectories of their vehicles. We use graph-theoretic tools to extract driver behavior features from the trajectories and machine learning to obtain a computational mapping between the extracted trajectory of a vehicle in traffic and the driver behaviors. Compared to prior approaches in this domain, we prove that our method is robust, general, and extendable to broad-ranging applications such as autonomous navigation. We evaluate our approach on real-world traffic datasets captured in the U.S., India, China, and Singapore, as well as in simulation.

We present a novel learning-based modal sound synthesis approach that includes a mixed vibration solver for modal analysis and a radiation network for acoustic transfer. Our mixed vibration solver consists of a 3D sparse convolution network and a Locally Optimal Block Preconditioned Conjugate Gradient (LOBPCG) module for iterative optimization. Moreover, we highlight the correlation between a standard numerical vibration solver and our network architecture. Our radiation network predicts the Far-Field Acoustic Transfer maps (FFAT Maps) from the surface vibration of the object. The overall running time of our learning-based approach for most new objects is less than one second on a RTX 3080 Ti GPU while maintaining a high sound quality close to the ground truth solved by standard numerical methods. We also evaluate the numerical and perceptual accuracy of our approach on different objects with various shapes and materials.

We present the design and results of an experiment investigating the occurrence of self-illusion and its contribution to realistic behavior consistent with a virtual role in virtual environments. Self-illusion is a generalized illusion about one's self in cognition, eliciting a sense of being associated with a role in a virtual world, despite sure knowledge that this role is not the actual self in the real world. We validate and measure self-illusion through an experiment where each participant occupies a non-human perspective and plays a non-human role using this role's behavior patterns. 77 participants were enrolled for the user study according to the priori power analysis. In the mixed-design experiment with different levels of manipulations, we asked the participants to play a cat (a non-human role) within an immersive VE and captured their different kinds of responses, finding that the participants with higher self-illusion can connect themselves to the virtual role more easily. Based on statistical analysis of questionnaires and behavior data, there is some evidence that self-illusion can be considered a novel psychological component of presence because it is dissociated from sense of embodiment (SoE), plausibility illusion (Psi), and place illusion (PI). Moreover, self-illusion has the potential to be an effective evaluation metric for user experience in a virtual reality system for certain applications.

We present LCollision, a learning-based method that synthesizes collision-free 3D human poses. At the crux of our approach is a novel deep architecture that simultaneously decodes new human poses from the latent space and predicts colliding body parts. These two components of our architecture are used as the objective function and surrogate hard constraints in a constrained optimization for collision-free human pose generation. A novel aspect of our approach is the use of a bilevel autoencoder that decomposes whole-body collisions into groups of collisions between localized body parts. By solving the constrained optimizations, we show that a significant amount of collision artifacts can be resolved. Furthermore, in a large test set of 2.5 × 10 6 randomized poses from SCAPE, our architecture achieves a collision-prediction accuracy of 94.1% with 80× speedup over exact collision detection algorithms. To the best of our knowledge, LCollision is the first approach that accelerates collision detection and resolves penetrations using a neural network.

We present a novel approach that improves the performance of reverberant speech separation. Our approach is based on an accurate geometric acoustic simulator (GAS) which generates realistic room impulse responses (RIRs) by modeling both specular and diffuse reflections. We also propose two training methods - pre-training and curriculum learning that significantly improve separation quality in the presence of reverberation. We also demonstrate that mixing the synthetic RIRs with a small number of real RIRs during training enhances separation performance. We eval-uate our approach on reverberant mixtures generated from real, recorded data (in several different room configurations) from the VOiCES dataset. Our novel approach (curriculum learning+pre-training+GAS) results in relative improvement of 123% in SI-SDRi over prior techniques based on image source method (ISM).

We present a method for improving the quality of synthetic room impulse responses for far-field speech recognition. We bridge the gap between the fidelity of synthetic room impulse responses (RIRs) and the real room impulse responses using our novel, TS-RIRGAN architecture. Given a synthetic RIR in the form of raw audio, we use TS-RIRGAN to translate it into a real RIR. We also perform real-world sub-band room equalization on the translated synthetic RIR. Our overall approach improves the quality of synthetic RIRs by compensating low-frequency wave effects, similar to those in real RIRs. We evaluate the performance of improved synthetic RIRs on a far-field speech dataset augmented by convolving the LibriSpeech clean speech dataset [1] with RIRs and adding back-ground noise. We show that far-field speech augmented using our improved synthetic RIRs reduces the word error rate by up to 19.9% in Kaldi far-field automatic speech recognition benchmark [2].

We present Affect2MM, a learning method for time- series emotion prediction for multimedia content. Our goal is to automatically capture the varying emotions depicted by characters in real-life human-centric situations and behaviors. We use the ideas from emotion causation theories to computationally model and determine the emotional state evoked in clips of movies. Affect2MM explicitly models the temporal causality using attention-based methods and Granger causality. We use a variety of components like facial features of actors involved, scene understanding, visual aesthetics, action/situation description, and movie script to obtain an affective-rich representation to understand and perceive the scene. We use an LSTM-based learning model for emotion perception. To evaluate our method, we analyze and compare our performance on three datasets, SENDv1, MovieGraphs, and the LIRIS-ACCEDE dataset, and observe an average of 10 – 15% increase in the performance over SOTA methods for all three datasets.

We present a new approach, that we call AdaGTCN, for identifying human reader intent from Electroencephalogram (EEG) and Eye movement (EM) data in order to help differentiate between normal reading and task-oriented reading. Understanding the physiological aspects of the reading process (the cognitive load and the reading intent) can help improve the quality of crowd-sourced annotated data. Our method, Adaptive Graph Temporal Convolution Network (AdaGTCN), uses an Adaptive Graph Learning Layer and Deep Neighborhood Graph Convolution Layer for identifying the reading activities using time-locked EEG sequences recorded during word-level eye-movement fixations. Adaptive Graph Learning Layer dynamically learns the spatial correlations between the EEG electrode signals while the Deep Neighborhood Graph Convolution Layer exploits temporal features from a dense graph neighborhood to establish the state of the art in reading task identification over other contemporary approaches. We compare our approach with several baselines to report an improvement of 6.29% on the ZuCo 2.0 dataset, along with extensive ablation experiments.

We present a domain- and user-preference-agnostic approach to detect highlightable excerpts from human-centric videos. Our method works on the graph-based representation of multiple observable human-centric modalities in the videos, such as poses and faces. We use an autoencoder network equipped with spatial-temporal graph convolutions to detect human activities and interactions based on these modalities. We train our network to map the activity- and interaction-based latent structural representations of the different modalities to per-frame highlight scores based on the representativeness of the frames. We use these scores to compute which frames to highlight and stitch contiguous frames to produce the excerpts. We train our network on the large-scale AVA-Kinetics action dataset and evaluate it on four benchmark video highlight datasets: DSH, TVSum, PHD2, and SumMe. We observe a 4–12% improvement in the mean average precision of matching the human-annotated highlights over state-of-the-art methods in these datasets, without requiring any user-provided preferences or dataset-specific fine-tuning.

We present a novel approach for estimating depth from a monocular camera as it moves through complex and crowded indoor environments, e.g., a department store or a metro station. Our approach predicts absolute scale depth maps over the entire scene consisting of a static background and multiple moving people, by training on dynamic scenes. Since it is difficult to collect dense depth maps from crowded indoor environments, we design our training framework without requiring depths produced from depth sensing devices. Our network leverages RGB images and sparse depth maps generated from traditional 3D reconstruction methods to estimate dense depth maps. We use two constraints to handle depth for non-rigidly moving people without tracking their motion explicitly. We demonstrate that our approach offers consistent improvements over recent depth estimation methods on the NAVERLABS dataset, which includes complex and crowded scenes.

We present a novel approach to robustly detect and perceive vehicles in different camera views as part of a cooperative vehicle-infrastructure system (CVIS). Our formulation is designed for arbitrary camera views and makes no assumptions about intrinsic or extrinsic parameters. First, to deal with multi-view data scarcity, we propose a part-assisted novel view synthesis algorithm for data augmentation. We train a part-based texture inpainting network in a self-supervised manner. Then we render the textured model into the background image with the target 6-DoF pose. Second, to handle various camera parameters, we present a new method that produces dense mappings between image pixels and 3D points to perform robust 2D/3D vehicle parsing. Third, we build the first CVIS dataset for bench-marking, which annotates more than 1540 images (14017 instances) from real-world traffic scenarios. We combine these novel algorithms and datasets to develop a robust approach for 2D/3D vehicle parsing for CVIS. In practice, our approach outperforms SOTA methods on 2D detection, in-stance segmentation, and 6-DoF pose estimation by 3.8%, 4.3%, and 2.9%, respectively.

Automatic speech emotion recognition (SER) is a challenging task that plays a crucial role in natural human-computer interaction. One of the main challenges in SER is data scarcity, i.e., insufficient amounts of carefully labeled data to build and fully explore complex deep learning models for emotion classification. This paper aims to address this challenge using a transfer learning strategy combined with spectrogram augmentation. Specifically, we propose a transfer learning approach that leverages a pre-trained residual network (ResNet) model including a statistics pooling layer from speaker recognition trained using large amounts of speaker-labeled data. The statistics pooling layer enables the model to efficiently process variable-length input, thereby eliminating the need for sequence truncation which is commonly used in SER systems. In addition, we adopt a spectrogram augmentation technique to generate additional training data samples by applying random time-frequency masks to log-mel spectrograms to mitigate overfitting and improve the generalization of emotion recognition models. We evaluate the effectiveness of our proposed approach on the interactive emotional dyadic motion capture (IEMOCAP) dataset. Experimental results indicate that the transfer learning and spectrogram augmentation approaches improve the SER performance, and when combined achieve state-of-the-art results.

We present a novel algorithm for self-supervised monocular depth completion. Our approach is based on training a neural network that requires only sparse depth measurements and corresponding monocular video sequences without dense depth labels. Our self-supervised algorithm is designed for challenging indoor environments with textureless regions, glossy and transparent surfaces, moving people, longer and diverse depth ranges and scenes captured by complex ego-motions. Our novel architecture leverages both deep stacks of sparse convolution blocks to extract sparse depth features and pixel-adaptive convolutions to fuse image and depth features. We compare with existing approaches in NYUv2, KITTI and NAVERLABS indoor datasets, and observe 5 - 34 % improvements in root- means-square error (RMSE) reduction.

An important aspect of dynamic urban coverage is how building collision avoidance is incorporated into the overall coverage mission. We consider a multi-agent urban dynamic coverage problem in which a team of flying agents uses downward facing cameras to observe the street-level environment outside of buildings. Cameras are assumed to be ineffective above a maximum altitude (lower than building height), such that agents must move around or over buildings to complete their mission. The main objective of this paper is to compare three different building avoidance strategies that are compatible with dynamic ergodic methods. To provide context for these results, we also compare our results to three other common coverage methods including: boustrophedon coverage (lawn-mower sweep), Voronoi region based coverage, and a naive grid method. All algorithms are evaluated in simulation with respect to four performance metrics (percent coverage, revisit count, revisit time, and the integral of area viewed over time), across team sizes ranging from 1 to 25 agents, and in five types of urban environments of varying density and height. We find that the relative performance of algorithms changes based on the ratio of team size to search area, as well the height and density characteristics of the urban environment.

We present a novel Deep Reinforcement Learning (DRL) based policy to compute dynamically feasible and spatially aware velocities for a robot navigating among mobile obstacles. Our approach combines the benefits of the Dynamic Window Approach (DWA) in terms of satisfying the robot’s dynamics constraints with state-of-the-art DRL-based navigation methods that can handle moving obstacles and pedestrians well. Our formulation achieves these goals by embedding the environmental obstacles’ motions in a novel low-dimensional observation space. It also uses a novel reward function to positively reinforce velocities that move the robot away from the obstacle’s heading direction leading to significantly lower number of collisions. We evaluate our method in realistic 3-D simulated environments and on a real differential drive robot in challenging dense indoor scenarios with several walking pedestrians. We compare our method with state-of-the-art collision avoidance methods and observe significant improvements in terms of success rate (up to 33% increase), number of dynamics constraint violations (up to 61% decrease), and smoothness. We also conduct ablation studies to highlight the advantages of our observation space formulation, and reward structure.

We present a novel geometric deep learning method to compute the acoustic scattering properties of geometric objects. Our learning algorithm uses a point cloud representation of objects to compute the scattering properties and integrates them with ray tracing for interactive sound propagation in dynamic scenes. We use discrete Laplacian-based surface encoders and approximate the neighborhood of each point using a shared multi-layer perceptron. We show that our formulation is permutation invariant and present a neural network that computes the scattering function using spherical harmonics. Our approach can handle objects with arbitrary topologies and deforming models, and takes less than 1ms per object on a commodity GPU. We have analyzed the accuracy and perform validation on thousands of unseen 3D objects and highlight the benefits over other point-based geometric deep learning methods. To the best of our knowledge, this is the first real-time learning algorithm that can approximate the acoustic scattering properties of arbitrary objects with high accuracy.

We present a Generative Adversarial Network (GAN) based room impulse response generator (IR-GAN) for generating realistic synthetic room impulse responses (RIRs). IR-GAN extracts acoustic parameters from captured real-world RIRs and uses these parameters to generate new synthetic RIRs. We use these generated synthetic RIRs to improve far-field automatic speech recognition in new environments that are different from the ones used in training datasets. In particular, we augment the far-field speech training set by convolving our synthesized RIRs with a clean LibriSpeech dataset. We evaluate the quality of our synthetic RIRs on the real-world LibriSpeech test set created using real-world RIRs from the BUT ReverbDB and AIR datasets. Our IR-GAN reports up to an 8.95% lower error rate than Geometric Acoustic Simulator (GAS) in far-field speech recognition benchmarks. We further improve the performance when we combine our synthetic RIRs with synthetic impulse responses generated using GAS. This combination can reduce the word error rate by up to 14.3% in far-field speech recognition benchmarks.

We present a decentralized collision avoidance method for dense environments based on buffered Voronoi cells (BVC) and reciprocal velocity obstacles (RVO). Our approach is designed for scenarios with a large number of agents in close proximity and provides passive-friendly collision avoidance guarantees. The Voronoi cells are superimposed with RVO cones to compute a suitable direction for each agent, and we use that direction to compute a local collision-free path. Our approach can also satisfy double-integrator dynamics, and we use the properties of the BVC to formulate a simple, decentralized deadlock resolution strategy. We demonstrate the benefits of V-RVO in complex scenarios with tens of agents in close proximity. In practice, V-RVO’s performance is comparable to prior velocity-obstacle methods, and the collision avoidance behavior is significantly less conservative than ORCA.

We present a novel sensor-based learning navigation algorithm to compute a collision-free trajectory for a robot in dense and dynamic environments with moving obstacles or targets. Our approach uses deep reinforcement learning-based expert policy that is trained using a sim2real paradigm. In order to increase the reliability and handle the failure cases of the expert policy, we combine with a policy extraction technique to transform the resulting policy into a decision tree format. We use properties of decision trees to analyze and modify the policy and improve performance of navigation algorithm including smoothness, frequency of oscillation, frequency of immobilization, and obstruction of target. Overall, we are able to modify the policy to design an improved learning algorithm without retraining. We highlight the benefits of our approach in simulated environments and navigating a Clearpath Jackal robot among moving pedestrians. (Videos at this url: https://gamma.umd.edu/researchdirections/xrl/navviper)

The data loss caused by unreliable network seriously impacts the results of remote visual SLAM systems. From our experiment, a loss of less than 1 second of data can cause a visual SLAM algorithm to lose tracking. We present a novel buffering method, ORBBuf, to reduce the impact of data loss on remote visual SLAM systems. We model the buffering problem as an optimization problem by introducing a similarity metric between frames. To solve the buffering problem, we present an efficient greedy algorithm to discard the frames that have the least impact on the quality of SLAM results. We implement our ORBBuf method on ROS, a widely used middleware framework. Through an extensive evaluation on real-world scenarios and tens of gigabytes of datasets, we demonstrate that our ORBBuf method can be applied to different state-estimation algorithms (DSO and VINS-Fusion), different sensor data (both monocular images and stereo images), different scenes (both indoor and outdoor), and different network environments (both WiFi networks and 4G networks). Our experimental results indicate that the network losses indeed affect the SLAM results, and our ORBBuf method can reduce the RMSE up to 50 times comparing with the Drop-Oldest and Random buffering methods.

We present BOhance, an efficient solution for optimizing digital content like images. Our approach enhances the standard and widely-used method for optimizing content, A/B testing, by using Bayesian Optimization. Our work effectively extends A/B testing in the continuous domain where A/B testing cannot efficiently test infinitely many variants. We test our approach on an image enhancement task where we use iterative human feedback on different variants of an image to arrive at the optimal variant. BOhance auto-generates candidate content variants to be tested based on the human feedback on prior variants. We demonstrate with user-studies conducted on Amazon Mechanical Turk that BOhance can be both time and cost-efficient; and a superior alternative to existing solutions. Furthermore, we conduct a Visual Turing Test to obtain human impressions on the optimum variants generated by BOhance. Our experiments show that given a human-enhanced image and an image generated by BOhance, 53% users think that the BOhance image was generated by a human expert.

We present a generative adversarial network to synthesize 3D pose sequences of co-speech upper-body gestures with appropriate affective expressions. Our network consists of two components: a generator to synthesize gestures from a joint embedding space of features encoded from the input speech and the seed poses, and a discriminator to distinguish between the synthesized pose sequences and real 3D pose sequences. We leverage the Mel-frequency cepstral coefficients and the text transcript computed from the input speech in separate encoders in our generator to learn the desired sentiments and the associated affective cues. We design an affective encoder using multi-scale spatial-temporal graph convolutions to transform 3D pose sequences into latent, pose-based affective features. We use our affective encoder in both our generator, where it learns affective features from the seed poses to guide the gesture synthesis, and our discriminator, where it enforces the synthesized gestures to contain the appropriate affective expressions. We perform extensive evaluations on two benchmark datasets for gesture synthesis from the speech, the TED Gesture Dataset and the GENEA Challenge 2020 Dataset. Compared to the best baselines, we improve the mean absolute joint error by 10-33%, the mean acceleration difference by 8-58%, and the Fréchet Gesture Distance by 21-34%. We also conduct a user study and observe that compared to the best current baselines, around 15.28% of participants indicated our synthesized gestures appear more plausible, and around 16.32% of participants felt the gestures had more appropriate affective expressions aligned with the speech.

We present Text2Gestures, a transformer-based learning method to interactively generate emotive full-body gestures for virtual agents aligned with natural language text inputs. Our method generates emotionally expressive gestures by utilizing the relevant biomechanical features for body expressions, also known as affective features. We also consider the intended task corresponding to the text and the target virtual agents' intended gender and handedness in our generation pipeline. We train and evaluate our network on the MPI Emotional Body Expressions Database and observe that our network produces state-of-the-art performance in generating gestures for virtual agents aligned with the text for narration or conversation. Our network can generate these gestures at interactive rates on a commodity GPU. We conduct a web-based user study and observe that around 91% of participants indicated our generated gestures to be at least plausible on a five-point Likert Scale. The emotions perceived by the participants from the gestures are also strongly positively correlated with the corresponding intended emotions, with a minimum Pearson coefficient of 0.77 in the valence dimension.

We present a novel hybrid sound propagation algorithm for interactive applications. Our approach is designed for dynamic scenes and uses a neural network-based learned scattered field representation along with ray tracing to generate specular, diffuse, diffraction, and occlusion effects efficiently. We use geometric deep learning to approximate the acoustic scattering field using spherical harmonics. We use a large 3D dataset for training, and compare its accuracy with the ground truth generated using an accurate wave-based solver. The additional overhead of computing the learned scattered field at runtime is small and we demonstrate its interactive performance by generating plausible sound effects in dynamic scenes with diffraction and occlusion effects. We demonstrate the perceptual benefits of our approach based on an audio-visual user study.

Increasing attention is being given to the modeling and simulation of traffic flow and crowd movement, two phenomena that both deal with interactions between pedestrians and cars in many situations. In particular, crowd simulation is important for understanding mobility and transportation patterns. In this paper, we propose an emotion-based crowd simulation model integrating physical strength consumption. Inspired by the theory of “the devoted actor,” the movements of each individual in our model are determined by modeling the influence of physical strength consumption and the emotion of panic. In particular, human physical strength consumption is computed using a physics-based numerical method. Inspired by the James-Lange theory, panic levels are estimated by means of an enhanced emotional contagion model that leverages the inherent relationship between physical strength consumption and panic. To the best of our knowledge, our model is the first method integrating physical strength consumption into an emotion-based crowd simulation model by exploiting the relationship between physical strength consumption and emotion. We highlight the performance on different scenarios and compare the resulting behaviors with real-world video sequences. Our approach can reliably predict changes in physical strength consumption and panic levels of individuals in an emergency situation.

We present a learning-based approach for generating binaural audio from mono audio using multi-task learning. Our formulation leverages additional information from two related tasks: the binaural audio generation task and the flipped audio classification task. Our learning model extracts spatialization features from the visual and audio input, predicts the left and right audio channels, and judges whether the left and right channels are flipped. First, we extract visual features using ResNet from the video frames. Next, we perform binaural audio generation and flipped audio classification using separate subnetworks based on visual features. Our learning method optimizes the overall loss based on the weighted sum of the losses of the two tasks. We train and evaluate our model on the FAIR-Play dataset and the YouTube-ASMR dataset. We perform quantitative and qualitative evaluations to demonstrate the benefits of our approach over prior techniques.

We present a data-driven algorithm for generating gaits of virtual characters with varying dominance traits. Our formulation utilizes a user study to establish a data-driven dominance mapping between gaits and dominance labels. We use our dominance mapping to generate walking gaits for virtual characters that exhibit a variety of dominance traits while interacting with the user. Furthermore, we extract gait features based on known criteria in visual perception and psychology literature that can be used to identify the dominance levels of any walking gait. We validate our mapping and the perceived dominance traits by a second user study in an immersive virtual environment. Our gait dominance classification algorithm can classify the dominance traits of gaits with ˜73 percent accuracy. We also present an application of our approach that simulates interpersonal relationships between virtual characters. To the best of our knowledge, ours is the first practical approach to classifying gait dominance and generate dominance traits in virtual characters.

Interactive multi-agent simulation algorithms are used to compute the trajectories and behaviors of different entities in virtual reality scenarios. However, current methods involve considerable parameter tweaking to generate plausible behaviors. We introduce a novel approach (Heter-Sim) that combines physics-based simulation methods with data-driven techniques using an optimization-based formulation. Our approach is general and can simulate heterogeneous agents corresponding to human crowds, traffic, vehicles, or combinations of different agents with varying dynamics. We estimate motion states from real-world datasets that include information about position, velocity, and control direction. Our optimization algorithm considers several constraints, including velocity continuity, collision avoidance, attraction, direction control. Other constraints are implemented by introducing a novel energy function to control the motions of heterogeneous agents. To accelerate the computations, we reduce the search space for both collision avoidance and optimal solution computation. Heter-Sim can simulate tens or hundreds of agents at interactive rates and we compare its accuracy with real-world datasets and prior algorithms. We also perform user studies that evaluate the plausible behaviors generated by our algorithm and a user study that evaluates the plausibility of our algorithm via VR.

We present a novel redirected walking controller based on alignment that allows the user to explore large and complex virtual environments, while minimizing the number of collisions with obstacles in the physical environment. Our alignment-based redirection controller, ARC, steers the user such that their proximity to obstacles in the physical environment matches the proximity to obstacles in the virtual environment as closely as possible. To quantify a controller's performance in complex environments, we introduce a new metric, Complexity Ratio (CR), to measure the relative environment complexity and characterize the difference in navigational complexity between the physical and virtual environments. Through extensive simulation-based experiments, we show that ARC significantly outperforms current state-of-the-art controllers in its ability to steer the user on a collision-free path. We also show through quantitative and qualitative measures of performance that our controller is robust in complex environments with many obstacles. Our method is applicable to arbitrary environments and operates without any user input or parameter tweaking, aside from the layout of the environments. We have implemented our algorithm on the Oculus Quest head-mounted display and evaluated its performance in environments with varying complexity. Our project website is available at https://ganuna.umd.edu/arc/.

We present a new approach for redirected walking in static and dynamic scenes that uses techniques from robot motion planning to compute the redirection gains that steer the user on collision-free paths in the physical space. Our first contribution is a mathematical framework for redirected walking using concepts from motion planning and configuration spaces. This framework highlights various geometric and perceptual constraints that tend to make collision-free redirected walking difficult. We use our framework to propose an efficient solution to the redirection problem that uses the notion of visibility polygons to compute the free spaces in the physical environment and the virtual environment. The visibility polygon provides a concise representation of the entire space that is visible, and therefore walkable, to the user from their position within an environment. Using this representation of walkable space, we apply redirected walking to steer the user to regions of the visibility polygon in the physical environment that closely match the region that the user occupies in the visibility polygon in the virtual environment. We show that our algorithm is able to steer the user along paths that result in significantly fewer resets than existing state-of-the-art algorithms in both static and dynamic scenes. Our project website is available at https://ganuna.umd.edu/vis.poly/.

We present a CPU-based real-time cloth animation method for dressing virtual humans of various shapes and poses. Our approach formulates the clothing deformation as a high-dimensional function of body shape parameters and pose parameters. In order to accelerate the computation, our formulation factorizes the clothing deformation into two independent components: the deformation introduced by body pose variation (Clothing Pose Model) and the deformation from body shape variation (Clothing Shape Model). Furthermore, we sample and cluster the poses spanning the entire pose space and use those clusters to efficiently calculate the anchoring points. We also introduce a sensitivity-based distance measurement to both find nearby anchoring points and evaluate their contributions to the final animation. Given a query shape and pose of the virtual agent, we synthesize the resulting clothing deformation by blending the Taylor expansion results of nearby anchoring points. Compared to previous methods, our approach is general and able to add the shape dimension to any clothing pose model. Furthermore, we can animate clothing represented with tens of thousands of vertices at 50+ FPS on a CPU. We also conduct a user evaluation and show that our method can improve a user's perception of dressed virtual agents in an immersive virtual environment (IVE) compared to a realtime linear blend skinning method.

We present a novel classifier network called STEP, to classify perceived human emotion from gaits, based on a Spatial Temporal Graph Convolutional Network (ST-GCN) architecture. Given an RGB video of an individual walking, our formulation implicitly exploits the gait features to classify the perceived emotion of the human into one of four emotions: happy, sad, angry, or neutral. We train STEP on annotated real-world gait videos, augmented with annotated synthetic gaits generated using a novel generative network called STEP-Gen, built on an ST-GCN based Conditional Variational Autoencoder (CVAE). We incorporate a novel push-pull regularization loss in the CVAE formulation of STEP-Gen to generate realistic gaits and improve the classification accuracy of STEP. We also release a novel dataset (E-Gait), which consists of 4,227 human gaits annotated with perceived emotions along with thousands of synthetic gaits. In practice, STEP can learn the affective features and exhibits classification accuracy of 88% on E-Gait, which is 14–30% more accurate over prior methods.

We present M3ER, a learning-based method for emotion recognition from multiple input modalities. Our approach combines cues from multiple co-occurring modalities (such as face, text, and speech) and also is more robust than other methods to sensor noise in any of the individual modalities. M3ER models a novel, data-driven multiplicative fusion method to combine the modalities, which learn to emphasize the more reliable cues and suppress others on a per-sample basis. By introducing a check step which uses Canonical Correlational Analysis to differentiate between ineffective and effective modalities, M3ER is robust to sensor noise. M3ER also generates proxy features in place of the ineffectual modalities. We demonstrate the efficiency of our network through experimentation on two benchmark datasets, IEMOCAP and CMU-MOSEI. We report a mean accuracy of 82.7% on IEMOCAP and 89.0% on CMU-MOSEI, which, collectively, is an improvement of about 5% over prior work.

We propose improving the cross-target and cross-scene generalization of visual navigation through learning an agent that is guided by conceiving the next observations it expects to see. This is achieved by learning a variational Bayesian model, called NeoNav, which generates the next expected observations (NEO) conditioned on the current observations of the agent and the target view. Our generative model is learned through optimizing a variational objective encompassing two key designs. First, the latent distribution is conditioned on current observations and the target view, leading to a model-based, target-driven navigation. Second, the latent space is modeled with a Mixture of Gaussians conditioned on the current observation and the next best action. Our use of mixture-of-posteriors prior effectively alleviates the issue of over-regularized latent space, thus significantly boosting the model generalization for new targets and in novel scenes. Moreover, the NEO generation models the forward dynamics of agent-environment interaction, which improves the quality of approximate inference and hence benefits data efficiency. We have conducted extensive evaluations on both real-world and synthetic benchmarks, and show that our model consistently outperforms the state-of-the-art models in terms of success rate, data efficiency, and generalization.

We present EmotiCon, a learning-based algorithm for context-aware perceived human emotion recognition from videos and images. Motivated by Frege's Context Principle from psychology, our approach combines three interpretations of context for emotion recognition. Our first interpretation is based on using multiple modalities (e.g.faces and gaits) for emotion recognition. For the second interpretation, we gather semantic context from the input image and use a self-attention-based CNN to encode this information. Finally, we use depth maps to model the third interpretation related to socio-dynamic interactions and proximity among agents. We demonstrate the efficiency of our network through experiments on EMOTIC, a benchmark dataset. We report an Average Precision (AP) score of 35.48 across 26 classes, which is an improvement of 7-8 over prior methods. We also introduce a new dataset, GroupWalk, which is a collection of videos captured in multiple real-world settings of people walking. We report an AP of 65.83 across 4 categories on GroupWalk, which is also an improvement over prior methods.

We present a real-time algorithm for emotion-aware navigation of a socially-assistive robot among pedestrians. Our approach estimates time-varying emotional dynamics and behaviors of people from multiple modalities (i.e., their faces and trajectories) using a combination of deep-learning, and affective features from the PAD (Pleasure-Arousal-Dominance) model from psychology. These PAD characteristics are used to predict pedestrian movement with proxemic constraints. We use a multi-channel model to classify pedestrian characteristics into four emotion categories (happy, sad, angry, neutral). In our validation results, we observe an emotion detection accuracy of 85.33parcent. We formulate emotion-based proxemic constraints to perform socially-aware robot navigation in low- to medium density environments. We demonstrate the benefits of our algorithm in simulated environments with tens of pedestrians as well as in a real-world setting with Pepper, a social humanoid robot.

We present an efficient and realistic geometric acoustic simulation approach for generating and augmenting training data in speech-related machine learning tasks. Our physically-based acoustic simulation method is capable of modeling occlusion, specular and diffuse reflections of sound in complicated acoustic environments, whereas the classical image method can only model specular reflections in simple room settings. We show that by using our synthetic training data, the same neural networks gain significant performance improvement on real test sets in far-field speech recognition by 1.58% and keyword spotting by 21%, without fine-tuning using real impulse responses.

We propose a method for generating low-frequency compensated synthetic impulse responses that improve the performance of farfield speech recognition systems trained on artificially augmented datasets. We design linear-phase filters that adapt the simulated impulse responses to equalization distributions corresponding to realworld captured impulse responses. Our filtered synthetic impulse responses are then used to augment clean speech data from LibriSpeech dataset. We evaluate the performance of our method on the real-world LibriSpeech test set. In practice, our low-frequency compensated synthetic dataset can reduce the word-error-rate by up to 8.8% for far-field speech recognition.

We present a novel algorithm (GraphRQI) to identify driver behaviors from road-agent trajectories. Our approach assumes that the road-agents exhibit a range of driving traits, such as aggressive or conservative driving. Moreover, these traits affect the trajectories of nearby road-agents as well as the interactions between road-agents. We represent these inter-agent interactions using unweighted and undirected traffic graphs. Our algorithm classifies the driver behavior using a supervised learning algorithm by reducing the computation to the spectral analysis of the traffic graph. Moreover, we present a novel eigenvalue algorithm to compute the spectrum efficiently. We provide theoretical guarantees for the running time complexity of our eigenvalue algorithm and show that it is faster than previous methods by 2 times. We evaluate the classification accuracy of our approach on traffic videos and autonomous driving datasets corresponding to urban traffic. In practice, GraphRQI achieves an accuracy improvement of up to 25% over prior driver behavior classification algorithms. We also use our classification algorithm to predict the future trajectories of road-agents.

We present a realtime tracking algorithm, Road-Track, to track heterogeneous road-agents in dense traffic videos. Our approach is designed for dense traffic scenarios that consist of different road-agents such as pedestrians, two-wheelers, cars, buses, etc. sharing the road. We use the tracking-by-detection approach where we track a road-agent by matching the appearance or bounding box region in the current frame with the predicted bounding box region propagated from the previous frame. Roadtrack uses a novel motion model called the Simultaneous Collision Avoidance and Interaction (SimCAI) model to predict the motion of road-agents by modeling collision avoidance and interactions between the road-agents for the next frame. We demonstrate the advantage of RoadTrack on a dataset of dense traffic videos and observe an accuracy of 75.8% on this dataset, outperforming prior state-of-the-art tracking algorithms by at least 5.2%. RoadTrack operates in realtime at approximately 30 fps and is at least 4× faster than prior tracking algorithms on standard tracking datasets.

Deep reinforcement learning has great potential to acquire complex, adaptive behaviors for autonomous agents automatically. However, the underlying neural network polices have not been widely deployed in real-world applications, especially in these safety-critical tasks (e.g., autonomous driving). One of the reasons is that the learned policy cannot perform flexible and resilient behaviors as traditional methods to adapt to diverse environments. In this paper, we consider the problem that a mobile robot learns adaptive and resilient behaviors for navigating in unseen uncertain environments while avoiding collisions. We present a novel approach for uncertainty-aware navigation by introducing an uncertainty-aware predictor to model the environmental uncertainty, and we propose a novel uncertainty-aware navigation network to learn resilient behaviors in the prior unknown environments. To train the proposed uncertainty-aware network more stably and efficiently, we present the temperature decay training paradigm, which balances exploration and exploitation during the training process. Our experimental evaluation demonstrates that our approach can learn resilient behaviors in diverse environments and generate adaptive trajectories according to environmental uncertainties.

We present a new method to generate optimal grasps for brittle and fragile objects using a novel stress- minimization (SM) metric. Our approach is designed for objects that are composed of homogeneous isotopic materials. Our SM metric measures the maximal resistible external wrenches that would not result in fractures in the target objects. In this paper, we propose methods to compute our new metric. We also use our SM metric to design optimal grasp planning algorithms. Finally, we compare the performance of our metric and conventional grasp metrics, including Q1,Q∞,QG11,QMSV,QVEW. Our experiments show that our SM metric takes into account the material characteristics and object shapes to indicate the fragile regions, where prior methods may not work well. We also show that the computational cost of our SM metric is on par with prior methods. Finally, we show that grasp planners guided by our metric can lower the probability of breaking target objects.

We present DenseCAvoid, a novel algorithm for navigating a robot through dense crowds and avoiding collisions by anticipating pedestrian behaviors. Our formulation uses visual sensors and a pedestrian trajectory prediction algorithm to track pedestrians in a set of input frames and compute bounding boxes that extrapolate to the pedestrian positions in a future time. Our hybrid approach combines this trajectory prediction with a Deep Reinforcement Learning-based collision avoidance method to train a policy to generate smoother, safer, and more robust trajectories during run-time. We train our policy in realistic 3-D simulations of static and dynamic scenarios with multiple pedestrians. In practice, our hybrid approach generalizes well to unseen, real-world scenarios and can navigate a robot through dense crowds (~1-2 humans per square meter) in indoor scenarios, including narrow corridors and lobbies. As compared to cases where prediction was not used, we observe that our method reduces the occurrence of the robot freezing in a crowd by up to 48%, and performs comparably with respect to trajectory lengths and mean arrival times to goal.

We present a novel high fidelity 3-D simulator that significantly reduces the sim-to-real gap for collision avoidance in dense crowds using Deep Reinforcement Learning (DRL). Our simulator models realistic crowd and pedestrian behaviors, along with friction, sensor noise and delays in the simulated robot model. We also describe a technique to incrementally control the randomness and complexity of training scenarios to achieve better convergence and generalization capabilities. We demonstrate the effectiveness of our simulator by training a policy that fuses data from multiple perception sensors such as a 2-D lidar and a depth camera to detect pedestrians and computes smooth, collision-free velocities. Our novel reward function and multi-sensor formulation results in smooth and unobtrusive navigation. We have evaluated the learned policy on two differential drive robots and evaluate its performance in new dense crowd scenarios, narrow corridors, T and L-junctions, etc. We observe that our algorithm outperforms prior dynamic navigation techniques in terms of metrics such as success rate, trajectory length, mean time to goal, and smoothness.

We present a new measure, CMetric, to classify driver behaviors using centrality functions. Our formulation combines concepts from computational graph theory and social traffic psychology to quantify and classify the behavior of human drivers. CMetric is used to compute the probability of a vehicle executing a driving style, as well as the intensity used to execute the style. Our approach is designed for realtime autonomous driving applications, where the trajectory of each vehicle or road-agent is extracted from a video. We compute a dynamic geometric graph (DGG) based on the positions and proximity of the road-agents and centrality functions corresponding to closeness and degree. These functions are used to compute the CMetric based on style likelihood and style intensity estimates. Our approach is general and makes no assumption about traffic density, heterogeneity, or how driving behaviors change over time. We present an algorithm to compute CMetric and demonstrate its performance on real-world traffic datasets. To test the accuracy of CMetric, we introduce a new evaluation protocol (called "Time Deviation Error") that measures the difference between human prediction and the prediction made by CMetric.

We present ProxEmo, a novel end-to-end emotion prediction algorithm for socially aware robot navigation among pedestrians. Our approach predicts the perceived emotions of a pedestrian from walking gaits, which is then used for emotion-guided navigation taking into account social and proxemic constraints. To classify emotions, we propose a multi-view skeleton graph convolution-based model that works on a commodity camera mounted onto a moving robot. Our emotion recognition is integrated into a mapless navigation scheme and makes no assumptions about the environment of pedestrian motion. It achieves a mean average emotion prediction precision of 82.47% on the Emotion-Gait benchmark dataset. We outperform current state-of-art algorithms for emotion recognition from 3D gaits. We highlight its benefits in terms of navigation in indoor scenes using a Clearpath Jackal robot.

We present a novel algorithm (DeepMNavigate) for global multi-agent navigation in dense scenarios using deep reinforcement learning (DRL). Our approach uses local and global information for each robot from motion information maps. We use a three-layer CNN that takes these maps as input to generate a suitable action to drive each robot to its goal position. Our approach is general, learns an optimal policy using a multi-scenario, multi-state training algorithm, and can directly handle raw sensor measurements for local observations. We demonstrate the performance on dense, complex benchmarks with narrow passages and environments with tens of agents. We highlight the algorithm’s benefits over prior learning methods and geometric decentralized algorithms in complex scenarios.

We present a novel autoregression network to generate virtual agents that convey various emotions through their walking styles or gaits. Given the 3D pose sequences of a gait, our network extracts pertinent movement features and affective features from the gait. We use these features to synthesize subsequent gaits such that the virtual agents can express and transition between emotions represented as combinations of happy, sad, angry, and neutral. We incorporate multiple regularizations in the training of our network to simultaneously enforce plausible movements and noticeable emotions on the virtual agents. We also integrate our approach with an AR environment using a Microsoft HoloLens and can generate emotive gaits at interactive rates to increase the social presence. We evaluate how human observers perceive both the naturalness and the emotions from the generated gaits of the virtual agents in a web-based study. Our results indicate around 89% of the users found the naturalness of the gaits satisfactory on a five-point Likert scale, and the emotions they perceived from the virtual agents are statistically similar to the intended emotions of the virtual agents. We also use our network to augment existing gait datasets with emotive gaits and will release this augmented dataset for future research in emotion prediction and emotive gait synthesis. Our project website is available at https://gamma.umd.edu/gen-emotive-gaits/.

Model sound synthesis is a physically-based sound synthesis method used to generate audio content in games and virtual worlds. We present a novel learning-based impact sound synthesis algorithm called Deep-Modal. Our approach can handle sound synthesis for common arbitrary objects, especially dynamic generated objects, in real-time. We present a new compact strategy to represent the mode data, corresponding to frequency and amplitude, as fixed-length vectors. This is combined with a new network architecture that can convert shape features of 3D objects into mode data. Our network is based on an encoder-decoder architecture with the contact positions of objects and external forces embedded. Our method can synthesize interactive sounds related to objects of various shapes at any contact position, as well as objects of different materials and sizes. The synthesis process only takes ~0.01s on a GTX 1080 Ti GPU. We show the effectiveness of Deep-Modal through extensive evaluation using different metrics, including recall and precision of prediction, sound spectrogram, and a user study.

We present a learning-based method for detecting real and fake deepfake multimedia content. To maximize information for learning, we extract and analyze the similarity between the two audio and visual modalities from within the same video. Additionally, we extract and compare affective cues corresponding to perceived emotion from the two modalities within a video to infer whether the input video is "real" or "fake". We propose a deep learning network, inspired by the Siamese network architecture and the triplet loss. To validate our model, we report the AUC metric on two large-scale deepfake detection datasets, DeepFake-TIMIT Dataset and DFDC. We compare our approach with several SOTA deepfake detection methods and report per-video AUC of 84.4% on the DFDC and 96.6% on the DF-TIMIT datasets, respectively. To the best of our knowledge, ours is the first approach that simultaneously exploits audio and video modalities and also perceived emotions from the two modalities for deepfake detection.

We present a novel random-access depth map compression algorithm (RANDM) for interactive rendering. Our compressed representation provides random access to the depth values and enables real-time parallel decompression on commodity hardware. Our method partitions the depth range captured in a given scene into equal-sized intervals and uses this partition to generate three separate components that exhibit higher coherence. Each of these components is processed independently to generate the compressed stream. Our decompression algorithm is simple and performs prefix-sum computations while also decoding the entropy compressed blocks. We have evaluated the performance on large databases on depth maps and obtain a compression ratio of 20 − 100 × with a root-means-square (RMS) error of 0.05 − 2 in the disparity values of the depth map. The decompression algorithm is fast and takes about 1 microsecond per block on a single thread on an Intel Xeon CPU. To the best of our knowledge, RANDM is the first depth map compression algorithm that provides random access capability for interactive applications.

We present a novel approach for generating plausible verbal interactions between virtual human-like agents and user avatars in shared virtual environments. Sense-Plan-Ask, or SPA, extends prior work in propositional planning and natural language processing to enable agents to plan with uncertain information, and leverage question and answer dialogue with other agents and avatars to obtain the needed information and complete their goals. The agents are additionally able to respond to questions from the avatars and other agents using natural-language enabling real-time multi-agent multi-avatar communication environments.Our algorithm can simulate tens of virtual agents at interactive rates interacting, moving, communicating, planning, and replanning. We find that our algorithm creates a small runtime cost and enables agents to complete their goals more effectively than agents without the ability to leverage natural-language communication. We demonstrate quantitative results on a set of simulated benchmarks and detail the results of a preliminary user-study conducted to evaluate the plausibility of the virtual interactions generated by SPA. Overall, we find that participants prefer SPA to prior techniques in 84% of responses including significant benefits in terms of the plausibility of natural-language interactions and the positive impact of those interactions.

We present a novel parallel algorithm for cloth simulation that exploits multiple GPUs for fast computation and the handling of very high resolution meshes. To accelerate implicit integration, we describe new parallel algorithms for sparse matrix-vector multiplication (SpMV) and for dynamic matrix assembly on a multi-GPU workstation. Our algorithms use a novel work queue generation scheme for a fat-tree GPU interconnect topology. Furthermore, we present a novel collision handling scheme that uses spatial hashing for discrete and continuous collision detection along with a non-linear impact zone solver. Our parallel schemes can distribute the computation and storage overhead among multiple GPUs and enable us to perform almost interactive simulation on complex cloth meshes, which can hardly be handled on a single GPU due to memory limitations. We have evaluated the performance with two multi-GPU workstations (with 4 and 8 GPUs, respectively) on cloth meshes with 0.5 -- 1.65 M triangles. Our approach can reliably handle the collisions and generate vivid wrinkles and folds at 2 -- 5 fps, which is significantly faster than prior cloth simulation systems. We observe almost linear speedups with respect to the number of GPUs.

We present a novel chi-squared progressive photon mapping algorithm (CPPM) that constructs an estimator by controlling the bandwidth to obtain superior image quality. Our estimator has parametric statistical advantages over prior nonparametric methods. First, we show that when a probability density function of the photon distribution is subject to uniform distribution, the radiance estimation is unbiased under certain assumptions. Next, the local photon distribution is evaluated via a chi-squared test to determine whether the photons follow the hypothesized distribution (uniform distribution) or not. If the statistical test deems that the photons inside the bandwidth are uniformly distributed, bandwidth reduction should be suspended. Finally, we present a pipeline with a bandwidth retention and conditional reduction scheme according to the test results. This pipeline not only accumulates sufficient photons for a reliable chi-squared test, but also guarantees that the estimate converges to the correct solution under our assumptions. We evaluate our method on various benchmarks and observe significant improvement in the running time and rendering quality in terms of mean squared error over prior progressive photon mapping methods.

Harnessing the power of modern multi-GPU architectures, we present a massively parallel simulation system based on the Material Point Method (MPM) for simulating physical behaviors of materials undergoing complex topological changes, self-collision, and large deformations. Our system makes three critical contributions. First, we introduce a new particle data structure that promotes coalesced memory access patterns on the GPU and eliminates the need for complex atomic operations on the memory hierarchy when writing particle data to the grid. Second, we propose a kernel fusion approach using a new Grid-to-Particles-to-Grid ( G2P2G ) scheme, which efficiently reduces GPU kernel launches, improves latency, and significantly reduces the amount of global memory needed to store particle data. Finally, we introduce optimized algorithmic designs that allow for efficient sparse grids in a shared memory context, enabling us to best utilize modern multi-GPU computational platforms for hybrid Lagrangian-Eulerian computational patterns. We demonstrate the effectiveness of our method with extensive benchmarks, evaluations, and dynamic simulations with elastoplasticity, granular media, and fluid dynamics. In comparisons against an open-source and heavily optimized CPU-based MPM codebase [Fang et al. 2019] on an elastic sphere colliding scene with particle counts ranging from 5 to 40 million, our GPU MPM achieves over 100x per-time-step speedup on a workstation with an Intel 8086K CPU and a single Quadro P6000 GPU, exposing exciting possibilities for future MPM simulations in computer graphics and computational science. Moreover, compared to the state-of-the-art GPU MPM method [Hu et al. 2019a], we not only achieve 2x acceleration on a single GPU but our kernel fusion strategy and Array-of-Structs-of-Array ( AoSoA ) data structure design also generalizes to multi-GPU systems. Our multi-GPU MPM exhibits near-perfect weak and strong scaling with 4 GPUs, enabling performant and large-scale simulations on a 1024 3 grid with close to 100 million particles with less than 4 minutes per frame on a single 4-GPU workstation and 134 million particles with less than 1 minute per frame on an 8-GPU workstation.

We conduct novel analyses of users' gaze behaviors in dynamic virtual scenes and, based on our analyses, we present a novel CNN-based model called DGaze for gaze prediction in HMD-based applications. We first collect 43 users' eye tracking data in 5 dynamic scenes under free-viewing conditions. Next, we perform statistical analysis of our data and observe that dynamic object positions, head rotation velocities, and salient regions are correlated with users' gaze positions. Based on our analysis, we present a CNN-based model (DGaze) that combines object position sequence, head velocity sequence, and saliency features to predict users' gaze positions. Our model can be applied to predict not only realtime gaze positions but also gaze positions in the near future and can achieve better performance than prior method. In terms of realtime prediction, DGaze achieves a 22.0% improvement over prior method in dynamic scenes and obtains an improvement of 9.5% in static scenes, based on using the angular distance as the evaluation metric. We also propose a variant of our model called DGaze_ET that can be used to predict future gaze positions with higher precision by combining accurate past gaze data gathered using an eye tracker. We further analyze our CNN architecture and verify the effectiveness of each component in our model. We apply DGaze to gaze-contingent rendering and a game, and also present the evaluation results from a user study.

We present a new method to capture the acoustic characteristics of real-world rooms using commodity devices, and use the captured characteristics to generate similar sounding sources with virtual models. Given the captured audio and an approximate geometric model of a real-world room, we present a novel learning-based method to estimate its acoustic material properties. Our approach is based on deep neural networks that estimate the reverberation time and equalization of the room from recorded audio. These estimates are used to compute material properties related to room reverberation using a novel material optimization objective. We use the estimated acoustic material characteristics for audio rendering using interactive geometric sound propagation and highlight the performance on many real-world scenarios. We also perform a user study to evaluate the perceptual similarity between the recorded sounds and our rendered audio.

To safely and efficiently navigate in complex urban traffic, autonomous vehicles must make responsible predictions in relation to surrounding traffic-agents (vehicles, bicycles, pedestrians, etc.). A challenging and critical task is to explore the movement patterns of different traffic-agents and predict their future trajectories accurately to help the autonomous vehicle make reasonable navigation decision. To solve this problem, we propose a long short-term memory-based (LSTM-based) realtime traffic prediction algorithm, TrafficPredict. Our approach uses an instance layer to learn instances’ movements and interactions and has a category layer to learn the similarities of instances belonging to the same type to refine the prediction. In order to evaluate its performance, we collected trajectory datasets in a large city consisting of varying conditions and traffic densities. The dataset includes many challenging scenarios where vehicles, bicycles, and pedestrians move among one another. We evaluate the performance of TrafficPredict on our new dataset and highlight its higher accuracy for trajectory prediction by comparing with prior prediction methods.

We present a novel, real-time algorithm, EVA, for generating virtual agents with various perceived emotions. Our approach is based on using Expressive Features of gaze and gait to convey emotions corresponding to happy, sad, angry, or neutral. We precompute a data-driven mapping between gaits and their perceived emotions. EVA uses this gait emotion association at runtime to generate appropriate walking styles in terms of gaits and gaze. Using the EVA algorithm, we can simulate gaits and gazing behaviors of hundreds of virtual agents in real-time with known emotional characteristics. We have evaluated the benefits in different multi-agent VR simulation environments. Our studies suggest that the use of expressive features corresponding to gait and gaze can considerably increase the sense of presence in scenarios with multiple virtual agents.

We present a new algorithm for predicting the near-term trajectories of road agents in dense traffic videos. Our approach is designed for heterogeneous traffic, where the road agents may correspond to buses, cars, scooters, bi-cycles, or pedestrians. We model the interactions between different road agents using a novel LSTM-CNN hybrid network for trajectory prediction. In particular, we take into account heterogeneous interactions that implicitly account for the varying shapes, dynamics, and behaviors of different road agents. In addition, we model horizon-based interactions which are used to implicitly model the driving behavior of each road agent. We evaluate the performance of our prediction algorithm, TraPHic, on the standard datasets and also introduce a new dense, heterogeneous traffic dataset corresponding to urban Asian videos and agent trajectories. We outperform state-of-the-art methods on dense traffic datasets by 30%.

We present a novel algorithm for point cloud segmentation. Our approach transforms unstructured point clouds into regular voxel grids, and further uses a kernel-based interpolated variational autoencoder (VAE) architecture to encode the local geometry within each voxel. Traditionally, the voxel representation only comprises Boolean occupancy information which fails to capture the sparsely distributed points within voxels in a compact manner. In order to handle sparse distributions of points, we further employ radial basis functions (RBF) to compute a local, continuous representation within each voxel. Our approach results in a good volumetric representation that effectively tackles noisy point cloud datasets and is more robust for learning. Moreover, we further introduce group equivariant CNN to 3D, by defining the convolution operator on a symmetry group acting on Z3 and its isomorphic sets. This improves the expressive capacity without increasing parameters, leading to more robust segmentation results. We highlight the performance on standard benchmarks and show that our approach outperforms state-of-the-art segmentation algorithms on the ShapeNet and S3DIS datasets.

We present a new approach to transfer grasp configurations from prior example objects to novel objects. We assume the novel and example objects have the same topology and similar shapes. We perform 3D segmentation on these objects using geometric and semantic shape characteristics. We compute a grasp space for each part of the example object using active learning. We build bijective contact mapping between these model parts and compute the corresponding grasps for novel objects. Finally, we assemble the individual parts and use local replanning to adjust grasp configurations while maintaining its stability and physical constraints. Our approach is general, can handle all kind of objects represented using mesh or point cloud and a variety of robotic hands.

We present a novel sound localization algorithm for a non-line-of-sight (NLOS) sound source in indoor environments. Our approach exploits the diffraction properties of sound waves as they bend around a barrier or an obstacle in the scene. We combine a ray tracing-based sound propagation algorithm with a Uniform Theory of Diffraction (UTD) model, which simulate bending effects by placing a virtual sound source on a wedge in the environment. We precompute the wedges of a reconstructed mesh of an indoor scene and use them to generate diffraction acoustic rays to localize the 3D position of the source. Our method identifies the convergence region of those generated acoustic rays as the estimated source position based on a particle filter. We have evaluated our algorithm in multiple scenarios consisting of static and dynamic NLOS sound sources. In our tested cases, our approach can localize a source position with an average accuracy error of 0.7m, measured by the L2 distance between estimated and actual source locations in a 7m×7m×3m room. Furthermore, we observe 37% to 130% improvement in accuracy over a state-of-the-art localization method that does not model diffraction effects, especially when a sound source is not visible to the robot.

We present an efficient algorithm for motion planning and controlling a robot system with a high number of degrees-of-freedom (DOF). These systems include high-DOF soft robots and articulated robots interacting with a deformable environment. We present a novel technique to accelerate the evaluations of the forward dynamics function by storing the results of costly computations in a hierarchical adaptive grid. Furthermore, we exploit the underactuated properties of the robot systems and build the grid in a low-dimensional space. Our approach approximates the forward dynamics function with guaranteed error bounds and can be used in optimization-based motion planning and reinforcement-learning-based feed-back control. We highlight the performance on two high-DOF robot systems: a line-actuated elastic robot arm and an underwater swimming robot in water. Compared to prior techniques based on exact dynamics evaluation, we observe one to two orders of magnitude improvement in the performance.

We present an algorithm for combining natural language processing (NLP) and fast robot motion planning to automatically generate robot movements. Our formulation uses a novel concept called Dynamic Constraint Mapping to transform complex, attribute-based natural language instructions into appropriate cost functions and parametric constraints for optimization-based motion planning. We generate a factor graph from natural language instructions called the Dynamic Grounding Graph (DGG), which takes latent parameters into account. The coefficients of this factor graph are learned based on conditional random fields (CRFs) and are used to dynamically generate the constraints for motion planning. We map the cost function directly to the motion parameters of the planner and compute smooth trajectories in dynamic scenes. We highlight the performance of our approach in a simulated environment and via a human interacting with a 7-DOF Fetch robot using intricate language commands including negation, orientation specification, and distance constraints.

We present a Pedestrian Dominance Model (PDM) to identify the dominance characteristics of pedestrians for robot navigation. Through a perception study on a simulated dataset of pedestrians, PDM models the perceived dominance levels of pedestrians with varying motion behaviors corresponding to trajectory, speed, and personal space. At runtime, we use PDM to identify the dominance levels of pedestrians to facilitate socially-aware navigation for the robots. PDM can predict dominance levels from trajectories with ~85% accuracy. Prior studies in psychology literature indicate that when interacting with humans, people are more comfortable around people that exhibit complementary movement behaviors. Our algorithm leverages this by enabling the robots to exhibit complementing responses to pedestrian dominance. We also present an application of PDM for generating dominance-based collision-avoidance behaviors in the navigation of autonomous vehicles among pedestrians. We demonstrate the benefits of our algorithm for robots navigating among tens of pedestrians in simulated environments.

We present a novel learning-based approach to estimate the direction-of-arrival (DOA) of a sound source using a convolutional recurrent neural network (CRNN) trained via regression on synthetic data and Cartesian labels. We also describe an improved method to generate synthetic data to train the neural network using state-of-the-art sound propagation algorithms that model specular as well as diffuse reflections of sound. We compare our model against three other CRNNs trained using different formulations of the same problem: classification on categorical labels, and regression on spherical coordinate labels. In practice, our model achieves up to 43% decrease in angular error over prior methods. The use of diffuse reflection results in 34% and 41% reduction in angular prediction errors on LOCATA and SOFA datasets, respectively, over prior methods based on image-source methods. Our method results in an additional 3% error reduction over prior schemes that use classification based networks, and we use 36% fewer network parameters.

Inter-process communication (IPC) is one of the core functions of modern robotics middleware. We propose an efficient IPC technique called TZC (Towards Zero-Copy). As a core component of TZC, we design a novel algorithm called partial serialization. Our formulation can generate messages that can be divided into two parts. During message transmission, one part is transmitted through a socket and the other part uses shared memory. The part within shared memory is never copied or serialized during its lifetime. We have integrated TZC with ROS and ROS2 and find that TZC can be easily combined with current open-source platforms. By using TZC, the overhead of IPC remains constant when the message size grows. In particular, when the message size is 4MB (less than the size of a full HD image), TZC can reduce the overhead of ROS IPC from tens of milliseconds to hundreds of microseconds and can reduce the overhead of ROS2 IPC from hundreds of milliseconds to less than 1 millisecond. We also demonstrate the benefits of TZC by integrating it with TurtleBot2 to be used in autonomous driving scenarios. We show that by using TZC, the braking distance can be 16% shorter than with ROS.

We present a pedestrian tracking algorithm, DensePeds, that tracks individuals in highly dense crowds (>2 pedestrians per square meter). Our approach is designed for videos captured from front-facing or elevated cameras. We present a new motion model called Front-RVO (FRVO) for predicting pedestrian movements in dense situations using collision avoidance constraints and combine it with state-of-the-art Mask R-CNN to compute sparse feature vectors that reduce the loss of pedestrian tracks (false negatives). We evaluate DensePeds on the standard MOT benchmarks as well as a new dense crowd dataset. In practice, our approach is 4.5 × faster than prior tracking algorithms on the MOT benchmark and we are state-of-the-art in dense crowd videos by over 2.6% on the absolute scale on average.

We present a learning-based method for representing grasp poses of a high-DOF hand using neural networks. Due to redundancy in such high-DOF grippers, there exists a large number of equally effective grasp poses for a given target object, making it difficult for the neural network to find consistent grasp poses. We resolve this ambiguity by generating an augmented dataset that covers many possible grasps for each target object and train our neural networks using a consistency loss function to identify a one-to-one mapping from objects to grasp poses. We further enhance the quality of neural-network-predicted grasp poses using a collision loss function to avoid penetrations. We use an object dataset that combines the BigBIRD Database, the KIT Database, the YCB Database, and the Grasp Dataset to show that our method can generate high-DOF grasp poses with higher accuracy than supervised learning baselines. The quality of the grasp poses is on par with the groundtruth poses in the dataset. In addition, our method is robust and can handle noisy object models such as those constructed from multi-view depth images, allowing our method to be implemented on a 25-DOF Shadow Hand hardware platform.

We present a new hierarchical compression scheme for encoding light field images (LFI) that is suitable for interactive rendering. Our method (RLFC) exploits redundancies in the light field images by constructing a tree structure. The top level (root) of the tree captures the common high-level details across the LFI, and other levels (children) of the tree capture specific low-level details of the LFI. Our decompressing algorithm corresponds to tree traversal operations and gathers the values stored at different levels of the tree. Furthermore, we use bounded integer sequence encoding which provides random access and fast hardware decoding for compressing the blocks of children of the tree. We have evaluated our method for 4D two-plane parameterized light fields. The compression rates vary from 0.08 -- 2.5 bits per pixel (bpp), resulting in compression ratios of around 200:1 to 20:1 for a PSNR quality of 40 to 50 dB. The decompression times for decoding the blocks of LFI are 1 -- 3 microseconds per channel on an NVIDIA GTX-960 and we can render new views with a resolution of 512 X 512 at 200 fps. Our overall scheme is simple to implement and involves only bit manipulations and integer arithmetic operations.1

We introduce a novel, perceptually derived metric (P - Reverb) that relates the just-noticeable difference (JND) of the early sound field (also called early reflections) to the late sound field (known as late reflections or reverberation). Early and late reflections are crucial components of the sound field and provide multiple perceptual cues for auditory displays. We conduct two extensive user evaluations that relate the JNDs of early reflections and late reverberation in terms of the mean-free path of the environment and present a novel P - Reverb metric. Our metric is used to estimate dynamic reverberation characteristics efficiently in terms of important parameters like reverberation time (RT60). We show the numerical accuracy of our P - Reverb metric in estimating RT60. Finally, we use our metric to design an interactive sound propagation algorithm and demonstrate its effectiveness on various benchmarks.

Abstract We present a new motion‐compensated hierarchical compression scheme (HMLFC) for encoding light field images (LFI) that is suitable for interactive rendering. Our method combines two different approaches, motion compensation schemes and hierarchical compression methods, to exploit redundancies in LFI. The motion compensation schemes capture the redundancies in local regions of the LFI efficiently (local coherence) and the hierarchical schemes capture the redundancies present across the entire LFI (global coherence). Our hybrid approach combines the two schemes effectively capturing both local as well as global coherence to improve the overall compression rate. We compute a tree from LFI using a hierarchical scheme and use phase shifted motion compensation techniques at each level of the hierarchy. Our representation provides random access to the pixel values of the light field, which makes it suitable for interactive rendering applications using a small run‐time memory footprint. Our approach is GPU friendly and allows parallel decoding of LF pixel values. We highlight the performance on the two‐plane parameterized light fields and obtain a compression ratio of 30–800× with a PSNR of 40–45 dB. Overall, we observe a ∼2–5× improvement in compression rates using HMLFC over prior light field compression schemes that provide random access capability. In practice, our algorithm can render new views of resolution 512 × 512 on an NVIDIA GTX‐980 at ∼200 fps.

Abstract We present a novel deep learning‐based method for fast encoding of textures into current texture compression formats. Our approach uses state‐of‐the‐art neural network methods to compute the appropriate encoding configurations for fast compression. A key bottleneck in the current encoding algorithms is the search step, and we reduce that computation to a classification problem. We use a trained neural network approximation to quickly compute the encoding configuration for a given texture. We have evaluated our approach for compressing the textures for the widely used adaptive scalable texture compression format and evaluate the performance for different block sizes corresponding to 4 × 4, 6 × 6 and 8 × 8. Overall, our method (TexNN) speeds up the encoding computation up to an order of magnitude compared to prior compression algorithms with very little or no loss in the visual quality.

Active vision is inherently attention-driven: an agent actively selects views to attend in order to rapidly perform a vision task while improving its internal representation of the scene being observed. Inspired by the recent success of attention-based models in 2D vision tasks based on single RGB images, we address multi-view depth-based active object recognition using an attention mechanism, by use of an end-to-end recurrent 3D attentional network. The architecture takes advantage of a recurrent neural network to store and update an internal representation. Our model, trained with 3D shape datasets, is able to iteratively attend the best views targeting an object of interest for recognizing it. To realize 3D view selection, we derive a 3D spatial transformer network. It is differentiable, allowing training with backpropagation, and so achieving much faster convergence than the reinforcement learning employed by most existing attention-based models. Experiments show that our method, with only depth input, achieves state-of-the-art next-best-view performance both in terms of time taken and recognition accuracy.

We present a novel, data-driven eye-head coordination model that can be used for realtime gaze prediction for immersive HMD-based applications without any external hardware or eye tracker. Our model (SGaze) is computed by generating a large dataset that corresponds to different users navigating in virtual worlds with different lighting conditions. We perform statistical analysis on the recorded data and observe a linear correlation between gaze positions and head rotation angular velocities. We also find that there exists a latency between eye movements and head movements. SGaze can work as a software-based realtime gaze predictor and we formulate a time related function between head movement and eye movement and use that for realtime gaze position prediction. We demonstrate the benefits of SGaze for gaze-contingent rendering and evaluate the results with a user study.

We present a real-time algorithm to infer the intention of a user's avatar in a virtual environment shared with multiple human-like agents. Our algorithm applies the Bayesian Theory of Mind approach to make inferences about the avatar's hidden intentions based on the observed proxemics and gaze-based cues. Our approach accounts for the potential irrationality in human behavior, as well as the dynamic nature of an individual's intentions. The inferred intent is used to guide the response of the virtual agent and generate locomotion and gaze-based behaviors. Our overall approach allows the user to actively interact with tens of virtual agents from a first-person perspective in an immersive setting. We systematically evaluate our inference algorithm in controlled multi-agent simulation environments and highlight its ability to reliably and efficiently infer the hidden intent of a user's avatar even under noisy conditions. We quantitatively demonstrate the performance benefits of our approach in terms of reducing false inferences, as compared to a prior method. The results of our user evaluation show that 68.18% of participants reported feeling more comfortable in sharing the virtual environment with agents simulated with our algorithm as compared to a prior inference method, likely as a direct result of significantly fewer false inferences and more plausible responses from the virtual agents.

We present a new approach for improving the friendliness and warmth of a virtual agent in an AR environment by generating appropriate movement characteristics. Our algorithm is based on a novel data-driven friendliness model that is computed using a user-study and psychological characteristics. We use our model to control the movements corresponding to the gaits, gestures, and gazing of friendly virtual agents (FVAs) as they interact with the user's avatar and other agents in the environment. We have integrated FVA agents with an AR environment using with a Microsoft HoloLens. Our algorithm can generate plausible movements at interactive rates to increase the social presence. We also investigate the perception of a user in an AR setting and observe that an FVA has a statistically significant improvement in terms of the perceived friendliness and social presence of a user compared to an agent without the friendliness modeling. We observe an increment of 5.71% in the mean responses to a friendliness measure and an improvement of 4.03% in the mean responses to a social presence measure.

We present a realtime virtual grasping algorithm to model interactions with virtual objects. Our approach is designed for multi-fingered hands and makes no assumptions about the motion of the user's hand or the virtual objects. Given a model of the virtual hand, we use machine learning and particle swarm optimization to automatically pre-compute stable grasp configurations for that object. The learning pre-computation step is accelerated using GPU parallelization. At runtime, we rely on the pre-computed stable grasp configurations, and dynamics/non-penetration constraints along with motion planning techniques to compute plausible looking grasps. In practice, our realtime algorithm can perform virtual grasping operations in less than 20ms for complex virtual objects, including high genus objects with holes. We have integrated our grasping algorithm with Oculus Rift HMD and Leap Motion controller and evaluated its performance for different tasks corresponding to grabbing virtual objects and placing them at arbitrary locations. Our user evaluation suggests that our virtual grasping algorithm can increase the user's realism and participation in these tasks and offers considerable benefits over prior interaction algorithms, such as pinch grasping and raycast picking.

We present a new method for training pedestrian detectors on an unannotated set of images. We produce a mixed reality dataset that is composed of real-world background images and synthetically generated static human-agents. Our approach is general, robust, and makes few assumptions about the unannotated dataset. We automatically extract from the dataset: i) the vanishing point to calibrate the virtual camera, and ii) the pedestrians' scales to generate a Spawn Probability Map, which is a novel concept that guides our algorithm to place the pedestrians at appropriate locations. After putting synthetic human-agents in the unannotated images, we use these augmented images to train a Pedestrian Detector, with the annotations generated along with the synthetic agents. We conducted our experiments using Faster R-CNN by comparing the detection results on the unannotated dataset performed by the detector trained using our approach and detectors trained with other manually labeled datasets. We showed that our approach improves the average precision by 5-13% over these detectors.

Many virtual reality applications let multiple users communicate in a multi-talker environment, recreating the classic cocktail-party effect. While there is a vast body of research focusing on the perception and intelligibility of human speech in real-world scenarios with cocktail party effects, there is little work in accurately modeling and evaluating the effect in virtual environments. Given the goal of evaluating the impact of virtual acoustic simulation on the cocktail party effect, we conducted experiments to establish the signal-to-noise ratio (SNR) thresholds for target-word identification performance. Our evaluation was performed for sentences from the coordinate response measure corpus in presence of multi-talker babble. The thresholds were established under varying sound propagation and spatialization conditions. We used a state-of-the-art geometric acoustic system integrated into the Unity game engine to simulate varying conditions of reverberance (direct sound, direct sound & early reflections, direct sound and early reflections and late reverberation) and spatialization (mono, stereo, and binaural). Our results show that spatialization has the biggest effect on the ability of listeners to discern the target words in multi-talker virtual environments. Reverberance, on the other hand, slightly affects the target word discerning ability negatively.

We present a novel, reflection-aware method for 3D sound localization in indoor environments. Unlike prior approaches, which are mainly based on continuous sound signals from a stationary source, our formulation is designed to localize the position instantaneously from signals within a single frame. We consider direct sound and indirect sound signals that reach the microphones after reflecting off surfaces such as ceilings or walls. We then generate and trace direct and reflected acoustic paths using inverse acoustic ray tracing and utilize these paths with Monte Carlo localization to estimate a 3D sound source position. We have implemented our method on a robot with a cube-shaped microphone array and tested it against different settings with continuous and intermittent sound signals with a stationary or a mobile source. Across different settings, our approach can localize the sound with an average distance error of 0.8 m tested in a room of 7 m by 7 m area with 3 m height, including a mobile and non-line-of-sight sound source. We also reveal that the modeling of indirect rays increases the localization accuracy by 40% compared to only using direct acoustic rays.

The complex physical properties of highly deformable materials such as clothes pose significant challenges for autonomous robotic manipulation systems. We present a novel visual feedback dictionary-based method for manipulating deformable objects towards a desired configuration. Our approach is based on visual servoing and we use an efficient technique to extract key features from the RGB sensor stream in the form of a histogram of deformable model features. These histogram features serve as high-level representations of the state of the deformable material. Next, we collect manipulation data and use a visual feedback dictionary that maps the velocity in the high-dimensional feature space to the velocity of the robotic end-effectors for manipulation. We have evaluated our approach on a set of complex manipulation tasks and human-robot manipulation tasks on different cloth pieces with varying material characteristics.

We present a method for planning the motion of arbitrarily-shaped volumetric deformable bodies or robots through complex environments. Such robots have very high-dimensional configuration spaces and we compute trajectories that satisfy the dynamics constraints using a two-stage learning method. First, we train a multitask controller parameterized using dynamic movement primitives (DMP), which encodes various locomotion or movement skills. Next, we train a neural-network controller to select the DMP task to navigate the robot through environments while avoiding obstacles. By combining the finite element method (FEM), model reduction, and contact invariant optimization (CIO), the DMP controller's parameters can be optimized efficiently using a gradient-based method, while the neural-network's parameters are optimized using Deep Q-Learning (DQL). This two-stage learning algorithm also allows us to reuse the trained DMP controller for different navigation tasks, such as moving through different environmental types and to different goal positions. Our results show that the learned motion planner can navigate swimming and walking deformable robots with thousands of DOFs at realtime.

We present a real-time, data-driven algorithm to enhance the social-invisibility of robots within crowds. Our approach is based on prior psychological research, which reveals that people notice and-importantly-react negatively to groups of social actors when they have high entitativity, moving in a tight group with similar appearances and trajectories. In order to evaluate that behavior, we performed a user study to develop navigational algorithms that minimize entitativity. This study establishes mapping between emotional reactions and multi-robot trajectories and appearances, and further generalizes the finding across various environmental conditions. We demonstrate the applicability of our entitativity modeling for trajectory computation for active surveillance and dynamic intervention in simulated robot-human interaction scenarios. Our approach empirically shows that various levels of entitative robots can be used to both avoid and influence pedestrians while not eliciting strong emotional reactions, giving multi-robot systems socially-invisibility.

We present a novel approach to automatically identify driver behaviors from vehicle trajectories and use them for safe navigation of autonomous vehicles. We propose a novel set of features that can be easily extracted from car trajectories. We derive a data-driven mapping between these features and six driver behaviors using an elaborate web-based user study. We also compute a summarized score indicating a level of awareness that is needed while driving next to other vehicles. We also incorporate our algorithm into a vehicle navigation simulation system and demonstrate its benefits in terms of safer realtime navigation, while driving next to aggressive or dangerous drivers.

We present a new time-integrator for modeling the frictional dynamics of articulated bodies. Our formulation represents the configuration of the articulated body using position variables and then uses those variables to model the friction forces between the articulated body and the environment. Our approach corresponds to a Newton-type optimization scheme that is guaranteed to converge so that it is stable with large timestep sizes. We evaluate the accuracy and stability of our time-integrator by comparing it with a conventional formulations based on the Newton-Euler equation and demonstrate the benefits on standard controller-optimization applications. We achieve 3-5 times speedup over a Newton-Euler-based simulator on a CPU. Our approach can be easily parallelized on a GPU and results in additional 4-15 times performance improvement.

We present a novel acoustic optimization algorithm to synthesize dynamic sound fields in a static scene. Our approach places new active loudspeakers or virtual sources in the scene so that the dynamic sound field in a region satisfies optimization criteria to improve speech and music perception. We use a frequency domain formulation of sound propagation and reduce the computation of dynamic sound field synthesis to solving a linear least squares problem, and do not impose any constraints on the environment or loudspeakers type, or loudspeaker placement. We highlight the performance on complex indoor scenes in terms of speech and music improvements. We evaluate the performance with a user study and highlight the perceptual benefits for virtual reality and multimedia applications.

We present an interactive algorithm to generate plausible movements for human-like agents interacting with other agents or avatars in a virtual environment. Our approach takes into account high-dimensional human motion constraints and bio-mechanical constraints to compute collision-free trajectories for each agent. We present a novel full-body movement constrained-velocity computation algorithm that can easily be combined with many existing motion synthesis techniques. Compared to prior local navigation methods, our formulation reduces artefacts that arise in dense scenarios and close interactions, and results in smoother and plausible locomotive behaviors. We have evaluated the benefits of our new algorithm in single-agent and multi-agent environments. We investigated the perception of a single agent's movements in dense scenarios and observed that our algorithm has a strong positive effect on the perceived quality of the simulation. Our approach also allows the user to interact with the agents from a first-person perspective in immersive settings. We conducted a study to investigate the perception of such avatar-agent interactions, and found that interactions generated using our approach lead to an increase in the user's sense of co-presence.

We present a data-driven algorithm to model and predict the socio-emotional impact of groups on observers. Psychological research finds that highly entitative i.e. cohesive and uniform groups induce threat and unease in observers. Our algorithm models realistic trajectory-level behaviors to classify and map the motion-based entitativity of crowds. This mapping is based on a statistical scheme that dynamically learns pedestrian behavior and computes the resultant entitativity induced emotion through group motion characteristics. We also present a novel interactive multi-agent simulation algorithm to model entitative groups and conduct a VR user study to validate the socio-emotional predictive power of our algorithm. We further show that model-generated high-entitativity groups do induce more negative emotions than low-entitative groups.

Abstract Bounding volume hierarchy (BVH) has been widely adopted as the acceleration structure in broad‐phase collision detection. Previous state‐of‐the‐art BVH‐based collision detection approaches exploited the spatio‐temporal coherence of simulations by maintaining a bounding volume test tree (BVTT) front. A major drawback of these algorithms is that large deformations in the scenes decrease culling efficiency and slow down collision queries. Moreover, for front‐based methods, the inefficient caching on GPU caused by the arbitrary layout of BVH and BVTT front nodes becomes a critical performance issue. We present a fast and robust BVH‐based collision detection scheme on GPU that addresses the above problems by ordering and restructuring BVHs and BVTT fronts. Our techniques are based on the use of histogram sort and an auxiliary structure BVTT front log, through which we analyze the dynamic status of BVTT front and BVH quality. Our approach efficiently handles inter‐ and intra‐object collisions and performs especially well in simulations where there is considerable spatio‐temporal coherence. The benchmark results demonstrate that our approach is significantly faster than the previous BVH‐based method, and also outperforms other state‐of‐the‐art spatial subdivision schemes in terms of speed.

We present a practical and inexpensive method to reconstruct 3D scenes that include transparent and mirror objects. Our work is motivated by the need for automatically generating 3D models of interior scenes, which commonly include glass. These large structures are often invisible to cameras or even to our human visual system. Existing 3D reconstruction methods for transparent objects are usually not applicable in such a room-sized reconstruction setting. Our simple hardware setup augments a regular depth camera (e.g., the Microsoft Kinect camera) with a single ultrasonic sensor, which is able to measure the distance to any object, including transparent surfaces. The key technical challenge is the sparse sampling rate from the acoustic sensor, which only takes one point measurement per frame. To address this challenge, we take advantage of the fact that the large scale glass structures in indoor environments are usually either piece-wise planar or a simple parametric surface. Based on these assumptions, we have developed a novel sensor fusion algorithm that first segments the (hybrid) depth map into different categories such as opaque/transparent/infinity (e.g., too far to measure) and then updates the depth map based on the segmentation outcome. We validated our algorithms with a number of challenging cases, including multiple panes of glass, mirrors, and even a curved glass cabinet.

We present a novel approach to perform probabilistic collision detection between a high-DOF robot and imperfect obstacle representations in dynamic and uncertain environments. Our formulation is designed for high-DOF robot trajectory planning in dynamic scenes, where the uncertainties are modeled using Gaussian distributions. We present an efficient algorithm to compute collision probabilities between the robot and the obstacles. Furthermore, we present a prediction algorithm for obstacle positions that takes into account spatial and temporal uncertainties and uses that for trajectory optimization. We highlight the performance of our trajectory planning algorithm in challenging simulated and real-world environments with robot arms operating next to dynamically moving human obstacles.

We present an incremental collision handling algorithm for GPU-based interactive cloth simulation. Our approach exploits the spatial and temporal coherence between successive iterations of an optimization-based solver for collision response computation. We present an incremental continuous collision detection algorithm that keeps track of deforming vertices and combine it with spatial hashing. We use a non-linear GPU-based impact zone solver to resolve the penetrations. We combine our collision handling algorithm with implicit integration to use large time steps. Our overall algorithm, I-Cloth, can simulate complex cloth deformation with a few hundred thousand vertices at 2 - 8 frames per second on a commodity GPU. We highlight its performance on different benchmarks and observe up to 7 - 10X speedup over prior algorithms.

We present a learning-based method to control a coupled 2D system involving both fluid and rigid bodies. Our approach is used to modify the fluid/rigid simulator's behavior by applying control forces only at the simulation domain boundaries. The rest of the domain, corresponding to the interior, is governed by the Navier-Stokes equation for fluids and Newton-Euler's equation for the rigid bodies. We represent our controller using a general neural-net, which is trained using deep reinforcement learning. Our formulation decomposes a control task into two stages: a precomputation training stage and an online generation stage. We utilize various fluid properties, e.g., the liquid's velocity field or the smoke's density field, to enhance the controller's performance. We set up our evaluation benchmark by letting controller drive fluid jets move on the domain boundary and allowing them to shoot fluids towards a rigid body to accomplish a set of challenging 2D tasks such as keeping a rigid body balanced, playing a two-player ping-pong game, and driving a rigid body to sequentially hit specified points on the wall. In practice, our approach can generate physically plausible animations.

Blue noise sampling has proved useful for many graphics applications, but remains underexplored in high-dimensional spaces due to the difficulty of generating distributions and proving properties about them. We present a blue noise sampling method with good quality and performance across different dimensions. The method, spoke-dart sampling, shoots rays from prior samples and selects samples from these rays. It combines the advantages of two major high-dimensional sampling methods: the locality of advancing front with the dimensionality-reduction of hyperplanes, specifically line sampling. We prove that the output sampling is saturated with high probability, with bounds on distances between pairs of samples and between any domain point and its nearest sample. We demonstrate spoke-dart applications for approximate Delaunay graph construction, global optimization, and robotic motion planning. Both the blue-noise quality of the output distribution and the adaptability of the intermediate processes of our method are useful in these applications.

We present an efficient spacetime optimization method to automatically generate animations for a general volumetric, elastically deformable body. Our approach can model the interactions between the body and the environment and automatically generate active animations. We model the frictional contact forces using contact invariant optimization and the fluid drag forces using a simplified model. To handle complex objects, we use a reduced deformable model and present a novel hybrid optimizer to search for the local minima efficiently. This allows us to use long-horizon motion planning to automatically generate animations such as walking, jumping, swimming, and rolling. We evaluate the approach on different shapes and animations, including deformable body navigation and combining with an open-loop controller for realtime forward simulation.

We present a contact planner for complex legged locomotion tasks: standing up, climbing stairs using a handrail, crossing rubble, and getting out of a car. The need for such a planner was shown at the DARPA Robotics Challenge, where such behaviors could not be demonstrated (except for egress). Current planners suffer from their prohibitive algorithmic complexity because they deploy a tree of robot configurations projected in contact with the environment. We tackle this issue by introducing a reduction property: the reachability condition. This condition defines a geometric approximation of the contact manifold, which is of low dimension, presents a Cartesian topology, and can be efficiently sampled and explored. The hard contact planning problem can then be decomposed into two subproblems: first, we plan a path for the root without considering the whole-body configuration, using a sampling-based algorithm; then, we generate a discrete sequence of whole-body configurations in static equilibrium along this path, using a deterministic contact-selection algorithm. The reduction breaks the algorithm complexity encountered in previous works, resulting in the first interactive implementation of a contact planner (open source). While no contact planner has yet been proposed with theoretical completeness, we empirically show the interest of our framework: in a few seconds, with high success rates, we generate complex contact plans for various scenarios and two robots: HRP-2 and HyQ. These plans are validated in dynamic simulations or on the real HRP-2 robot.

We present a novel method to generate plausible diffraction effects for interactive sound propagation in dynamic scenes. Our approach precomputes a diffraction kernel for each dynamic object in the scene and combines them with interactive ray tracing algorithms at runtime. A diffraction kernel encapsulates the sound interaction behavior of individual objects in the free field and we present a new source placement algorithm to significantly accelerate the precomputation. Our overall propagation algorithm can handle highly-tessellated or smooth objects undergoing rigid motion. We have evaluated our algorithm's performance on different scenarios with multiple moving objects and demonstrate the benefits over prior interactive geometric sound propagation methods. We also performed a user study to evaluate the perceived smoothness of the diffracted field and found that the auditory perception using our approach is comparable to that of a wave-based sound propagation method.

We present a novel algorithm to generate virtual acoustic effects in captured 3D models of real-world scenes for multimodal augmented reality. We leverage recent advances in 3D scene reconstruction in order to automatically compute acoustic material properties. Our technique consists of a two-step procedure that first applies a convolutional neural network (CNN) to estimate the acoustic material properties, including frequency-dependent absorption coefficients, that are used for interactive sound propagation. In the second step, an iterative optimization algorithm is used to adjust the materials determined by the CNN until a virtual acoustic simulation converges to measured acoustic impulse responses. We have applied our algorithm to many reconstructed real-world indoor scenes and evaluated its fidelity for augmented reality applications.

We present a novel approach for collision-free global navigation for continuous-time multi-agent systems with general linear dynamics. Our approach is general and can be used to perform collision-free navigation in 2D and 3D workspaces with narrow passages and crowded regions. As part of pre-computation, we compute multiple bridges in the narrow or tight regions in the workspace using kinodynamic RRT algorithms. Our bridge has certain geometric properties that enable us to calculate a collision-free trajectory for each agent using simple interpolation at runtime. Moreover, we combine interpolated bridge trajectories with local multi-agent navigation algorithms to compute global collision-free paths for each agent. The overall approach combines the performance benefits of coupled multi-agent algorithms with the precomputed trajectories of the bridges to handle challenging scenarios. In practice, our approach can perform global navigation for tens to hundreds of agents on a single CPU core in 2D and 3D workspaces.

We present new algorithms to perform fast probabilistic collision queries between convex as well as non-convex objects. Our approach is applicable to general shapes, where one or more objects are represented using Gaussian probability distributions. We present a fast new algorithm for a pair of convex objects, and extend the approach to non-convex models using hierarchical representations. We highlight the performance of our algorithms with various convex and non-convex shapes on complex synthetic benchmarks and trajectory planning benchmarks for a 7-DOF Fetch robot arm.

We present a real-time algorithm to automatically classify the behavior or personality of a pedestrian based on his or her movements in a crowd video. Our classification criterion is based on Personality Trait theory. We present a statistical scheme that dynamically learns the behavior of every pedestrian and computes its motion model. This model is combined with global crowd characteristics to compute the movement patterns and motion dynamics and use them for crowd prediction. Our learning scheme is general and we highlight its performance in identifying the personality of different pedestrians in low and high density crowd videos. We also evaluate the accuracy by comparing the results with a user study.

We present an interactive and stable multi-contact dynamic simulation algorithm for rigid bodies. Our approach is based on fast frictional dynamics (FFD) [14], which is designed for large sets of non-convex rigid bodies. We use a new friction model that performs velocity-level multi-contact simulation using impulse decomposition. Moreover, we accurately handle friction at each contact point using contact distribution and frictional impulse solvers, which also account for relative motion. We evaluate our algorithm's performance on many complex multi-body benchmarks with thousands of contacts. In practice, our dynamics simulation algorithm takes a few milliseconds per timestep and exhibits more stable behaviors.

We present a real-time algorithm, SocioSense, for socially-aware navigation of a robot amongst pedestrians. Our approach computes time-varying behaviors of each pedestrian using Bayesian learning and Personality Trait theory. These psychological characteristics are used for long-term path prediction and generating proxemic characteristics for each pedestrian. We combine these psychological constraints with social constraints to perform human-aware robot navigation in low- to medium-density crowds. The estimation of time-varying behaviors and pedestrian personalities can improve the performance of long-term path prediction by 21%, as compared to prior interactive path prediction algorithms. We also demonstrate the benefits of our socially-aware navigation in simulated environments with tens of pedestrians.

We present AutonoVi, a novel algorithm for autonomous vehicle navigation that supports dynamic maneuvers and integrates traffic constraints and norms. Our approach is based on optimization-based maneuver planning that supports dynamic lane-changes, swerving, and braking in all traffic scenarios and guides the vehicle to its goal position. We take into account various traffic constraints, including collision avoidance with other vehicles, pedestrians, and cyclists using control velocity obstacles. We use a data-driven approach to model the vehicle dynamics for control and collision avoidance. Furthermore, our trajectory computation algorithm takes into account traffic rules and behaviors, such as stopping at intersections and stoplights, based on an arc-spline representation. We have evaluated our algorithm in a simulated environment and tested its interactive performance in urban and highway driving scenarios with tens of vehicles, pedestrians, and cyclists. These scenarios include jaywalking pedestrians, sudden stops from high speeds, safely passing cyclists, a vehicle suddenly swerving into the roadway, and high-density traffic where the vehicle must change lanes to progress more effectively.

We present PRVO, a probabilistic variant of Reciprocal Velocity Obstacle (RVO) for decentralized multi-robot navigation under uncertainty. PRVO characterizes the space of velocities that would allow each robot to fulfill its share in collision avoidance with a specified probability. PRVO is modeled as chance constraints over the velocity level constraints defined by RVO and takes into account the uncertainty associated with both state estimation as well as the actuation of each robot. Since chance constraints are in general computationally intractable, we propose a series of reformulations which when combined with time scaling based concepts leads to a closed form characterization of solution space of PRVO for a given probability of collision avoidance. We validate our formulation through numerical simulations in which we highlight the advantages of PRVO over the related existing formulations.

We present a novel motion planning algorithm for pouring a liquid body from a source to a target container. Our approach uses a receding-horizon optimization strategy that considers liquid dynamics and various other constraints. To handle liquid dynamics without costly fluid simulations, we use a neural network to infer a set of key liquid-related parameters from the observation of the current liquid configuration. To train the neural network, we generate a dataset of successful pouring examples using stochastic optimization in a problem-specific search space. These parameters are then used in the objective function for trajectory optimization. Our feedback motion planner achieves real-time performance, and we observe a high success rate in our simulated 2D and 3D liquid pouring benchmarks.

We present a new method, Motion Picture Texture Compression (MPTC), to compress a series of video frames into a compressed video-texture format, such that the decoded output can be rendered using commodity texture mapping hardware. Our approach reduces the file size of compressed textures by exploiting redundancies in both the temporal and spatial domains. Furthermore, we ensure that the overall rendering quality of the compressed video-textures is comparable to that of compressed image-textures. At runtime, we render each frame of the video decoded from the MPTC format using texture mapping hardware on GPUs. Overall, MPTC improves the bandwidth from CPU to GPU memory up to 4−6× ona desktop and enables rendering of high-resolution videos (2K or higher) on current Head Mounted Displays (HMDs). We observe 3 − 4× improvement in rendering speeds for rendering high-resolution videos on desktop. Furthermore, we observe 9 − 10× improvement in frame rate on mobile platforms using a series of compressed-image textures for rendering high-resolution videos.

Abstract We present an efficient and accurate algorithm for self‐collision detection in deformable models. Our approach can perform discrete and continuous collision queries on triangulated meshes. We present a simple and linear time algorithm to perform the normal cone test using the unprojected 3D vertices, which reduces to a sequence point‐plane classification tests. Moreover, we present a hierarchical traversal scheme that can significantly reduce the number of normal cone tests and the memory overhead using front‐based normal cone culling. The overall algorithm can reliably detect all (self) collisions in models composed of hundreds of thousands of triangles. We observe considerable performance improvement over prior continuous collision detection algorithms.

We present a new method for editing smoke animations by directly deforming the grid used for simulation. We present a modification to the widely used semi-Lagrangian advection operator and use it to transfer the deformation from the grid to the smoke body. Our modified operator bends the smoke particle streamlines according to the deformation gradient. We demonstrate that the controlled smoke animation preserves the fine-grained vortical velocity components and incompressibility constraints, while conforming to the deformed grid. Moreover, our approach enables interactive 3D smoke animation editing by using a reduced-dimensional subspace. Overall, our method makes it possible to use current mesh editing tools to control the smoke body.

Abstract Current 3D capture and modeling technology can rapidly generate highly photo‐realistic 3D avatars of human subjects. However, while the avatars look like their human counterparts, their movements often do not mimic their own due to existing challenges in accurate motion capture and retargeting. A better understanding of factors that influence the perception of biological motion would be valuable for creating virtual avatars that capture the essence of their human subjects. To investigate these issues, we captured 22 subjects walking in an open space. We then performed a study where participants were asked to identify their own motion in varying visual representations and scenarios. Similarly, participants were asked to identify the motion of familiar individuals. Unlike prior studies that used captured footage with simple “point‐light” displays, we rendered the motion on photo‐realistic 3D virtual avatars of the subject. We found that self‐recognition was significantly higher for virtual avatars than with point‐light representations. Users were more confident of their responses when identifying their motion presented on their virtual avatar. Recognition rates varied considerably between motion types for recognition of others, but not for self‐recognition. Overall, our results are consistent with previous studies that used recorded footage and offer key insights into the perception of motion rendered on virtual avatars.

We present a novel algorithm to control the physically-based animation of smoke. Given a set of keyframe smoke shapes, we compute a dense sequence of control force fields that can drive the smoke shape to match several keyframes at certain time instances. Our approach formulates this control problem as a spacetime optimization constrained by partial differential equations. In order to compute the locally optimal control forces, we alternatively optimize the velocity fields and density fields using an alternating direction method of multiplier (ADMM) optimizer. In order to reduce the high complexity of multiple passes of fluid resimulation during velocity field optimization, we utilize the coherence between consecutive fluid simulation passes. We demonstrate the benefits of our approach by computing accurate solutions on 2D and 3D benchmarks. In practice, we observe up to an order of magnitude improvement over prior optimal control methods.

We present an approach to generate plausible acoustic effects at interactive rates in large dynamic environments containing many sound sources. Our formulation combines listener-based backward ray tracing with sound source clustering and hybrid audio rendering to handle complex scenes. We present a new algorithm for dynamic late reverberation that performs high-order ray tracing from the listener against spherical sound sources. We achieve sublinear scaling with the number of sources by clustering distant sound sources and taking relative visibility into account. We also describe a hybrid convolution-based audio rendering technique that can process hundreds of thousands of sound paths at interactive rates. We demonstrate the performance on many indoor and outdoor scenes with up to 200 sound sources. In practice, our algorithm can compute more than 50 reflection orders at interactive rates on a multicore PC, and we observe a 5x speedup over prior geometric sound propagation algorithms.

We present a rapidly exploring-random-tree-based parallel motion planning algorithm that uses the maximal Poisson-disk sampling scheme. Our approach exploits the free-disk property of the maximal Poisson-disk samples to generate nodes and perform tree expansion. Furthermore, we use an adaptive scheme to generate more samples in challenging regions of the configuration space. The Poisson-disk sampling results in improved parallel performance and we highlight the performance benefits on multicore central processing units as well as manycore graphics processing units on different benchmarks.

No abstract available.

We present a novel real-time algorithm to predict the path of pedestrians in cluttered environments. Our approach makes no assumption about pedestrian motion or crowd density, and is useful for short-term as well as long-term prediction. We interactively learn the characteristics of pedestrian motion and movement patterns from 2D trajectories using Bayesian inference. These include local movement patterns corresponding to the current and preferred velocities and global characteristics such as entry points and movement features. Our approach involves no precomputation and we demonstrate the real-time performance of our prediction algorithm on sparse and noisy trajectory data extracted from dense indoor and outdoor crowd videos. The combination of local and global movement patterns can improve the accuracy of long-term prediction by 12-18% over prior methods in high-density videos.

We present a novel algorithm for real-time collision-free navigation between elliptical agents. Each robot or agent is represented using a tight-fitting 2D ellipse in the plane. We extend the reciprocal velocity obstacle formulation by using conservative linear approximations of ellipses and derive sufficient conditions for collision-free motion based on low-dimensional linear programming. We use precomputed Minkowski Sum approximations for real-time and conservative collision avoidance in large multi-agent environments. Finally, we present efficient techniques to update the orientation to compute collision-free trajectories. Our algorithm can handle thousands of elliptical agents in real-time on a single core and provides significant speedups over prior algorithms for elliptical agents. We compare the runtime performance and behavior with circular agents on different benchmarks.

We present a decentralized algorithm for group-based coherent and reciprocal multi-agent navigation. In addition to generating collision-free trajectories for each agent, our approach is able to simulate macroscopic group movements and proxemic behaviors that result in coherent navigation. Our approach is general, makes no assumptions about the size or shape of the group, and can generate smooth trajectories for the agents. Furthermore, it can dynamically adapt to obstacles or the behavior of other agents. The additional overhead of generating proxemic group behaviors is relatively small and our approach can simulate hundreds of agents in real-time. We highlight its benefits on different benchmarks.

We present an optimization-based motion planning algorithm to compute a smooth, collision-free trajectory for a manipulator used to transfer a liquid from a source to a target container. We take into account fluid dynamics constraints as part of the trajectory computation. In order to avoid the high complexity of exact fluid simulation, we introduce a simplified dynamics model based on physically inspired approximations and system identification. Our optimization approach can incorporate various other constraints such as collision avoidance with obstacles, kinematic and dynamics constraints of the manipulator, and fluid dynamics characteristics. We demonstrate the performance of our planner on different benchmarks corresponding to various obstacles and container shapes. We also evaluate its accuracy by validating the motion plan using an accurate but computationally costly Navier-Stokes fluid simulation.

We present an interactive motion planning algorithm to compute plausible trajectories for high-DOF human-like characters. Given a discrete sequence of contact configurations, we use a three-phase optimization approach to ensure that the resulting trajectory is collision-free, smooth, and satisfies dynamic balancing constraints. Our approach can directly compute dynamically balanced and natural-looking motions at interactive frame rates and is considerably faster than prior methods. We highlight its performance on complex human motion benchmarks corresponding to walking, climbing, crawling, and crouching, where the discrete configurations are generated from a kinematic planner or extracted from motion capture datasets.

We present novel techniques to accelerate the computation of impulse responses for interactive sound rendering. Our formulation is based on geometric acoustic algorithms that use ray tracing to compute the propagation paths from each source to the listener in large, dynamic scenes. In order to accelerate generation of realistic acoustic effects in multi-source scenes, we introduce two novel concepts: the impulse response cache and an adaptive frequency-driven ray tracing algorithm that exploits psychoacoustic characteristics of the impulse response length. As compared to prior approaches, we trace relatively fewer rays while maintaining high simulation fidelity for real-time applications. Furthermore, our approach can handle highly reverberant scenes and high-dynamic-range sources. We demonstrate its application in many scenarios and have observed a 5x speedup in computation time and about two orders of magnitude reduction in memory overhead compared to previous approaches. We also present the results of a preliminary user evaluation of our approach.

We present an adaptive data-driven algorithm for interactive crowd simulation. Our approach combines realistic trajectory behaviors extracted from videos with synthetic multi-agent algorithms to generate plausible simulations. We use statistical techniques to compute the movement patterns and motion dynamics from noisy 2D trajectories extracted from crowd videos. These learned pedestrian dynamic characteristics are used to generate collision-free trajectories of virtual pedestrians in slightly different environments or situations. The overall approach is robust and can generate perceptually realistic crowd movements at interactive rates in dynamic environments. We also present results from preliminary user studies that evaluate the trajectory behaviors generated by our algorithm.

We present an adaptive data-driven algorithm for interactive crowd simulation. Our approach combines realistic trajectory behaviors extracted from videos with synthetic multi-agent algorithms to generate plausible simulations. We use statistical techniques to compute the movement patterns and motion dynamics from noisy 2D trajectories extracted from crowd videos. These learned pedestrian dynamic characteristics are used to generate collision-free trajectories of virtual pedestrians in slightly different environments or situations. The overall approach is robust and can generate perceptually realistic crowd movements at interactive rates in dynamic environments. We also present results from preliminary user studies that evaluate the trajectory behaviors generated by our algorithm.

We present a novel interactive approach, PedVR, to generate plausible behaviors for a large number of virtual humans, and to enable natural interaction between the real user and virtual agents. Our formulation is based on a coupled approach that combines a 2D multi-agent navigation algorithm with 3D human motion synthesis. The coupling can result in plausible movement of virtual agents and can generate gazing behaviors, which can considerably increase the believability. We have integrated our formulation with the DK-2 HMD and demonstrate the benefits of our crowd simulation algorithm over prior decoupled approaches. Our user evaluation suggests that the combination of coupled methods and gazing behavior can considerably increase the behavioral plausibility.

Abstract Recent advances in computer graphics have relied on high‐quality textures in order to generate photorealistic real‐time images. Texture compression standards meet these growing demands for data, but current texture compression schemes use fixed‐rate methods where statically sized blocks of pixels are represented using the same numbers of bits irrespective of their data content. In order to account for the natural variation in detail, we present an alternative format that allows variable bit‐rate texture compression with minimal changes to texturing hardware. Our proposed scheme uses one additional level of indirection to allow the variation of the block size across the same texture. This single change is exploited to both vary the amount of bits allocated to certain parts of the texture and to duplicate redundant texture information across multiple pixels. To minimize hardware changes, the method picks combinations of block sizes and compression methods from existing fixed‐rate standards. With this approach, our method is able to demonstrate energy savings of up to 50%, as well as higher quality compressed textures over current state of the art techniques.

Abstract We present a novel GPU‐based approach to robustly and efficiently simulate high‐resolution and complexly layered cloth. The key component of our formulation is a parallelized matrix assembly algorithm that can quickly build a large and sparse matrix in a compressed format and accurately solve linear systems on GPUs. We also present a fast and integrated solution for parallel collision handling, including collision detection and response computations, which utilizes spatio‐temporal coherence. We combine these algorithms as part of a new cloth simulation pipeline that incorporates contact forces into implicit time integration for collision avoidance. The entire pipeline is implemented on GPUs, and we evaluate its performance on complex benchmarks consisting of 100 – 300K triangles. In practice, our system takes a few seconds to simulate one frame of a complex cloth scene, which represents significant speedups over prior CPU and GPU‐based cloth simulation systems.

As sound propagation algorithms become faster and more accurate, the question arises as to whether the additional efforts to improve fidelity actually offer perceptual benefits over existing techniques. Could environmental sound effects go the way of music, where lower-fidelity compressed versions are actually favored by listeners? Here we address this issue with two acoustic phenomena that are known to have perceptual effects on humans and that, accordingly, might be expected to heighten their experience with simulated environments. We present two studies comparing listeners’ perceptual response to both accurate and approximate algorithms simulating two key acoustic effects: diffraction and reverberation. For each effect, we evaluate whether increased numerical accuracy of a propagation algorithm translates into increased perceptual differentiation in interactive virtual environments. Our results suggest that auditory perception does benefit from the increased accuracy, with subjects showing better perceptual differentiation when experiencing the more accurate rendering method: the diffraction experiment shows a more linearly decaying sound field (with respect to the diffraction angle) for the accurate diffraction method, whereas the reverberation experiment shows that more accurate reverberation, after modest user experience, results in near-logarithmic response to increasing room volume.

Discusses the current state of the journal, reports on current and future areas of exploration and research, and presents new editors.

In this paper, we present a novel method to recognize the types of crowd movement from crowd trajectories using agent-based motion models (AMMs). Our idea is to apply a number of AMMs, referred to as exemplar-AMMs, to describe the crowd movement. Specifically, we propose an optimization framework that filters out the unknown noise in the crowd trajectories and measures their similarity to the exemplar-AMMs to produce a crowd motion feature. We then address our real-world crowd movement recognition problem as a multilabel classification problem. Our experiments show that the proposed feature outperforms the state-of-the-art methods in recognizing both simulated and real-world crowd movements from their trajectories. Finally, we have created a synthetic dataset, SynCrowd, which contains two-dimensional (2D) crowd trajectories in various scenarios, generated by various crowd simulators. This dataset can serve as a training set or benchmark for crowd analysis work.

We introduce Bidirectional Sound Transport (BST) , a new algorithm that simulates sound propagation by bidirectional path tracing using multiple importance sampling. Our approach can handle multiple sources in large virtual environments with complex occlusion, and can produce plausible acoustic effects at an interactive rate on a desktop PC. We introduce a new metric based on the signal-to-noise ratio (SNR) of the energy response and use this metric to evaluate the performance of ray-tracing-based acoustic simulation methods. Our formulation exploits temporal coherence in terms of using the resulting sample distribution of the previous frame to guide the sample distribution of the current one. We show that our sample redistribution algorithm converges and better balances between early and late reflections. We evaluate our approach on different benchmarks and demonstrate significant speedup over prior geometric acoustic algorithms.

Modern GPUs supporting compressed textures allow interactive application developers to save scarce GPU resources such as VRAM and bandwidth. Compressed textures use fixed compression ratios whose lossy representations are significantly poorer quality than traditional image compression formats such as JPEG. We present a new method in the class of supercompressed textures that provides an additional layer of compression to already compressed textures. Our texture representation is designed for endpoint compressed formats such as DXT and PVRTC and decoding on commodity GPUs. We apply our algorithm to commonly used formats by separating their representation into two parts that are processed independently and then entropy encoded. Our method preserves the CPU-GPU bandwidth during the decoding phase and exploits the parallelism of GPUs to provide up to 3X faster decode compared to prior texture supercompression algorithms. Along with the gains in decoding speed, our method maintains both the compression size and quality of current state of the art texture representations.

We present an algorithm to accelerate resolution independent curve rendering on mobile GPUs. Recent trends in graphics hardware have created a plethora of compressed texture formats specific to GPU manufacturers. However, certain implementations of platform independent path rendering require generating grayscale textures on the CPU containing the extent that each pixel is covered by the curve. In this paper, we demonstrate that generating a compressed grayscale texture prior to uploading it to the GPU creates faster rendering times in addition to the memory savings. We implement a real-time compression technique for coverage masks and compare our results against the GPU-based implementation of the highly optimized Skia rendering library. We also analyze the worst case properties of our compression algorithms. We observe up to a 2 × speed improvement over the existing GPU-based methods in addition to up to a 9:1 improvement in GPU memory gains. We demonstrate the performance on multiple mobile platforms.

The physical world consists of spatially varying media, such as the atmosphere and the ocean, in which light and sound propagates along non-linear trajectories. This presents a challenge to existing ray-tracing based methods, which are widely adopted to simulate propagation due to their efficiency and flexibility, but assume linear rays. We present a novel algorithm that traces analytic ray curves computed from local media gradients, and utilizes the closed-form solutions of both the intersections of the ray curves with planar surfaces, and the travel distance. By constructing an adaptive unstructured mesh, our algorithm is able to model general media profiles that vary in three dimensions with complex boundaries consisting of terrains and other scene objects such as buildings. Our analytic ray curve tracer with the adaptive mesh improves the efficiency considerably over prior methods. We highlight the algorithm's application on simulation of visual and sound propagation in outdoor scenes.

Recent research in sound simulation has focused on either sound synthesis or sound propagation, and many standalone algorithms have been developed for each domain. We present a novel technique for coupling sound synthesis with sound propagation to automatically generate realistic aural content for virtual environments. Our approach can generate sounds from rigid-bodies based on the vibration modes and radiation coefficients represented by the single-point multipole expansion. We present a mode-adaptive propagation algorithm that uses a perceptual Hankel function approximation technique to achieve interactive runtime performance. The overall approach allows for high degrees of dynamism - it can support dynamic sources, dynamic listeners, and dynamic directivity simultaneously. We have integrated our system with the Unity game engine and demonstrate the effectiveness of this fully-automatic technique for audio content creation in complex indoor and outdoor scenes. We conducted a preliminary, online user-study to evaluate whether our Hankel function approximation causes any perceptible loss of audio quality. The results indicate that the subjects were unable to distinguish between the audio rendered using the approximate function and audio rendered using the full Hankel function in the Cathedral, Tuscany, and the Game benchmarks.

We present a practical and inexpensive method to reconstruct 3D scenes that include piece-wise planar transparent objects. Our work is motivated by the need for automatically generating 3D models of interior scenes, in which glass structures are common. These large structures are often invisible to cameras or even our human visual system. Existing 3D reconstruction methods for transparent objects are usually not applicable in such a room-size reconstruction setting. Our approach augments a regular depth camera (e.g., the Microsoft Kinect camera) with a single ultrasonic sensor, which is able to measure distance to any objects, including transparent surfaces. We present a novel sensor fusion algorithm that first segments the depth map into different categories such as opaque/transparent/infinity (e.g., too far to measure) and then updates the depth map based on the segmentation outcome. Our current hardware setup can generate only one additional point measurement per frame, yet our fusion algorithm is able to generate satisfactory reconstruction results based on our probabilistic model. We highlight the performance in many challenging indoor benchmarks.

We present a novel, real-time algorithm to extract the trajectory of each pedestrian in moderately dense crowd videos. In order to improve the tracking accuracy, we use a hybrid motion model that combines discrete and continuous flow models. The discrete model is based on microscopic agent formulation and is used for local navigation, interaction, and collision avoidance. The continuum model accounts for macroscopic behaviors, including crowd orientation and flow. We use our hybrid model with particle filters to compute the trajectories at interactive rates. We demonstrate its performance in moderately-dense crowd videos with tens of pedestrians and highlight the improved accuracy on different datasets.

We present a new hybrid approach to computing penetration depth (PD) for general polygonal models. Our approach exploits both local and global approaches to PD computation and can compute error-bounded PD approximations for both deep and shallow penetrations. We use a two-step formulation: the first step corresponds to a global approximation approach that samples the configuration space with bounded error using support vector machines; the second step corresponds to a local optimization that performs a projection operation refining the penetration depth. We have implemented this hybrid algorithm on a standard PC platform and tested its performance with various benchmarks. The experimental results show that our algorithm offers significant benefits over previously developed local-only and global-only methods used to compute the PD.

We present an interactive approach for analyzing crowd videos and generating content for multimedia applications. Our formulation combines online tracking algorithms from computer vision, non-linear pedestrian motion models from computer graphics, and machine learning techniques to automatically compute the trajectory-level pedestrian behaviors for each agent in the video. These learned behaviors are used to detect anomalous behaviors, perform crowd replication, augment crowd videos with virtual agents, and segment the motion of pedestrians. We demonstrate the performance of these tasks using indoor and outdoor crowd video benchmarks consisting of tens of human agents, moreover, our algorithm takes less than a tenth of a second per frame on a multi-core PC. The overall approach can handle dense and heterogeneous crowd behaviors and is useful for realtime crowd scene analysis applications.

We present a fast algorithm for continuous collision detection between deformable models. Our approach performs no precomputation and can handle general triangulated models undergoing topological changes. We present a fast decomposition algorithm that represents the mesh boundary using hierarchical clusters and only needs to perform inter-cluster collision checks. The key idea is to compute such clusters quickly and merge them to generate a dynamic bounding volume hierarchy. The overall approach reduces the overhead of computing the hierarchy and also reduces the number of false positives. We highlight the the algorithm's performance on many complex benchmarks generated from medical simulations and crash analysis. In practice, we observe 1.4 to 5 times speedup over prior CCD algorithms for deformable models in our benchmarks.

We present an algorithm to accelerate resolution independent curve rendering on mobile GPUs. The bottleneck for certain platform independent GPU implementations is in generating grayscale textures on the CPU containing the amount that each pixel is covered by the curve. In this paper, we demonstrate that generating a compressed grayscale texture prior to uploading it to the GPU creates faster rendering times in addition to the memory savings. We implement a real-time compression technique for coverage masks and compare our results against the GPU-based implementation of the highly optimized Skia rendering library. We observe up to a 2X speed improvement over the existing GPU-based methods in addition to up to a 9:1 improvement in GPU memory gains. We demonstrate the performance on multiple mobile platforms.

We present a multi-agent simulation algorithm to compute the trajectories and full-body motion of human-like agents. Our formulation uses a coupled approach that combines 2D collision-free navigation with high-DOF human motion simulation using a behavioral finite state machine. In order to generate plausible pedestrian motion, we use a closed-loop hierarchical planner that satisfies dynamic stability, biomechanical, and kinematic constraints, and is tightly integrated with multi-agent navigation. Furthermore, we use motion capture data to generate natural looking human motion. The overall system is able to generate plausible motion with upper and lower body movements and avoid collisions with other human-like agents. We highlight its performance in indoor and outdoor scenarios with tens of human-like agents.

http://gamma.cs.unc.edu/BSC/ We present a realtime and reliable continuous collision detection (CCD) algorithm between triangulated models that exploits the floating point hardware capability of current CPUs and GPUs. Our formulation is based on Bernstein Sign Classification that takes advantage of the geometry properties of Bernstein basis and Bézier curves to perform Boolean collision queries. We derive tight numerical error bounds on the computations and employ those bounds to design an accurate algorithm using finite-precision arithmetic. Compared with prior floatingpoint CCD algorithms, our approach eliminates all the false negatives and 90–95% of the false positives. We integrated our algorithm (TightCCD) with physically-based simulation system and observe speedups in collision queries of 5–15X compared with prior reliable CCD algorithms. Furthermore, we demonstrate its benefits in terms of improving the performance or robustness of cloth simulation systems.

Simultaneous multi-impact simulation is a challenging problem that frequently arises in physically-based modeling of rigid bodies. There are several physical criteria that should be satisfied for rigid body collision handling, but existing methods generally fail to meet one or more of them. In order to capture the inner process of potential energy variation, which is the physical foundation of collisions in a multi-impact system, we present a novel quadratic contact energy model for rigid body simulation. By constructing quadratic energy functions with respect to the impulses, post-impact reactions of rigid bodies can be computed efficiently. Our model can satisfy the physical criteria and can simulate various natural phenomena including the wave effect. Also, our model can be easily combined with Linear Complementary Problem (LCP) and can provide feasible results with any restitution coefficient. In practice, our model can solve the simultaneous multi-impact problem efficiently and robustly, and we highlight its performance on different benchmarks.

We present a multiple-person tracking algorithm, based on combining particle filters (PFs) and reciprocal velocity obstacle (RVO), an agent-based crowd model that infers collision-free velocities so as to predict a pedestrian's motion. In addition to position and velocity, our tracking algorithm can estimate the internal goals (desired destination or desired velocity) of the tracked pedestrian in an online manner, thus removing the need to specify this information beforehand. Furthermore, we leverage the longer term predictions of RVO by deriving a higher order PF, which aggregates multiple predictions from different prior time steps. This yields a tracker that can recover from short-term occlusions and spurious noise in the appearance model. Experimental results show that our tracking algorithm is suitable for predicting pedestrians' behaviors online without needing scene priors or hand-annotated goal information, and improves tracking in real-world crowded scenes under low frame rates.

We present an interactive wave-based sound propagation system that generates accurate, realistic sound in virtual environments for dynamic (moving) sources and listeners. We propose a novel algorithm to accurately solve the wave equation for dynamic sources and listeners using a combination of precomputation techniques and GPU-based runtime evaluation. Our system can handle large environments typically used in VR applications, compute spatial sound corresponding to listener's motion (including head tracking) and handle both omnidirectional and directional sources, all at interactive rates. As compared to prior wave-based techniques applied to large scenes with moving sources, we observe significant improvement in runtime memory. The overall sound-propagation and rendering system has been integrated with the Half-Life 2 game engine, Oculus-Rift head-mounted display, and the Xbox game controller to enable users to experience high-quality acoustic effects (e.g., amplification, diffraction low-passing, high-order scattering) and spatial audio, based on their interactions in the VR application. We provide the results of preliminary user evaluations, conducted to study the impact of wave-based acoustic effects and spatial audio on users' navigation performance in virtual environments.

We present a hybrid architecture, inspired by asynchronous BVH construction [1], for ray tracing animated scenes. Our hybrid architecture utilizes heterogeneous hardware resources: dedicated ray-tracing hardware for BVH updates and ray traversal and a CPU for BVH reconstruction. We also present a traversal scheme using a primitive's axis-aligned bounding box (PrimAABB). This scheme reduces ray-primitive intersection tests by reusing existing BVH traversal units and the primAABB data for tree updates; it enables the use of shallow trees to reduce tree build times, tree sizes, and bus bandwidth requirements. Furthermore, we present a cache scheme that exploits consecutive memory access by reusing data in an L1 cache block. We perform cycle-accurate simulations to verify our architecture, and the simulation results indicate that the proposed architecture can achieve real-time Whitted ray tracing animated scenes at 1,920 × 1,200 resolution. This result comes from our high-performance hardware architecture and minimized resource requirements for tree updates.

We present a novel, real time algorithm to compute the trajectory of each pedestrian in moderately dense crowd scenes. Our formulation is based on an adaptive particle filtering scheme that uses a multi-agent motion model based on velocity-obstacles, and takes into account local interactions as well as physical and personal constraints of each pedestrian. Our method dynamically changes the number of particles allocated to each pedestrian based on different confidence metrics. Additionally, we use a new high-definition crowd video dataset to evaluate the performance of different pedestrian tracking algorithms. This dataset consists of videos of indoor and outdoor scenes recorded at different locations, each with 30-80 pedestrians. Using this dataset, we highlight the performance benefits of our algorithm over prior techniques. In practice, our algorithm can compute trajectories of tens of pedestrians on a multi-core desktop CPU at interactive rates (27-30 frames per second). To the best of our knowledge, our approach is 4-5 times faster than prior methods that provide similar accuracy.

We present a novel realtime algorithm to compute the trajectory of each pedestrian in a crowded scene. Our formulation is based on an adaptive scheme that uses a combination of deterministic and probabilistic trackers to achieve high accuracy and efficiency simultaneously. Furthermore, we integrate it with a multi-agent motion model and local interaction scheme to accurately compute the trajectory of each pedestrian. We highlight the performance and benefits of our algorithm on well-known datasets with tens of pedestrians.

We present an RRT-based motion planning algorithm that uses the maximal Poisson-disk sampling scheme. Our approach exploits the free-disk property of the maximal Poisson-disk samples to generate nodes and perform tree expansion. Furthermore, we use an adaptive scheme to generate more samples in challenging regions of the configuration space. Our approach can be easily parallelized on multi-core CPUs and many-core GPUs. We highlight the performance of our algorithm on different benchmarks.

Accurate rendering of 3D spatial audio for interactive virtual auditory displays requires the use of personalized head-related transfer functions (HRTFs). We present a new approach to compute personalized HRTFs for any individual using a method that combines state-of-the-art image-based 3D modeling with an efficient numerical simulation pipeline. Our 3D modeling framework enables capture of the listener's head and torso using consumer-grade digital cameras to estimate a high-resolution non-parametric surface representation of the head, including the extended vicinity of the listener's ear. We leverage sparse structure from motion and dense surface reconstruction techniques to generate a 3D mesh. This mesh is used as input to a numeric sound propagation solver, which uses acoustic reciprocity and Kirchhoff surface integral representation to efficiently compute an individual's personalized HRTF. The overall computation takes tens of minutes on multi-core desktop machine. We have used our approach to compute the personalized HRTFs of few individuals, and we present our preliminary evaluation here. To the best of our knowledge, this is the first commodity technique that can be used to compute personalized HRTFs in a lab or home setting.

We present a real-time algorithm for compressing textures based on low frequency signal modulated (LFSM) texture compression. Our formulation is based on intensity dilation and exploits the notion that the most important features of an image are those with high contrast ratios. We present a simple two pass algorithm for propagating the extremal intensity values that contribute to these contrast ratios into the compressed encoding. We use our algorithm to compress PVRTC textures in real-time and compare our performance with prior techniques in terms of speed and quality. http://gamma.cs.unc.edu/FasTC

Understanding and modeling how a crowd behaves in a wide variety of situations is an important problem in many areas. For example, during the planning stages, city, traffic, and evacuation engineers use crowd behavior modeling to predict usage patterns and to do safety analysis of their designs. Several research areas benefit from realistic simulation of crowds such as augmented reality, animation, games, virtual therapy, and virtual training. Not only is a realistic rendering of a virtual environment required for these applications, but also a realistic simulation of virtual humans is essential to providing an immersive experience for the users.

With recent advances in distributed virtual worlds, online users have access to larger and more immersive virtual environments. Sometimes the number of users in virtual worlds is not large enough to make the virtual world realistic. In our paper, we present a crowd simulation algorithm that allows a large number of virtual agents to navigate around the virtual world autonomously by sequentially selecting the goals. Our approach is based on our sequential goal selection model (SGS) which can learn goal-selection patterns from synthetic sequences. We demonstrate our algorithm's simulation results in complex scenarios containing more than 20 goals.

Abstract We present the surface area traversal order (SATO) metric to accelerate shadow ray traversal. Our formulation uses the surface area of each child node to compute the TO. In this metric, we give a traversal priority to the child node with the larger surface area to quickly find occluders. Our algorithm reduces the pre‐processing overhead significantly, and is much faster than other metrics. Overall, the SATO is useful for ray tracing large and complex dynamic scenes (e.g. a few million triangles) with shadows.

Abstract We present a novel framework to evaluate multi‐agent crowd simulation algorithms based on real‐world observations of crowd movements. A key aspect of our approach is to enable fair comparisons by automatically estimating the parameters that enable the simulation algorithms to best fit the given data. We formulate parameter estimation as an optimization problem, and propose a general framework to solve the combinatorial optimization problem for all parameterized crowd simulation algorithms. Our framework supports a variety of metrics to compare reference data and simulation outputs. The reference data may correspond to recorded trajectories, macroscopic parameters, or artist‐driven sketches. We demonstrate the benefits of our framework for example‐based simulation, modeling of cultural variations, artist‐driven crowd animation, and relative comparison of some widely‐used multi‐agent simulation algorithms.

We present novel algorithms for modeling interactive diffuse reflections and higher-order diffraction in large-scale virtual environments. Our formulation is based on ray-based sound propagation and is directly applicable to complex geometric datasets. We use an incremental approach that combines radiosity and path tracing techniques to iteratively compute diffuse reflections. We also present algorithms for wavelength-dependent simplification and visibility graph computation to accelerate higher-order diffraction at runtime. The overall system can generate plausible sound effects at interactive rates in large, dynamic scenes that have multiple sound sources. We highlight the performance in complex indoor and outdoor environments and observe an order of magnitude performance improvement over previous methods.

We present fast algorithms to perform accurate CCD queries between triangulated models. Our formulation uses properties of the Bernstein basis and Bézier curves and reduces the problem to evaluating signs of polynomials. We present a geometrically exact CCD algorithm based on the exact geometric computation paradigm to perform reliable Boolean collision queries. Our algorithm is more than an order of magnitude faster than prior exact algorithms. We evaluate its performance for cloth and FEM simulations on CPUs and GPUs, and highlight the benefits.

We present RayCore, a mobile ray-tracing hardware architecture. RayCore facilitates high-quality rendering effects, such as reflection, refraction, and shadows, on mobile devices by performing real-time Whitted ray tracing. RayCore consists of two major components: ray-tracing units (RTUs) based on a unified traversal and intersection pipeline and a tree-building unit (TBU) for dynamic scenes. The overall RayCore architecture offers considerable benefits in terms of die area, memory access, and power consumption. We have evaluated our architecture based on FPGA and ASIC evaluations and demonstrate its performance on different benchmarks. According to the results, our architecture demonstrates high performance per unit area and unit energy, making it highly suitable for use in mobile devices.

We present an approach to model dynamic, data-driven source and listener directivity for interactive wave-based sound propagation in virtual environments and computer games. Our directional source representation is expressed as a linear combination of elementary spherical harmonic (SH) sources. In the preprocessing stage, we precompute and encode the propagated sound fields due to each SH source. At runtime, we perform the SH decomposition of the varying source directivity interactively and compute the total sound field at the listener position as a weighted sum of precomputed SH sound fields. We propose a novel plane-wave decomposition approach based on higher-order derivatives of the sound field that enables dynamic HRTF-based listener directivity at runtime. We provide a generic framework to incorporate our source and listener directivity in any offline or online frequency-domain wave-based sound propagation algorithm. We have integrated our sound propagation system in Valve's Source game engine and use it to demonstrate realistic acoustic effects such as sound amplification, diffraction low-passing, scattering, localization, externalization, and spatial sound, generated by wave-based propagation of directional sources and listener in complex scenarios. We also present results from our preliminary user study.

We present fast CCD algorithm for general rigid and articulated models based on conservative advancement. We have implemented the CCD algorithm with two different acceleration techniques which can handle rigid models, and have extended one of them to articulated models. The resulting algorithms take a few milliseconds for rigid models with tens of thousands of triangles, and a few milliseconds for articulated models with tens of links. We show that the performance of our algorithms is much faster than existing CCD algorithms for polygon-soup models and it is also comparable to competing CCD algorithms that are limited to manifold models.

Most prior techniques for proximity computations are designed for synthetic models and assume exact geometric representations. However, real robots construct representations of the environment using their sensors, and the generated representations are more cluttered and less precise than synthetic models. Furthermore, this sensor data is updated at high frequency. In this paper, we present new collision- and distance-query algorithms, which can efficiently handle large amounts of point cloud sensor data received at real-time rates. We present two novel techniques to accelerate the computation of broad-phase data structures: 1) we present a progressive technique that incrementally computes a high-quality dynamic AABB tree for fast culling, and 2) we directly use an octree representation of the point cloud data as a proximity data structure. We assign a probability value to each leaf node of the tree, and the algorithm computes the nodes corresponding to high collision probability. In practice, our new approaches can be an order of magnitude faster than previous methods. We demonstrate the performance of the new methods on both synthetic data and on sensor data collected using a Kinect™ for motion planning for a mobile manipulator robot.

We present a novel algorithm to compute collision-free trajectories in dynamic environments. Our approach is general and does not require a priori knowledge about the obstacles or their motion. We use a replanning framework that interleaves optimization-based planning with execution. Furthermore, we describe a parallel formulation that exploits a high number of cores on commodity graphics processors (GPUs) to compute a high-quality path in a given time interval. We derive bounds on how parallelization can improve the responsiveness of the planner and the quality of the trajectory.

We present a new algorithm for encoding low dynamic range images into fixed-rate texture compression formats. Our approach provides orders of magnitude improvements in speed over existing publicly-available compressors, while generating high quality results. The algorithm is applicable to any fixed-rate texture encoding scheme based on Block Truncation Coding and we use it to compress images into the OpenGL BPTC format. The underlying technique uses an axis-aligned bounding box to estimate the proper partitioning of a texel block and performs a generalized cluster fit to compute the endpoint approximation. This approximation can be further refined using simulated annealing. The algorithm is inherently parallel and scales with the number of processor cores. We highlight its performance on low-frequency game textures and the high frequency Kodak Test Image Suite.

Abstract We present a GPU‐based streaming algorithm to perform high‐resolution and accurate cloth simulation. We map all the components of cloth simulation pipeline, including time integration, collision detection, collision response, and velocity updating to GPU‐based kernels and data structures. Our algorithm perform intra‐object and inter‐object collisions, handles contacts and friction, and is able to accurately simulate folds and wrinkles. We describe the streaming pipeline and address many issues in terms of obtaining high throughput on many‐core GPUs. In practice, our algorithm can perform high‐fidelity simulation on a cloth mesh with 2M triangles using 3GB of GPU memory. We highlight the parallel performance of our algorithm on three different generations of GPUs. On a high‐end NVIDIA Tesla K20c, we observe up to two orders of magnitude performance improvement as compared to a single‐threaded CPU‐based algorithm, and about one order of magnitude improvement over a 16‐core CPU‐based parallel implementation.

We present a novel approach for wave-based sound propagation suitable for large, open spaces spanning hundreds of meters, with a small memory footprint. The scene is decomposed into disjoint rigid objects. The free-field acoustic behavior of each object is captured by a compact per-object transfer function relating the amplitudes of a set of incoming equivalent sources to outgoing equivalent sources. Pairwise acoustic interactions between objects are computed analytically to yield compact inter-object transfer functions. The global sound field accounting for all orders of interaction is computed using these transfer functions. The runtime system uses fast summation over the outgoing equivalent source amplitudes for all objects to auralize the sound field for a moving listener in real time. We demonstrate realistic acoustic effects such as diffraction, low-passed sound behind obstructions, focusing, scattering, high-order reflections, and echoes on a variety of scenes.

We present a new method for efficiently approximating the global penetration depth between two rigid objects using machine learning techniques. Our approach consists of two phases: offline learning and performing run-time queries. In the learning phase, we precompute an approximation of the contact space of a pair of intersecting objects from a set of samples in the configuration space. We use active and incremental learning algorithms to accelerate the precomputation and improve the accuracy. During the run-time phase, our algorithm performs a nearest-neighbor query based on translational or rotational distance metrics. The run-time query has a small overhead and computes an approximation to global penetration depth in a few milliseconds. We use our algorithm for collision response computations in Box2D or Bullet game physics engines and complex 3D models and observe more than an order of magnitude improvement over prior PD computation techniques.

We present a novel hybrid approach that couples geometric and numerical acoustic techniques for interactive sound propagation in complex environments. Our formulation is based on a combination of spatial and frequency decomposition of the sound field. We use numerical wave-based techniques to precompute the pressure field in the near-object regions and geometric propagation techniques in the far-field regions to model sound propagation. We present a novel two-way pressure coupling technique at the interface of near-object and far-field regions. At runtime, the impulse response at the listener position is computed at interactive rates based on the stored pressure field and interpolation techniques. Our system is able to simulate high-fidelity acoustic effects such as diffraction, scattering, low-pass filtering behind obstruction, reverberation, and high-order reflections in large, complex indoor and outdoor environments and Half-Life 2 game engine. The pressure computation requires orders of magnitude lower memory than standard wave-based numerical techniques.

We present an efficient algorithm to compute spatially-varying, direction-dependent artificial reverberation and reflection filters in large dynamic scenes for interactive sound propagation in virtual environments and video games. Our approach performs Monte Carlo integration of local visibility and depth functions to compute directionally-varying reverberation effects. The algorithm also uses a dynamically-generated rectangular aural proxy to efficiently model 2-4 orders of early reflections. These two techniques are combined to generate reflection and reverberation filters which vary with the direction of incidence at the listener. This combination leads to better sound source localization and immersion. The overall algorithm is efficient, easy to implement, and can handle moving sound sources, listeners, and dynamic scenes, with minimal storage overhead. We have integrated our approach with the audio rendering pipeline in Valve's Source game engine, and use it to generate realistic directional sound propagation effects in indoor and outdoor scenes in real-time. We demonstrate, through quantitative comparisons as well as evaluations, that our approach leads to enhanced, immersive multi-modal interaction.

We present an efficient algorithm to extract the manifold surface that approximates the boundary of a solid represented by a Binary Space Partition (BSP) tree. Our polygonization algorithm repeatedly performs clipping operations on volumetric cells that correspond to a spatial convex partition and computes the boundary by traversing the connected cells. We use point-based representations along with finite-precision arithmetic to improve the efficiency and generate the B-rep approximation of a BSP solid. The core of our polygonization method is a novel clipping algorithm that uses a set of logical operations to make it resistant to degeneracies resulting from limited precision of floating-point arithmetic. The overall BSP to B-rep conversion algorithm can accurately generate boundaries with sharp and small features, and is faster than prior methods. At the end of this paper, we use this algorithm for a few geometric processing applications including Boolean operations, model repair, and mesh reconstruction.

We present a new Bi-level LSH algorithm to perform approximate k-nearest neighbor search in high dimensional spaces. Our formulation is based on a two-level scheme. In the first level, we use a RP-tree that divides the dataset into sub-groups with bounded aspect ratios and is used to distinguish well-separated clusters. During the second level, we compute a single LSH hash table for each sub-group along with a hierarchical structure based on space-filling curves. Given a query, we first determine the sub-group that it belongs to and perform k-nearest neighbor search within the suitable buckets in the LSH hash table corresponding to the sub-group. Our algorithm also maps well to current GPU architectures and can improve the quality of approximate KNN queries as compared to prior LSH-based algorithms. We highlight its performance on two large, high-dimensional image datasets. Given a runtime budget, Bi-level LSH can provide better accuracy in terms of recall or error ration. Moreover, our formulation reduces the variation in runtime cost or the quality of results.

This paper presents LQG-Obstacles, a new concept that combines linear-quadratic feedback control of mobile robots with guaranteed avoidance of collisions with obstacles. Our approach generalizes the concept of Velocity Obstacles [3] to any robotic system with a linear Gaussian dynamics model. We integrate a Kalman filter for state estimation and an LQR feedback controller into a closed-loop dynamics model of which a higher-level control objective is the “control input”. We then define the LQG-Obstacle as the set of control objectives that result in a collision with high probability. Selecting a control objective outside the LQG-Obstacle then produces collision-free motion. We demonstrate the potential of LQG-Obstacles by safely and smoothly navigating a simulated quadrotor helicopter with complex non-linear dynamics and motion and sensing uncertainty through three-dimensional environments with obstacles and narrow passages.

We present a new collision and proximity library that integrates several techniques for fast and accurate collision checking and proximity computation. Our library is based on hierarchical representations and designed to perform multiple proximity queries on different model representations. The set of queries includes discrete collision detection, continuous collision detection, separation distance computation and penetration depth estimation. The input models may correspond to triangulated rigid or deformable models and articulated models. Moreover, FCL can perform probabilistic collision checking between noisy point clouds that are captured using cameras or LIDAR sensors. The main benefit of FCL lies in the fact that it provides a unified interface that can be used by various applications. Furthermore, its flexible architecture makes it easier to implement new algorithms within this framework. The runtime performance of the library is comparable to state of the art collision and proximity algorithms. We demonstrate its performance on synthetic datasets as well as motion planning and grasping computations performed using a two-armed mobile manipulation robot.

In this paper we introduce a new bounding box method for footstep planning for humanoid robots. Similar to the classic bounding box method (which uses a single rectangular box to encompass the robot) it is computationally efficient, easy to implement and can be combined with any rigid body motion planning library. However, unlike the classic bounding box method, our method takes into account the stepping over capabilities of the robot, and generates precise leg trajectories to avoid obstacles on the ground. We demonstrate that this method is well suited for footstep planning in cluttered environments.

It is a common artifact of multi-agent motion planning for groups of agents, following similar paths, to converge to a line. This occurs because the agents' immediate goals, also known as waypoints, are frequently a common point in space. Contention for the point goal causes agents to line up and generally interferes with agent motion. By extending the definition of an immediate point goal to a line segment, which we call a "way portal", we enable the agents to better utilize the space available to them in responding to dynamic constraints. We present a novel multi-agent navigation algorithm to efficiently compute the trajectories of autonomous agents using these way portals. We have incorporated the concept into a velocity obstacle-based local navigation model and present a new segment optimization algorithm that efficiently computes a new agent velocity with respect to the way portal. We show how way portal data is extracted from current global navigation data structures, such as navigation meshes. The algorithm is relatively simple to implement and has a small run-time cost (approximately 3 μs per agent.) We highlight its performance in different game-like scenarios and observe improved agent behavior and better utilization of free space.

We propose a new technique to simulate dynamic patterns of crowd behaviors using stress modeling. Our model accounts for permanent, stable disposition and the dynamic nature of human behaviors that change in response to the situation. The resulting approach accounts for changes in behavior in response to external stressors based on well-known theories in psychology. We combine this model with recent techniques on personality modeling for multi-agent simulations to capture a wide variety of behavioral changes and stressors. The overall formulation allows different stressors, expressed as functions of space and time, including time pressure, positional stressors, area stressors and inter-personal stressors. This model can be used to simulate dynamic crowd behaviors at interactive rates, including walking at variable speeds, breaking lane-formation over time, and cutting through a normal flow. We also perform qualitative and quantitative comparisons between our simulation results and real-world observations.

We present a novel algorithm to compute collision-free trajectories in dynamic environments. Our approach is general and makes no assumption about the obstacles or their motion. We use a replanning framework that interleaves optimization-based planning with execution. Furthermore, we describe a parallel formulation that exploits high number of cores on commodity graphics processors (GPUs) to compute a high-quality path in a given time interval. Overall, we show that search in configuration spaces can be significantly accelerated by using GPU parallelism.

We present a new algorithm to simulate a variety of crowd behaviors using the Discrete Choice Model (DCM). DCM has been widely studied in econometrics to examine and predict customers' or households' choices. Our DCM formulation can simulate virtual agents' goal selection and we highlight our algorithm by simulating heterogeneous crowd behaviors: evacuation, shopping, and rioting scenarios.

We present an interactive sound propagation algorithm that can compute high orders of specular and diffuse reflections as well as edge diffractions in response to moving sound sources and a moving listener. Our formulation is based on a precomputed acoustic transfer operator, which we compactly represent using the Karhunen-Loeve transform. At runtime, we use a two-pass approach that combines acoustic radiance transfer with interactive ray tracing to compute early reflections as well as higher-order reflections and late reverberation. The overall approach allows accuracy to be traded off for improved performance at runtime, and has a low memory overhead. We demonstrate the performance of our algorithm on different scenarios, including an integration of our algorithm with Valve's Source game engine.

We present an information-theoretic method to measure the similarity between a given set of observed, real-world data and visual simulation technique for aggregate crowd motions of a complex system consisting of many individual agents. This metric uses a two-step process to quantify a simulator's ability to reproduce the collective behaviors of the whole system, as observed in the recorded real-world data. First, Bayesian inference is used to estimate the simulation states which best correspond to the observed data, then a maximum likelihood estimator is used to approximate the prediction errors. This process is iterated using the EM-algorithm to produce a robust, statistical estimate of the magnitude of the prediction error as measured by its entropy (smaller is better). This metric serves as a simulator-to-data similarity measurement. We evaluated the metric in terms of robustness to sensor noise, consistency across different datasets and simulation methods, and correlation to perceptual metrics.

We present a simple algorithm to compute continuous penalty forces to determine collision response between rigid and deformable models bounded by triangle meshes. Our algorithm computes a well-behaved solution in contrast to the traditional stability and robustness problems of penalty methods, induced by force discontinuities. We trace contact features along their deforming trajectories and accumulate penalty forces along the penetration time intervals between the overlapping feature pairs. Moreover, we present a closed-form expression to compute the continuous and smooth collision response. Our method has very small additional overhead compared to previous penalty methods, and can significantly improve the stability and robustness. We highlight its benefits on several benchmarks.

This paper presents algorithms for optimal grasping forces. The previous work reveals that contact forces with minimal sum or maximum of normal force components can be written as positive combinations of primitive contact forces, to which the corresponding primitive contact wrenches express the required resultant wrench as their positive combination with minimum coefficients in terms of theL1orL∞metric. On this basis, we first propose an algorithm to compute such a set of primitive contact forces and the minimum contact forces. Moreover, considering the uncertainty in the actual friction coefficient, we develop another algorithm to determine the minimum required friction coefficient and the corresponding minimum contact forces within a given limit on their magnitude so that such contact forces are more reliable in practice. As our algorithms do not rely on any general optimization technique, they are very efficient and easy to implement. These algorithms can also be used in grasp quality evaluation and optimal grasp planning.

We present an efficient algorithm for simulating diffuse reflections of sound in a static scene. Our approach is built on recent advances in precomputed light transport techniques for visual rendering and uses them to develop an improved acoustic radiance transfer technique. We precompute a direct-to-indirect acoustic transfer operator for a scene, and use it to map direct sound incident on the surfaces of the scene to multibounce diffuse indirect sound, which is gathered at the listener to compute the final impulse response. Our algorithm decouples the transfer operator from the source position so we can efficiently update the acoustic response at the listener when the source moves. We highlight its performance on various benchmarks and observe significant speedups over prior methods based on acoustic radiance transfer.

We present a novel method for tuning geometric acoustic simulations based on ray tracing. Our formulation computes sound propagation paths from source to receiver and exploits the independence of visibility tests and validation tests to dynamically guide the simulation to high accuracy and performance. Our method makes no assumptions of scene layout and can account for moving sources, receivers, and geometry. We combine our guidance algorithm with a fast GPU sound propagation system for interactive simulation. Our implementation efficiently computes early specular paths and first order diffraction with a multiview tracing algorithm. We couple our propagation simulation with an audio output system supporting a high order interpolation scheme that accounts for attenuation, cross fading, and delay. The resulting system can render acoustic spaces composed of thousands of triangles interactively.

One of the most ubiquitous AI applications is vehicle route planning. While state-of-the-art systems take into account current traffic conditions or historic traffic data, current planning approaches ignore the impact of their own plans on the future traffic conditions. We present a novel algorithm for self-aware route planning that uses the routes it plans for current vehicle traffic to more accurately predict future traffic conditions for subsequent cars. Our planner uses a roadmap with stochastic, time-varying traffic densities that are defined by a combination of historical data and the densities predicted by the planned routes for the cars ahead of the current traffic. We have applied our algorithm to large-scale traffic route planning, and demonstrated that our self-aware route planner can more accurately predict future traffic conditions, which results in a reduction of the travel time for those vehicles that use our algorithm.

We present an efficient GPU-based parallel LSH algorithm to perform approximate k-nearest neighbor computation in high-dimensional spaces. We use the Bi-level LSH algorithm, which can compute k-nearest neighbors with higher accuracy and is amenable to parallelization. During the first level, we use the parallel RP-tree algorithm to partition datasets into several groups so that items similar to each other are clustered together. The second level involves computing the Bi-Level LSH code for each item and constructing a hierarchical hash table. The hash table is based on parallel cuckoo hashing and Morton curves. In the query step, we use GPU-based work queues to accelerate short-list search, which is one of the main bottlenecks in LSH-based algorithms. We demonstrate the results on large image datasets with 200,000 images which are represented as 512 dimensional vectors. In practice, our GPU implementation can obtain more than 40X acceleration over a single-core CPU-based LSH implementation.

We present an approach for collision avoidance for mobile robots that takes into account acceleration constraints. We discuss both the case of navigating a single robot among moving obstacles, and the case of multiple robots reciprocally avoiding collisions with each other while navigating a common workspace. Inspired by the concept of velocity obstacles [3], we introduce the acceleration-velocity obstacle (AVO) to let a robot avoid collisions with moving obstacles while obeying acceleration constraints. AVO characterizes the set of new velocities the robot can safely reach and adopt using proportional control of the acceleration. We extend this concept to reciprocal collision avoidance for multi-robot settings, by letting each robot take half of the responsibility of avoiding pairwise collisions. Our formulation guarantees collision-free navigation even as the robots act independently and simultaneously, without coordination. Our approach is designed for holonomic robots, but can also be applied to kinematically constrained non-holonomic robots such as cars. We have implemented our approach, and we show simulation results in challenging environments with large numbers of robots and obstacles.

We present a fast GPU-based streaming algorithm to perform collision queries between deformable models. Our approach is based on hierarchical culling and reduces the computation to generating different streams. We present a novel stream registration method to compact the streams and efficiently compute the potentially colliding pairs of primitives. We also use a deferred front tracking method to lower the memory overhead. The overall algorithm has been implemented on different GPUs and we have evaluated its performance on non-rigid and deformable simulations. We highlight our speedups over prior CPU-based and GPU-based algorithms. In practice, our algorithm can perform inter-object and intra-object computations on models composed of hundreds of thousands of triangles in tens of milliseconds.

We present a novel culling algorithm to perform fast and robust continuous collision detection between deforming volume meshes. This includes a continuous separating axis test that can conservatively check whether two volume meshes overlap during a given time interval. In addition, we present efficient methods to eliminate redundant elementary tests between the features (e.g., vertices, edges, and faces) of volume elements (e.g., tetrahedra, hexahedra, triangular prisms, etc.). Our approach is applicable to various deforming meshes, including those with changing topologies, and efficiently computes the first time of contact. We are able to perform inter-object and intra-object collision queries in models represented with tens of thousands of volume elements at interactive rates on a single CPU core. Moreover, we observe more than an order of magnitude performance improvement over prior methods.

We present the hybrid reciprocal velocity obstacle for collision-free and oscillation-free navigation of multiple mobile robots or virtual agents. Each robot senses its surroundings and acts independently without central coordination or communication with other robots. Our approach uses both the current position and the velocity of other robots to compute their future trajectories in order to avoid collisions. Moreover, our approach is reciprocal and avoids oscillations by explicitly taking into account that the other robots sense their surroundings as well and change their trajectories accordingly. We apply hybrid reciprocal velocity obstacles to iRobot Create mobile robots and demonstrate direct, collision-free, and oscillation-free navigation.

Recent GPU algorithms for constructing spatial hierarchies have achieved promising performance for moderately complex models by using the breadth-first search (BFS) construction order. While being able to exploit the massive parallelism on the GPU, the BFS order also consumes excessive GPU memory, which becomes a serious issue for interactive applications involving very complex models with more than a few million triangles. In this paper, we propose to use the partial breadth-first search (PBFS) construction order to control memory consumption while maximizing performance. We apply the PBFS order to two hierarchy construction algorithms. The first algorithm is for kd-trees that automatically balances between the level of parallelism and intermediate memory usage. With PBFS, peak memory consumption during construction can be efficiently controlled without costly CPU-GPU data transfer. We also develop memory allocation strategies to effectively limit memory fragmentation. The resulting algorithm scales well with GPU memory and constructs kd-trees of models with millions of triangles at interactive rates on GPUs with 1 GB memory. Compared with existing algorithms, our algorithm is an order of magnitude more scalable for a given GPU memory bound. The second algorithm is for out-of-core bounding volume hierarchy (BVH) construction for very large scenes based on the PBFS construction order. At each iteration, all constructed nodes are dumped to the CPU memory, and the GPU memory is freed for the next iteration's use. In this way, the algorithm is able to build trees that are too large to be stored in the GPU memory. Experiments show that our algorithm can construct BVHs for scenes with up to 20 M triangles, several times larger than previous GPU algorithms.

We present a method for procedurally modeling general complex 3D shapes. Our approach can automatically generate complex models of buildings, man-made structures, or urban data sets in a few minutes based on user-defined inputs. The algorithm attempts to generate complex 3D models that resemble a user-defined input model and satisfy various dimensional, geometric, and algebraic constraints to control the shape. These constraints are used to capture the intent of the user and generate shapes that look more natural. We also describe efficient techniques to handle complex shapes and highlight its performance on many different types of models. We compare model synthesis algorithms with other procedural modeling techniques, discuss the advantages of different approaches, and describe as close connection between model synthesis and context-sensitive grammars.

We present a novel approach to direct and control virtual crowds using navigation fields. Our method guides one or more agents toward desired goals based on guidance fields. The system allows the user to specify these fields by either sketching paths directly in the scene via an intuitive authoring interface or by importing motion flow fields extracted from crowd video footage. We propose a novel formulation to blend input guidance fields to create singularity-free, goal-directed navigation fields. Our method can be easily combined with the most current local collision avoidance methods and we use two such methods as examples to highlight the potential of our approach. We illustrate its performance on several simulation scenarios.

We present a novel concept, Virtualized Traffic, to reconstruct and visualize continuous traffic flows from discrete spatiotemporal data provided by traffic sensors or generated artificially to enhance a sense of immersion in a dynamic virtual world. Given the positions of each car at two recorded locations on a highway and the corresponding time instances, our approach can reconstruct the traffic flows (i.e., the dynamic motions of multiple cars over time) between the two locations along the highway for immersive visualization of virtual cities or other environments. Our algorithm is applicable to high-density traffic on highways with an arbitrary number of lanes and takes into account the geometric, kinematic, and dynamic constraints on the cars. Our method reconstructs the car motion that automatically minimizes the number of lane changes, respects safety distance to other cars, and computes the acceleration necessary to obtain a smooth traffic flow subject to the given constraints. Furthermore, our framework can process a continuous stream of input data in real time, enabling the users to view virtualized traffic events in a virtual world as they occur. We demonstrate our reconstruction technique with both synthetic and real-world input.

We present novel randomized algorithms for solving global motion planning problems that exploit the computational capabilities of many-core GPUs. Our approach uses thread and data parallelism to achieve high performance for all components of sample-based algorithms, including random sampling, nearest neighbor computation, local planning, collision queries and graph search. The approach can efficiently solve both the multi-query and single-query versions of the problem and obtain considerable speedups over prior CPU-based algorithms. We demonstrate the efficiency of our algorithms by applying them to a number of 6DOF planning benchmarks in 3D environments. Overall, this is the first algorithm that can perform real-time motion planning and global navigation using commodity hardware.

We present an approach to reciprocal collision avoidance, where multiple mobile agents must avoid collisions with each other while moving in a common workspace. Each agent acts fully independently, and does not communicate with others. Yet our approach guarantees that all agents will be collision-free for at least a fixed amount of time. Our approach provides a sufficient condition for collision-free motion. Given the agent's objective, the optimal collision-free action can be computed very efficiently, as it is the solution to a two-dimensional linear program. We show our approach on dense and complex simulation scenarios involving thousands of agents at fast real-time running times.

We present a new retraction algorithm for high DOF articulated models and use our algorithm to improve the performance of RRT planners in narrow passages. The retraction step is formulated as a constrained optimization problem and performs iterative refinement on the boundary of C-Obstacle space. We also combine the retraction algorithm with decomposition planners to handle very high DOF articulated models. The performance of our approach is analyzed using Voronoi diagrams and we show that our retraction algorithm provides a good approximation to the ideal RRT-extension in constrained environments. We have implemented our algorithm and tested its performance on robots with more than 40 DOFs in complex environments. In practice, we observe significant performance (2-80X) improvement over prior RRT planners on challenging scenarios with narrow passages.

We present an algorithm for collision-free navigation of multiple flying robots in three-dimensional workspace. Our approach extends the model of a simple car to a simple-airplane, which has constraints on speed and steering angle and includes a configuration variable for the altitude. We use a locally optimal reciprocal collision avoidance scheme that computes the trajectory without any collisions or oscillations for each airplane independently. In addition, our algorithm explicitly considers the kinematic and dynamic constraints of a simple-airplane and uses the notion of variable reciprocity when choosing velocities to ensure that simple-airplanes that are less constrained take more responsibility for avoiding collisions. We test our approach in two simulations and compute collision-free and oscillation-free trajectories that satisfy the kinematic and dynamic constraints of each simple-airplane.

We present a novel algorithm to perform continuous collision detection(CCD) between non-rigid, deformable models using local advancement. Given the initial and final configurations of a deformable model, our algorithm computes linear deformation by interpolating the vertices from the initial to the final configurations with a straight line path and checks for collision along that path. Our approach is applicable to polygon-soup models with arbitrary topology, handles self-collisions and makes no assumption about the underlying non-rigid motion. We accelerate the algorithm by computing motion bounds on the primitives and their bounding volumes. These bounds are combined with hierarchical culling techniques and used for fast collision checking. In practice, we have observed up to four times improvement in running time because of local advancement.

We present an efficient algorithm for solving the ray-shooting problem on high dimensional sets. Our algorithm computes the intersection of the boundary of a compact convex set with a ray emanating from an interior point of the set and represents the intersection point as a convex combination of a set of affinely independent points. We use our intersection algorithm to compute two types of optimal grasping forces, where either the sum or the maximum of normal force components is minimized. In our simulation, the algorithm converges well and performs the computations in tens of milliseconds on a laptop.

We present a novel k-nearest neighbor search algorithm (KNNS) for proximity computation in motion planning algorithm that exploits the computational capabilities of many-core GPUs. Our approach uses locality sensitive hashing and cuckoo hashing to construct an efficient KNNS algorithm that has linear space and time complexity and exploits the multiple cores and data parallelism effectively. In practice, we see magnitude improvement in speed and scalability over prior GPU-based KNNS algorithm. On some benchmarks, our KNNS algorithm improves the performance of overall planner by 20-40 times for CPU-based planner and up to 2 times for GPU-based planner.

We present a method for smooth and collision-free navigation for multiple independent robots under differential-drive constraints. Our algorithm is based on the optimal reciprocal collision avoidance formulation and guarantees both smoothness in the trajectories of the robots and locally collision-free paths. We provide proofs of these guarantees and demonstrate the effectiveness of our method in experimental scenarios using iRobot Create mobile robots navigating amongst each other.