Dexterous grasping remains a fundamental yet challenging problem in robotics. A general-purpose robot must be capable of grasping diverse objects in arbitrary scenarios. However, existing research typically relies on restrictive assumptions, such as single-object settings or limited environments, showing constrained generalization. We present DexGraspVLA, a hierarchical framework for robust generalization in language-guided general dexterous grasping and beyond. It utilizes a pre-trained Vision-Language model as the high-level planner and learns a diffusion-based low-level Action controller. The key insight to achieve generalization lies in iteratively transforming diverse language and visual inputs into domain-invariant representations via foundation models, where imitation learning can be effectively applied due to the alleviation of domain shift. Notably, our method achieves a 90+% dexterous grasping success rate under thousands of challenging unseen cluttered scenes. Empirical analysis confirms the consistency of internal model behavior across environmental variations, validating our design. DexGraspVLA also, for the first time, simultaneously demonstrates free-form long-horizon prompt execution, robustness to adversarial objects and human disturbance, and failure recovery. Extended application to nonprehensile grasping further proves its generality.

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.

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 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.

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.

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.

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 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 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 present a visual and inertial-based terrain classification network (VINet) for robotic navigation over different traversable surfaces. We use a novel navigation-based labeling scheme for terrain classification and generalization on unknown surfaces. Our proposed perception method and adaptive scheduling control framework can make predictions according to terrain navigation properties and lead to better performance on both terrain classification and navigation control on known and unknown surfaces. Our VINet can achieve 98.37% in terms of accuracy under supervised setting on known terrains and improve the accuracy by 8.51% on unknown terrains compared to previous methods. We deploy VINet on a mobile tracked robot for trajectory following and navigation on different terrains, and we demonstrate an improvement of 10.3% compared to a baseline controller in terms of RMSE.

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 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.

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.

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.

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.