Unsupervised Domain Adaptation (UDA) is essential for enabling semantic segmentation in new domains without requiring costly pixel-wise annotations. State-of-the-art (SOTA) UDA methods primarily use self-training with architecturally identical teacher and student networks, relying on Exponential Moving Average (EMA) updates. However, these approaches face substantial performance degradation with lightweight models due to inherent architectural inflexibility leading to low-quality pseudo-labels. To address this, we propose Distilled Unsupervised Domain Adaptation (DUDA), a novel framework that combines EMA-based self-training with knowledge distillation (KD). Our method employs an auxiliary student network to bridge the architectural gap between heavyweight and lightweight models for EMA-based updates, resulting in improved pseudo-label quality. DUDA employs a strategic fusion of UDA and KD, incorporating innovative elements such as gradual distillation from large to small networks, inconsistency loss prioritizing poorly adapted classes, and learning with multiple teachers. Extensive experiments across four UDA benchmarks demonstrate DUDA's superiority in achieving SOTA performance with lightweight models, often surpassing the performance of heavyweight models from other approaches.

We propose Graph2Nav, a real-time 3D object-relation graph generation framework, for autonomous navigation in the real world. Our framework fully generates and exploits both 3D objects and a rich set of semantic relationships among objects in a 3D layered scene graph, which is applicable to both indoor and outdoor scenes. It learns to generate 3D semantic relations among objects, by leveraging and advancing state-of-the-art 2D panoptic scene graph works into the 3D world via 3D semantic mapping techniques. This approach avoids previous training data constraints in learning 3D scene graphs directly from 3D data. We conduct experiments to validate the accuracy in locating 3D objects and labeling object-relations in our 3D scene graphs. We also evaluate the impact of Graph2Nav via integration with SayNav, a state-of-the-art planner based on large language models, on an unmanned ground robot to object search tasks in real environments. Our results demonstrate that modeling object relations in our scene graphs improves search efficiency in these navigation tasks.

Machine Learning (ML) for Mineral Prospectivity Mapping (MPM) is important for automating and accelerating the workflow to critical mineral assessment. Recent MPM works have explored Deep Learning (DL) as a modeling tool with more representation capacity. However, these overparameterized methods may be more prone to overfitting due to their reliance on scarce labeled data which is naturally inherent in mineral-relevant data sources. DL predictions can also be challenging to be interpreted and validated from a geological perspective to industrial experts. We propose a new unified framework that addresses these DL difficulties for MPM, using self-supervised learning, explainable artificial intelligence, and epistemic uncertainty modeling. Our approach uses a masked image modeling framework to pretrain a backbone neural network in a self-supervised manner using unlabeled geospatial data alone. After pretraining, the backbone network provides feature extraction for downstream MPM tasks. We conducted rigorous evaluations against existing techniques to assess mineral prospectivity for Mississippi Valley-type (MVT) and Clastic-Dominated (CD) Zn–Pb deposits in North America and Australia. Our results demonstrate that self-supervision promotes robustness in learned features, improving prospectivity predictions. Furthermore, we incorporate and demonstrate how explainable artificial intelligence techniques can provide interpretable predictions, offering valuable geological insights for industry professionals involved in high-stakes resource exploration and development.

Semantic reasoning and dynamic planning capabilities are crucial for an autonomous agent to perform complex navigation tasks in unknown environments. It requires a large amount of common-sense knowledge, that humans possess, to succeed in these tasks. We present SayNav, a new approach that leverages human knowledge from Large Language Models (LLMs) for efficient generalization to complex navigation tasks in unknown large-scale environments. SayNav uses a novel grounding mechanism, that incrementally builds a 3D scene graph of the explored environment as inputs to LLMs, for generating feasible and contextually appropriate high-level plans for navigation. The LLM-generated plan is then executed by a pre-trained low-level planner, that treats each planned step as a short-distance point-goal navigation sub-task. SayNav dynamically generates step-by-step instructions during navigation and continuously refines future steps based on newly perceived information. We evaluate SayNav on multi-object navigation (MultiON) task, that requires the agent to utilize a massive amount of human knowledge to efficiently search multiple different objects in an unknown environment. We also introduce a benchmark dataset for MultiON task employing ProcTHOR framework that provides large photo-realistic indoor environments with variety of objects. SayNav achieves state-of-the-art results and even outperforms an oracle based baseline with strong ground-truth assumptions by more than 8% in terms of success rate, highlighting its ability to generate dynamic plans for successfully locating objects in large-scale new environments. The code, benchmark dataset and demonstration videos are accessible at https://www.sri.com/ics/computer-vision/saynav.

Deep learning-based solutions for semantic segmentation suffer from significant performance degradation when tested on data with different characteristics than what was used during the training. Adapting the models using annotated data from the new domain is not always practical. Unsupervised Domain Adaptation (UDA) approaches are crucial in deploying these models in the actual operating conditions. Recent state-of-the-art (SOTA) UDA methods employ a teacher-student self-training approach, where a teacher model is used to generate pseudo-labels for the new data which in turn guide the training process of the student model. Though this approach has seen a lot of success, it suffers from the issue of noisy pseudo-labels being propagated in the training process. To address this issue, we propose an auxiliary pseudo-label refinement network (PRN) for online refining of the pseudo labels and also localizing the pixels whose predicted labels are likely to be noisy. Being able to improve the quality of pseudo labels and select highly reliable ones, PRN helps self-training of segmentation models to be robust against pseudo label noise propagation during different stages of adaptation. We evaluate our approach on benchmark datasets with three different domain shifts, and our approach consistently performs significantly better than the previous state-of-the-art methods.

Unsupervised domain adaptation (UDA) approaches focus on adapting models trained on a labeled source domain to an unlabeled target domain. In contrast to UDA, source-free domain adaptation (SFDA) is a more practical setup as access to source data is no longer required during adaptation. Recent state-of-the-art (SOTA) methods on SFDA mostly focus on pseudo-label refinement based self-training which generally suffers from two issues: i) inevitable occurrence of noisy pseudo-labels that could lead to early training time memorization, ii) refinement process requires maintaining a memory bank which creates a significant burden in resource constraint scenarios. To address these concerns, we propose C-SFDA, a curriculum learning aided self-training framework for SFDA that adapts efficiently and reliably to changes across domains based on selective pseudo-labeling. Specifically, we employ a curriculum learning scheme to promote learning from a restricted amount of pseudo labels selected based on their reliabilities. This simple yet effective step successfully prevents label noise propagation during different stages of adaptation and eliminates the need for costly memory-bank based label refinement. Our extensive experimental evaluations on both image recognition and semantic segmentation tasks confirm the effectiveness of our method. C-SFDA is also applicable to online test-time domain adaptation and outperforms previous SOTA methods in this task.

Precise estimation of global orientation and location is critical to ensure a compelling outdoor Augmented Reality (AR) experience. We address the problem of geo-pose estimation by cross-view matching of query ground images to a geo-referenced aerial satellite image database. Recently, neural network-based methods have shown state-of-the-art performance in cross-view matching. However, most of the prior works focus only on location estimation, ignoring orientation, which cannot meet the requirements in outdoor AR applications. We propose a new transformer neural network-based model and a modified triplet ranking loss for joint location and orientation estimation. Experiments on several benchmark cross-view geo-localization datasets show that our model achieves state-of-the-art performance. Furthermore, we present an approach to extend the single image query-based geo-localization approach by utilizing temporal information from a navigation pipeline for robust continuous geo-localization. Experimentation on several large-scale real-world video sequences demonstrates that our approach enables high-precision and stable AR insertion.

This paper presents a novel approach for the Vision-and-Language Navigation (VLN) task in continuous 3D environments, which requires an autonomous agent to follow natural language instructions in unseen environments. Existing end-to-end learning-based VLN methods struggle at this task as they focus mostly on utilizing raw visual observations and lack the semantic spatio-temporal reasoning capabilities which is crucial in generalizing to new environments. In this regard, we present a hybrid transformer-recurrence model which focuses on combining classical semantic mapping techniques with a learning-based method. Our method creates a temporal semantic memory by building a top-down local ego-centric semantic map and performs cross-modal grounding to align map and language modalities to enable effective learning of VLN policy. Empirical results in a photo-realistic long-horizon simulation environment show that the proposed approach outperforms a variety of state-of-the-art methods and baselines with over 22% relative improvement in SPL in prior unseen environments.

Understanding the geometric relationships between objects in a scene is a core capability in enabling both humans and autonomous agents to navigate in new environments. A sparse, unified representation of the scene topology will allow agents to act efficiently to move through their environment, communicate the environment state with others, and utilize the representation for diverse downstream tasks. To this end, we propose a method to train an autonomous agent to learn to accumulate a 3D scene graph representation of its environment by simultaneously learning to navigate through said environment. We demonstrate that our approach, GraphMapper, enables the learning of effective navigation policies through fewer interactions with the environment than vision-based systems alone. Further, we show that GraphMapper can act as a modular scene encoder to operate alongside existing Learning-based solutions to not only increase navigational efficiency but also generate intermediate scene representations that are useful for other future tasks.

Class imbalance is a fundamental problem in computer vision applications such as semantic segmentation. Specifically, uneven class distributions in a training dataset often result in unsatisfactory performance on under-represented classes. Many works have proposed to weight the standard cross entropy loss function with pre-computed weights based on class statistics, such as the number of samples and class margins. There are two major drawbacks to these methods: 1) constantly up-weighting minority classes can introduce excessive false positives in semantic segmentation; 2) a minority class is not necessarily a hard class. The consequence is low precision due to excessive false positives. In this regard, we propose a hard-class mining loss by reshaping the vanilla cross entropy loss such that it weights the loss for each class dynamically based on instantaneous recall performance. We show that the novel recall loss changes gradually between the standard cross entropy loss and the inverse frequency weighted loss. Recall loss also leads to improved mean accuracy while offering competitive mean Intersection over Union (IoU) performance. On Synthia dataset11Synthia-Sequence Summer split, recall loss achieves 9% relative improvement on mean accuracy with competitive mean IoU using DeepLab-ResNet18 compared to the cross entropy loss. Code available at https://github.com/PotatoTian/recall-semseg.

As progress is made on training machine learning models on incrementally expanding classification tasks (i.e., incremental learning), a next step is to translate this progress to industry expectations. One technique missing from incremental learning is automatic architecture design via Neural Architecture Search (NAS). In this paper, we show that leveraging NAS for incremental learning results in strong performance gains for classification tasks. Specifically, we contribute the following: first, we create a strong baseline approach for incremental learning based on Differentiable Architecture Search (DARTS) and state-of-the-art incremental learning strategies, outperforming many existing strategies trained with similar-sized popular architectures; second, we extend the idea of architecture search to regularize architecture forgetting, boosting performance past our proposed baseline. We evaluate our method on both RF signal and image classification tasks, and demonstrate we can achieve up to a 10% performance increase over state-of-the-art methods. Most importantly, our contribution enables learning from continuous distributions on real-world application data for which the complexity of the data distribution is unknown, or the modality less explored (such as RF signal classification).

Ranging information from ultra-wideband (UWB) ranging radios can be used to improve estimated navigation accuracy of a ground robot with other on-board sensors. However, all ranging-aided navigation methods demand the locations of ranging nodes to be known, which is not suitable for time-pressed situations, dynamic cluttered environments, or collaborative navigation applications. This paper describes a new ranging-aided navigation approach that does not require the locations of ranging radios. Our approach formulates relative pose constraints using ranging readings. The formulation is based on geometric relationships between each stationary ranging node and two ranging antennas on the moving robot across time. Our experiments show that estimated navigation accuracy of the ground robot is substantially enhanced with ranging information using our approach under a variety of scenarios, when ranging nodes are placed at unknown locations. We analyze and compare our performance with a traditional ranging-aided method, which requires mapping the positions of ranging nodes. We also demonstrate the applicability of our approach for collaborative navigation in large-scale unknown environments, by using ranging information from one mobile robot to improve navigation estimation of the other robot. This application does not require the installation of ranging nodes at fixed locations.

Understanding the perceived scene during navigation enables intelligent robot behaviors. Current vision-based semantic SLAM (Simultaneous Localization and Mapping) systems provide these capabilities. However, their performance decreases in visually-degraded environments, that are common places for critical robotic applications, such as search and rescue missions. In this paper, we present SIGNAV, a real-time semantic SLAM system to operate in perceptually-challenging situations. To improve the robustness for navigation in dark environments, SIGNAV leverages a multi-sensor navigation architecture to fuse vision with additional sensing modalities, including an inertial measurement unit (IMU), LiDAR, and wheel odometry. A new 2.5D semantic segmentation method is also developed to combine both images and LiDAR depth maps to generate semantic labels of 3D mapped points in real time. We demonstrate that the navigation accuracy from SIGNAV in a variety of indoor environments under both normal lighting and dark conditions. SIGNAV also provides semantic scene understanding capabilities in visually-degraded environments. We also show the benefits of semantic information to SIGNAV’s performance.

Visual navigation for autonomous agents is a core task in the fields of computer vision and robotics. Learning-based methods, such as deep reinforcement learning, have the potential to outperform the classical solutions developed for this task; however, they come at a significantly increased computational load. Through this work, we design a novel approach that focuses on performing better or comparable to the existing learning-based solutions but under a clear time/computational budget. To this end, we propose a method to encode vital scene semantics such as traversable paths, unexplored areas, and observed scene objects–alongside raw visual streams such as RGB, depth, and semantic segmentation masks—into a semantically informed, top-down egocentric map representation. Further, to enable the effective use of this information, we introduce a novel 2-D map attention mechanism, based on the successful multi-layer Transformer networks. We conduct experiments on 3-D reconstructed indoor PointGoal visual navigation and demonstrate the effectiveness of our approach. We show that by using our novel attention schema and auxiliary rewards to better utilize scene semantics, we outperform multiple baselines trained with only raw inputs or implicit semantic information while operating with an 80% decrease in the agent’s experience.

Proper occlusion based rendering is very important to achieve realism in all indoor and outdoor Augmented Reality (AR) applications. This paper addresses the problem of fast and accurate dynamic occlusion reasoning by real objects in the scene for large scale outdoor AR applications. Conceptually, proper occlusion reasoning requires an estimate of depth for every point in augmented scene which is technically hard to achieve for outdoor scenarios, especially in the presence of moving objects. We propose a method to detect and automatically infer the depth for real objects in the scene without explicit detailed scene modeling and depth sensing (e.g. without using sensors such as 3D-LiDAR). Specifically, we employ instance segmentation of color image data to detect real dynamic objects in the scene and use either a top-down terrain elevation model or deep learning based monocular depth estimation model to infer their metric distance from the camera for proper occlusion reasoning in real time. The realized solution is implemented in a low latency real-time framework for video-see-though AR and is directly extendable to optical-see-through AR. We minimize latency in depth reasoning and occlusion rendering by doing semantic object tracking and prediction in video frames.

We study an important, yet largely unexplored problem of large-scale cross-modal visual localization by matching ground RGB images to a geo-referenced aerial LIDAR 3D point cloud (rendered as depth images). Prior works were demonstrated on small datasets and did not lend themselves to scaling up for large-scale applications. To enable large-scale evaluation, we introduce a new dataset containing over 550K pairs (covering 143 km2 area) of RGB and aerial LIDAR depth images. We propose a novel joint embedding based method that effectively combines the appearance and semantic cues from both modalities to handle drastic cross-modal variations. Experiments on the proposed dataset show that our model achieves a strong result of a median rank of 5 in matching across a large test set of 50K location pairs collected from a 14km^2 area. This represents a significant advancement over prior works in performance and scale. We conclude with qualitative results to highlight the challenging nature of this task and the benefits of the proposed model. Our work provides a foundation for further research in cross-modal visual localization.

We propose a new approach that utilizes semantic information to register 2D monocular video frames to the world using 3D georeferenced data, for augmented reality driving applications. The geo-registration process uses our predicted vehicle pose to generate a rendered depth map for each frame, allowing 3D graphics to be convincingly blended with the real world view. We also estimate absolute depth values for dynamic objects, up to 120 meters, based on the rendered depth map and update the rendered depth map to reflect scene changes over time. This process also creates opportunistic global heading measurements, which are fused with other sensors, to improve estimates of the 6 degrees-of- freedom global pose of the vehicle over state-of-the-art outdoor augmented reality systems [5]-, [19]. We evaluate the navigation accuracy and depth map quality of our system on a driving vehicle within various large-scale environments for producing realistic augmentations.

This paper proposes a real-time navigation approach that is able to integrate many sensor types while fulfilling performance needs and system constraints. Our approach uses a plug-and-play factor graph framework, which extends factor graph formulation to encode sensor measurements with different frequencies, latencies, and noise distributions. It provides a flexible foundation for plug-and-play sensing, and can incorporate new evolving sensors. A novel constrained optimal selection mechanism is presented to identify the optimal subset of active sensors to use, during initialization and when any sensor condition changes. This mechanism constructs candidate subsets of sensors based on heuristic rules and a ternary tree expansion algorithm. It quickly decides the optimal subset among candidates by maximizing observability coverage on state variables, while satisfying resource constraints and accuracy demands. Experimental results demonstrate that our approach selects subsets of sensors to provide satisfactory navigation solutions under various conditions, on large-scale real data sets using many sensors.

This paper proposes a novel vision-aided navigation approach that continuously estimates precise 3D absolute pose for aerial vehicles, using only inertial measurements and monocular camera observations. Our approach is able to provide accurate navigation solutions under long-term GPS outage, by tightly incorporating absolute geo-registered information into two kinds of visual measurements: 2D-3D tie-points, and geo-registered feature tracks. 2D-3D tie-points are established by finding feature correspondences to align an aerial video frame to a 2D geo-referenced image rendered from the 3D terrain database. These measurements provide global information to correct accumulated error in navigation estimation. Geo-registered feature tracks are generated by associating features across consecutive frames. They enable the propagation of 3D geo-referenced values to further improve the pose estimation. All sensor measurements are fully optimized in a smoother-based inference framework, which achieves efficient relinearization and real-time estimation of navigation states and their covariances over a constant-length of sliding window. Experimental results demonstrate that our approach provides accurate and consistent aerial navigation solutions on several large-scale GPS-denied scenarios.

This paper proposes a navigation algorithm that provides a low-latency solution while estimating the full nonlinear navigation state. Our approach uses Sliding-Window Factor Graphs, which extend existing incremental smoothing methods to operate on the subset of measurements and states that exist inside a sliding time window. We split the estimation into a fast short-term smoother, a slower but fully global smoother, and a shared map of 3D landmarks. A novel three-stage visual feature model is presented that takes advantage of both smoothers to optimize the 3D landmark map, while minimizing the computation required for processing tracked features in the short-term smoother. This three-stage model is formulated based on the maturity of the estimation of the 3D location of the underlying landmark in the map. Long-range associations are used as global measurements from matured landmarks in the short-term smoother and loop closure constraints in the long-term smoother. Experimental results demonstrate our approach provides highly-accurate solutions on large-scale real data sets using multiple sensors in GPS-denied settings.

Visual landmark matching with a pre-built landmark database is a popular technique for localization. Traditionally, landmark database was built with visual odometry system, and the 3D information of each visual landmark is reconstructed from video. Due to the drift of the visual odometry system, a global consistent landmark database is difficult to build, and the inaccuracy of each 3D landmark limits the performance of landmark matching. In this paper, we demonstrated that with the use of precise 3D Li-dar range data, we are able to build a global consistent database of high precision 3D visual landmarks, which improves the landmark matching accuracy dramatically. In order to further improve the accuracy and robustness, landmark matching is fused with a multi-stereo based visual odometry system to estimate the camera pose in two aspects. First, a local visual odometry trajectory based consistency check is performed to reject some bad landmark matchings or those with large errors, and then a kalman filtering is used to further smooth out some landmark matching errors. Finally, a disk-cache-mechanism is proposed to obtain the real-time performance when the size of the landmark grows for a large-scale area. A week-long real time live marine training experiments have demonstrated the high-precision and robustness of our proposed system.

In this paper, we present a unified approach for a drift-free and jitter-reduced vision-aided navigation system. This approach is based on an error-state Kalman filter algorithm using both relative (local) measurements obtained from image based motion estimation through visual odometry, and global measurements as a result of landmark matching through a pre-built visual landmark database. To improve the accuracy in pose estimation for augmented reality applications, we capture the 3D local reconstruction uncertainty of each landmark point as a covariance matrix and implicity rely more on closer points in the filter. We conduct a number of experiments aimed at evaluating different aspects of our Kalman filter framework, and show our approach can provide highly-accurate and stable pose both indoors and outdoors over large areas. The results demonstrate both the long term stability and the overall accuracy of our algorithm as intended to provide a solution to the camera tracking problem in augmented reality applications.

Our goal is to grasp 3D objects given a single image, by using prior 3D shape models of object classes. The shape models, defined as a collection of oriented primitive shapes centered at fixed 3D positions, can be learned from a few labeled images for each class. The 3D class model can then be used to estimate the 3D shape of a detected object, including occluded parts, from a single image. The estimated 3D shape is used as to select one of the target grasps for the object. We show that our 3D shape estimation is sufficiently accurate for a robot to successfully grasp the object, even in situations where the part to be grasped is not visible in the input image.

In this paper, we present a unified approach for a camera tracking system based on an error-state Kalman filter algorithm. The filter uses relative (local) measurements obtained from image based motion estimation through visual odometry, as well as global measurements produced by landmark matching through a pre-built visual landmark database and range measurements obtained from radio frequency (RF) ranging radios. We show our results by using the camera poses output by our system to render views from a 3D graphical model built upon the same coordinate frame as the landmark database which also forms the global coordinate system and compare them to the actual video images. These results help demonstrate both the long term stability and the overall accuracy of our algorithm as intended to provide a solution to the GPS denied ubiquitous camera tracking problem under both vision-aided and vision-impaired conditions.

Our goal is to circumvent one of the roadblocks to using existing approaches for single-view recognition for achieving multi-view recognition, namely, the need for sufficient training data for many viewpoints. We show how to construct virtual training examples for multi-view recognition using a simple model of objects (nearly planar facades centered at fixed 3D positions). We also show how the models can be learned from a few labeled images for each class.