Sophisticated learning architectures, e.g., Transformers, present a unique opportunity for robots to understand complex vehicle-terrain kinodynamic interactions for off-road mobility. While internet-scale data are available for Natural Language Processing (NLP) and Computer Vision (CV) tasks to train Transformers, real-world mobility data are difficult to acquire with physical robots navigating off-road terrain. Furthermore, training techniques specifically designed to process text and image data in NLP and CV may not apply to robot mobility. In this paper, we propose VertiFormer, a novel data-efficient multi-task Transformer model trained with only one hour of data to address such challenges of applying Transformer architectures for robot mobility on extremely rugged, vertically challenging, off-road terrain. Specifically, VertiFormer employs a new learnable masked modeling and next token prediction paradigm to predict the next pose, action, and terrain patch to enable a variety of off-road mobility tasks simultaneously, e.g., forward and inverse kinodynamics modeling. The non-autoregressive design mitigates computational bottlenecks and error propagation associated with autoregressive models. VertiFormer's unified modality representation also enhances learning of diverse temporal mappings and state representations, which, combined with multiple objective functions, further improves model generalization. Our experiments offer insights into effectively utilizing Transformers for off-road robot mobility with limited data and demonstrate our efficiently trained Transformer can facilitate multiple off-road mobility tasks onboard a physical mobile robot.

Most existing social robot navigation techniques either leverage hand-crafted rules or human demonstrations to connect robot perception to socially compliant actions. However, there remains a significant gap in effectively translating perception into socially compliant actions, much like how human reasoning naturally occurs in dynamic environments. Considering the recent success of Vision-Language Models (VLMs), we propose using language to bridge the gap in human-like reasoning between perception and socially aware robot actions. We create a vision-language dataset, Social robot Navigation via Explainable Interactions (SNEI), featuring 40K human-annotated Visual Question Answers (VQAs) based on 2K human-robot social interactions in unstructured, crowded public spaces, spanning perception, prediction, chain-of-thought reasoning, action, and explanation. We fine-tune a VLM, Social-LLaVA, using SNEI to demonstrate the practical application of our dataset. Social-LLaVA outperforms state-of-the-art models like GPT-4V and Gemini, based on the average of fifteen different human-judge scores across 50 VQA. Deployed onboard a mobile robot, Social-LLaVA enables human-like reasoning, marking a promising step toward socially compliant robot navigation in dynamic public spaces through language reasoning.

Long-duration, off-road, autonomous missions require robots to continuously perceive their surroundings regardless of the ambient lighting conditions. Most existing autonomy systems heavily rely on active sensing, e.g., LiDAR, RADAR, and Time-of-Flight sensors, or use (stereo) visible light imaging sensors, e.g., color cameras, to perceive environment geometry and semantics. In scenarios where fully passive perception is required and lighting conditions are degraded to an extent that visible light cameras fail to perceive, most downstream mobility tasks such as obstacle avoidance become impossible. To address such a challenge, this paper presents a Multi-Modal Passive Perception dataset, M2P2, to enable off-road mobility in low-light to no-light conditions. We design a multi-modal sensor suite including thermal, event, and stereo RGB cameras, GPS, two Inertia Measurement Units (IMUs), as well as a high-resolution LiDAR for ground truth, with a novel multi-sensor calibration procedure that can efficiently transform multi-modal perceptual streams into a common coordinate system. Our 10-hour, 32 km dataset also includes mobility data such as robot odometry and actions and covers well-lit, low-light, and no-light conditions, along with paved, on-trail, and off-trail terrain. Our results demonstrate that off-road mobility is possible through only passive perception in extreme low-light conditions using end-to-end learning and classical planning. The project website can be found at https://cs.gmu.edu/~xiao/Research/M2P2/

Most traversability estimation techniques divide off-road terrain into traversable (e.g., pavement, gravel, and grass) and non-traversable (e.g., boulders, vegetation, and ditches) regions and then inform subsequent planners to produce trajectories on the traversable part. However, recent research demonstrated that wheeled robots can traverse vertically challenging terrain (e.g., extremely rugged boulders comparable in size to the vehicles themselves), which unfortunately would be deemed as non-traversable by existing techniques. Motivated by such limitations, this work aims at identifying the traversable from the seemingly non-traversable, vertically challenging terrain based on past kinodynamic vehicle-terrain interactions in a data-driven manner. Our new Traverse the Non-Traversable(TNT) traversability estimator can efficiently guide a down-stream sampling-based planner containing a high-precision 6-DoF kinodynamic model, which becomes deployable onboard a small-scale vehicle. Additionally, the estimated traversability can also be used as a costmap to plan global and local paths without sampling. Our experiment results show that TNT can improve planning performance, efficiency, and stability by 50%, 26.7%, and 9.2% respectively on a physical robot platform.

Abstract-- Wheeled robots have recently demonstrated superior mechanical capability to traverse vertically challenging terrain (e.g., extremely rugged boulders comparable in size to the vehicles themselves). Negotiating such terrain introduces significant variations of vehicle pose in all six Degrees-of-Freedom (DoFs), leading to imbalanced contact forces, varying momentum, and chassis deformation due to non-rigid tires and suspensions. To autonomously navigate on vertically challenging terrain, all these factors need to be efficiently reasoned within limited onboard computation and strict real-time constraints. For highly articulated systems, Despite their wide availability, wheeled mobile robots are efficient decomposition is possible to break down the modeling usually limited in terms of mobility, mostly moving in 2D of the vehicle chassis and actuators (e.g., legs and active flat environments. After dividing their planar workspaces into suspensions) so that the chassis trajectory can be computed free spaces and obstacles, those robots are assumed to be separately in parallel and the low-level actuation solved using rigid bodies and efficiently find collision-free paths to move fast control and optimization techniques [6].

Humans excel at efficiently navigating through crowds without collision by focusing on specific visual regions relevant to navigation. However, most robotic visual navigation methods rely on deep learning models pre-trained on vision tasks, which prioritize salient objects -- not necessarily relevant to navigation and potentially misleading. Alternative approaches train specialized navigation models from scratch, requiring significant computation. On the other hand, self-supervised learning has revolutionized computer vision and natural language processing, but its application to robotic navigation remains underexplored due to the difficulty of defining effective self-supervision signals. Motivated by these observations, in this work, we propose a Self-Supervised Vision-Action Model for Visual Navigation Pre-Training (VANP). Instead of detecting salient objects that are beneficial for tasks such as classification or detection, VANP learns to focus only on specific visual regions that are relevant to the navigation task. To achieve this, VANP uses a history of visual observations, future actions, and a goal image for self-supervision, and embeds them using two small Transformer Encoders. Then, VANP maximizes the information between the embeddings by using a mutual information maximization objective function. We demonstrate that most VANP-extracted features match with human navigation intuition. VANP achieves comparable performance as models learned end-to-end with half the training time and models trained on a large-scale, fully supervised dataset, i.e., ImageNet, with only 0.08% data.

While the workspace of traditional ground vehicles is usually assumed to be in a 2D plane, i.e., SE(2), such an assumption may not hold when they drive at high speeds on unstructured off-road terrain: High-speed sharp turns on high-friction surfaces may lead to vehicle rollover; Turning aggressively on loose gravel or grass may violate the non-holonomic constraint and cause significant lateral sliding; Driving quickly on rugged terrain will produce extensive vibration along the vertical axis. Therefore, most offroad vehicles are currently limited to drive only at low speeds to assure vehicle stability and safety. In this work, we aim at empowering high-speed off-road vehicles with competence awareness in SE(3) so that they can reason about the consequences of taking aggressive maneuvers on different terrain with a 6-DoF forward kinodynamic model. The model is learned from visual and inertial Terrain Representation for Off-road Navigation (TRON) using multimodal, self-supervised vehicle-terrain interactions. We demonstrate the efficacy of our Competence-Aware High-Speed Off-Road (CAHSOR) navigation approach on a physical ground robot in both an autonomous navigation and a human shared-control setup and show that CAHSOR can efficiently reduce vehicle instability by 62% while only compromising 8.6% average speed with the help of TRON.

Abstract-- Most conventional wheeled robots can only move in flat environments and simply divide their planar workspaces into free spaces and obstacles. Deeming obstacles as nontraversable significantly limits wheeled robots' mobility in realworld, extremely rugged, off-road environments, where part of the terrain (e.g., irregular boulders and fallen trees) will be treated as non-traversable obstacles. To improve wheeled mobility in those environments with vertically challenging terrain, we present two wheeled platforms with little hardware modification compared to conventional wheeled robots; we collect datasets of our wheeled robots crawling over previously non-traversable, vertically challenging terrain to facilitate data-driven mobility; we also present algorithms and their experimental results to show that conventional wheeled robots have previously unrealized potential of moving through vertically challenging terrain. I. INTRODUCTION Building mobile robots that are capable of reaching as workspaces into free spaces (traversable) or obstacles (nontraversable), many places as possible has long been a dream for many Indeed, autonomous mobile robots have in the real world, especially outdoor off-road environments ventured into remote deserts for scientific exploration [1], explored where vertical protrusions from the ground are not uncommon. Achieving reliable and robust in conquering requires driving wheeled robots over irregular mobility in these environments is challenging due to the and complex obstacles and is therefore much more difficult intricate nature of the terrain, the complex vehicle-terrain compared to simply driving on non-flat environments.

Autonomous mobile robots need to perceive the environments with their onboard sensors (e.g., LiDARs and RGB cameras) and then make appropriate navigation decisions. In order to navigate human-inhabited public spaces, such a navigation task becomes more than only obstacle avoidance, but also requires considering surrounding humans and their intentions to somewhat change the navigation behavior in response to the underlying social norms, i.e., being socially compliant. Machine learning methods are shown to be effective in capturing those complex and subtle social interactions in a data-driven manner, without explicitly hand-crafting simplified models or cost functions. Considering multiple available sensor modalities and the efficiency of learning methods, this paper presents a comprehensive study on learning social robot navigation with multimodal perception using a large-scale real-world dataset. The study investigates social robot navigation decision making on both the global and local planning levels and contrasts unimodal and multimodal learning against a set of classical navigation approaches in different social scenarios, while also analyzing the training and generalizability performance from the learning perspective. We also conduct a human study on how learning with multimodal perception affects the perceived social compliance. The results show that multimodal learning has a clear advantage over unimodal learning in both dataset and human studies. We open-source our code for the community's future use to study multimodal perception for learning social robot navigation.

Humans are well-adept at navigating public spaces shared with others, where current autonomous mobile robots still struggle: while safely and efficiently reaching their goals, humans communicate their intentions and conform to unwritten social norms on a daily basis; conversely, robots become clumsy in those daily social scenarios, getting stuck in dense crowds, surprising nearby pedestrians, or even causing collisions. While recent research on robot learning has shown promises in data-driven social robot navigation, good-quality training data is still difficult to acquire through either trial and error or expert demonstrations. In this work, we propose to utilize the body of rich, widely available, social human navigation data in many natural human-inhabited public spaces for robots to learn similar, human-like, socially compliant navigation behaviors. To be specific, we design an open-source egocentric data collection sensor suite wearable by walking humans to provide multi-modal robot perception data; we collect a large-scale (~100 km, 20 hours, 300 trials, 13 humans) dataset in a variety of public spaces which contain numerous natural social navigation interactions; we analyze our dataset, demonstrate its usability, and point out future research directions and use cases.