Autonomous Robots最新文献

Natural environments such as forests and grasslands are challenging for robotic navigation because of the false perception of rigid obstacles from high grass, twigs, or bushes. In this work, we present Wild Visual Navigation (WVN), an online self-supervised learning system for visual traversability estimation. The system is able to continuously adapt from a short human demonstration in the field, only using onboard sensing and computing. One of the key ideas to achieve this is the use of high-dimensional features from pre-trained self-supervised models, which implicitly encode semantic information that massively simplifies the learning task. Further, the development of an online scheme for supervision generator enables concurrent training and inference of the learned model in the wild. We demonstrate our approach through diverse real-world deployments in forests, parks, and grasslands. Our system is able to bootstrap the traversable terrain segmentation in less than 5 min of in-field training time, enabling the robot to navigate in complex, previously unseen outdoor terrains.

In this paper, we propose a learning-free approach for an autonomous robotic system to grasp, hand over, and regrasp previously unseen objects. The proposed framework includes two main components: a novel grasping detector to predict grasping poses directly from the point cloud and a reachability-aware handover planner to select the exchange pose and grasping poses for two robots. In the grasping detection stage, multiple superquadrics are first recovered at different positions within the object, representing the local geometric feature of the object. Our algorithm then exploits the tri-symmetry feature of superquadrics and synthesizes a list of antipodal grasps from each recovered superquadric. An evaluation model is designed to assess and quantify the quality of each grasp candidate. In the handover planning stage, the planner first selects grasping candidates that have high scores and a larger number of collision-free partners. Then the exchange location is computed by utilizing two signed distance fields (SDF) which model the reachability space for the pair of two robots. To evaluate the performance of the proposed method, we first run experiments on isolated and packed scenes to corroborate the effectiveness of our grasping detection method. Then the handover experiments are conducted on a dual-arm system with two 7 degrees of freedom (DoF) manipulators. The results indicate that our method shows better performance compared with the state-of-the-art, without the need for large amounts of training.

We present the design, implementation and testing of a multi-robot exploration algorithm for NASA’s upcoming Cooperative Autonomous Distributed Robotic Exploration (CADRE) lunar technology demonstration mission. The CADRE mission, among its various objectives, entails utilizing a trio of autonomous mobile robots to collaboratively explore and construct a map of a designated area of the lunar surface. Given the mission’s inherent constraints, including limited mission duration, constrained power resources, and restricted communication capabilities, we formulate an exploration algorithm to improve exploration efficiency, facilitate equitable workload distribution among individual agents, and minimize inter-robot communication. To achieve these requirements, we employ a semi-centralized exploration algorithm that partitions the unexplored area, regardless of its shape and size, into a series of non-overlapping partitions, assigning each partition to a specific robot for exploration. Each robot autonomously explores its designated region without intervention from other robots. We explore the design space of the proposed algorithm and evaluate its performance under diverse conditions in simulations. Finally, we validate the algorithm’s functionality through two sets of hardware experiments: the first utilizes prototype rovers using a ROS-based navigation software stack for feasibility testing, while the second employs high-fidelity development model rovers running CADRE’s custom flight-software stack for flight-like performance validation. Both sets of experiments are conducted in the Jet Propulsion Laboratory’s lunar-simulated rover testing facilities, demonstrating the algorithm’s robustness and readiness for lunar deployment.

Distributed multi-target tracking is a canonical task for multi-robot systems, encompassing applications from environmental monitoring to disaster response to surveillance. In many situations the unknown distribution of the targets in a search area is non-uniform, e.g., herds of animals moving together. This paper develops a novel distributed multi-robot multi-target tracking algorithm to effectively search for and track clustered targets. There are two key features. First, there are two parallel estimators, one to provide the best guess of the current states of targets and a second to provide a coarse, long-term distribution of clusters. Second, robots use the power diagram to divide the search space between agents in a way that effectively trades off between tracking detected targets within high density areas and searching for other potential targets. Extensive simulation experiments demonstrate the efficacy of the proposed method and show that it outperforms other approaches in tracking accuracy of clustered targets while maintain good performance for uniformly distributed targets.

Three-dimensional navigation of legged robots is crucial for field exploration and post-disaster rescue. Existing optimization-based local trajectory planners predominantly focus on obstacle avoidance, neglecting negative obstacles (e.g., pits) and varying ground features (e.g., different terrain types). Additionally, non-overlapping areas between the planned space in three-dimensional trajectory planning and the robot’s actual reachable space lead to decision-making issues between crossing and obstacle avoidance, making it challenging to differentiate between passable and hazardous areas, thus impacting navigation safety and stability. To address these limitations, we propose a novel visual local planner, LSF-Planner (Visual Local Planner for Legged Robots Based on Ground Structure and Feature Information). The LSF-Planner employs a multi-layer local perception map that integrates ground feature semantics, sensor range, and negative obstacles (e.g., voids detected by depth sensors) to construct a ground reliability representation. The Label2Grad method is introduced to convert this representation into gradient layers, incorporating a ground reliability penalty function into trajectory optimization. By incorporating constraints on the center of mass height and crossing angles, LSF-Planner effectively differentiates between traversable and hazardous areas. Experimental results show that LSF-Planner significantly outperforms existing methods in 3D trajectory planning, enhancing the navigation performance of legged robots in unstructured environments.

We study the problem of determining minimum-length coordinated motions for two axis-aligned square robots translating in an obstacle-free plane: Given feasible start and goal configurations (feasible in the sense that the two squares are interior disjoint), find a continuous motion for the two squares from start to goal, comprising only robot-robot collision-free configurations, such that the total Euclidean distance traveled by the two squares is minimal among all possible such motions. In this paper we present an adaptation of the tools developed for the case of disks to the case of squares. We show that in certain aspects the case of squares is more complicated, requiring additional and more involved arguments over the case of disks.

A novel approach for the fast onboard detection of isolated markers for visual relative localisation of multiple teammates in agile UAV swarms is introduced in this paper. As the detection forms a key component of real-time localisation systems, a three-fold innovation is presented, consisting of an optimised procedure for CPUs, a GPU shader program, and a functionally equivalent FPGA streaming architecture. For the proposed CPU and GPU solutions, the mean processing time per pixel of input camera frames was accelerated by two to three orders of magnitude compared to the unoptimised state-of-the-art approach. For the localisation task, the proposed FPGA architecture offered the most significant overall acceleration by minimising the total delay from camera exposure to detection results. Additionally, the proposed solutions were evaluated on various 32-bit and 64-bit embedded platforms to demonstrate their efficiency, as well as their feasibility for applications using low-end UAVs and MAVs. Thus, it has become a crucial enabling technology for agile UAV swarming.

We present an approach to ensure safe and deadlock-free navigation for decentralized multi-robot systems operating in constrained environments, including doorways and intersections. Although many solutions have been proposed that ensure safety and resolve deadlocks, optimally preventing deadlocks in a minimally invasive and decentralized fashion remains an open problem. We first formalize the objective as a non-cooperative, non-communicative, partially observable multi-robot navigation problem in constrained spaces with multiple conflicting agents, which we term as social mini-games. Formally, we solve a discrete-time optimal receding horizon control problem leveraging control barrier functions for safe long-horizon planning. Our approach to ensuring liveness rests on the insight that there exists barrier certificates that allow each robot to preemptively perturb their state in a minimally-invasive fashion onto liveness sets i.e. states where robots are deadlock-free. We evaluate our approach in simulation as well on physical robots using F1/10 robots, a Clearpath Jackal, as well as a Boston Dynamics Spot in a doorway, hallway, and corridor intersection scenario. Compared to both fully decentralized and centralized approaches with and without deadlock resolution capabilities, we demonstrate that our approach results in safer, more efficient, and smoother navigation, based on a comprehensive set of metrics including success rate, collision rate, stop time, change in velocity, path deviation, time-to-goal, and flow rate.

In the domain of multi-robot systems, cooperative systems that are highly attuned and connected to their surroundings are becoming increasingly significant. This surge in interest highlights various challenges, especially regarding system integration and safety constraints. Our research contributes to the assurance of fault tolerance to avert abnormal behaviors and sustain reliable robot localization. In this paper, a mixed approach between data-driven and model-based for fault detection is introduced, within a decentralized architecture, thereby strengthening the system’s capacity to handle simultaneous sensor faults. Information theory-based fault indicators are developed by computing the Jensen-Shannon divergence ((D_{JS})) between state predictions and sensor-obtained corrections. This initiates a two-tiered data-driven mechanism: one layer employing Machine Learning for fault detection, and another distinct layer for fault isolation. The methodology’s efficacy is assessed using real data from the Turtlebot3 platform.

Developing effective reward functions is crucial for robot learning, as they guide behavior and facilitate adaptation to human-like tasks. We present Human2Bot (H2B), advancing the learning of such a generalized multi-task reward function that can be used zero-shot to execute unknown tasks in unseen environments. H2B is a newly designed task similarity estimation model that is trained on a large dataset of human videos. The model determines whether two videos from different environments represent the same task. At test time, the model serves as a reward function, evaluating how closely a robot’s execution matches the human demonstration. While previous approaches necessitate robot-specific data to learn reward functions or policies, our method can learn without any robot datasets. To achieve generalization in robotic environments, we incorporate a domain augmentation process that generates synthetic videos with varied visual appearances resembling simulation environments, alongside a multi-scale inter-frame attention mechanism that aligns human and robot task understanding. Finally, H2B is integrated with Visual Model Predictive Control (VMPC) to perform manipulation tasks in simulation and on the xARM6 robot in real-world settings. Our approach outperforms previous methods in simulated and real-world environments trained solely on human data, eliminating the need for privileged robot datasets.