Peng Xu

Legged robots are designed to perform operational tasks in complex, rugged terrains. These tasks generally encompass two fundamental modalities: locomotion and manipulation. Humanoid robots, benefiting from their distinctive morphological configuration (independent arm and leg structures), can decouple the control of these two task types. In contrast, conventional quadruped/hexapod robots, lacking dedicated end effectors, are typically optimized for locomotion. While manipulation capabilities can be added via auxiliary robotic arms, this approach leads to unnecessary energy consumption and sacrifices payload capacity. Adopting a limb reuse architecture offers a promising solution. This work proposes Legs-as-Arms, a multitask reinforcement learning framework for hexapod robots that enables simultaneous locomotion and manipulation. The framework implements effective limb reuse, enabling any leg(s) to track desired trajectories while maintaining stable body movement. By leveraging a multihead critic network and value normalization, we integrate three key capabilities into a single control policy: 1) base locomotion control; 2) stationary whole-body manipulation; and 3) dynamic mobile manipulation. Our method demonstrates fault-tolerant gaits, workspace expansion, centimeter-level trajectory tracking accuracy, and significant load adaptability. We validate the framework’s effectiveness and practical application value through comprehensive hardware experiments across diverse multimodal scenarios.

Legged robots face significant challenges in navigating complex environments, as they require precise real-time decisions for foothold selection and contact planning. While existing research has explored methods to select footholds based on terrain geometry or kinematics, a critical gap remains: few existing methods efficiently validate the existence of a non-collision swing trajectory. This article addresses this gap by introducing KCFRC, a novel approach for efficient foothold reachability analysis. We first formally define the foothold reachability problem and establish a sufficient condition for foothold reachability. Based on this condition, we develop the KCFRC algorithm, which enables robots to validate foothold reachability in real time. Our experimental results demonstrate that KCFRC achieves remarkable time efficiency, completing foothold reachability checks for a single leg across 900 potential footholds in an average of 2 ms. Furthermore, we show that KCFRC can accelerate trajectory optimization and is particularly beneficial for contact planning in confined spaces, enhancing the adaptability and robustness of legged robots in challenging environments.

Contact planning for multi-legged robots is a challenging sequential decision-making problem due to the interplay of gaits, footholds, configurations, and physical constraints from both the robot and the environment. Existing multi-contact planners often fail to find feasible sequences within a limited time in complex scenarios and to ensure physical possibility. We propose a parallel Monte Carlo Tree Search (MCTS)-based planner that leverages multi-constraint reachability to efficiently generate physically valid contact sequences. The method accelerates planning through a hash-driven parallel approach, prioritizing promising candidates while pruning trapped nodes via valueless node evaluation. It employs depth-first backup for long-horizon planning and uses virtual loss to balance parallel exploration. To ensure feasible transitions between contact states, we establish comprehensive reachability conditions for multi-legged robots, incorporating stability, collision avoidance, kinematics, joint torques, and contact constraints into the planning framework. In experiments in sparse foothold environments, our planner outperforms mainstream contact planning approaches in traversability, solution quality and physical feasibility, while achieving a competitive planning speed. Furthermore, simulation and hardware validation on hexapod and humanoid robots exhibit successful locomotion across various terrains while satisfying constraints.

Reinforcement learning (RL) has demonstrated impressive performance in legged locomotion over various challenging environments. However, due to the sim-to-real gap and lack of explainability, unconstrained RL policies deployed in the real world still suffer from inevitable safety issues, such as joint collisions, excessive torque, or foot slippage in low-friction environments. These problems limit its usage in missions with strict safety requirements, such as planetary exploration, nuclear facility inspection, and deep-sea operations. In this paper, we design a hierarchical optimization-based whole-body follower, which integrates both hard and soft constraints into RL framework to make the robot move with better safety guarantees. Leveraging the advantages of model-based control, our approach allows for the definition of various types of hard and soft constraints during training or deployment, which allows for policy fine-tuning and mitigates the challenges of sim-to-real transfer. Meanwhile, it preserves the robustness of RL when dealing with locomotion in complex unstructured environments. The trained policy with introduced constraints was deployed in a hexapod robot and tested in various outdoor environments, including snow-covered slopes and stairs, demonstrating the great traversability and safety of our approach.

Deep reinforcement learning (DRL) methods have shown significant promise in controlling the movement of quadruped robots. However, for systems like hexapod robots, which feature a higher-dimensional action space, it remains challenging for an agent to devise an effective control strategy directly. Currently, no hexapod robots have demonstrated highly dynamic motion. To address this, we propose imitation-enhanced reinforcement learning (IERL), a two-stage approach enabling hexapod robots to achieve dynamic motion through direct control using RL methods. Initially, imitation learning (IL) replicates a basic positional control method, creating a pre-trained policy for basic locomotion. Subsequently, the parameters from this model are utilized as the starting point for the reinforcement learning process to train the agent. Moreover, we incorporate a smooth transition (ST) method to make IERL overcome the changes in network inputs between two stages, and adaptable to various complex network architectures incorporating latent features. Extensive simulations and real-world experiments confirm that our method effectively tackles the high-dimensional action space challenges of hexapod robots, significantly enhancing learning efficiency and enabling more natural, efficient, and dynamic movements compared to existing methods.

Specialized robots, such as legged robots and unmanned aerial vehicles (UAVs), are commonly regarded as effective platforms for aiding in search and rescue (SAR) missions. However, existing approaches often decouple the tasks between UAVs and legged robots, for instance, using UAVs for mapping and legged robots for operation. Such feature hinders the realization of collaborative advantages and consequently reduces task completion efficiency. This letter proposes a perception-planning framework for collaborative SAR missions involving UAVs and legged robots. The framework primarily utilizes UAVs to efficiently search for target objects in unknown environments, while also determining a feasible path for the legged robots to reach the targets. More specifically, the UAV employs an optimized frontier-based exploration method to explore unknown environments. Simultaneously, a depth-first reachable search tree is utilized to swiftly assess the traversability of the guided path of the legged robot, thereby accelerating the exploration process. After locating the target, the method further constrains the exploration area based on permissible ellipses until an the shortest path is found. The proposed approach is evaluated through experiments in both simulation and real-world scenarios and the results demonstrate that our approach exhibits significantly higher exploration efficiency compared to state-of-the-art methods, reducing completion times in SAR tasks by over 18.3%.

Legged robots have been widely used in unstructured field environments, where unforeseen nongeometric hazards, such as sinkage or slippage might compromise their mobility. Such uncertain terrain-related risks may be very challenging to mitigate by employing point or ball-shaped feet, however, determining the appropriate principles for foot design remains an open-research problem. This study presents a systematic pipeline of design and evaluations for the sensorized terrain-adaptive foot (STAF) aiming to remodel the underlying soft, sloped granular media. The advanced foot configuration enables explicit characterization of complex foot-terrain interaction responses, thereby facilitating evidence-based evaluations using proposed terrain physical metrics and terrain traveling metrics. These terrain-aware metrics are interpretable, physics-informed, and versatile for various robot prototypes. Compared to traditional empirical foot design methods, our model-based guideline can reliably identify physical terrain properties and explain contact behavior to optimize the foot sole's primary parameters. We conduct multiple experiments on both the single-foot testbed and a quadruped robot. The findings demonstrate that our developed STAF system significantly enhances mobility and in situ sensing capabilities over loose deformable slopes.