Autonomous Loco-Manipulation for Hybrid Wheeled-Legged Robots in Cluttered and Unknown Environments

Despite the recent progress in robot mobility, autonomous locomotion in cluttered and completely unknown environments remains a significant challenge for both wheeled and legged mobile robots. To fulfill this task, robots have to leverage their kinematics to negotiate the most effective locomotion strategy to adapt their configuration to the terrain conformation. This task becomes even more challenging when working with redundant hybrid wheeled-legged platforms, which combine both types of locomotion to enhance efficiency and safety during the traverse. Although the hybrid mobility system offers higher flexibility when traversing difficult terrains, effective hybrid locomotion planners that transparently combine different locomotion modes have not been extensively explored. Moreover, to further enhance the autonomy level during locomotion tasks, robots need to be equipped with the additional ability to reason about the scene composition and react autonomously to failures. This research project addresses these challenges by focusing on the implementation of multiple frameworks that can also be combined together to achieve autonomous loco-manipulation, ensuring safe traverses through irregular and cluttered environments. The first module introduces an online hybrid path planner for autonomous locomotion with wheeled-legged robots utilizing a set of parametrized motion primitives to adapt the robot's configuration to the treated scenario. The planned solution is continuously updated to incorporate changes in the scene while keeping track of possible failures that may arise during the execution. This allows the robot to effectively and autonomously react to such failures, without the need for failing the task or requiring human intervention. However, traversing unstructured environments is not always guaranteed to be feasible, especially in cluttered scenarios where multiple objects may prevent navigation. For this reason, an additional framework was developed to increase the robots' reasoning capabilities, enabling them to plan for a sequence of pushing actions to move the objects in the scene and rearrange the scenario. The goal is to allow the robot to overcome the environmental constraints and mobility limits, creating autonomously new pathways in the case of failures during locomotion. The final part of this thesis focuses on improving the mobility of planetary rovers by integrating the previously introduced hybrid path planner and implementing an improved velocity-based traction controller. This controller aims to reduce slip, minimize wheel fighting, and optimize locomotion during planetary exploration while providing metrics to dynamically update the terrain cost. In summary, this research contributes to effectively enhancing the autonomous capabilities of mobile robots enabling them to freely navigate in complex environments, further reducing the gap that prevents wheeled-legged robotic systems from achieving full autonomy in real-world applications.