Reactive Configuration Space Collision-Free Motion Planning for Mobile Manipulator in Unforeseen Scenarios

This article investigates reactive, collision-free motion planning for mobile manipulator (MM) in unknown, cluttered environments. In these environments, real-time perception data guides local planning and nonconvex obstacles can cause optimization-based methods to become trapped in local minima. While sampling-based methods can overcome local minima, their high collision detection cost limits their ability to support high-frequency, real-time planning, particularly in configuration space (C-space) with dynamic, point cloud-based obstacles. To address these issues, we propose a hierarchical planning architecture in C-space. This approach dynamically adjusts the perception range based on the MM’s motion trends and limits extraneous areas to enhance the frequency of collision information updates. For the mobile base, we introduce a tangential velocity optimization to provide heuristic guidance for whole-body planning and improve the ability to escape local minima. We also apply a control postprocessing method to refine the base’s path and reduce oscillations. This framework is tested on a 9-degree-of-freedom mobile manipulator. Simulations in various scenarios show its robustness, with success rates of approximately 80% even with dynamic obstacles. Real-world experiments further confirm its effectiveness and applicability.