Trajectory Tracking and Reinforcement Learning Based Obstacle Avoidance for a Cable-Driven Parallel Robot With Aerial-Ground Mobile Bases

A Cable-Driven Parallel Robot (CDPR) is a robot driven by a set of cables and has advantages such as large workspace and low inertia. Scholars have proposed to apply Aerial Mobile Bases (AMBs) and Ground Mobile Bases (GMBs) to a CDPR to achieve a large workspace and fully constrain the moving platform. However, trajectory tracking and obstacle avoidance for CDPR-AGMBs remain open and challenging problems due to their heterogeneous actuation and complex dynamics. This paper proposes a trajectory tracking planner based on optimization and a real-time obstacle avoidance planner based on reinforcement learning (RL) for CDPR-AGMBs. Both planners are developed for a CDPR-AGMB system consisting of four length-variable cables connected to four GMBs and one fixed-length cable connected to an AMB. A physical prototype of the CDPR-AGMB has been built to evaluate the proposed methods. The trajectory tracking planner is implemented according to the prototype’s physical parameters, while the obstacle avoidance planner is trained in simulation using MuJoCo and then deployed on the prototype for real-world validation.