An Integrated Tracking Control Approach Based on Reinforcement Learning for a Continuum Robot in Space Capture Missions

Abstract

In this paper, an integrated tracking control approach was developed for a continuum robot in space capture missions. For the configuration of a three-module cable-driven continuum robot, the nonlinear dynamics equations were derived. The uncertain movement of noncooperative debris requires a real-time trajectory planning solution. Therefore, an adaptive controller based on deep reinforcement learning (DRL) is proposed to generate a dynamic controller in continuous action space, where the trajectory planning function is simultaneously integrated into the dynamic solution. To obtain an efficient policy network for the highly nonlinear dynamics model, the rolling optimization method was combined in the DRL method of the deep deterministic policy gradient (DDPG). The DRL controller generated an appropriate control sequence according to the long-term control performance of the robot system and then executed optimal control input according to the rolling optimization. The simulation result shows that the proposed policy network of the improved DDPG controller can reasonably provide the tracking control solution in the noncooperative debris capture mission.