An Improved RRT Path Planning Method Incorporating Deep Reinforcement Learning for Space Multi-arm Robot

Aiming at the complex internal structure of the satellite to be repaired in the on-orbit maintenance mission and the high requirements of the maintenance mission on the precision of robotic arm path planning, a multi-robotic arm path planning algorithm Deep Deterministic Policy Gradient RRT (DDPG-RRT) combined with deep reinforcement learning is proposed. Firstly, the initial environment is established. Then, on the basis of the traditional RRT algorithm, the idea of DDPG algorithm is introduced to set the dynamic step size to search the collision-free path between the start position and the target position of each robotic arm departing at the same time. Finally, a path optimization algorithm is introduced to smooth the planning path. The results show that compared with the traditional RRT and RRT-Connect obstacle avoidance algorithms, the average planning time, the total average path length, and the success rate of this method are reduced, which is feasible.