Motion Planning of Six-DOF Arm Robot Based on Improved DDPG Algorithm
Zhuang Li, Hongbin Ma, Yazhe Ding, Chen Wang, Ying Jin
Abstract
This paper presents an improved deep deterministic policy gradient algorithm based on a six-DOF(six multi-degree-of- freedom) arm robot. First, we build a robot model based on the DH(Denavit-Hartenberg) parameters of the UR5 arm robot. Then, we improved the experience pool of the traditional DDPG(deep deterministic policy gradient) algorithm by adding a success experience pool and a collision experience pool. Next, the reward function is improved to increase the degree of successful reward and the penalty of collision. Finally, the training is divided into segments, the front three axes are trained first, and then the six axes. The simulation results in ROS(Robot Operating System) show that the improved DDPG algorithm can effectively solve the problem that the six-DOF arm robot moves too far in the configuration space. The trained model can reach the target area in five steps. Compared with the traditional DDPG algorithm, the improved DDPG algorithm has fewer training episodes, but achieves better results.