Vision-Based Deep Reinforcement Learning For UR5 Robot Motion Control

Recently, Deep Reinforcement Learning (DRL) has been widely used for robot manipulation skill learning. However, learning directly from the high-dimensional sensory observations is always inefficient, which makes it impractical to apply DRL methods to real-world robots. This paper designed a vision-based DRL method based on Deep Deterministic Policy Gradients (DDPG). To improve the learning efficiency, we construct an asymmetric actor-critic structure and add an auxiliary-task branch to the actor network in the method. A reaching-task based UR5 robot is designed to evaluate the performance of the method. The results indicate that the vision-based DRL method proposed in the paper can successfully learn the reaching-task skill and the utilization of asymmetric actor-critic structure and auxiliary-task objective can improve the learning efficiency and the final performance of the DRL method effectively.