Robot Skill Learning Algorithm Based on Dynamic Motion Primitive

Abstract

In order to realize the complex operation skills learning of a UR 10 collaborative robot, we propose a dynamic-motion-primitive robot skill learning algorithm based on reinforcement learning and imitation learning. Shapes of demonstrated trajectories is re-trained with dynamic motion primitives, and the robot arm replaces the explicit coordinates to reach the target point in an exploratory manner. Experiment results show that the optimized trajectory of the robot can preserves the shape of the teach-and-fit track well.