Skill-Learning Method of Dual Peg-in-Hole Compliance Assembly for Micro-Device

Abstract

For the dual peg-in-hole compliance assembly task of upper and lower double-hole structural micro-devices, a skill-learning method is proposed. This method combines offline training in a simulation space and online training in a realistic space. In this paper, a dual peg-in-hole model is built according to the results of a force analysis, and contact-point searching methods are provided for calculating the contact force. Then, a skill-learning framework is built based on deep reinforcement learning. Both expert action and incremental action are used in training, and a reward system considers both efficiency and safety; additionally, a dynamic exploration method is provided to improve the training efficiency. In addition, based on experimental data, an online training method is used to optimize the skill-learning model continuously so that the error caused by the deviation in the offline training data from reality can be reduced. The final experiments demonstrate that the method can effectively reduce the contact force while assembling, improve the efficiency and reduce the impact of the change in position and orientation.

Keywords

• skill learning
• peg-in-hole assembly task
• reinforcement learning
• deep deterministic policy gradient (DDPG)
• micro-device