Robotic Skill Acquisition in Peg-in-hole Assembly Tasks Based on Deep Reinforcement Learning

Abstract

Robot assembly skill learning has gradually become a research focus in the field of industrial robots. To improve the learning efficiency and adaptability of robot peg-in-hole assembly strategy, a deep reinforcement learning (DRL) method combining PD force controller is proposed in this paper. Firstly, the twin delayed deep deterministic policy gradient (TD3) algorithm based on prioritized experience replay (PER) is used to perceive the environment and make assembly skill decisions. Then, according to the PD force controller, the action output of the DRL policy network is adjusted to modify the robot's trajectory. Finally, the curriculum learning strategy is adopted to gradually increase the task difficulty in the process of algorithm training, thus improving the convergence speed and generalization ability of the model. The proposed method is verified in two different peg-in-hole assembly tasks, which realizes the autonomous acquisition of robot assembly skills, and effectively improves the learning efficiency and adaptability of assembly strategy.