Research on Intelligent Peg-in-Hole Assembly Strategy Based on Deep Reinforcement Learning

Due to the error of visual positioning and camera calibration in the peg_in_hole assembly method based on visual positioning, there is a certain position and angle deviation between shafts and holes, which leads to the failure of the assembly. In this paper, a flexible assembly algorithm with variable admittance parameters, which incorporates reinforcement learning, is proposed to improve the flexibility of the assembly process in unknown environments. First, Archimedes spiral hole searching method with admittance control is used to search the hole on the workpiece surface. And then, the DDPG algorithm is used to train the admittance control parameters in the process of assembly to improve the safety of the robot inserting hole. We generalize the robot inserting operation and enable the robot to safely handle more assembly problems in unknown environments. Finally, the reliability and safety of the methods are verified in the Pybullet.