Industrial Insert Robotic Assembly Based on Model-based Meta-Reinforcement Learning

Abstract

The high-rigidity working environment in the industrial insert robotic assembly is likely to cause damage to the robot and parts, so it is difficult to use traditional position control methods under these circumstances. The combination of compliance control and reinforcement learning brings new ideas to this problem. In this paper, a robotic industrial insertion learning method based on model-based meta-reinforcement learning is proposed. By interacting with the environment and updating the dynamic model in real-time, the online adaptation of assembly tasks is realized. A simulation environment is used to train a neural network dynamics model, with a model-based reinforcement learning method. Finally, we compare our method with other methods in the real environment to verify the effectiveness of the algorithm.