Data-Driven Dynamic Decision-Making Strategy for Gear-Shaft Robotic Assembly Process

Gear-shaft assembly is a common process in assembling process of products. It can be divided into rough alignment process, precise alignment and insertion process. Inspired by operator’s gear-shaft assembly process, visual data is utilized to detect the relative pose between the shaft and the gear, while force/torque data is utilized to handle their interaction. In this paper, Proximal Policy Optimization (PPO) algorithm is employed to make dynamic decisions during the gear-shaft robotic assembly. First, the framework of the proposed method is studied. Afterwards, Convolutional Neural Network (CNN) and PPO are employed to make dynamic decisions in rough alignment process, while PPO is employed to make dynamic decisions in precise alignment and insertion process. Finally, case study is carried out to verify the proposed method. The results show the converged PPO model could dynamically generate optimal strategy to complete the gear-shaft robotic assembly.