A Novel Robotic Skill Learning Approach for Assembly Task With Dynamical System and Broad Learning

Abstract

This study introduces a strategy for robotic peg-hole assembly designed to efficiently transfer skills from a human instructor to a robot, achieved through a series of demonstrations and training sessions. Moving the peg fixed to the robotic arm’s end from the distal position to the top of the hole, which is represented as an autonomous dynamical system. During the modeling and generalization phases, a fast modeling learning method based on an extreme learning machine is proposed, and an energy function is used to improve the generalization accuracy. This approach enables the system to maintain high tracking accuracy and stability while resisting disturbances. Additionally, an obstacle avoidance strategy is implemented to prevent the peg from contacting obstacles during movement. The incremental learning-based broad learning system (BLS) is used for peg-hole assembly, where peg-hole position and attitude are adjusted by continuous neural network training during insertion until 1–3 adjustments are made to complete the assembly. Finally, the developed skill-learning strategy’s effectiveness is confirmed through the validation of simulations and experiments.