A Novel DS-GMR Coupled Primitive for Robotic Motion Skill Learning

Abstract

Imitation learning is a promising paradigm for enabling robots to autonomously perform new tasks, which is similar to the procedure of human's motion skill acquirement. In the paper, we present a novel DS-GMR coupled primitive (DGCP) for robotic motion skill learning based on imitation learning. DGCP comprises a dominated linear ordinary differential dynamic component and a GMR based forcing component. Furthermore, we carefully design the linkage mechanism of hyper parameters to achieve spatiotemporal coupling synchronically. In this way an intelligent trajectory planning in similar scenario (fulfilling target within different time and positon) could be generated spontaneously. Finally, simulation that robot perform a trajectory planning with min-jerk criteria in various duration demonstrates practical capability and efficiency of the presented method.