Imitation Learning of Human Operation Based on Visual Demonstration

Abstract

In this paper, we propose an imitation learning method based on the optimization of the dynamic movement primitives (DMPs). The DMPs framework is one of the common methods of imitation learning because of its adaptability in time and space. On the basis of the traditional framework, we add the consideration of the dynamic performance of the robot during the reproduction process. We optimize the learned parameters (via DMPs) to reduce the average torque of robot joints while keeping the trajectory error small. The proposed method is evaluated with an experiment where a 6-degrees of freedom (6-DOF) robot learned a pick-place task from the visual demonstration of a human teacher. The experimental results show that our method reduced the average torque of robot joints by 73.4 N.m, which proves the effectiveness of the proposed method.