LSTM-Based Imitation Learning of Robot Manipulator Using Impedance Control

This paper proposes an imitation learning method based on long short-term memory (LSTM) to demonstrate robot manipulators using impedance control. An impedance controller controls the force and position of the robot manipulator. In this study, direct demonstrated position and force data for imitation learning of the robot were designed to be the reference input of the impedance controller. LSTM-based imitation learning methods enabled the robot to function as intended, even when its initial position was changed or other contact forces were applied according to the environment. The proposed method was verified by applying the writing task of the actual industrial robot manipulator that functions as the expert’s intention.