Human-Robot Interaction System Design for Manipulator Control Using Reinforcement Learning

Abstract

In this article, a novel human-robot interaction (HRI) system is presented and applied in the robotic arm coordinated operation control task. The presented HRI system includes two parts, the impedance model controller and the robotic arm controller, which allows the operator to manipulate the robotic arm to accomplish the given task with minimal human effort. First, the model-based reinforcement learning (RL) method is applied in the impedance model for operator adaptation. The impedance model controller can transform human input into the specific signal for the manipulator. Second, a novel adaptive manipulator controller is designed. In contrast to existing controllers, a velocity-free filter is implemented in our controller, which is developed to replace the manipulator actuator's speed signal. The effectiveness of the presented HRI system is verified by the simulation based on real manipulator parameters.