Development of a pneumatic artificial muscle driven drum playing robot for reinforcement learning using real robot

Abstract

Pneumatic artificial muscles are advantageous owing to dynamic stiffness. However, the drawback of this is their poor controllability. In this paper, a reinforcement learning-based control system using a real robot is proposed to improve controllability. The proposed system is aimed to generate dynamic motion patterns that utilize the body structure and variable stiffness characteristics. We propose a reinforcement learning-based realistic drum-playing robot, and confirm its effectiveness through experiments.