Control of single-stroke movement of a drum-playing robot by reinforcement learning using a realistic artificial muscle-driven robot

Artificial muscles are advantageous owing to dynamic stiffness; however, the drawback of this is their poor controllability. Here, a reinforcement learning-based control system is proposed and exemplified on a realistic artificial muscle-driven robot. The proposed system suppresses the modeling error, and allows to generate dynamic motion patterns that utilize the body structure and variable stiffness characteristics. As an example application, we consider drum playing. We propose a reinforcement learning-based realistic drum-playing robot. We propose a reinforcement learning-based strategy for this realistic robot and confirm its effectiveness through simulations and experiments.