Deep Reinforcement Learning in Soft Viscoelastic Actuator of Dielectric Elastomer

Abstract

Dielectric elastomer actuators (DEAs) have been widely employed as artificial muscles in soft robots. Due to material viscoelasticity and nonlinear electromechanical coupling, it is challenging to accurately model a viscoelastic DEA, especially when the actuator is of a complex or irregular configuration. Control of DEAs is thus challenging but significant. In this letter, we propose a model-free method for control of DEAs, based on deep reinforcement learning. We perform dynamic feedback control by considering the time-dependent behavior of DEAs. Our method is generic in that it does not require task-specific knowledge about the structure or material parameters of the DEA. The experiments show that our method is robust to achieve accurate control for the DEAs of different configurations, different prestretches, and at different times (the material property usually changes due to viscoelasticity effects). To the best of our knowledge, this letter is the first effort to explore deep reinforcement learning for control of DEAs.