Koopman-based improvement of the closed-loop response of controlled articulated soft robots

Abstract

To handle the complex nature of robots built from soft material, the data-based Koopman operator theory recently has been proposed as an alternative to model-based controller designs. In this paper we investigate the use of this theory for articulated soft robots, in which the elasticity is concentrated at the joints. In particular, we propose to apply the Koopman operator theory to the residual dynamics of an underlying model-based controller. For the inner loop control we utilize the Elastic Structure Preserving (ESP) control approach that has been successfully applied to a wide range of articulated soft robots. However, the control performance of the ESP control is clearly limited by the accuracy of the model, which motivates the combination with an outer data-based control approach. The concept is experimentally verified using antagonistically-driven elastic actuators with highly nonlinear stiffness characteristics.