Design of Nonlinear Predictive Control for Pneumatic Muscle Actuator Based on Echo State Gaussian Process

Abstract

Recently, the application of Pneumatic Muscle Actuators (PMAs) for driving rehabilitation robots has become a matter of great concern. A traditional control algorithm, such as PID, cannot achieve satisfactory high-precision performance in trajectory tracking problem for PMAs, due to PMAs’ features of nonlinear effects, slow response time, time-varying parameters. In this study we proposed a nonlinear predictive control strategy SNN-ESGP, which is comprised of a novel model called echo state Gaussian process (ESGP) that is suitable for modeling nonlinear unknown systems as well as measuring their uncertainties, and a single neural network (SNN) serves as the controller of the system. To analyze the system convergence, we utilize the gradient descent algorithm and deduce the iterative rules of the weights of SNN. The stability of the closed-loop system is analyzed by using Lyapunov theorem. Finally, studies of physical experiments illustrate the validity of the control strategy with high-precision trajectory performance.