Data-based optimal tracking control of unknown nonlinear systems with constrained-input via echo state network

Abstract

This paper designs a data-based optimal tracking control policy for the unknown nonlinear systems with constrained-input by using integral reinforcement learning (IRL) method, which is implemented by echo state network (ESN). An augmented system is firstly generated by combining the tracking error and the desired trajectory. Because the system dynamics are unknown, the IRL method is designed to obtain the optimal tracking control policy by using the system data. The ESN is employed to approximate the critic and actor function, which the hidden layers do not require elaborate design. To reduce the approximation error, the output weights are tuned by using the least squares and only system data. At last, a nonlinear system tracking task is simulated to verify the availability of the proposed method.