Adaptive Optimal Control for Continuous-Time System: A Value Iteration Approach

This study presents a value iteration based output feedback adaptive optimal control algorithm for continuous-time linear system under completely unknown dynamics and partially observable states. A key strategy is that the unobservable states are estimated by a radial basis function neural network observer. Furthermore, compared with the policy iteration method, there is no need to assume the knowledge of an initial stabilizing control policy. Instead, the value iteration method starts from an arbitrary real symmetric and positive definite matrix. The stability and convergence analysis of the algorithm are completed by the Lyapunov direct method. Finally, the efficacy of the radial basis function neural network based observer and the value iteration method are demonstrated through case simulation.