Generalised Policy Learning Based Adaptive Optimal Tracking Control of Nonlinear Servomechanisms

This paper develops an optimal tracking control strategy for nonlinear servomechanisms via generalised policy learning (GPL) algorithm, where the actor neural network (ANN) and an initial stabilising control policy are avoided. To this end, we first design a steady-state control to guarantee the tracking performance, while an optimal control is derived to further regulate the tracking error via minimizing the cost function. Then, a critic neural network (CNN) is used to reformulate the defined optimal cost function, this contributes to updating the critic with Bellman approach. To obtain CNN weights with guaranteed convergence, a novel adaptive law is designed via adopting the sliding mode technique. Because of its strong online learning ability, the adaptive critic and the control action update each other continuously and simultaneously without any iterative steps. Finally, simulation results are performed to verify the effectiveness of presented control method.