Autonomous Control of Lift System based on Actor-Critic Learning for Air Cushion Vehicle

In this article, considering the autonomous lift problem of air cushion vehicle (ACV), an optimal control method based on reinforcement learning (RL) is investigated. A neural network (NN) observer is first developed to estimate the unknown state and uncertain nonlinear part. Then a kind of non-quadratic utility function and Hamilton-Jacobi-Bellman (HJB) equation is described to guarantee long-term system performance index and optimal control input, simultaneously. Critic and action NNs are designed to approximate the unavailable non-quadratic utility function and HJB equation. Rigorous theoretical analysis is provided to show the uniformly ultimately bounded (UUB). Finally, the Simulation results are provided to verify the effectiveness of the devised method.