Supplemental Control of Hydraulic Systems with Adaptive/Approximate Dynamic Programming

For reinforcement learning based tracking control which utilizes neural network approximations, the critical issue is how to search the solution quickly from the huge spaces to enable system states follow specific reference trajectories. We therefore develop backstepping scheme guided reinforcement learning supplemental control for hydraulic systems in this paper. In this algorithm, backstepping feedback control is designed firstly to ensure the roughly tracking result, then adaptive/approximate dynamic programming (ADP) is synthesized to compensate unknown system dynamic in backstepping framework. This proposed control is data driven, not model based, and only reference output trajectory is needed to fulfill the tracking task. By using Lyapunov approach, the stability analysis is accomplished, which shows that the proposed algorithm can guarantee uniformly ultimately boundedness result.