Adaptive trajectory tracking of the unmanned surface vessel based on improved AC-MPC method

Abstract

Trajectory tracking is one of the key technologies for ensuring safe navigation of the unmanned surface vessel (USV), whose main challenges are the low control accuracy and uncertainties caused by environmental disturbances. In this study, an adaptive model predictive control (MPC) method is proposed based on the Actor-Critic (AC) reinforcement learning strategy. First, a traditional MPC method is used for USV trajectory tracking. Then, the AC strategy is integrated to continuously adjust the state weight coefficients in the model predictive controller to address the increased error problem resulting from suboptimal control parameters. Finally, the prediction step size in the MPC is adaptively improved based on changes in reinforcement learning reward values and the selection of state-weight parameters. Simulation results show that the AC-MPC method adjusts the state-weight parameters quickly, and selects the appropriate prediction step size according to the change of the reward value. The proposed method effectively addresses the challenge of parameter adjustment when using the model predictive controller for trajectory tracking, thereby enhancing the adaptability of the controller in performing trajectory tracking tasks across various environments.