Self-supervised learning with high-stable guidance law and label generation for USV trajectory tracking control

Unmanned Surface Vehicle (USV) Systems confront significant challenges in achieving precise trajectory tracking, primarily attributed to their high coupling, nonlinear relationships, and external disturbances from environmental factors such as winds and currents. Addressing these obstacles is imperative for advancing the autonomy and performance of USVs. This paper introduces a self-supervised learning (SSL) based framework for USV trajectory tracking control. Firstly, we propose an adaptive look-ahead distance, which enhances the guidance law that exhibits remarkable stability, even at minimal look-ahead distances. Therefore, elevating the upper limit of guidance performance. Secondly, leveraging this refined guidance law, we develop a novel control label generation methodology specifically designed for USV trajectory tracking applications. This methodology facilitates the training of controllers via self-supervised learning, thereby circumventing the need for extensive and labor-intensive manual labeling processes. Finally, the proposed method is tested in multiple tracking scenarios, including simple and complex trajectories, and compared with the previous state-of-the-art (SOTA) approach. Simulation results demonstrate its effectiveness in achieving accurate trajectory tracking control for USVs.