PID-Based Target Tracking for Underactuated ASVs With Online Reinforcement Learning

This article studies the target tracking problem for uncertain underactuated autonomous surface vehicles with unknown target velocity. The proposed solution consists of three key components. First, a predefined-time observer is developed to estimate the target velocity. Second, a predefined-time guidance law is designed to generate the desired surge velocity and yaw angle. By integrating these two components, we establish a unified predefined-time observer-guidance framework that enhances the stability of the reinforcement learning (RL) process. Finally, based on the predefined-time observer-guidance framework, an adaptive proportional-integral-derivative controller is proposed to ensure effective tracking performance, with its parameters dynamically optimized through an actor-critic based online RL algorithm. The effectiveness of the proposed control scheme is validated through simulation and experimental results.