Reinforcement learning event-triggered energy-based control for unmanned surface vessel with disturbances

Abstract

This paper proposes a novel event-triggered energy-based controller for Unmanned Surface Vessels (USVs) operating in complex scenarios, integrating reinforcement learning techniques with an energy-based framework. Model uncertainties are captured via actor-critic neural networks (NNs), where actor NNs generate control actions and critic NNs assess their performance. To address disturbances, a self-learning nonlinear disturbance observer with an adaptive learning factor is developed, enhancing the accuracy of disturbance estimation. A state-error port-controlled Hamiltonian (PCH) strategy ensures trajectory tracking, complemented by variable damping techniques to optimize the closed-loop system’s dynamic response. The design incorporates event-triggered mechanisms and adaptive control methods to ensure boundedness of all closed-loop signals. Stability analysis demonstrates convergence of the tracking error to a neighborhood of the origin, and simulation results validate the controller’s feasibility and efficacy.