Reinforcement learning-driven autonomous navigation strategy for rotor-assisted vehicles via integral event-triggered mechanism

Abstract

To exploit wind resources for maritime decarbonization, this study proposes an autonomous navigation strategy driven by reinforcement learning (RL) for rotor-assisted vehicles (RAVs) equipped with rotor-sails—a pivotal wind-assisted propulsion technology. For achieving the optimal solution in each backstepping step, a concise and robust path following optimization algorithm using RL is designed in the form of actor-critic neural networks (AC-NNs), where the actor NNs generate control commands through real-time state feedback and the critic NNs perform state value assessment based on environmental information. Furthermore, an integral dynamic variable threshold event-triggered mechanism dependent on output state errors is proposed to avoid excessive actuator wear caused by frequent updates and transmission of control commands. Finally, theoretical validation and numerical experiments are illustrated to confirm the viability of the proposed RL-driven navigation strategy and its significant contribution to energy conservation in maritime transport.