Resilient train formation control for ASVs under intermittent communication environment via actor–critic reinforcement learning

Abstract

To address the train formation task of autonomous surface vehicles (ASVs) operating under intermittent communication environment, this paper investigates the problem of resilient formation robust control for ASVs. First, a resilient distributed leader predictor (RDLP) is designed. By introducing a time-varying adjustment mechanism with communication state awareness into the communication topology, autonomous estimation and dynamic compensation of the leader’s states during communication interruptions are realized. Meanwhile, a resilient train formation mechanism is integrated to ensure smooth waypoint-based course transitions for each ASV. In the control module, a reinforcement learning (RL) controller based on actor–critic neural networks (AC-NNs) is constructed, which uses the critic NN to evaluate the cost function and drives the actor NN to optimize control strategies online, thereby adaptively compensating for model uncertainties and environmental disturbances. Theoretical analysis demonstrates that the proposed algorithm guarantees the convergence of prediction errors under intermittent communication conditions and ensures the semi-global uniform ultimate bounded (SGUUB) stability of closed-loop ASVs system. Finally, two sets of numerical simulations verify the effectiveness and superiority of the proposed strategy.