Objectives In view of the problem of insufficient autonomy and coordination of unmanned multi-agent collaborating on the surface environment, a digital twin-based multi-unmanned agent collaborative cruising strategy optimization method is proposed.

Methods Firstly, according to the construction of digital twin model of unmanned multi-agent physical entity, the mathematical model of unmanned multi-agent collaborative cruising is established to analyze the motion characteristics of the collaborative process of unmanned multi-agent. Then, aiming at the mutual influence and collaborative relationship of unmanned multi-agent, the unmanned multi-agent proximal strategy optimization algorithm is adopted to carry out the strategy optimization of unmanned multi-agent, and improve the cooperative cruising efficiency among unmanned multi-agent. Finally, the proposed water-surface multi-agent digital twin model is utilized to validate the training effectiveness of multiple autonomous agents.

Results Compared with multi-agent deep deterministic policy gradient (MADDPG), the proposed (muti-agent proximal policy optimization, MAPPO) algorithm achieves a 14.7% improvement in average reward and demonstrates more stable convergence. The unmanned agents are capable of forming a uniformly distributed formation centered around the patrol target, thereby providing more comprehensive information for coordinated cruising.

Conclusions The study provides significant theoretical and practical support for the research on the optimization of cooperative cruising strategies of unmanned multi-agent on the water surface.