Artificial Intelligence-Empowered Unmanned Surface Vehicles: A Review and Prospects

Unmanned surface vehicles (USVs) constitute a critical component of maritime unmanned systems and serve as essential platforms for missions including maritime safety and security, resource exploitation, and scientific exploration. As maritime missions grow increasingly complex, classical approaches grounded in physical models, expert rules, geometric search, and classical optimization may face limitations in handling coupled multi-constraint problems, cross-scenario adaptation, and model mismatch conditions, rendering them insufficient to independently satisfy the demand for high autonomy and reliability in USVs. In recent years, advances in artificial intelligence techniques, particularly deep learning and reinforcement learning, have opened new pathways toward intelligent operation of USVs in complex marine environments. Taking USVs as the object of review, this paper examines the applicability boundaries of classical methods and the functional mechanisms of AI approaches, organized across four dimensions: environmental perception, decision-making and planning, autonomous control, and sim-to-real transfer with evaluation. On this basis, the paper synthesizes key open challenges in environmental adaptability, decision interpretability, safety and reliability, sim-to-real fidelity, and engineering deployment, and outlines future research directions, with the aim of providing a reference for intelligent USV research.