Research on Intelligent Collision Avoidance Decision-Making Algorithm for Multi-Unmanned Surface Vehicles (USVs) Based on COLREGs

Abstract

With the rapid advancement of USV technology, autonomous maritime systems are increasingly employed in marine monitoring, resource exploration, and environmental protection. However, collision avoidance remains a critical challenge, exceptionally in complex and dynamic multi-vessel maritime scenarios that demand legal compliance and real-time responsiveness. This paper proposes an International Regulations for Preventing Collisions at Sea (COLREGs)-compliant intelligent collision avoidance decision-making algorithm for multi-USV systems. The framework integrates Deep Reinforcement Learning (DRL), game-theoretic reasoning, distributed coordination, and multi-sensor fusion. A Collision Risk Indicator (CRI) is introduced, based on Distance at Closest Point of Approach (DCPA) and Time to Closest Point of Approach (TCPA), enhanced by differences in relative position, velocity, and heading, to support real-time risk assessment. To ensure regulatory adherence, the system classifies vessel encounter types (head-on, crossing, overtaking) and determines corresponding COLREGs roles (give-way or stand-on) using game-theoretic logic. These symbolic roles and encounter types are embedded into the DRL state vector as discrete semantic features. This semantic embedding enables the learning agent to generate behavior strategies that are both adaptive and COLREGs-compliant. A dynamic path planning and heading adjustment strategy is proposed, continuously modifying course and speed in response to evolving collision risk and regulatory constraints. A distributed multi-agent decision-making mechanism allows USVs to cooperate through local information exchange and decentralized negotiation. Simulation results demonstrate that the proposed approach achieves superior collision avoidance performance, faster response times, and improved compliance with COLREGs in complex maritime scenarios.