Cooperative collision avoidance decision-making for intelligent ships in intersection waters driven by enhanced large language models

Objectives To address the complex cooperative collision avoidance problem among multiple intelligent ships in intersection waters, this paper proposes an enhanced large language models(LLM)-driven decision-making method for multi-ship cooperative collision avoidance in intersection waters. Methods By analyzing the navigation characteristics of ships in intersection waters, the multi-ship cooperative collision avoidance problem is modeled as a Partially Observable Markov Decision Process (POMDP). A central-distributed dual-layer decision architecture is designed: Central layer: An LLM coordinator collects multi-ship situational information, integrates navigation rules, and conflict severity to determine passage priority sequences； Distributed layer: LLM-empowered intelligent ships perform chain-of-thought prompting-based progressive decision-making, synthesizing scenario descriptions, coordination instructions, and navigation experience to generate collision avoidance strategies. To overcome the inherent limitations of LLM in precise computation and continual learning—and to mitigate their potential hallucination risks—we integrate a decision‐augmentation module to enhance the LLM’ decision‐making capabilities. Results Simulation experiments demonstrate that the proposed enhanced LLM-driven method implemented in DeepSeek-v3 achieves safe and efficient cooperative collision avoidance in typical intersection scenarios involving two, three, and four ships. The system maintains a minimum maneuvering speed of over 3 knots throughout and ensures a safety margin exceeding twice the ship length Conclusions This method advances the engineering application of LLMs in maritime decision-making and provides a new pathway for realizing highly autonomous shipboard artificial intelligence in complex operational environments.