CORALL: A COLREGs-Guided Risk-Aware LLM for Decision-Making in Maritime Autonomous Surface Ships

The paper introduces a novel approach to utilising Large Language Models (LLMs) for real-time COLREGs-based decision-making in collision encounters. The COLREGs (collision regulations), published by the International Maritime Organisation (IMO), have served as baseline regulations for empirical maritime collision avoidance in open seas to ensure vehicle safety. As COLREGs were originally defined for human seafarers, translating them into computer language for autonomous operations has been extensively investigated in recent decades, attracting significant interest but lacking a universal solution. A key aspect is the need for explainable decision-making, either for autonomous operations or as captain assistance. This study leverages the power of LLMs to develop such a decision-making algorithm. The proposed COLREGs-guided Risk-Aware LLM (CORALL) algorithm consists of a high-level, risk-aware decision-maker and an execution layer that implements path tracking and collision avoidance in the presence of external disturbances. A framework for risk-aware, COLREGs-compliant LLM-based decision-making is devised, and its performance in online operation on a high-speed ship model is demonstrated. The LLM processes navigation outputs and risk indices, identifies the COLREGs encounter type, and makes decisions with accompanying explanations. The proposed solution is tested on all the 22 Imazu's benchmark problems, showcasing its effectiveness. This risk-aware AI solution, thought to be a first of its kind, opens the door to future research on advanced decision-making algorithms in maritime and similar operational environments.