Reliability of maritime collision avoidance systems algorithms in the implementation of COLREGs
本文提出一个验证框架和分类矩阵，评估了855篇海上避碰系统论文，发现没有算法完全符合《国际海上避碰规则》的12项关键操作规则，并识别出29种反复出现的实施错误，对海事监管和船级社有重要参考价值。
• Proposed a verification framework to formally assess algorithm compliance. • Introduced a classification matrix for MCPS based on COLREGs compliance levels. • Evaluated 855 MCPS papers; only 464 attempted to implement COLREGs rules. • No algorithm achieved full compliance with key COLREGs operational rules. • Identified 29 recurring implementation errors compromising navigational safety. Maritime Collision Prevention Systems (MCPS) are being developed to enhance navigational safety and mitigate human error, yet their operational trustworthiness is contingent upon strict adherence to the International Regulations for Preventing Collisions at Sea (COLREGs). To evaluate these possible systemic deficiencies, this paper introduces a two-tiered contribution: (1) a structured verification framework with explicit indicators to formally assess COLREGs compliance, and (2) a novel regulatory classification matrix to categorise algorithms based on their demonstrated capabilities. With these tools the regulatory compliance in MCPS algorithms was evaluated, a frequently overlooked aspect of system safety. The analysis reveals a critical and systemic failure: no algorithm achieved full compliance with the 12 key operational rules evaluated. Compliance was null for rules governing geographical restrictions and vessel priorities, and critically low for common encounters. Furthermore, 29 recurring error patterns were identified, stemming from fundamental misinterpretations of navigational parameters that directly compromise safety. This work provides a robust methodology for the verification and validation of algorithms, useful for administrations and classification societies, laying the foundations for future autonomous systems.