Towards an approach to define transparency requirements for maritime collision avoidance

Abstract

This study discusses an approach to support human supervision of autonomous maritime collision avoidance systems by disclosing the system’s perceived information, internal reasoning, decisions, and planned actions as layers of transparency. Information requirements, identified through a cognitive task analysis, were structured using the information processing model by Parasuraman, Sheridan, and Wickens (2000). This model was contextualized to the maritime collision avoidance setting such that the information from the analysis could be structured into unique and distinct layers. A set of minimum information requirements was identified depicting the system’s decisions and planned action, supported by additional layers to reveal its internal reasoning. This approach aims at supporting humans in effectively supervising autonomous collision avoidance systems in their operational context by providing understandability and predictability about what the system is doing, why it is doing it, and what it will do next, i.e., transparency.