Navigating tomorrow: Designing a Simulator Scenario for Generation of AI Training Data

This paper presents a model for creating a simulator scenario aimed at generating training data for an artificial intelligence (AI) to learn good seamanship, as referred to by the International Maritime Organization's Collision Regulations (COLREGs). The model is used to design a scenario for collecting high quality data intended for the training of the AI in the AI-Navigator project, which aims to develop an Advanced Navigation Assistance System (ADNAV) that is compliant with COLREGs, including the principles of good seamanship. ADNAV is intended to guide ship navigators by providing them with relevant options in demanding situations where cognitive overload is a risk. The model suggested in this paper ensures a simulation that challenges human navigators to demonstrate proactive decision-making, situational awareness, and contextual compliance, which are all identified as key components of good seamanship. The three key phases of the model are: objective identification, scenario design, and testing, each critical for ensuring the efficiency of a scenario and the realism of the situations. This study contributes to ongoing efforts in maritime safety research, exploring the synergy between human expertise and AI support in navigating increasingly congested and complex waterways.