Explainable Artificial Intelligence for Autonomous Surface Vessels by Fuzzy-Based Collision Avoidance System

Autonomy at sea relies on algorithms (often local) to make the decisions. One approach to create these algorithms are through the use of artificial intelligence. Numerous black-box machine learning-based algorithms are proposed for autonomous surface vessels (ASV) to make decisions like changing the speed or changing the route in order to reach the operational goal in an optimal way with respect to cost (fuel, time, etc.) and safety (avoid collisions or dangerous situations). Hence, the algorithms must take into account many constraints and are influenced by several varying factors such as other vessels, weather, etc. The objective of this paper is to propose a model that provides the reason behind the ASV’s decision when it is on a predefined path and changes speed or route. Fuzzy logic is used to record the expert knowledge based on COLREGs to steer the vessel and take the decision during the collision course. Data has been captured based on expert knowledge and used to train an explainable model. The explainable model predicts the reason behind the decision. The focus of the paper is on local explainability instead of global decisions. The structured abstracts of the paper are (1) Background: Several AI-enabled algorithms have been proposed for implementing autonomy to avoid the collision. These black-box techniques provide good predictions at the same time, they fail to explain the reason behind the decision, which makes the model less trustworthy; (2) Methods: Expert knowledge (COLREGs) has been captured using fuzzy rules and applied when ASV progresses, the decision has been recorded. (3) Results: The explainable model provides the reason behind the action taken by the collision avoidance system; (4) Conclusion: A model has been proposed that explains the collision avoidance system to make it transparent and trustworthy.