Validation of ship intention model for maritime collision avoidance control using historical AIS data

This article tests the method for inferring and modeling ship intentions presented in [1] on real ship encounters gathered through the automatic identification system (AIS) that all larger ships are required to use. Empirical distributions on how early ships tend to perform avoidance maneuvers and how close they tend to come are evaluated. These are used by the intention model to identify when a ship’s behavior is outside normal behavior. Running the intention model on the historical ship encounters demonstrates that the intention model is able to correctly infer the intentions of ships in real collision encounters. The model is able to distinguish between different types of incompliant behavior such as a ship not giving way when it should, the wrong ship giving way, or a ship giving way in the wrong direction. Some improvement potentials are identified, mainly with respect to understanding when the situation starts.