Identification of hazardous encounter scenarios using AIS data for collision avoidance system testing

The rise of artificial intelligence and advanced automation techniques have supported the development of Maritime Autonomous Surface Ships (MASS). Countries and companies are competing and collaborating to become leaders in this arising market. The Collision Avoidance System (CAS) replicates the human operator with its decision-making ability to ensure navigational safety of MASS. The CAS employs advanced algorithms to implement a wide spectrum of functions from collision avoidance to route optimization. However, the verification of the CAS dependability is highly reliant on the coverage of implemented scenarios during testing, which directly influences its trustworthiness. Scenarios in previous research from manually designed approaches have a limited coverage, while those from simulation-based approaches based on algorithms are disconnected from the scenarios occurring in the actual operational contexts. Others from real data-based approaches using Automatic Identification System (AIS) data propose an unbearably large number of scenarios. Considering that critical risk scenarios can constitute the basis for the development of CAS testing, this study proposes a method for identifying critical encounter scenarios based on AIS data. The method uses safety indices to identify hazardous encounter scenarios. Then, a muti-ship encounter scenario classification method based on COLREGs is proposed to categorize these scenarios. For each category, the risk value of each scenario is evaluated by Time-varying Risk Vectors (TRV). Scenarios with the lowest and highest risk are then used as representative for the whole. In this study, AIS data from Singapore Strait covering one month of operation is used for scenario identification. The results are discussed indicating good effectiveness in identifying critical scenarios in water area.