Ship encounter scenario generation for collision avoidance algorithm testing based on AIS data

Abstract

Collision avoidance algorithms play a crucial role in assisting onboard officers and ensuring navigation safety. To ensure the effectiveness of these algorithms, comprehensive testing with a wide range of scenarios is necessary before widespread deployment. However, there are currently few systematic methods available for generating such testing scenarios. In this paper, a general framework is proposed to generate testing scenarios based on Automatic Identification System (AIS) data, consisting of three modules: encounter scenario extraction, scenario importance evaluation, and testing scenario sampling. Firstly, ship encounter scenarios are extracted from the AIS data according to spatial-temporal proximity relationships between ships. Subsequently, a two-factor model is developed to evaluate the importance of extracted scenarios, considering both exposure frequency and collision risk. Finally, the unequal probability sampling method, which takes into account the importance of each scenario, is utilized to construct the testing scenario set. The proposed framework is validated through a systematic experiment using real AIS data. The results demonstrate that the framework effectively extracts diverse encounter scenarios from AIS data and rationally evaluates their importance. On this basis, an appropriate scenario set can be generated for collision avoidance algorithm testing. The generated scenario set ensures comprehensive coverage while prioritizing high-importance scenarios.