Distribution-driven generation model of collision risk scenarios for MASS collision avoidance system verification

In real-world maritime operations, collision-risk encounters occur infrequently, creating inherent limitations in securing a sufficiently large set of scenarios for validating collision avoidance systems in maritime autonomous surface ships. To overcome this constraint, it is necessary to systematically generate expanded sets of risk scenarios that reflect actual encounter patterns. In this study, AIS-derived collision-risk encounters are organized into a Bag-of-Encounters representation, and the characteristics of similar encounter groups are modeled using probability density functions. By probabilistically sampling from these distributions, we propose a scenario generation method capable of producing scenarios that balance realism with variability. The generated scenarios were found to exhibit patterns consistent with their original scenarios, demonstrating that the distribution-based approach effectively reproduces the statistical properties of real encounters. Ultimately, this study introduces a data-driven scenario generation framework that preserves the underlying distribution of real encounters while compensating for the limited availability of empirical data, thereby improving the realism and applicability of MASS collision-avoidance algorithm verification.