Verification of MASS collision avoidance systems ABSTRACT using a distribution-driven generating model of collision risk scenarios

Abstract

ollision-risk interactions are rare in real-world 
maritime operations, which makes it difficult to secure a 
large enough set of cases to validate collision avoidance 
systems in maritime autonomous surface ships. 
Expanded collections of risk scenarios that represent 
real encounter patterns must be methodically created in 
order to get around this restriction. In this study, 
collision-risk encounters produced from AIS are 
arranged into a Bag-of-Encounters representation, and 
probability density functions are used to characterize 
the features of comparable encounter groups. We 
suggest a scenario generating technique that can 
generate situations that strike a compromise between 
realism and variability by probabilistically sampling 
from these distributions. The distribution-based 
approach successfully replicates the statistical 
characteristics of actual encounters, as evidenced by the 
created scenarios exhibiting patterns consistent with 
their original scenarios. In order to improve the realism 
and applicability of MASS collision-avoidance 
algorithm verification, this study ultimately presents a 
data-driven scenario generation framework that 
maintains the underlying distribution of actual 
encounters while making up for the scarcity of 
empirical data.