Implementing a dissimilarity metric for scenarios categorization and selection for automated driving systems

To ensure road traffic safety, the safety verification and validation of automated vehicles are of the utmost importance. Scenario-based testing is one of the most popular approaches, as it is cheaper, safer, and faster than on-road testing. The number of possible scenarios encountered by an automated driving system could be virtually infinite due to the complexity and uncertainty of the driving environment. Hence a framework is needed which expresses the degree of dissimilarity between two driving scenarios quantitatively. This work first develops a dissimilarity metric that compares different driving scenarios and secondly, categorizes them to identify the most critical ones in each category. This way, a finite set of non-redundant scenarios are identified which can be used for validating the safety of an automated driving system.