Defining Safety-Critical Scenarios for Simulation-Based Automated Vehicle Evaluation
Simulation-based evaluation of automated vehicles (AVs) is an essential part of making sure that AVs meet certain safety standards before being allowed to operate widely on public roads. Defining scenarios for simulation-based evaluation is key and challenging. The objective of this study is to make use of publicly available historical human-driven vehicle (HDV) crash databases and identify safety-critical scenarios AV evaluation. This study will identify both common and the rare safety-critical scenarios that can be used for AV safety evaluation.
Clustering will be applied as the primary method to identify groups of crash cases. The common safety-critical scenarios are those larger clusters with more cases and high severity levels. The rare safety-critical scenarios are those smaller clusters (or outliers) with fewer cases and high severity levels. The functions and capabilities of AVs will be considered in the clustering process to ensure the applicability of identified scenarios to AV safety evaluation. The perception-reaction and decision-making mechanism of AVs are different from human drivers. It is expected that AVs have a much wider range of perception, a much faster reaction, and can make their decisions more rationally. Thus, these aspects of the AVs are what need to be tested with the safety-critical scenarios. In developing the scenarios, input variables related to the challenges for these aspects of the AVs will be included in the cluster analysis.