Analyzing Real-world Accidents for Test Scenario Generation for Automated Vehicles

Intersection two-vehicle crash scenario specification for automated vehicle safety evaluation using sequence analysis and Bayesian networks

“…Therefore, such a scenario library would complement scenario generating methodologies developed based on vehicle kinematic data sources such as naturalistic driving data and crash reconstruction data. Comparing with previous efforts in systematically generating scenarios for AV testing using crash data, such as the work by Nitsche et al (2017), Sander and Lubbe (Sander & Lubbe, 2018), and the Safety Pool scenario library of Warwick University (Esenturk et al, 2021(Esenturk et al, , 2022, the approach we proposed in this paper is a significant improvement. That is because 1) the crash sequence modeling captures the interactive dynamics and 2) the Bayesian network provides a comprehensive but interpretable representation of the relationships between all types of factors involved in crashes.…”

“…Prior efforts in developing test scenarios using historical crash data have developed characterization of crashes to be used as representative scenarios for the evaluation of ADAS or ADS. The crash characterization was developed by summarizing and mining patterns in crash attributes (Najm et al, 2007;Nitsche et al, 2017;Sander & Lubbe, 2018;Sui et al, 2019;Watanabe et al, 2019;Esenturk et al, 2021Esenturk et al, , 2022. The end products from prior efforts -representative scenarios -lack considerations of crash progression, dynamics, and mechanisms, which are important information to distinguish crashes and their outcomes (Song et al, 2021;Wu et al, 2016).…”

Leveraging the GIDAS Database for the Criticality Analysis of Automated Driving Systems

“…Te previous work of Esenturk et al [26,27] analyzed the UK accident data collection STATS19 (https://www.data. gov.uk/dataset/cb7ae6f0-4be6-4935-9277-47e5ce24a11f/ road-safety-data) with the goal of generating valuable test scenarios for ADSs.…”

Vehicle-to-Everything (V2X) in Scenarios: Extending Scenario Description Language for Connected Vehicle Scenario Descriptions

“…There exist a number of scenario generation methods: data-driven approaches (e.g. accident analysis [6]) and knowledge based approaches (hazard analysis [3] or ontological approaches [7]). However, in order to communicate with various stakeholders in the ADS landscape, a common language for scenario definition is required.…”