Intersection two-vehicle crash scenario specification for automated vehicle safety evaluation using sequence analysis...

Area of Science:

• Road Safety Engineering
• Traffic Accident Analysis
• Computational Modeling

Background:

• Intersection crashes pose significant safety risks.
• Existing methods for test scenario specification lack detail in crash dynamics.
• Accurate modeling of crash contributing factors is crucial for developing effective safety interventions.

Purpose of the Study:

• To introduce a novel procedure for test scenario specification using crash sequence analysis and Bayesian network modeling.
• To characterize intersection two-vehicle crashes into distinct types based on sequence patterns.
• To develop a Bayesian network model for understanding interrelationships between crash factors and outcomes.

Main Methods:

• Utilized National Highway Traffic Safety Administration (NHTSA) Crash Report Sampling System (CRSS) data (2016-2018).
• Developed a crash sequence encoding method detailing pre-crash and collision events.
• Characterized crashes into 55 types and built a Bayesian network to model crash dynamics, human factors, and environmental conditions.
• Specified test scenarios by querying the Bayesian network and analyzing operational design domain (ODD) attribute distributions.

Main Results:

• Successfully categorized intersection crashes into 55 distinct types based on sequence analysis.
• Developed a comprehensive Bayesian network model illustrating complex interdependencies.
• Demonstrated a method for specifying ODD attribute distributions based on crash sequence types and outcomes.

Conclusions:

• The proposed procedure offers a robust framework for generating realistic and informative test scenarios for autonomous and human-driven vehicles.
• Bayesian network modeling effectively captures the probabilistic relationships in crash causation.
• This approach enhances the fidelity of simulation-based testing for road safety validation.