Automated vehicle crash sequences: Patterns and potential uses in safety testing

Area of Science:

• Road Safety
• Autonomous Systems
• Transportation Engineering

Background:

• Automated vehicles (AVs) require rigorous safety testing.
• California Department of Motor Vehicles (DMV) collects AV collision and disengagement reports.
• Understanding crash sequences is crucial for AV safety.

Purpose of the Study:

• To investigate patterns in AV crash sequences using California DMV data.
• To explore the use of these patterns in developing AV test scenarios.
• To propose a scenario-based AV safety testing framework.

Main Methods:

• Sequence analysis and clustering of 168 AV crashes (2015-2019).
• Extraction and analysis of temporal sequences of events leading to AV collisions.
• Cross-tabulation analysis of crash sequence groups with outcomes and environmental conditions.

Main Results:

• The most representative crash pattern was 'collision following AV stop'.
• Disengagement preceded collision in 24% of crashes, with a 68% transition probability to immediate collision.
• Seven distinct AV crash sequence groups were identified, associated with crash outcomes and conditions.

Conclusions:

• AV crash sequences provide valuable insights for developing realistic test scenarios.
• A scenario-based safety testing framework incorporating event sequences is proposed.
• Analyzing crash event sequences enhances the safety validation of automated vehicles.