Exploration on prior driving modes for automated vehicle collisions

ABSTRACT

The emergence of automated vehicles (AV) has been occurring rapidly as these vehicles have the potential to reduce/eradicate human driving faults and related collisions. To enhance AV safety, the NHTSA recommends the continuous presence of a backup human driver that has a reasonable understanding of AV technologies to ensure disengagement when manual overtake is required. However, due to several AV-related traffic crashes during roadway testing and extensive media interest, AV safety has become a critical issue. This study collected 255 crash reports filed by different manufacturers testing AVs in California from September 2014 to April 2020. The crash dataset was analyzed using two data mining algorithms (association rule mining and text network analysis) to identify the key AV-related crash attributes and their associations based on the vehicle’s prior driving mode (conventional or automated). The results show that the manner of collision and the prior movement of the testing vehicle are strongly connected with prior driving mode. For example, AV crashes in a manual driving mode often result in a sideswipe collision during moving status, whereas AV crashes in an automated driving mode are highly associated with rear-end collisions when AVs are stopped in traffic. The findings of this study can help policymakers and AV engineers improve AV deployment strategies to support the adoption of AVs and promote potential safety benefits for AV technologies.

KEYWORDS:

• Automated vehicle
• crash pattern
• crash data analysis
• prior driving mode
• data mining

Acknowledgements

The authors like to thank two anonymous reviewers for their excellent feedback.

Disclosure statement

No potential conflict of interest was reported by the author(s).