Exploring autonomous vehicle crash risk: system coupling effects and key causal factors

Area of Science:

• Automotive Engineering
• Traffic Safety
• Artificial Intelligence

Background:

• Autonomous driving technology is rapidly advancing, setting new benchmarks for vehicle intelligence and safety.
• However, autonomous vehicle crashes have raised public and governmental concerns, necessitating a thorough analysis of crash causes.
• Understanding crash causation is vital for technological progress, occupant safety, and the future of the automotive industry.

Purpose of the Study:

• To construct an autonomous vehicle crash causation system and risk coupling metric model.
• To investigate the interplay between risk systems and crash coupling effects in complex traffic environments.
• To identify key causal factors contributing to autonomous vehicle accidents.

Main Methods:

• Utilized the AVOID global autonomous vehicle operation event dataset.
• Employed the hiking optimization algorithm, recursive feature elimination, N-K model, and Bayesian networks.
• Developed a crash causation system and risk coupling metric model.

Main Results:

• The developed crash causation system demonstrates high stability and robustness.
• Crash risk positively correlates with the number of coupled systems; key factors include non-engagement, depreciation, speeding, abrupt stops, unrestrained occupants, and environmental conditions like darkness.
• Vehicle defects and poor environmental conditions significantly increase crash risk, especially under the coupling of multiple systems.

Conclusions:

• Identified the mechanism of risk system coupling in autonomous vehicle crashes.
• Pinpointed critical causal factors contributing to autonomous vehicle accidents.
• Provided theoretical support for enhancing autonomous vehicle safety and informing traffic safety policies.