In-depth investigation for identifying autonomous vehicle crash causations: New insights from system functions,...

Understanding the root causes of autonomous vehicle (AV) crashes is crucial for enhancing road safety and advancing the commercialization of autonomous driving. This study aims to investigate the relative contributions of system functions, driver behaviors, and kinematics to AV crash causations, with a focus on how they affect crash injury severity. To accurately capture the thorough growth processes of crashes near injuries, we analyzed 19 AV crash cases obtained from the Traffic Accident Digital In-depth Investigation and Research in China (TADIIRC). A unified methodology was used to categorize crashes into distinct scenarios, with a focus on the typical vehicle-obstacle crash scenario for detailed analysis. Injury severity of the incident, classified as none, possible, or serious, was adopted as the dependent variable. An ordered logit/proportional odds model was employed to identify critical factors influencing injury severity. The results showed that the most frequent crash cluster was "other" (42.11 %), followed by longitudinal (26.32 %), crossing (26.32 %), and turning crashes (5.26 %). Furthermore, the modeling results revealed that injury severity was significantly associated with critical risk factors, such as autonomous emergency braking (AEB), adaptive cruise control (ACC), brake pedal force, and longitudinal acceleration. The model identified AEB (Coefficient = -16.343, p < 0.05), ACC (Coefficient = -20.816, p < 0.05), brake pedal force (Coefficient = -0.032, p < 0.05), and longitudinal acceleration (Coefficient = -1.160, p < 0.01) as significant predictors, highlighting their importance in mitigating injury severity. The findings improve our understanding of AV crash causal mechanisms and provide new insights for developing safer autonomous driving technologies.