Exploring the endogeneity between the autonomous vehicle takeover and crash severity: comparative analysis of structural equation modeling and generalized linear logit model.

Objectives: Understanding the factors influencing crash severity of autonomous vehicles is important for increasing road safety. This study focuses on a multi-source accident dataset of vehicles equipped with autonomous driving systems to explore the endogenous relationship between manual takeover of autonomous vehicles and the severity of crash, as well as the influencing factors.

Methods: By screening and summarizing data on autonomous vehicle accidents. We choose self-driving car takeover and crash severity as potential variables to build a structural equation model to explore the influences of crash severity through continuous variable updating and path improvement. We select autonomous vehicle takeover and crash severity as potential variables and designed a structural equation model to explore the factors affecting crash severity through continuous variable updating and path improvement. Meanwhile, we establish a generalized linear logit model to analyze the factors affecting manual takeover. Finally, the intrinsic link between crash severity and manual takeover is discussed through path analysis and comparison of model results.

Results: Cloudy and rainy weather, left rear of vehicle contact area, and daylight lighting significantly impact manual takeover and crash severity. Specifically, wet road surface, rainy weather, and daylight have relatively more significant effects on takeover in the structural equation model. And takeover, roadway type including non-freeway and intersection can significantly impact crash severity. Additionally, the study demonstrates the endogeneity between crash severity and takeover at the time of autonomous vehicle crash.

Conclusions: This study analyzes the potential relationships and influencing factors between takeover events of autonomous vehicles and crash severity. It is found that the frequency of takeover events significantly increases when driving in rainy weather and at night. It is suggested that a real-time monitoring module for adverse weather or lighting conditions should be added to the autonomous driving system to provide early warnings and reduce the occurrence of takeover events, thereby enhancing the safety and reliability of autonomous vehicles.