Characteristics analysis of autonomous vehicle pre-crash scenarios

To date, hundreds of crashes have occurred in open-road testing of autonomous vehicles (AVs), highlighting the need for improving AV reliability and safety. However, current studies predominantly analyze crash data based on oversimplified classification schemes that lack clear scenario definitions. Consequently, they impede an in-depth investigation of crash characteristics. Pre-crash scenario typology classifies crashes based on vehicle dynamics and kinematics features. Building on this, characteristics analysis can identify similar features under comparable crashes, offering a more effective reflection of general crash patterns and providing more targeted recommendations for enhancing AV performance. In this paper, we initially collected the latest 774 California AV crash reports, then selected 384 autonomous mode crashes, and used the newly revised pre-crash scenario typology to identify AV pre-crash scenarios. To improve the efficiency of scenario identification and adaptability to future updates in scenario typology, we proposed a set of mapping rules to extract pre-crash scenarios automatically. We successfully identified 27 types of AV pre-crash scenarios with an accuracy of 98.1%. Through detailed analysis, we obtained two key groups of AV pre-crash scenarios: rear-end scenarios and intersection scenarios. Based on the abundance of crash data, we adopted different analysis methods to analyze the features of key scenarios. Association analysis of rear-end scenarios showed that the significant environmental influencing factors were traffic control type, location type, light, etc. For intersection scenarios prone to severe crashes with detailed descriptions, we employed causal analysis to obtain the significant causal factors: habitual violations and temporary obstruction of view. The extracted scenarios in this paper and their features can assist in constructing the AV simulation test with precise environmental parameters and realistic interactions with other traffic parties. The resulting optimization recommendations can inform regulators and reveal control-algorithm weaknesses across diverse real-world conditions, thereby enhancing the AV safety.