Crash Patterns in Level 2 & 4 Autonomous Vehicles: Exploratory & Clustering Analysis

Abstract: Autonomous driving systems have brought revolutionary changes to traffic safety and Intelligent Transportation Systems (ITS). As autonomous vehicles (AVs) continue to evolve and reach higher levels of automation, there has been an increase in reported AV-related crashes. Unlike traditional crashes, AV crashes involve distinct factors due to varying levels of automation, necessitating a comprehensive analysis to understand their complexity. While previous research investigated AV crashes, most of them addressed very small datasets involving limited geographical study areas and did not include multi-level automation systems. This study addresses these gaps by analyzing SAE Level 2 and Level 4 AV crash data across the USA through exploratory analysis and clustering. The results show exciting insights into AV crash trends of different autonomous levels considering weather factors, road surface, speed limit, pre-crash condition, etc. Key findings show that most crashes occur during midday to evening rush hours (1 PM to 5 PM) and from July to December. Additionally, approximately 85% of crashes occurred in clear weather and daylight conditions. This study highlights significant variables like pre-crash movement, driving speed, speed limits, and road conditions, presenting their relationships across multiple clusters. These insights provide policymakers and stakeholders with a clearer understanding of crash trends for Level 2 and Level 4 AVs, offering valuable guidance for improving system designs and implementation strategies to enhance traffic safety sustainably.