HARNESSING CRASH AND DISENGAGEMENT DATA TO ANALYZE THE SAFETY PERFORMANCES OF AUTOMATED VEHICLES

Highly automated vehicles or autonomous vehicles are projected to reform the transportation system and improve transportation safety by eliminating drivers’ errors, as safety studies reveal humans contribute to more than 90% of the crashes. This dissertation examines the full spectrum of automated vehicle topics, including related literature, enacted state policies, and manufacturer-reported crash and disengagements reports to understand more about this emerging and transformative technology. First, an extensive review of safety-related literature was conducted, followed by a text analysis to identify areas of research needed to advance the knowledge of automated vehicles. California is the only transparent state as the Department of Motor Vehicles requires permit-holding manufacturers to publicly share automated vehicle safety performance data – crashes, disengagements, and automated vehicle miles traveled – of the vehicles on the public roadways. By harnessing this significant and emerging field data, associated risk factors of crashes and disengagements were identified and quantified using rigor statistical analyses. Contributing roadway, environmental, and vehicle factors were modeled with the most frequent type of crash, rear-end, and stated injury crashes. Disengagements, the transition from autonomous mode to conventional mode, are considered near-miss crashes as the vehicle operator is mitigating the likelihood of a crash by regaining control of the vehicle. These safety-critical events of disengagements were then identified and quantified by using the 5 W’s – who (disengagement initiator), when (the maturity of automated driving system, where (location of disengagement), what and why (the facts causing the disengagement). The assessment of the effects of time on disengagement frequency at the manufacturer- level was conducted. With the rapid technological advancements in automated vehicles, the technology poses a threat to crafting legislation. By using a novel approach of conducting a manual review and synthesis against the only federal guidance available for states, USDOT Model State Policy, married with a text mining analysis, insights on uneven research, development, and deployment of automated vehicle technologies at the state level were revealed. Finally, the implications of the findings and future research areas are discussed to a comprehensive understanding of automated vehicle associated risk factors on the transportation network.