A detailed analysis of California Department of Motor Vehicle crash records examines the most frequent types of automated vehicle crashes.
Crash Themes in Automated Vehicles: A Topic Modeling Analysis of the California Department of Motor Vehicles Automated Vehicle Crash Database
Researchers at Texas A&M University analyzed one of the most comprehensive crash databases in the nation, the California Department of Motor Vehicles Database (CDMV) to better understand what types of crashes PAVs are most commonly involved in. The CDMV crash database is particularly robust and contains extensive detail about the crashes recorded. The database contained records relating to 167 PAV crashes, recorded between October 2014 and June 2019. All these crashes occurred in or around the San Francisco Bay Area. Of these 167 crashes, the team used 114 crashes in their final analysis as they eliminated any crashes where the driver was operating the vehicle in manual mode.
After cleaning and preparing the data, the research team used a form of semantic analysis called probabilistic topic modeling to identify themes in each of the crashes and then classify these crashes based on the identified themes.
- Further research, testing, and safety features should be developed to assist drivers as they transition from automated to manual control of a vehicle. Analysis of automated vehicle crashes in California shows that 30 percent of crashes occur while transitioning from automated to manual control or vice versa. Of the crashes associated with transitioning, the majority occurred when the driver initiated a transition and not when the vehicle initiated the transition. These findings may partially validate concerns over the extensive workload placed on drivers in automated mode who are tasked with monitoring both challenging roadway conditions and the automated systems.