Future of autonomous vehicles: spatiotemporal insights from collision data

OBJECTIVE: With the recent expansion of Autonomous Vehicles (AV) testing on open roads, observations from AV collision data will likely guide public acceptance and large-scale adoption of the technology. This study sits at the cusp of a recent release of AV collision data spanning the United States. The dataset covers high-automation driving systems (ADS) as well as low-automation advanced driver assistance systems (ADAS). METHODS: The study aims to analyze the spatiotemporal trends in the observed collisions using Moran's I statistical test and time-series modeling to predict future collision trends. Further investigations are also conducted using the ADS collision data reported in California to utilize the exclusive mileage data as an exposure measure. RESULTS: The frequency of ADS collisions seems to increase at a faster rate than ADAS collisions, which reflects the rapid increase in market share. Spatial analysis revealed clusters of increasing ADS and ADAS collisions, including California and Arizona. Still, several states showed a reduction in collisions over time compared to neighboring states, including Florida and Wyoming in terms of ADS collisions, and New York, Vermont, and New Mexico in terms of ADAS collisions. Time-series modeling showed ADS collision spikes occurring repeatedly between April and June. Collision forecasts further indicated that ADS collisions will likely exceed ADAS collisions toward the end of 2025. Data from California indicated that despite the increase in ADS collisions, the rate of collisions per miles traveled will likely remain stable. CONCLUSION: Overall, the reliable reporting of AV collisions across the US provided a wealth of information that was explored in this study. Continued efforts to track AV collision trends and investigate their unique characteristics are essential to improve the safety of network-wide deployment and foster public support.