Rise of the machines: crash experiences of highly automated vehicles and human drivers

Teoh, Eric R. / Kidd, David G. / Riexinger, Luke E.
Insurance Institute for Highway Safety
July 2026

Abstract
Introduction: As highly automated (SAE Level 4 [L4]) vehicles accrue more mileage and operate in more areas, it is important to understand how their crash involvement rates compare with human drivers. Doing so is challenging, largely because of differences in crash-reporting requirements and practices.
Method: Data on L4 vehicles’ crashes were obtained from federally required reports on crashes involving automation. Human drivers’ crash involvements were obtained from state databases of police-reported crashes. Regional data on vehicle miles traveled (VMT) were obtained from the Federal Highway Administration and from Waymo. Incident narratives about driverless vehicle crash involvements were manually coded to determine whether a reasonable person would have reported such an incident to police and the crash type and circumstances. Rates of police-reportable Waymo L4 crash involvements and human driver police-reported crash involvements per million VMT were computed and compared.
Results: Overall, 22% of L4 crash involvements were deemed police-reportable or maybe police-reportable, and two thirds of these occurred in driverless operation. L4 vehicles were unlikely to be the striking vehicle or primary contributor to crashes. Waymo L4 vehicles in driverless operation had police-reportable crash involvement rates 68% lower than human drivers in the same areas and years. Waymo’s rate of rear-ending another vehicle was 91% lower; its vehicles’ rate of being rear-ended was 40% lower. These results apply only to the L4 vehicles in driverless operation that were studied.
Conclusions: The results provide further evidence that Waymo’s current L4 vehicles have lower crash involvement rates than human drivers. The method of coding narratives is unsustainable as these deployments and the number of incidents continue to grow. Practical application: This study identifies ways in which national crash and VMT data collection for L4 vehicles can be improved for more timely and accurate safety evaluations.
Abstract Introduction: As highly automated (SAE Level 4 [L4]) vehicles accrue more mileage and operate in more areas, it is important to understand how their crash involvement rates compare with human drivers. Doing so is challenging, largely because of differences in crash-reporting requirements and practices.
Method: Data on L4 vehicles’ crashes were obtained from federally required reports on crashes involving automation. Human drivers’ crash involvements were obtained from state databases of police-reported crashes. Regional data on vehicle miles traveled (VMT) were obtained from the Federal Highway Administration and from Waymo. Incident narratives about driverless vehicle crash involvements were manually coded to determine whether a reasonable person would have reported such an incident to police and the crash type and circumstances. Rates of police-reportable Waymo L4 crash involvements and human driver police-reported crash involvements per million VMT were computed and compared.
Results: Overall, 22% of L4 crash involvements were deemed police-reportable or maybe police-reportable, and two thirds of these occurred in driverless operation. L4 vehicles were unlikely to be the striking vehicle or primary contributor to crashes. Waymo L4 vehicles in driverless operation had police-reportable crash involvement rates 68% lower than human drivers in the same areas and years. Waymo’s rate of rear-ending another vehicle was 91% lower; its vehicles’ rate of being rear-ended was 40% lower. These results apply only to the L4 vehicles in driverless operation that were studied.
Conclusions: The results provide further evidence that Waymo’s current L4 vehicles have lower crash involvement rates than human drivers. The method of coding narratives is unsustainable as these deployments and the number of incidents continue to grow. Practical application: This study identifies ways in which national crash and VMT data collection for L4 vehicles can be improved for more timely and accurate safety evaluations.