Comparing autonomous vehicles with human drivers may appear straightforward: measure crashes per mile and compare the results. In reality, the circumstances in which those miles are driven can have a major influence on risk. New research from Waymo highlights why location, road type, time of day and day of the week should be considered when evaluating automated driving safety.

The research combines human crash records with detailed traffic-volume data to create more precise benchmarks. This makes it possible to compare autonomous and human driving under similar conditions rather than relying on broad averages that can hide important differences between locations and driving patterns.

The geographical differences are substantial. Across the 50 most populous US urban areas examined, the fatal crash involvement rate for human drivers on surface streets in Memphis was 8.4 times higher than in Boston. The researchers also found that surface streets had a 2.3-times higher fatal crash rate than freeways across the regions studied.

Time matters just as much. Human crash risk was highest during late-night hours, particularly between midnight and 4 a.m. and during weekends. Although these periods represent only a small share of total human mileage, they contain disproportionately high levels of risk associated with factors such as darkness, fatigue and impaired driving.

Applying these more detailed benchmarks to Waymo's autonomous fleet produced a notable result. Across 127 million autonomous miles, the researchers estimate that the Waymo vehicles were involved in 359 fewer injury crashes than would have been expected from comparable human driving. More than half of the avoided crashes occurred between 8 p.m. and 4 a.m.—precisely when human risk is elevated.

The broader lesson extends beyond Waymo. Safety evaluation of automated driving will increasingly require exposure data that reflects the real operating environment. A vehicle driving mostly in daylight on familiar roads should not necessarily be compared with one operating extensively at night in dense urban areas. More context-sensitive benchmarks could therefore become an important part of developing a common scientific basis for assessing autonomous vehicle safety.
 
WAYMO. Safety benchmarks for autonomous vehicles need to reflect where and when they drive. Waymo Blog [online]. July 2026 [cited 2026-09-09]. Available from:  https://waymo.com/blog/2026/07/time-geo-crash-risk-effect/
 
Picture: chat gpt