Autonomous vehicles' blind spot
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Sarah Grillo/Axios
Everyone agrees the standard for self-driving cars is that they have to perform better than a human driver. Proving that is not so simple, however.
Why it matters: Inconsistent data reporting methods mean there's no common yardstick to compare safety claims, adding to the confusion and distrust surrounding autonomous vehicles.
State of play: Waymo says its robotaxis are already "more than 10 times safer than the average human driver."
- Its peer-reviewed research — based on more than 220 million miles without a human driver — shows a 94% reduction in crashes involving serious injuries or worse, and 93% fewer crashes involving pedestrian injuries.
- A recent independent study by the Insurance Institute for Highway Safety (IIHS) generally concurs, finding Waymo had 81% fewer injury crashes per mile than human drivers in four cities it studied.
Yes, but: IIHS highlighted a major blind spot in wider AV research and said current data collection isn't good enough to monitor safety on an ongoing basis.
The big picture: The National Highway Traffic Safety Administration has required companies to report incidents involving automated and assisted-driving systems since 2021, primarily to spot potential safety defects and enforce recalls.
- The database is the only national archive for tracking AV safety, but it's insufficient as a research tool because it lacks context — there's no correlation to the number of vehicle miles traveled.
- Waymo voluntarily shares that data, known as VMT, which allowed for the analysis by IIHS. But other companies do not.
- Without that critical denominator, researchers can't calculate meaningful crash rates or compare performance between companies or technologies.
Between the lines: Even with the mileage data that Waymo provides, it's difficult to compare AV crash rates to those involving human drivers.
- IIHS says about half of all crashes — and one-third of injury crashes — involving human drivers go unreported. That's in part because drivers involved in fender-benders want to avoid a hike in their insurance rate.
- Plus, human crash databases categorize injuries and crash severity differently, making apples-to-apples comparisons difficult.
Zoom in: When IIHS researchers analyzed Waymo's crash rates in San Francisco, Phoenix, Los Angeles and Austin, Texas, for example, they had to discard roughly 25% of reported AV crashes because they were duplicates, occurred off public roads or would not have been reported to police.
- Using this method, IIHS found Waymo vehicles were involved in 68% fewer "police-reportable crashes" per mile than human drivers in those cities.
What's next: Researchers say robotaxi crash data should mirror the standards in other federal crash databases, such as tracking airbag deployments and logging crashes according to established injury severity scales.
- NHTSA should also require all AV companies to disclose VMT in order to determine crash frequency rates, IIHS said.
What they're saying: Waymo supports the recommendations for more robust reporting requirements.
- "More high-quality data is fundamental for building public trust in this technology and ultimately making roads safer for everyone," the company said in a statement.
What to watch: The Autonomous Vehicle Industry Association has recommended a national data repository as part of its proposed AV policy framework, but congressional legislation remains stalled.
