IIHS Finds Lower Waymo Crash Rates
The Insurance Institute for Highway Safety reported on July 23 that Waymo's driverless Level 4 vehicles had 68% fewer police-reportable crash involvements per mile than human drivers in four U.S. cities. The comparison covered San Francisco, Phoenix, Los Angeles and Austin, while IIHS warned that federal reporting and mileage data are inadequate for continuous monitoring as autonomous deployments grow.
The Insurance Institute for Highway Safety (IIHS) reported July 23 that Waymo's driverless Level 4 vehicles recorded 68% fewer police-reportable crash involvements per vehicle mile traveled than human drivers in San Francisco, Phoenix, Los Angeles and Austin.
The study compared Waymo's driverless operations with human-driver crash records in the same regions and years. IIHS researchers manually reviewed federal automated-driving crash narratives to identify incidents that a reasonable person would have reported to police, then calculated rates per million vehicle miles traveled. The study applies only to the Waymo Level 4 vehicles and operating conditions examined, not to all automated-driving systems.
What the comparison found
According to the IIHS study, only 22% of reported Level 4 crash involvements were judged police-reportable or possibly police-reportable, and about two-thirds of those incidents occurred during driverless operation. The institute reported that Waymo vehicles were unlikely to be the striking vehicle or the primary contributor in crashes.
The study found that, compared with human drivers:
- •Waymo's rate of rear-ending another vehicle was 91% lower.
- •Its rate of being rear-ended was 40% lower.
- •Axios reported that single-vehicle crashes were 85% lower and injury crashes 81% lower on a per-mile basis.
Electrek reported the underlying rates as 1.28 police-reportable crashes per million miles for Waymo and 4.06 for human drivers. Axios reported that the city-level result ranged from 76% fewer crashes in Phoenix to 35% fewer in San Francisco; Austin showed a slightly higher Waymo crash rate, although IIHS cited a small sample there.
IIHS President David Harkey said the results show that, "on a limited scale, these driverless cars are safer than human drivers - who can be impaired or drowsy or suffer lapses in attention," while warning that the present collection system cannot support continuous monitoring during a large-scale expansion.
Data limitations remain material
The central methodological difficulty is that automated-vehicle and conventional-vehicle crashes enter different reporting systems under different thresholds. California requires self-driving companies testing on public roads to report crashes involving any property damage, injury or fatality. Human-driver comparison data, however, comes from state police-reported crash databases.
IIHS addressed that mismatch through manual narrative coding. Its paper concludes that this method is not sustainable as deployments and reported incidents grow. Axios noted that researchers also discarded roughly one-quarter of reported automated-vehicle crashes because they were duplicates, occurred off public roads, or did not meet the study's criteria.
The exposure-data problem is equally important. Federal reporting requires companies to report certain automated-driving incidents, but it does not require most companies to publish vehicle miles traveled. Axios reported that Waymo was the only major robotaxi operator voluntarily releasing driverless mileage data sufficient for the IIHS rate calculation. That limits equivalent, independent comparisons across operators.
Scope for autonomy evaluation
Waymo's vehicles are classified as SAE Level 4, meaning the system performs the entire driving task without human supervision within defined operating conditions. This differs from consumer Level 2 driver-assistance systems such as GM Super Cruise and Ford BlueCruise, which require an attentive human driver ready to intervene. Safety rates from a geofenced Level 4 robotaxi service therefore should not be generalized to supervised consumer automation.
For ML and autonomy practitioners, the report illustrates why model capability alone is not a sufficient safety metric. Credible fleet-level evaluation requires matched exposure data, consistent event definitions, operational-design-domain context, and independently auditable incident records. Industry assessments of comparable automated systems often face the same denominator and reporting-threshold problems identified by IIHS.
Key Points
- 1IIHS calculated a 68% lower police-reportable crash rate for Waymo's driverless Level 4 fleet across four operating cities.
- 2Manual incident coding reconciled incompatible reporting thresholds, but IIHS concluded that approach cannot scale with wider autonomous-vehicle deployment.
- 3Comparable autonomy safety evaluation generally requires public mileage denominators, matched operating domains, and consistent crash definitions across fleets.
Scoring Rationale
The IIHS analysis provides an independent, mileage-normalized safety comparison for a major U.S. driverless robotaxi deployment with public exposure data. Its methodological caveats are highly relevant to teams building, evaluating, and governing autonomous systems, although the results remain limited to Waymo's Level 4 operating domains.
Sources
Primary source and supporting public references used for this report.
Practice with real Ride-Hailing data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Ride-Hailing problems
