Two important caveats should be considered when interpreting the findings.
* the distance accumulated by self-driving vehicles is still relatively low (about 1.2 million miles, compared with about 3 trillion annual miles in the U.S. by conventional vehicles).
* selfdriving vehicles were thus far driven only in limited (and generally less demanding) conditions (e.g., avoiding snowy areas). Therefore, their exposure has not yet been representative of the exposure for conventional vehicles.
With these caveats in mind, there were four main findings.
1. The current best estimate is that self-driving vehicles have a higher crash rate per million miles traveled than conventional vehicles, and similar patterns were evident for injuries per million miles traveled and for injuries per crash.
2. The corresponding 95% confidence intervals overlap. Therefore, we currently cannot rule out, with a reasonable level of confidence, the possibility that the actual rates for selfdriving vehicles are lower than for conventional vehicles.
3. Self-driving vehicles were not at fault in any crashes they were involved in.
4. The overall severity of crash-related injuries involving self-driving vehicles has been lower than for conventional vehicles
Humans averaged just 1.2 crashes per million kilometres during the same period compared to just over 5.5 accidents per million kilometres for self driving cars.
Self driving cars had a total of 11 accidents involving self-driving cars between 2012 and September 2015, two of which resulted in injuries. With the cars having driven about 1.9 million kilometres on public roads (the overwhelming majority by Google’s fleet).
Taking underreporting into account only brings the accident rate for normal vehicles up to around 2.5 crashes per million kilometres – less than half that of self-driving cars.