The study was commissioned by Alphabet Inc’s Google unit, which has reported a series of minor crashes involving its self-driving fleet. It looked only at Google’s fleet of more than 50 self-driving cars, which has logged 1.3 million miles in Texas and California in self-driving mode.
The test fleet has reported 17 crashes over the last six years, although none were the fault of the self-driving cars, Google said.
After adjusting for severity and accounting for crashes not reported to police, the study estimated cars with drivers behind the wheel are involved in 4.2 crashes per million miles, versus 3.2 crashes per million miles for self-driving cars in autonomous mode.
Crash rates for conventional vehicles at all severity levels were higher than self-driving crash rates, the study found.
A 2015 National Highway Traffic Safety Administration study found about 60 percent of property-damage-only crashes and 24 percent of all injury crashes are not reported to the police.
California law requires all crashes involving self-driving vehicles be reported to police.
Google spokesman Johnny Luu said the company asked Virginia Tech “to look into the topic given the interest and develop a robust methodology to be able to make meaningful comparison between regular cars on the road as well as our self-driving cars.”
Luu said the study “will be helpful making apples-to-apples comparisons moving forward.”
Google, eager to demonstrate its vehicles are safe, criticized the proposed rule, which it said would maintain “the same old status quo and falls short on allowing this technology to reach its full potential, while excluding those who need to get around but cannot drive.”