Although their study finds that the accident rate of robocars from Google and two other companies is a bit worse than that for human-driven cars, they add that not a single accident has been blamed on a robocar—and those that did occur tended to be less severe than those of conventional cars. Nobody was killed.
Authors Brandon Schoettle and Michael Sivak of the University of Michigan's Transportation Research Institute offer a lot of caveats upfront. The study covered 50 cars that logged a mere 2 million kilometers in tests run by three of the 10 companies working on robocars. And the bulk of the data came from Google, with the rest drawn from Audi and Delphi, the auto supplier.
Finally, most of the trips took place in California, where road conditions are easier than in the snowier states—and where accident rates are lower than in the United States as a whole. Even when Google cars ventured elsewhere, they mainly stayed in southerly parts—notably Austin, Texas.
The robocars never had front-end collisions, but they were 50 percent more likely than conventional cars to get rear-ended. This makes sense: a robotic car is particularly good at watching where it goes, and the carbon-based driver in the following car is not.
But the sample was too small to be sure. Front-end collisions account for only 4 percent of all accidents, and four percent of 11 accidents rounds to zero.