With autonomous vehicle operators now required to report their crashes, we finally have some data to compare robot drivers to human drivers when it comes to road safety. Here's one good argument for a robot-driving future: Human drivers are more likely to get in crashes that hurt or kill other humans.
For a new study, the University of Michigan's Transportation Research Institute took the crash records of three self-driving vehicles currently on US roads -- Google, Audi, and Delphi -- and compared those to the crash records of all vehicles on US roads in 2013.
Geographic coverage of the three self-driving vehicle fleets in the study. While Delphi and Audi have one-time long-distance trips, Google operates on a daily basis in their regions
We've reported on most of the self-driving data from the study before. About 70 per cent of all self-driving car crashes occurred while the cars were stopped or going less than 8km/h. No human has been seriously hurt. And self-driving cars have never been at fault in any reported incident -- so far.
If you do a pure data comparison, self-driving vehicles do indeed have a higher crash rate per million miles travelled than conventional vehicles. However -- and this is a big part of the study -- regular human-piloted car crashes are very underreported, according to the NHTSA: 60% of property-damage crashes and 24% of injury crashes go unreported. So if you take those numbers into consideration, as authors Brandon Schoettle and Michael Sivak note, it's probably about as safe, or a little safer, to be in a self-driving car: "We currently cannot rule out, with a reasonable level of confidence, the possibility that the actual [crash] rates for self-driving vehicles are lower than for conventional vehicles."
What's most interesting, however, is the type of crash that conventional vs autonomous cars tend to have. When you compare most of the crashes that happen with human drivers behind the wheel, a full third of all collisions are not with other cars but with "fixed objects" (medians or buildings) or "non-fixed objects" (sadly, these are probably people). Self-driving cars, so far, don't experience those kinds of crashes at all.
So even if you might dispute the overall crash rate for human-operated vs autonomous cars based on the reporting rate, the data proves something that's kind of irrefutable: Humans driving cars are much more likely to cause property damage or hit other humans, causing injury or death. The study puts it this way: "The overall severity of crash-related injuries involving self-driving vehicles has been lower than for conventional vehicles."
Of course there are tons of caveats here. Self-driving cars have only driven about 1.9 million kilometres total; America drives its regular cars about 4.8 trillion kilometres a year. And self-driving vehicles aren't subjected to the same driving situations as human-operated ones: None of the autonomous cars are currently being tested in places that have wintry weather, for example.
Still, it's important to note the current shortcomings in our human-piloted vehicles and see where better tech can fill in the gaps for now. Today, Bosch announced a "pedestrian avoidance" technology which it wants to have in all cars by 2018. And of course designing better bike lanes and footpaths to protect street users who aren't in cars can go along way in preventing crashes. While we wait for all the cars on the road to become autonomous, there are ways to help human drivers operate a bit more safely.