Since Google’s self-driving cars first rolled onto public roads in 2014, they have had to cede control to their drivers roughly 350 times, mostly because their sensors failed to correctly recognise objects in the real-world.
But, contrary to what you might assume, the men and women who took the wheel in those cases weren’t Google engineers.
Turns out, the company purposely recruits non-software people to cruise around in its self-driving cars, because it wants them to analyse their experience as an average driver would, not through the lens of an engineer, according to a deep-dive into Google’s autonomous car facilities by BackChannel’s Steven Levy.
Google started hiring non-engineer drivers way back in 2009, and, today, there’s a formal, full-time training process that takes four weeks for each operator. Candidates prove their skills in a large testing facility Google built that includes a rain simulator, a traffic circle, and other test hazards.
The drivers Levy talked to had little in common — one had a marketing background another worked in a bakery — besides that they enjoyed driving and could do it well. Many had wound up at Google through a friend.
Once drivers pass their training, the job of not-driving a car isn’t as luxurious as you might think. Drivers can’t text, eat, talk on the phone, or smoke. Instead, their hands and feet are supposed to be positioned to be ready to take over control at any second.
And they have been good at it: Google revealed in its recent report to the Department of Motor Vehicles that its drivers were able to take control after a car switched out of autonomous mode in an average of .84 seconds.
You can read the rest of Levy’s piece, where he chronicles his own experience “driving” one of the autos, here.