The fact that Google’s bubble-like self-driving car, unveiled this week, lacks a steering wheel might be seen as evidence the company’s software is close to mastering the challenges of piloting a vehicle. But the car’s design is just as much a consequence of what Google’s existing fleet of automated Lexus SUVs revealed about human laziness.
Google’s engineers had been focused on perfecting how well those modified cars could handle freeway driving, and they imagined their technology hitting the market in a way that left humans sharing driving duties with their vehicle. “The idea was that the human drives onto the freeway, engages the system, [and] it takes them on the bulk of the trip—the boring part—and then they re-engage,” said Nathaniel Fairfield, a technical lead on the project, speaking at the Embedded Vision Summit in Santa Clara, California, on Thursday.
That approach had to be scrapped after tests showed that human drivers weren’t trustworthy enough to be co-pilots to Google’s software. When people began riding in one of the vehicles, they paid close attention to what the car was doing and to activity on the road around them, which meant the hand-off between person and machine was smooth. But that interest faded to indifference over weeks and months as people became too trusting of the car’s abilities. “Humans are lazy,” says Fairfield. “People go from plausible suspicion to way overconfidence.”
That convinced Google it had to give up on switching between human and machine control, says Fairfield. That also ruled out building on top of conventional car designs, because they assume a human is on hand and ready to take over in the event of an emergency.
“If the car detects there’s something wrong with the power steering, it cuts it out and trusts you,” said Fairfield. “Same with [power-assisted] brakes—they go limp but you can still stop the vehicle.” Google’s tests suggest that anyone accustomed to a self-driving car would be unlikely to be ready take over in the event of such a failure.
And so Google’s new vehicle design takes a leaf out of NASA’s design book to cope with such eventualities. “It doesn’t have a fallback to human—it has redundant systems,” said Fairfield. “It has two steering motors, and we have various ways we can bring it to a stop.”
The car is powered by an electric motor roughly equivalent to that used in the Fiat 500e and has a range of around 100 miles. Its maximum speed is 25 miles per hour to reduce the severity of injuries to pedestrians in the event of a crash, says Fairfield. The front of the vehicle is made of plastic material intended to cushion any impact with a human. Google says it intends to build up a fleet of 100 of the vehicles for testing this summer. However, existing laws mean that only versions that have had conventional controls installed can be tested on public roads.
Fairfield says that the decision to make a purely autonomous vehicle also had the benefit of bringing Google more in line with the company’s original vision of something that could “drive everywhere for everybody.” That more strongly differentiates Google’s approach to automated vehicles from those of conventional automakers, who have pledged to keep humans in ultimate control of their vehicles.
Last week I shared some photos from inside Google’s prototype of a self-driving car. The reactions were about as polarized as I’ve ever seen on a blog post I’ve written. The first group of readers were amazed, and they wanted to know where they could buy one. Everyone else was terrified of the interior. A car that they could never steer? No thanks.
The problem with including a steering wheel in a self-driving car is that human drivers can’t be trusted to effectively take over in sticky situations. So the makers of driverless cars can’t responsibly include a steering wheel.
Reid Hoffman, the renowned Silicon Valley investor and LinkedIn co-founder, relayed some telling research from Stefan Heck, the founder of Nauto, a self-driving car start-up:
Research that Stanford has done shows that drivers resuming control from Level 3 vehicles functioning in autonomous mode take 10 seconds just to attain the level of ability that a drunk driver possesses. And to get back to full driving competence takes 60 seconds.
That’s a full 10 seconds just to function at the level of someone who shouldn’t be behind the wheel. A self-driving car cruising at 70 mph is unlikely to have 60 or even 10 seconds to wait for a human to take over.
“A huge fraction of the time, nothing bad will happen because the vehicles really are reliable, hundreds of kilometers at a time before something really terrible happens,” said Ed Olson, a University of Michigan professor researching self-driving car technology. “You can’t just be tootling along on the freeway and the car says — ‘Oh your turn!’ Meanwhile the person is like ‘What?’ ”
Hoffman’s entire post on LinkedIn is worth your time. It’s full of interesting insights, such as how cars will turn into places for marketing and advertising. A coffee shop, for example, might even pay your self-driving car company to take you a block or two out of your way so that you drive by it. Accepting flexible, out-of-your way routes might mean a cheaper fare, and you might get offered a coupon to that coffee shop.