Autonomous cars, once the stuff of science fiction, are here. Google’s self-driving Toyota Prius hybrids have racked up more than 140,000 miles on public roads. Audi has sent a driverless TTS racing to the summit of Pikes Peak. Just about every major automaker has similar projects underway.
Despite the rapid advances — four years ago the most sophisticated autonomous vehicles were capable of only short distances and low speeds — we’re still a long way from the day we’re all letting the robot drive.
Wired recently talked to half a dozen experts about the road ahead. Their views are presented here in a roundtable-type format.
Wired: Google’s self-driving cars are logging more miles per year on California freeways than the average driver. What needs to happen — technologically, legally, culturally — to get autonomous vehicles on the road?
Sven Beiker, co-director, Center for Automotive Research, Stanford University: Are we talking about a car where the driver can sit back and doesn’t have to do anything? I’d be very, very surprised if we see anything like that in this decade, maybe even next decade.
Think of supposedly automated airplanes. I haven’t seen the situation where the pilot and copilot come back to the cabin and say, “Hey get a me a brandy,” where the whole thing is on autonomous mode. You have two highly trained professionals monitoring the system, reading out loud to each what they’re going to do next, very diligently checking the mobility apparatus before and after each flight. How often do we even check tire pressure?
Ralf Herrtwich, head of Telematic Research Laboratory, Daimler: I would say it comes down to two things: the availability of technology and the certifiability of vehicles. Of course, to some extent, that goes hand and hand. The more mature technology becomes, the more willing authorities are to grant you the ability to put such a product on the road.
Stefan Liske, auto industry consultant, PCH: We have to respect the fact that we will never be able in the next 10 or 15 years to come up with a budget that would be needed in order to turn cities into environments for autonomous driving, so I would like to give you a very differentiated answer.
I think for safety reasons [autonomous vehicles are] very helpful; also for traffic efficiency for reducing pollution, for reducing traffic jams, for making the commute more efficient — like for people living in Los Angeles who have a one-and-a-half-hour drive in the morning and a one-and-a-half-hour drive in the evening.
If autonomous driving possibilities can reduce this commute every morning, every evening by 30 minutes or an hour, I think it’s a solution that makes sense. It’s a roadmap that makes sense but we as customers, we as drivers, will take a long time to get used to this. And we will have a huge customer group and a huge number of cities or even rural areas where we won’t see autonomous driving on a large scale in the next 10 or 15 years. If you really want to introduce autonomous driving on a large scale, you have to fulfill so many prerequisites. Like for example the connectivity between the cars, the connectivity between cars and infrastructure. That would be a given in order to introduce autonomous driving on a large scale.
Junior, the autonomous Passat that Volkswagen entered in the 2007 Darpa Urban Challenge, during an early test run. Photo: Volkswagen
Paul Saffo, futurist: There’s the question of customer acceptance. How much does the machine control, and how much does the human control? You have to find that sweet spot, keeping it device-centric while giving humans the illusion they have some control. At Procter and Gamble I heard a story about a new cake mix the company had created. The scientists said, “This is perfect, we can make cake mix that doesn’t require eggs or anything else.” They came up with the product, taste-tasted the result and testers loved it. But it was an absolute dud in the marketplace. A marketing guy said, “We need to add one sentence to the directions: ‘Add one egg.’” The engineers said it doesn’t need an egg. But if all you do is add water, the housewife doesn’t consider it her cake. If you add an egg, it becomes her cake.
Clifford Nass, co-director, Center for Automotive Research, Stanford University: The first cars had no intelligence. Then we developed cars with hidden intelligence. If I told you, “You know how ABS works? When you press your brakes, that’s a suggestion for the car to stop. It can take it or not as it judges appropriate,” people wouldn’t have bought antilock brakes.
No one ever said the car is now in control. Think of fuel injection. You think that when you press on the gas pedal, gas goes and when you take up the gas pedal, it stops? No, it’s an incredibly complicated thing. Automakers have been able to keep that all secret. The advantage is I don’t have to think about it or know how it works. But compare that to functions in the semi-autonomous car, where it does adaptive cruise but not auto lanekeeping, or it does lanekeeping and not adaptive cruise control. There’s a weird mishmash of functions. I think the industry has done itself a disservice by not having standards in how they roll this out. One car company has adaptive cruise control but not lanekeeping. One has auto parking but not adaptive cruise control. People don’t get practice with these technologies; they don’t get to build mental models of what the car is doing.
Wired: What’s the significance of Google essentially grabbing pole position for autonomous driving? Is it a shot across the bow of the auto industry, or a sexy Trojan horse for a Google application?
Saffo: The genius of Sebastian [Thrun]’s approach is to put the emphasis on software over hardware. Good software can compensate at a much more cost-effective level for limitations of hardware than good hardware can ever do. He’s taking this weak info approach, with lots of sources of weak info rather than a single bullet of data. That’s the way to build safe systems.
The way I think about what they’re doing is they’re making it happen now, with the environment as it is. If more intelligence gets built [into the road environment], they can accommodate. We don’t have to wait for highways to get intelligent to have robots roll on them.
Beiker: I could imagine that they have a leading edge in the whole field of computer vision and decision making, and dealing with huge amount of data in real time. This is just what Google is very, very good at. The auto industry is still learning. Dealing with huge amounts of data is something that’s still a growing field in the auto industry.
Wired: What are the risks in ceding greater, or even full, control to the car?
Herrtwich: I always tend to say the road is longer than it is wide. Steering is the most dangerous thing.
Typically, what many of the emergency systems that are out there do is give the driver a warning. If nothing happens, the system kicks in. If you go into the pre-safe brake system, you get a few beeps. If you don’t pay attention to the beeps, the car will brake automatically. In parallel [highway] traffic, you have all the time in the world to go through this cascade. But when you have obstacles crossing your path, that’s so spontaneous there’s little room for such a cascade of warnings. Even if you did it, you’d hardly be able to initiate something by yourself. Originally, the driver was considered the safety net, and now the system is the safety net because even if the driver were to take over, things are happening so fast that human reaction time does not allow that to happen.
Donald Norman, consultant and author, The Design of Future Things: The hardest part about automation is when it fails and you have to immediately take over. In aviation, you have highly skilled pilots. You’re five miles up. You may even have a minute or two before you hit the ground. There’s time to analyze and see what happens. In the automobile, you have unskilled drivers who don’t have a clue, they may not have been paying attention — and they may have a second in which they have to respond.
Beiker: The biggest challenge is the mix of human and autonomous vehicles on the road. You might end up with an autonomous vehicle that absolutely obeys all traffic rules. It might be sitting forever at a four-way intersection because no one else is coming to a stop. It might be the most expensive but slowest vehicle in traffic. It might behave kind of funny.
Some people in the industry argue that once you combine lane keeping assist and adaptive cruise control with the latest level of sophistication — like stop and go assist — then you essentially have an autonomous car. But it needs be integrated more, which is another significant step toward highly automated vehicle. That’s really something that we must not underestimate: How much the driving task already is changing through this.