Self-driving cars: from 2020 you will become a permanent backseat driver
Original article can be found on theguardian.com
Driverless cars will revolutionise motoring, claim the manufacturers. But is the greatest danger that they will be too safe?
In the BMW museum at the company’s solidly futuristic headquarters, next to the old Olympic stadium site in Munich, you can view a century of evolving mechanical desire. BMW has long prided itself in creating “ultimate driving machines” and all that Bavarian engineering pride is dramatised in the decade-by-decade progression of engines that harness ever more efficient power in steel, and car bodies that have moved with the ergonomic times. Each sequence of cars on show leaves a gap at one end, ready to showcase the next generation of technical advancement. Over the past century, innovation has smoothly followed innovation; it is likely, however, that the next stage will be a paradigm shift rather than a marginal gain. The next empty space, or the one after, is likely to be filled by the ultimate driverless machine.
The person leading BMW’s prototype efforts to make that car a reality, Michael Aeberhard, does not want to see it in those terms. As he takes me for a drive in what seems a regular 5 Series, he is at pains to suggest that the new model now in gestation is simply another improved iteration of what has gone before.
When we reach the autobahn on the way to the airport, however, and he presses a button near the gear stick, the change feels a little more revolutionary than that. Aeberhard takes his hands from the steering wheel, and, Herbie-like, the car takes on a sudden, apparently joyous, life of its own. It shifts lanes in dense traffic – a lorry slightly alarmingly close to my passenger window – and accelerates up to its optimum speed of 120kph.
Aeberhard lets it go through the gears, confident, after five years of refining this minor miracle through simulator and test track, that the car knows precisely what it is doing. And for the next half hour we sit back and enjoy the ride.
Or at least I do. Aeberhard remains alert for the unexpected: roadworks, lane alterations, emergency vehicles, whatever. When any of these are on the horizon the car signals its approaching confusion and he returns his hands to the wheel and takes over. He does this seamlessly three or four times, in a spirit that suggests his close knowledge of the limitations of what he calls the “highly automated driving project” and in a way that seems to prefigure a new kind of relationship between driver and car – one of mutual co-operation rather than continuous control.
In contrast to some other autonomous projects, the BMW car does not have conspicuous sensors and radar bolted on to its exterior. “The idea at the very beginning was that we wanted the car to look like a normal 5 Series,” Aeberhard says. “Unless you know where you are looking, it is difficult to spot the 12 sensors.” Most are repurposed versions of technology already available. The radar sensor in the front bumper is the same one that currently enables “active cruise control”. The camera system is similar to the lane departure cameras already in use (though the software allows it to recognise speed limit signs as well as lanes). Lasers are in the bodywork near road level; there are, as a result, also only four, as opposed to 64 in some versions of the roof-mounted technology, because they “see” more laterally than downward.
BMW is limiting its driverless ambitions to the autobahn (except for a valet parking innovation it recently demonstrated that will allow you to leave a car at the entrance to a car park and happily allow it to find its own space). They have, for the moment, “no real interest in junctions and turning”, Aeberhard says. The key challenge in mostly straight-line driving is simply speed. “Getting from 130kph down to zero with a good amount of braking still takes 80 to 100 metres so you have to have detection with your sensor at least to that range very reliably,” he says. “That is near the limit of what the sensors can do at this speed. But still it is quite successful… ”
I turn over that word “quite” in my head momentarily and scan the speeding traffic ahead. “How far can it ‘see’?” I wonder. “Between 150 to 200 metres to the front,” Eberhardt says, “And to the rear and the side 50 to 60 metres.” “How does that compare with human perception?”
“Oh,” he says cheerfully, “human perception is way better. We have a plan to put a black wall at that distance on the simulator 150 metres away and see how that really feels. But if there is an incident ahead the braking distance is good enough…”
“But only just?”
Aeberhard looks ahead calmly.
“Have they had any crashes?”
“None at all,” he says.
There is something appropriate in test driving this particular car on these particular roads because it was here that the idea of autonomous driving had its authentic Eureka moment. Almost 30 years ago a German engineer, Ernst Dickmanns, working out of Munich University in association with Daimler-Benz, piloted a project called the “Programme for a European Traffic of Highest Efficiency and Unprecedented Safety” (or “Prometheus”).
Dickmanns and his team incorporated several dozen “transputers” into a car, pioneering the first man-made three-dimensional mobile vision system. By 1995 they were able to have their car drive autonomously 95% of a 1,700km trip from Munich to Odense in Denmark and back. On the autobahn, the system achieved speeds of 175kph. The car was, however, perhaps 20 years ahead of its time. The level of computing power required made the prototype too expensive and cumbersome for commercial production.
Two decades and many technological advances later, and almost every large car manufacturer is pursuing a version of Dickmann’s idea. There is, Aeberhard suggests – as we bomb along the autobahn – not that much sharing of technology between manufacturers, beyond some limited European-wide initiatives. Each manufacturer is trying to find its own way of creating autonomy, while mindful of going too far too fast. Toyota, for example, has learned lessons from its experience in 2005, when it introduced the world’s first self-parking car; far from being greeted as a game-changing advance, it was mostly criticised for being over-engineered as well as expensive, and was not widely copied.
Volvo has made many advances in incorporating automated safety systems, with its stated ambition of achieving “no accidents involving Volvo cars by 2020”. But so far this is mainly with an emphasis on keeping the driver alert. For example, if a car crosses a lane line without the driver using an indicator signal, the dashboard flashes the image of a steaming coffee cup and the words “time for a break”. If it is still drivers that cause accidents, we would like to believe it is also careful drivers that prevent them.
The concept of driverless vehicles is even more complicated for companies such as Mercedes and BMW which sell cars on the basis that people will love the experience of driving them. As Aeberhard suggests, BMW has put a great deal of its design effort into making hands-free feel like a natural driving experience. “The reality of the world is that a lot of driving we do is not fun driving. It is driving to work in traffic. Even so, we want it to feel as it should feel if they were driving it themselves. Or like maybe a chauffeur is driving. A very comfortable do-not-disturb type of driving.”
This is the second time in a few months that I have experienced a version of this much advertised future. On the first occasion, near its Mountain View headquarters in California, I ventured out in one of Google’s remodelled Lexus cars, and then around a test track in its homegrown bubble car just before it was allowed on the road. The experience could hardly have been more of a contrast. Google is attempting autonomy in city driving, a challenge, as Aeberhard acknowledges, at a different order of magnitude from what BMW and most of the rest are aiming for. Never shy of hubris, Google wants not only to reinvent the car but to replace the whole idea of driving. In some ways this is primarily a mapping challenge. The BMW mapping system is comparatively basic – a mix of GPS and sensory observation of lane markings and other vehicles; the map itself of the routes it follows in Munich hasn’t been updated in two years.
“One thing we have noticed about maps,” Aeberhard says, “is that the more detailed they are, the more the world they describe is constantly changing. The sensors themselves are getting better. The better the sensors get, the less you need the map.”
Google, which of course loves its cartography skills, is trying something altogether more sophisticated. Its system, aiming at full autonomy, not only has to take account of traffic but of every eventuality on a city road.
Dmitri Dolgov is the systems engineer responsible for imagining and interpreting this landscape: “We continuously get better at classifying things and predicting how they will behave,” he explained. “If you have a cyclist coming towards you the wrong way in a cycle lane for example. Or if a car in front is about to make a U-turn as opposed to a normal turn, the system can now recognise slight changes in behaviour. We don’t have to worry about birds, they will avoid us. We do have to worry about the likelihood of people opening doors from parked cars. We can predict when someone is going to run a red light. When cyclists do it. Then there is a long tail of very weird events…”
Google has a dedicated group of employees, as part of the project, whose job is to put the bubble car into the most unusual situations it can conjure up to prove it does the right thing. They spend days tossing beach balls in front of the prototype, or letting scooters loose, and seeing what happens.
Some intuitive behaviour is still well beyond it. At four-way junctions, which in American suburban streets rely on eye contact between drivers and a system of edging forward, the car can become paralysed.
In general – in the somewhat jerky drive I enjoyed – it is exceedingly cautious, in a driving culture where excessive caution creates its own difficulties. The accidents the Google car has had have almost all been when it has been rear-ended by other vehicles when it was following the letter of highway law perhaps too closely, or stopping too abruptly. It behaves a little like a nervy learner driver. “The real problem is that the car is too safe,” Donald Norman, director of the design lab at the University of California, San Diego, who studies autonomous vehicles, has observed. “They have to learn to be aggressive in the right amount, and the right amount depends on the culture.”
There is a kind of ambitious genius in what Google is aiming for – and of course the implications, positive and negative, are debatable – but the spirit of it seems to go against the grain of what people are ready for. I asked a couple of the biggest brains at Mountain View what would happen if kids realised that a fun game would be strolling out into the road in front of Google cars, which of course have no choice but to stop in their tracks. They hadn’t apparently considered the possibility.
“We want to fundamentally change the world with this,” Sergey Brin, the co-founder of Google, likes to say, with the authority of somebody who has already achieved that particular ambition at least once. It’s likely, however, that in this instance, the world is not quite ready for that change. Certainly that would seem to be the sceptical message of the attitude survey in Tech Monthly. The wariness displayed about the new technology is the kind of understanding that the marketing people at BMW – and at Toyota and Mercedes and Volvo and the rest – know only too well. Too much revolution alienates their customers. Google’s blank sheet of paper approach betrays the advantages and disadvantages of not having made cars for 100 years already. It remains to be seen if Apple, which is reportedly also building a self-driving car and has been scouting for secure locations in the San Francisco Bay area to test it, will attempt something equally ambitious, but it would be a surprise if it did not. And as such there is a stark divergence between what the tech giants believe is possible with autonomy and what traditional car-makers think can be achieved.
Much of this comes down to a schism over understanding how people adopt the new, new thing. Google has, as a core belief, the messianic attitude that human behaviour is redeemable by the universal application of its technology.
In this sense, Chris Urmson, affable head of Google’s car project, believes that some choices are too important to be left to individuals. At Mountain View he reeled off statistics about fatalities on the roads – “worldwide 1.2 million people are killed each year. In the US alone 33,000 people are killed, the equivalent of a 737 falling out of the sky five days a week. We make incremental progress but we accept it as the status quo” – before going on to question the pitch of traditional car-makers: “Driving in adverts is made to took like fun, but really driving is actually mundane, or hoping to get to a meeting on time. Traffic is getting worse. Doing basic maths, the average commute is 50 minutes a day. That’s six billion collective minutes a day, 162 lifetimes wasted in America alone…”
As a result, Google works on the basis that an all-or-nothing solution is the only solution. People and their errors need eventually to be removed from making bad choices behind steering wheels. Cars should be another element of our lives that rely on the data.
“We don’t think it will work incrementally,” Urmson says. His reasoning is that Google has taken a step that BMW, for example, has not yet attempted. They have put real people – that is to say, general Google employees rather than people intimately involved in the driverless project – behind the wheel of their cars. In 2013, they gave 100 people driver-assisted cars – with a capability similar to the BMW 5 Series prototype – to use in this way, let them go out on the public road and drive home.
“They loved it,” Urmson says. “It was transformational. People weren’t using their energy in the car, so they had more for other activities. One woman said that when she got home, she ran every evening and unlike before – when she had been tired from the commute – she cooked every day. We had another guy who drives a Porsche normally. To begin with he thought the self-driving idea was stupid, he loves driving, but after a few days he came back and said: ‘I get it. Most of my driving sucks. It drives better than me and it certainly drives better than anyone else…’”
So far so good. But then Google set up cameras and looked at what people were actually doing in the cars while they were being driven on the freeway. Urmson was frankly alarmed; far from approaching the experiment with trepidation and due vigilance, as I had watched Aeberhard doing on the autobahn, the human drivers were almost immediately putting their faith in the robotics of the car: “We saw that despite being told this was a prototype, despite moving at high speed on the freeway, they were over-trusting it. We had a guy who was sitting in the front seat, he pulls out his phone charger from the back seat, then turns back again for his laptop sets it up on the seat, and does all this without looking out the windshield. The whole time the thing has been moving fast down the freeway. He believed in the technology enough that he just trusted it. And as these things get more capable we think that faith will only grow. That really worries us. That is why it is extremely tough to get from incremental improvement to full self-driving.”
As a result, Google does not believe the half-way house solutions of the European and Asian car manufacturers will be workable. It puts its faith in full scale revolution (which is the only way of avoiding the kinds of four-way junction paralysis that the current prototype experience; if there were only autonomous cars on the road there would be no need for a nod and a wink about right of way. The cars would “talk” to each other.) Will driving therefore become another aspect of life colonised by the ubiquitous data-collectors of Silicon Valley? For all Google’s (and Apple’s) resources, you guess, in the foreseeable future at least, it will not.
Aeberhard and his colleagues at BMW think like car salesmen, not systems engineers; he believes that incremental solutions are the only solutions, and therefore, implicitly that driver behaviour will adapt to driverless parameters in the same way as previous generations of BMW owners have taken on new technology down the years. He knocks down some of the reservations that buyers might have, the kinds of reservations explicit in our survey, one by one. “Anxieties about how people will accept this are not new in the automobile industry,” he says.
“We have always had new technology. Take the air bag, for example. They had to sell the idea that a potential explosion in front of the driver was a good thing. There are cases where an air bag can injure a driver, but such accidents are so infrequent and the safety benefits so clear that we have eventually accepted the research and the value of air bags. I think automated driving will go down a similar path. It is a step-by-step introduction. It is not going to be all of a sudden you buy an automated car.”
How about the question of insurance liability. If the car is “driving” itself, how can the driver be at fault in an accident?
“I don’t think it will be too long before we have a black box in the car like you have in planes so that if something does happen you will be able to find out from the data where the fault lay,” he says. “I think the law in Nevada already requires that data in partially automatic cars.”
And what about the worries about hacking. Couldn’t an automated highway could also be a prime target for cyberterrorism? (At one technology conference a pair of hackers made headlines by showing how, by sending commands from their laptop, they’d been able to make a Toyota Prius sound its horn, and brake abruptly at eighty miles an hour.) “Hacking is a concern for any technology company,” Aeberhard says. “In BMWs, to get access to the systems that control the steering wheel and so on, you have to have physical access to the car. You need to be physically attached to the wires and then you need to know the protocols for the system to do anything, and still then it is very hard to get the car into a state where it is responsive. In automated cars there will be further layers of security.”
Does that kind of imperative mean a culture change at the car manufacturer; are they thinking more of robotics and less of mechanics? “Yes, I suppose there is a change happening in our culture in that it is becoming more software oriented,” he says. “A lot of the people we are hiring, you are looking for a systems background rather than a mechanical engineering background.” As Aeberhard speaks – his eyes always on the road, though his hands are not on the steering wheel – the car that is driving us seems to go up a gear and finds a gap in the traffic up ahead. It feels like an instinctive move that a real driver would make. Is that actually what has happened?
“It think that is more by chance that it looks that way,” he says. “The car just wanted to get up to its speed of 120kph. The odd thing is that the car is usually doing something much simpler than the thing you think it is doing. Complex human behavioural decision making is very hard to programme into the machine. We do that manoeuvre without thinking because our brains have an insane amount of experience and knowledge that they can convert into instinctive choices. A robot is not good at this. But it has other advantages. The sensors never get tired, they never get distracted, they always work at the same level.”
How soon does he think those qualities will be available to buy from a BMW catalogue?
“We think sometime after 2020 we will be ready for the first highly automated function, which means that the driver will be actually able to do something other than monitor the system – read emails, call somebody, check the news, whatever. But even in this mode, there are situations when the car knows it will reach its limit. Then it will tell the driver, with 15 or 20 second warning: sorry you need to take over.”
At which point you can only hope that the driver is awake, and sober, and remembers exactly how to drive.
