Driverless cars are coming. They will overtake, give way, stop at red lights, allow pedestrians to cross, drive in their lanes - all of this without human intervention. Some cars already have features like automatic parking and cruise control. And one company has taken a lead in the race to build a car that will drive itself , a company that has never made a car in the past - Google.
Sebastian Thrun is a Google fellow - a title reserved for the company's most distinguished scientists - and an expert in artificial intelligence. Thrun wants to transform the way humans commute. "As a boy, I loved cars. But when I was 18, I lost my best friend in a car accident," Thrun said at a
TED (Technology, Entertainment and Design) talk in
Long Beach,
California , last year. "I decided then that I would dedicate my life to saving one million people (who die in accidents) every year."
Thrun's mission led him to a driverless car competition organized by DARPA, the US military's top
research agency. In 2004, the car that came in first travelled about 11 miles before spluttering to a stop. The following year, the vehicle designed by Thrun's team completed the challenge, driving 212 kms in the Mojave Desert in a little under seven hours.
Google, which had got the same idea as Thrun, asked him to come to the company's Mountain View headquarters and start work on cars that could drive themselves. So Thrun put together a team of engineers, including Chris Urmson, who had taken part in the
DARPA challenges.
The result was a fleet of seven modified Toyota Prius cars (see box) that roam San Francisco and the Mountain View area - without any hands-on help from a driver.
The cars drive themselves through everyday traffic, without killing anybody . (That bit is important.) "The heart of the system is the data we get from the spinning laser mounted on top of the car. This data allows the car to see the world around it," Urmson told the audience at a conference on Intelligent Robots and Systems last year.
"Our car sees the world on two levels . The static world and the live world," he said. The static world consists of objects that don't move or change. Like driving lanes and road barriers. But people, vehicles, traffic lights change or move. Once the car identifies the moving objects around it, it starts tracking their movements in real time. This allows the car to make driving adjustments like giving way or overtaking.
Despite never making a mistake on its 200,000-mile run - a Google claim - the driverless car is not yet ready for the market. "But," says the Google spokesperson, "we think the cars are years away, not decades." To begin with, there are technical challenges. "We're still working on handling severe weather conditions like snow."
But even when Google irons out all the kinks, the driverless cars will still require a semblance of order on the roads to be able to function. Which means for countries like India with notoriously chaotic roads, it could be decades before the cars are let loose.
Then there are the legal and social challenges. Like who is to be held responsible should an accident occur?
Rohit Baluja, president of the Institute for Road Traffic Education, says, "The vehicles will obviously cut down on human error and that will make our roads safer. But how will these well-behaved robotic cars deal with human drivers? What if somebody indicates he's turning left, but at the very last moment turns right?" As for who would actually be responsible , Baluja says, "At the moment, we don't have the laws needed to decide that."
Just as tough to crack will be the social problem. Google says that when the driverless cars do arrive on the market , "some hesitancy is to be expected ." But adds, "The cars don't take control away from the driver... " People can take back the reins from their cars any time they wish to. That's comforting.
Look no hands Google says, for now, before putting the car on a particular route, a human driver first 'maps' the way, including lane markers and traffic signs, by driving 'manually' . This allows the driverless car to familiarise itself with the 'route environment and its characteristics' in advance The car uses its spinning laser, camera and sensors to create a 360-degree image of its surroundings. That allows the car to create a 'safe bubble' between itself and all objects The data received is processed by an off-the-shelf computer with a quadcore processor, which can process 1.3 million measurements and make about 20 driving decisions every second. Which means, if there is an obstacle on the road, say a pedestrian, the driverless car can step on the brakes Data like speed limits are already known to the system.
The driverless car always stays within the speed limit and obeys all traffic rules, which makes him a consistently safe driver Four radar sensors, three up front and one at the back, determine how near or far surrounding objects are A spinning laser on the roof of the car takes 360-degree measurements A video camera near the rear-view mirror detects changing traffic lights Computer makes 20 driving decisions a second
