DIESEL engine idling, Alice peers left, right, then straight at me. With her headlights on in the hazy morning, the imposing grey Ford van emits a loud beep, warning the world she is about to set off with no one at the wheel.
We’re at the starting line of this year’s DARPA Urban Challenge (UC), a 6-hour, 100-kilometre race along the roads of a simulated city organised by the Pentagon’s Defense Advanced Research Projects Agency (DARPA). Alice’s laser eye swings cautiously around. A robot can’t be nervous, but its human creators, a team of researchers from the California Institute of Technology in Pasadena, certainly are. They hope to prove that cars like Alice are a glimpse of the future.
It’s 26 October, the first day of the race’s qualifying round, a week-long test of robot road skills. Those that pass will take part in the race, in which they must navigate roads autonomously, obey the laws of traffic and, hardest of all, avoid each other. The winning team gets $2 million and a spot in robotics history.
Advertisement
The US government hopes to turn the winners into military supply vehicles for war zones, while some companies hope to make robotic cars a feature of our cities, and maybe reduce road deaths into the bargain.
As it turns out, Alice won’t be the one to take us into this brave new world. A later test reveals that beneath her cautious exterior lies an affinity for accelerating into oncoming traffic. She does not advance to the finals.
“Robotic cars could become a feature of our cities, and maybe reduce road deaths too”
Nonetheless, her brethren do the robot world proud. Of the 11 teams that race in the final on 3 November, six cross the finish line. It’s a gruelling race through unpredictable moving traffic, parking lots, construction sites and even an excursion down a dirt road into southern California’s high desert. Despite a couple of fender benders, the results are encouraging. Driverless cars on our highways might be closer than you think.
The UC isn’t the Pentagon’s first foray into robot racing. In the Grand Challenge (GC) of 2004 and 2005, robot cars had to navigate a dirt course across the Mojave desert (New Scientist, 19 November 2005, p 48). That race, in which robots had to avoid stationary obstacles and follow a list of GPS coordinates known as waypoints, was simple by comparison, though. According to veteran robot racer William Whittaker, whose Carnegie Mellon University team built the UC’s winning car, Boss, it consisted of nothing more than “barbaric, flat-out charges with big rooster tails of dust”. The UC, on the other hand, introduces much more sophisticated challenges.
As the vehicles race together around the track, they must detect and avoid moving objects. If another vehicle stops, the robot cars must work out why and then decide whether to change route, drive around it or simply wait for it to move again. They must also obey the rules of the road, such as who has right of way at a four-way stop sign, and demonstrate skills such as parallel parking.
Although Boss – a burly Chevrolet Tahoe – was the winner, most of the UC finalists have the same basic make-up. An array of lasers, radar emitters, sensors and stereo cameras function as the robot’s eyes. This visual data is fed into software which sews it together to create a 3D model of the car’s surroundings that is constantly updated.
On the starting line, the cars are loaded with a map of the area and a list of waypoints they have to pass to ensure they cover challenging parts of the course. To decide on the best route, the vehicles start by combining the starting position, the waypoints and the map, and then modify this plan in response to the surroundings.
For example, if a car encounters a large object in the road, it assumes it is a stopped vehicle. The car stops and checks how close this vehicle is to a traffic light, stop sign or intersection, which might indicate a reason for the delay. If the car doesn’t move for a while or is not near an intersection, the robot determines that it has broken down or is parked, and overtakes. To do this, it simulates several possible routes, checks to see if any require breaking the rules of the road – such as jumping a curb – and then picks the shortest allowable route.
The real test of the robots’ mettle came in “zones” on the course that were blanked out on the map. When cars enter these, not only must they navigate using sensors alone, they are usually instructed to carry out a mission like “park safely in a spot”.
Parking is one thing that robots find easier than we do. One of the zones resembled a shopping mall-style parking lot. Most people would have driven straight into a spot and then backed in and out to straighten up, but Boss chose a manoeuvre that hit the spot straight on by performing a neat pirouette first. Although this trick is faster, people don’t often use it.
In the end money and experience won the day. Boss cruised over the course at an average of 22 kilometres per hour, narrowly beating a team from Stanford University in California, which won the GC in 2005. Both teams spent more than $1 million.
Before driverless cars start driving alongside ordinary traffic, some challenges remain. Cars have to learn to tell the difference between a car and other moving objects, such as people. Although there are no formal plans for another DARPA race, roboticists are speculating that this ability will be the next skill that DARPA tests. The UC also didn’t test speeds that would be required on a real highway.
Once the technology is ready, the US army will be first to take advantage of it. Within a few years, it says, one-third of ground vehicles could go driverless. The move should save troops’ lives by automating logistics operations. “Twenty per cent of our transport capacity in Iraq is taken up shipping water to the troops,” says Chuck Jacobus, who led team Cybernet of the University of Michigan at Ann Arbor. “That’s a niche that could be all robotic and remove a lot of people from exposure to bombs.”
Driverless cars won’t be reaching our streets any time soon, however. One of the main reasons is that people might find accidents caused by driverless vehicles particularly unacceptable. That means robotic navigation skills and obstacle avoidance will have to be as near to perfect as possible before we see any driving around. “The first person that gets killed by an autonomous vehicle is going to set the field back 18 months,” says William Kehaly, team leader of Axion Racing. Apart from anything else, who would be liable in such an accident?
Robot cars will also demand changes to infrastructure. If they drive without someone in the back seat and need to stop for fuel, people will need to know how to respond to them. It will also be a long, and maybe unwinnable, battle to persuade people to give up their love affair with driving.
Still, the biggest argument in favour of robots seizing the wheel is safety. Automobile accidents killed 43,000 people in the US last year, while the World 91ɫƬ Organization reported that road traffic collisions killed 1.2 million people worldwide in 2002. By 2030, the WHO projects they will be the 8th leading cause of death. “We just accept this as reality, and that’s really sad,” says Jesse Levinson of the Stanford team.
That robots would make better drivers is something some roboticists clearly believe. Distracted by chatting, using cellphones or eating, human drivers are erratic, whereas a robot’s attention never wavers from the wheel. “Boss has been a better driver than me for a while,” says Whittaker, beaming like a proud father. Would he trust it to drive him on the highway? “Sure. Absolutely.”
“The car has been a better driver than me for a while. I would trust it on the highway”
But they can only dramatically lower the gruesome statistics if they replace today’s cars entirely. A few robots on the road wouldn’t make much difference. Roboticists are sure this is the future. Kehaly thinks the next generation might even view today’s cars as dinosaurs: “Our kids are going to say, ‘Remember when we had to drive our own cars around? That was so weird.'”
Cars and Motoring – Learn more about the latest technologies in our comprehensive .
Handing over control
The air-con is on in my Saab Sedan but I’m sweating like a pig. An Opel Saloon is about to attempt to crash into me.
A game of chicken? Not quite. I’m testing Saab’s prototype car-to-car (C2C) radio communications system, designed to prevent collisions. Cars communicate with each other and apply the brakes when necessary.
While DARPA dreams of driverless cars, partial control via C2C might be closer to reality. The US Federal Communications Commission has already allocated radio spectrum for cars, and Japan and Europe have plans to do the same.
As well as preventing crashes between cars, it might also safeguard pedestrians, provided they could carry transponders too. Emergency services could also use the technology to signal their imminent arrival.
Meanwhile technology is already chipping away at driver autonomy. Several high-end vehicles come with electronic stability control (ESC), which keeps a car from skidding or rolling over by controlling the brakes on each wheel independently and using sensors to determine how the car’s chassis is tilting. The US government now requires that ESC be installed on at least 50 per cent of cars made in 2009 and on all cars by 2011. “ESC could save up to 10,000 lives a year [in the US],” says Philip Hedley of Continental Automotive Systems in Auburn Hills, Michigan.
Also trickling down from the high end is adaptive cruise control, in which front-mounted radar systems monitor the area ahead. Another feature showing up on some Lexus, Cadillac, Volvo and BMW models is a lane departure warning. Most sound an alarm combined with a warning light, but some simply apply the brakes.
These safety features can also be a distraction, however (New Scientist, 20 October 2006, p 30). So Hedley says the future is to integrate them and do more behind the scenes. The first attempt at such a system, called Presafe, is available on Mercedes S and E-class cars. When the car’s radar system senses a front-end collision is imminent, it tightens the seat belts, rolls up the windows and adjusts the seats to maximise the effectiveness of airbags.
Additional reporting by Paul Marks
