Adjust font size:
LONDON, England (CNN) -- A car that can drive itself is the fantasy of any designated driver, but the dream of owning a vehicle that does all the driving while you sit back and relax is one step closer to reality, as in-car artificial intelligence being developed by a team at Stanford University is ready to be used on city streets in the ultimate test of robot cars.
Winning the Defense Advanced Research Projects Agency (DARPA) Grand Challenge last year with a car called Stanley, Sebastian Thrun and his team at the Stanford Artificial Intelligence Laboratory developed a form of robotics that went beyond being purely reactive.
Rather than simply processing data and reacting accordingly, the in-car A.I. could evaluate data in milliseconds and decide whether it was correct or not. Many of the early prototypes of robot-controlled cars were literally stopped in their tracks because of faulty information - mistaking tumbleweed for a rock for example.
"What we have in Stanley is a revolution in the field of artificial intelligence. We now have ways to make robots understand the environment and make decisions about it, even if that environment is really complex," Thrun told CNN.
Thrun and his team mounted a series of sensors on the roof of their Volkswagen Touareg that included a radar, laser range-finders, stereo cameras and GPS receivers. Crucially it was also equipped with a machine that learnt algorithms to mimic the behavior of a human driver.
Sending data to the bank of computers housed in the trunk of the car ten times every second, information was processed and sent to the brake, throttle and steering wheel controlled by tiny motors.
Thrun's car completed the 132-mile desert course in just less than seven hours. Only four other cars out of the 23 that were competing for the prize finished the course.
The rugged terrain of the Mojave Desert proved tough enough, but the next challenge will push in-car A.I. even further. Next year's DARPA Urban Challenge will pit robot racers against each other in negotiating a 60-mile course through a simulated city environment.
"After the success of last year's event, we believe the robotics community is ready to tackle vehicle operation inside city limits," said DARPA Director Dr. Tony Tether.
The contestants will have to obey traffic regulations and cope with all the aspects of city driving, from merging with other vehicles and changing lanes to observing stop signs and parking.
Thrun's team will be competing again, as will ten other teams from A.I. research institutes, including last year's Grand Challenge runners-up from Cornell University. All will be trying to perfect the driver-less car that can cope with an urban environment without remote control.
"The urban environment is very complex, the car really has to understand what is around it -- pedestrians, buses, bicycles -- and understand how these things interact with the car and make decisions in a split second," said Thrun.
Traffic in a city is hectic at the best of times, but Thrun is not daunted by the task of trying to bring order to this chaos.
"There is actually a lot of order, even with urban traffic. There are road markings and traffic rules and there's the behavior of other road users that can be anticipated. We're currently coding all this information into an artificially intelligent robot. I can't say it's an easy ride, but I'm confident that it can be achieved."
Ultimately Thrun and the other teams competing at next year's Urban Challenge hope to develop cars that exceed the capabilities of human-driven vehicles and dramatically increase their safety.
Every year around 43,000 people are killed in road traffic accidents in the United States, 90 percent of which are caused by human error.
Many of today's cars have elements of A.I., from radar-guided cruise control systems and GPS collision sensors that enable cars fitted with the same equipment to communicate with each other. Toyota has developed technology that uses cameras to detect the curb when parking and turns that wheel automatically to reverse into the right spot.
While the technology is being developed that could one day make the car autonomous and hopefully safer, there will still need to be a great leap in faith among road-users to trust cars to drive themselves.
"There is clearly an issue of making people trust their lives to the hands of a robot. We are already doing this in the field of aviation with auto-pilot, but we have to get the same confidence for cars," said Thrun.
As well as consumer confidence issues a further hurdle to the adoption of robot cars is the legal aspect. The team behind Stanley ultimately envisage a car that has a button on the dash board that would switch the vehicle from manual to computer controlled driving, so drivers can choose between the experience of driving themselves or having a computer-driven limousine.
However, there is no legal precedent to what would happen if a computer driven car was to be involved in an accident. The main problem would be to discern where responsibility lies should an accident occur.
The consensus among traffic analyst is that more autonomous road vehicles would be safer than their human-driven versions. Traffic safety expert Chris Wright from Middlesex University in the UK is among them and believes that in 20 years time car will be much more like robots than the cars we drive today.
Thrun is even more confident that the technological hurdles will be overcome.
"I would say that in 10 years we can have cars that can drive themselves on highways, and before that we will have cars that can park themselves at low speed. The progression to a commercial product may take a little longer, but it's going to happen."
"Stanely" won the 2005 DARPA Grand Challenge and has laid the foundation for robot-controlled cars.
Quick Job Search