Last Updated: Friday, 16 November 2007, 14:06 GMT
Driverless cars on city streets
Spencer Kelly
By Spencer Kelly
Click presenter

Driving a car is a bit different from an aircraft autopilot where you are surrounded by miles of empty air. When driving you follow the road with your eyes, watch out for obstacles and try to predict the actions of other drivers.

Boss crosses the finish line
Boss crossed the finish line and won the challenge

Can a computer cope with all that? Darpa, the military research folks behind things like the internet, are trying to find out.

In 2005, Stanford University's computer-controlled vehicle, Stanley, won the Darpa Grand Challenge to be the fastest to successfully navigate its way around a 160 mile (257 km)course through the Mojave desert. No human drivers, no remote controls, just sensors, artificial intelligence and jolly good suspension.

Only five of the 23 bots made it.

But the world is not all desert so Darpa came up with the Urban Challenge.

Teams from around the world took on the task of building a robot car that can navigate an urban environment, avoid other traffic, follow road markings and obey the rules of the road.

Sensing obstacles

"The big difference between the Desert Grand Challenge and the Urban Grand Challenge is that now you have to worry about the movement of other vehicles," said Michael Montemerlo of Team Stanford.

Things that humans can normally see and easily interpret are very hard for an ordinary camera to do
Ben Upcroft, Sydney-Berkeley

"And not only are there moving objects that you have to track and avoid but you also have to predict their behaviour as well.

"It's hard enough having to predict the behaviour of other people, imagine having to predict the behaviour of other robots who are not very good drivers yet."

The team at Sydney-Berkeley were one of those hoping to make an impression.

Bristling with sensors their car crept tentatively round the university campus test track, trying to make sense of the road and other objects ahead.

The team worked hard to make sure the computer program could tell the difference between a moving obstacle, the empty road and a falling leaf.

Ben Upcroft, co-team leader said: "To be able to detect where lines are in the road we have to deal with changes in light due to the trees and buildings. There are shadows - things that humans can normally see and easily interpret are very hard for an ordinary camera to do."

Laser guidance

After winning the Grand Challenge in 2005, the weight of expectation was on Stanford and their new car Junior.

Stanford's Junior completed the race in just over 269 minutes

Although apparently still a very conservative driver, Junior already drives much faster than Sydney-Berkeley. Because of this, the car had a baby-sitter behind the wheel, just in case.

That said, Junior's sensors handled junctions at its own test track very well - stopping, giving way and then pulling out.

There are two lasers in the front for finding cars in front of Junior, there is one for looking forward and finding the road markings and there is a spinning laser that it uses to find cars in every direction.

Then there are two rear lasers looking left and right to find road markings and finally two spinning lasers on the back for finding cars when Junior is reversing or changing lanes.

Predictive behaviour

As sensor technology and programming improves it is not inconceivable that robot cars of the future will be able to navigate a city environment.

But there is more to driving than just reacting to things you can see - you need to be able to predict the things you cannot see too.

Initially we might buy a car with a chauffeur button and that would cost maybe $2,000 or $3,000 more
Sebastian Thrun, Team Stanford.

A robot could easily avoid a ball bouncing into the road but would it be smart enough to expect a child to run out after it?

"I think there are differences between how a car and a human will drive," said Sebastian Thrun of Team Stanford.

"Humans are much smarter at grasping situations very quickly, for example a rolling ball that a child might follow.

"A robot is much better at precision measurements. It can measure the exact distance to the ball and how fast it rolls much faster than a human could ever do. So when it comes to swerving around the child it will be much more precise than a human can be."

Chauffeur mode

Mr Thrun points out that most accidents and road deaths are caused by human error, so it is important not to pretend that "being human" is something computers need to live up to.

"Initially we might buy a car with a chauffeur button and that would cost maybe $2,000 or $3,000 more. And you push it and it drives you home and wakes you up when you are in your garage.

"Maybe a few years later we'll have the self park button that you push and it drops you off at the office and finds a parking spot. And you have a cell phone number that calls the car back when you need it."

But that is all in the future. Until recently the teams had been concentrating on the Urban Challenge and the $2m (1m) prize for the fastest to cross the finish line.

The winner

The task for the robot vehicles was to drive a 60 mile (100 km) course around a small Californian town, which I am sure we are all glad to hear has a population of zero.

After a series of qualifying events over the previous days, 11 teams made it to the final.

Sydney-Berkeley failed to make the grade but the favourites Stanford and their robot Junior did.

Each car was required to drive safely round the roads, completing a number of manoeuvres and missions along the way.

They navigated junctions, avoided robots and manned vehicles. And all their human teams could do was watch anxiously from mission control.

Six cars completed the event - five were deactivated after getting stuck around the course.

The results: in third place team Victor Tango, completing the challenge in just over 276 minutes, in second place Stanford's Junior in just over 269 minutes, and the winners, in 250 minutes and 20 seconds were Tartan Racing and their car Boss.

Not the fastest 60 miles ever driven but certainly one of the most impressive drives in history.

Robot cars race around California
05 Nov 07 |  Technology
Look, no hands
19 Jul 07 |  Magazine

The BBC is not responsible for the content of external internet sites

Has China's housing bubble burst?
How the world's oldest clove tree defied an empire
Why Royal Ballet principal Sergei Polunin quit

banner watch listen bbc sport Americas Africa Europe Middle East South Asia Asia Pacific