The Speculist

In the first year of the DARPA Grand Challenge, none of the robotic vehicles finished the desert course. In Year 2, five robot cars finished a different but equally difficult off-road course. The picture is of "Stanley," the Stanford University DARPA Grand Challenge winner.

I guess DARPA decided that last year was just too easy. To keep the "Grand Challenge" ... challenging, they're taking the race to town:

The robot racers will face a "simulated" urban course 96 kilometres (60 miles) in length on November 2007. The course will feature urban obstacles, such as trees and buildings, traffic signs and other moving vehicles.

After last year's challenge blogger Ivan Kirigin pointed out that the 2005 Grand Challenge didn't mean that robotic cars were ready to drive Ms. Daisy to the store:

Let me just point out that there is a world of difference between off-road dessert driving and freeway driving. There are also huge differences between freeway and city driving.

Dessert driving is very hard [sic.; keeping an ice-cream sundae balanced on your knee in traffic is tough, but I don't think that's what Ivan meant], even for humans. Driving on a rocky, dirt road that you’ve never seen before, for those used to nice, paved roads, is a challenge. In most cars, you couldn’t top 20mph, making the average speed of Stanley, 19.1 mph, seem pretty good.

Off-road, there are few mobile obstacles, many large obstacles, many obscured obstacles, very tight turns, steep inclines, no road signs, no lanes (or lane markers).

On paved roads, you have signs, smooth curves, lane markers, barriers on the road edge, reflectors in the lanes and on the side of the road. Unfortunately, you also have other drivers, and sometimes pedestrians. You are also traveling, on average, at much higher speeds. This changes something called “look-ahead distance”, i.e. how far ahead you need to look for threats to avoid them.

Good points. But, as different as these two types of driving are - off-road and crowded Interstate - they are alike in one way. Both types of driving are within the capabilities of an experienced human driver.

Somewhere today, some guy will jump in his SUV 4x4, and drive from the city to go off-roading in the desert. He'll back from his driveway and hit the surface streets to the Interstate. Along the way he'll have to swerve around some guy that started to pull out on him from a stop sign.

Once on the Interstate it's crowded until he gets well out of the city. Cars will be merging in and out. He gets cut off once or twice.

After his exit he has a little uncrowded surface street driving. Then, he switches into four wheeled drive and hits the desert for the off-road fun he came for.

There's perhaps four different categories of driving involved there - city surface steet driving, Interstate driving, country road driving, and off-road driving. One processor - a human brain - can handle it all. And, usually, both the driver and the SUV return safely to home and garage at the end of the day.

This human capability/flexibility is what DARPA is aiming to duplicate with machines. I predict that the year after a successful urban race DARPA will combine different types of driving (perhaps off-road and urban) to see if a single machine can be as flexible as some guy trying to escape the city for a day in the desert.

I'd love to see next month's Urban Challenge live. This would be perfect for ESPN 8 - you know - "The Ocho."