Last night I watched NOVA | The Great Robot Race | PBS on TV. I'd heard about the challenge two years ago when none of the robots did very well.
You see, DARPA decided to host a challenge to anyone who could build a robot that could complete a 130 mile trek through the Mojave desert. And last year, the "best" team made it 7 miles.
The show (the entire hour is available on-line - PBS rocks) follows a number of the teams, discussing their different approaches, and then covers the race itself.
CMU has two robots, and Stanford has one - they were the main focus of the show. While I like CMU a lot, I think the Standford team's accomplishment was much more impressive. Of course, they won the competition, but it's not that they won, it's how they did it.
Stanford's robot, Stanley, is pretty much just a VW fitted with robotic controls and some off the shelf electronic components. On the other end of the spectrum, the CMU vehicles are Humvees with a lot of custom-made electronics. Stanford's robot was the only one to complete the qualification round without touching any of the obstacles (pretty impressive). And, when given the race course (a set of GPS coordinates), the Stanford team just uploaded those and let the robot go. The CMU team on the other hand, had a team of 30-50 people break the course down into detailed positions and contours, basically programming the path and speed for the robot for every inch of the race. So Stanford's approach seems more general and thus more powerful.