Carnegie Mellon Wins Urban Challenge

ThinkingInBinary writes "The results from the Urban Challenge are in! Carnegie Mellon's Tartan Racing team came in first (earning a $2 million prize), followed by Stanford's Stanford Racing team in second (earning $1 mil) and Virginia Tech's Victor Tango in third (earning $500k). Cornell's Team Cornell, University of Pennsylvania and Lehigh University's Ben Franklin Racing Team, and MIT, also finished the race in that order."

Comments

Re:The importance of this race cannot be overstate

[erlehmann]

If, on the other hand, *every* car was automated, it would be so much easier to implement.

also, if every car was automated and the controls were built into the road, there would be a massive single point of failure.

Re:Open source ...if only.

[SnowZero]

The most important thing is that the algorithms are written up and published in peer-reviewed journals. That understanding is more important than the code itself. My RoboCup robootic soccer team published all of its source code one year, and not much came of it; Some people used it but they didn't really understand it. It's also hard to take the code and make it work with a different robot, as all the customized hardware on the robot means a lot of porting, and uncovering bugs and design limitations. Also, competition code often has its design stretched to the limit by the time the competition arrives, and if you redid it you might design it differently to make it cleaner. So, our papers have probably helped many more people than our code ever did. I did release a library along with some papers explaining it, and that worked well. But that's just a small part of the overall codebase.

Of course, it would be nice to see the code out there, but the science is more important than the implementation. However, if we were talking about an off-the-shelf robot such as a roomba or aibo, the situation is quite different.

MIT pimp ride

[guacamole]

I almost laughed out loud when I saw pictures of MIT's pimped out Land Rover. Besides the numerous external sensors and other gear mounted on the vehicle, I read that there is so much internal equipment to manage everything that they had real heating issues that were solved by installing an additional air conditioner and a power generator to power the AC. This is what happens when you give some money and parts to a bunch of bright geeks with too much time.

Re:MIT pimp ride

[Anonymous Coward]

Indeed, it's truly geeks running that show. Way in the beginning, they were thinking of putting some equipment in the engine bay... neglecting to realize that it gets f*cking hot in there while it's going. The LR3 is the second vehicle that MIT built up (the first was a Ford Escape.) They tested out the equipment in MA, where it's much cooler, so they got away with underrating the power supply for everything in the vehicle (there's somewhere between 5 and 10 kW worth of electrical generation underneath the vehicle) but it turned out that in the southwest they'd need more cooling. The solution? Stick yet *another* 3kW generator on the roof just for the A/C. Mmmm....

The MIT computer setup was probably way overkill, the people who spec'd it didn't even consider electrical requirements; there's an obscene amount of computing power in the back (10 blades, 4x Xeons each.) All the extra weight has had serious tolls on the car itself as well; there's been multiple transmission issues and 3 sets of tires gone through.

The software also suffered... it uses a system written entirely in C, using UDP multicast between processes (sensor drivers, fusion and such). It was also purpose written for this application. I think what happened is that they'd rather stick with the devil that they knew than try anything else... but seriously. C? They've got this excessive amount of computing power, and yet they insisted upon using C. Oooookay...