Slashdot Mirror
Teach A Robot To Drive, Win A Million Bucks
An Anonymous Reader writes "DARPA has released the details of a 'Grand Challenge,' with a $1 million prize. The challenge is to build an autonomous vehicle which can 'navigate on its own over a 250-mile desert course in less than 10 hours.' from L.A. to Vegas, 'without external communication or human control.' The contest is to be conducted in March 2004, and is open to all comers. Can we get at least one entry to represent slashdot?" We've mentioned this contest a few times before: any intended entrants out there want to disclose your secret plans?
The challenge is to build an autonomous vehicle which can 'navigate on its own over a 250-mile desert course in less than 10 hours.' from L.A. to Vegas, 'without external communication or human control.'
:)
Somehow I have the feeling that 99% of the teams competing will try to figure out inventive, creative ways of using and obfuscating 'external communication or human control' as the first step.
Cover your eyes and click this link!
"For example, an extremely large vehicle that simply travels on a straight line between two points by climbing over or breaking through everything in its path (and destroying what cannot support that movement) is not the type of intelligent solution that is sought." - DARPA Site
:/
Obstacle intimidation algorithms not allowed.
But the devil ends up being in the details. While it probably would not be too hard to design a vehicle that could do this in a couples day or so, 250 miles in 10 hours means the vehicle would have to be averageing 25 mph. At 25 mph, there is not a whole lot of room for error. You would need a system that could react to environmental issues that came up very quickly such as obstacles, or dead end routes. You would also need to a system that could actually sense/see far enough ahead to steer the vehicle in the correct direction without running into things.
The complexities in this problem are enormous so trivialising them is apparently interesting.
Maybe we'll just hook up slashdot to an artificial neural network and spend few hours teaching/programing it to do moderating and throw in a cpu and run a rule based system with a well defined set of rules (for detecting trolls, redundancies, etc.) and we're set. Piece of cake.
Can anybody imagine how much R&D would have to go into even an attempt, much less a successful one? $1 million doesn't even scratch the surface...
--sdem
You're just looking at the problem all wrong.
You're not providing the guidance so the missle kills someone, you're providing the guidance so the missle DOESN'T kill all the doctors and patients in the hospital next door to the target.
paintball
Lemme guess, you'd also like to slam DARPA for builing that blasted ARPAnet? I mean, what good ever came of that?
Marxism is the opiate of dumbasses
People have been working on "smart cars" for decades, thowing every technique you mention (and quite a few more) at the problem, and I don't think we're close to having a robot car that could be trusted to drive unsupervised in real traffic...
However, feel free to prove me wrong by winning the contest!
If a Field Judge is unable to continue in the Safety Vehicle, and that inability is of a temporary nature expected to take less than ten minutes to resolve, the Field Judge shall instruct the Safety Vehicle driver to stop, and shall record the time of stopping so that the time stopped may be subtracted from the elapsed time for the team.
Perhaps that should read:
If a Field Judge needs to take a leak, then teams must stop when asked. Team members, on the other hand, are expected to be able to control their bladders.
I am artificially intelligent.
The heavier the vehicle is the more kinetic energy it has so the more problems you have negotiating obstacles. Furthermore, the heavier it is the more likelihood that it will damage itself in collisions or rollovers. Also, the heavier the vehicle is, the more energy it will consume, so the more fuel it needs to carry, so the heavier it is...
The solution to this problem, from a chasis point of view, is to build the lightest machine possible consistent with carrying a laptop computer, two video cameras and a small radar. If I were building it I'd aim for a lightweight carbon-fibre moncoque shell with a generally curved shape; large, lightweight wheels like mountain bike wheels; a small air-cooled four-stroke engine - say 100 to 250cc; a cone type continuously variable transmission; and a robot wars style self righting mechanism. I'd aim for at least 100 miles per gallon on-road fuel economy and carry four gallons of fuel in an underslung fuel tank for a fully fueled up weight of under 150 pounds.
Structurally the key thing would be to protect the cameras and the radar. Not only do you not want them to be damaged, you don't want their mounts to get bent even the slightest bit out of alignment.
On the road sections of the course you'd use stereoscopic vision to establish road position as with the Italian ARGO project mentioned earlier, possibly with the object detection assisted with radar. You'd go as fast as you possibly could on road sections to build up average speed.
Off road you'd use primarily radar to assess forward obstacles. The strategy would be to steer a near direct course deviating around small obstacles. If a large obstacle was encountered, you'd backtrack 100 yards, turn 30 degrees one way, and go forward; if that didn't work you'd recursively back up more, turn the other way, and try, until you had passed the obstruction, at which point you'd plot a new direct course and carry on.
But the key things, it seems to me, are keep it small, keep it light, keep it simple.
I'm old enough to remember when discussions on Slashdot were well informed.