Slashdot Mirror
DARPA Announces 2005 Grand Challenge Semifinalists
Mockingbird writes "DARPA announced 40 semifinalists for the 2005 Grand Challenge autonomous robot race today. Notable remaining teams include the Carnegie Mellon University Red Team, Stanford Racing and a high school team, the Palos Verde Road Warriors. 78 teams missed the cut. The race, which will take place on Oct. 8, 2005 features a $2 million prize for the first team whose robot crosses 175 miles of the Mojave in under ten hours. The robots must be fully autonomous, with no team intervention allowed once the vehicle is launched. The first race was held in 2003, when the most successful team managed to log only 7.4 miles."
Hey, don't joke too much: DARPA's end goal with challenges like this is to build fully robotic semi-autonomous armed combat vehicles.
;)
May I propose a new slogan for the contest: "DARPA Grand Challenge: Because An Army of Kill-Bots Will Make Them Like America Again!"
I think that the first autonomous kill-bot that we make should have a big "We've Come To Bring You Freedom!" sign placed directly beneath its main gun turret, for the irony.
We should start dealing in those black-market beagles.
For those that don't know, it's quite amazing how much the average car does this already. As mentioned, things like RPM, individual wheel speeds, clutch slip, internal pressure sensors, temperature sensors, and just about anything else you can think of that could be remotely useful is monitored by the computer which can kick the drivetrain into various "limp home" modes and set off that damn check engine light.
You gotta find first gear in your giant robot car
If it wasn't a Challenge, anybody could do it.
/. Hell with 'em. They got no idea of the gumption or stamina these teams are putting in, and the problems they are tackling every day.
2004 was Alpha. 2005 is Beta. 2006 will be Release.
Even Microsoft takes 2-3 tries to get something right.
Different teams are taking different approaches, but the more and different sensors you have, the more info you have to determine your environment.
Some of it is Brute Force, for this problem. But, it still takes Smarts to finish at a decent speed. And last year, people finally got a clue as to how much Smarts it can actually take.
So this year, it really will be more of a software problem. How fast can you process the sensors you do have, understanding limits of those sensors, tell the vehicle to change it's trajectory, and then perform the actions; all while following a course to an end goal.
It has been amusing to read all the other recent posts from half-dots (those who toss off 1/2 a thought without even a smidgen of researching or thinking through, but have got to pump an opinion out) The lines of code being used, the processing to be performed, and the harsh computing environment are just some of the variables that make this so darn interesting.
The Baja drivers DARPA hired to check the course and follow the vehicles, they talk about how difficult the problem is for them on this (relatively) easy terrain and constrained environment. To do it at the speed necessary to complete the course is hard. 17 mph average gets you there. 30 mph wins big, but is a much much harder driving task.
Sure, no one finished the race last year. That shows how hard the problem is! The Whining Armchair Quarterbacks(TM) can't seem to even imagine how you learn from failures and try try again. Sure seem to be more bunches of negative nabobs out there this year, even on
This is pushing the state of the art in robotics. And I for one am tickled to be doing my small part in the whole thing, and supporting one team as best I can.
See, it's pretty easy to fix the bug that destroyed your vehicle last year. What's 1000x harder is to anticipate the bug that will destroy your vehicle this year.
I'm studying a course in 3D Computer Vision right now, at TUHH. It's part of the Erasmus exchange program I'm having here - the eigth and last semester (excluding the thesiswork) of my master of engineering in automation and mechatronics at Chalmers in Gothenburg. I can easily say this course is the most difficult one of all I've been taking for all of my study time, hopefully the three weeks I have between that exam and the last of my others, will be enough to learn what doesn't stay in my head during the lectures...
In fact, I have the course book right beside me. To begin, the description of it would be more or less along the lines "an orgy in linear algebra, mathematical statistics, with some flavouring of image processing, geometry, optimization and algorithms". Basically, it's 30-40% mathematical formulas, 650 pages, some containing things not even all MSc even learn like tensor notations etc. Not something I'm even sure is a good thing to recommend to very many slashdotters, even. You'll get its name though - "Multiple View Geometry in Computer Vision", by Hartley & Zisserman. ISBN 0-521-54051-8.
What I see as problems in the book, is that almost everything is working on corner detection. This is great, if you want to make 3D-models of houses or other man-made objects (at least half of the examples in the book are architectural, I would say). It's not so great if you want to image bushes, rocks and other things with not so obvious corners on them. Also, the process involves quite heavy processing - both image processing, finding all those corners, statistical processing (to sort out outliers, which there will be), and optimization to find the best fitting backprojection of the image planes). I don't have a sure grip on the needed processing power but I doubt, when considering realtime demands in a car, that it'll hardly be easy to get it working.
Also, it's still to a big deal itself an area under research. The situation with using 5+ images (from different cameras och just consecutive images from the same, moved camera), isn't very well known. Using more images, of course would mean a bigger chance to get a decent 3D model of the scene...
And still, you would at least need two cameras to do anything useful. You can't reconstruct 3D space without having at least two images of the object to reconstruct. And probably you will need more - you would probably want to reconstruct all the way around (ie more cameras on the sides and backwards), and add extra sensors like radar etc for extra checks.
And then you really haven't solved the problem of driving the car. You have only built a decent mapping of the 3D surroundings of it. You have to add AI/some kind of steering logic, which only in itself is a demanding task. Just look at all FPS games out there - if it would be easy to construct good AI, with a known 3D-world, tailormade for the figures, would we really be seeing that many games with crap-AI? I'm happy I ain't taking an AI course too, for sure!
I have a really elegant proof for Fermat's last theorem. If this sig was only a bit longer...
You really think that the armed forces would allow their budgets to be threatened by unmanned combat vehicles? Commanders only think about how many men they command.
Oh, and your reference to killbots reminds me of the Simpsons, where Kent Brockman used ridiculously overhyped language to try and scare viewers. The Simpsons was making fun of the attitude you display in your post.
Shutting down free speech with violence isn't fighting fascism. It IS fascism!
DARPA, and the DOD would *love* to have semi to full autonomous "kill bots" - in a way, today, they already have them for some tasks - they are called "cruise missles", which can be launched, told to stay on "hold" above possible targets, then commanded to strike on located targets. I would assume "located" likely means some form of lat/lon coordinates or painted with a laser (either by troops or from the air).
The exact same thing could be done with a kill bot: send it to a predetermined position, and tell it to "hold fire" unless acted upon agressively, or if non-friendly comes into position (at which point it could bark a series of commands in different languages to the offender - think of it as an active landmine with intelligence that can move on command), which if not heeded, shoots a warning, then if continued, shoots to kill. Friendlies are identified by RFID or similar tags. Equip them with the ability to identify each other, as well as to flock or coordinate efforts with one another. Other commands could be something like "fire on ident", where they could be set up, then when a target is painted with a laser (perhaps from a troop's rifle), it fires on that target.
You better bet that the DOD and DARPA would be all over such a system if it was proven field safe (to our troops) and easy/quick to use, and rugged. They are half way there with the TALON robots already, they just lack the rest of the package, which the Grand Challenge is dealing with...
Of course, one can also easily see the potential of scaled up versions - robotic Humvees and M1A tanks, as well as robotic quads, and perhaps legged versions...
BTW - this last was actually funded by DARPA back in the 1980's, which culminated in the Odetics, Inc. (now known as Iteris, Inc. - based in Anaheim, California - interesting the strange things going on at this company, whatwith name changes, etc - plus, they are developers of an "electronic highway" concept - I am sure there is no relation to the Grand Challenge - wink, wink) ODEX-1 legged walker - a very unique leg design that proved to be fairly robust and strong, while keeping outboard weight (on the legs) to an absolute minimum by moving all the electric motors inward toward the torso of the machine.
Think about it - if you could, in addition to GPS coordinates, vision systems, etc - also bury in the ground or nearby some form of active or passive "locator" beacons, such as what Odetics - oops, I mean Iteris - is developing - wouldn't the problem become just a little bit simpler...?
Nah - DARPA hasn't been thinking about this, not at all, not at all...
Reason is the Path to God - Anon
"DARPA Grand Challenge: Because An Army of Kill-Bots Will Make Them Like America Again!"
I don't want our enemies to like us, I want them dead.
Fuck it