Physics Forums

Physics Forums (http://www.physicsforums.com/index.php)
-   Computing & Technology (http://www.physicsforums.com/forumdisplay.php?f=188)
-   -   Robotics AI competition, million dollar prize (http://www.physicsforums.com/showthread.php?t=3488)

schwarzchildradius Jul1-03 02:00 AM

Robotics AI competition, million dollar prize
 
Whoa, it was news to me. The robots have to be completely autonomous, and travel from LA to Las Vegas in 10 hours using only GPS guidance systems and computer brain-power (no remote control!). The prize is a million bucks, but the blueprints will become public knowledge, they say. If no winner, the contest repeats until 2007.
There are some genius roboticists in this country, and it seems likely that someone will make it before the 2007 deadline.
But it raises dubious questions - will the military (who is sponsoring the contest) be using semi-autonomous tanks and battle-groups controlled by a War-Games style "WOPR" supercomputer? Why would the military need anything that was autonomous, unless they planned to use the 'bots to do things that humans just wouldn't do?!
Decode for yourself.
Story
Darpa dot MIL grand challenge

eNtRopY Jul1-03 11:34 AM

Re: Robotics AI competition, million dollar prize
 
Honestly, have we learned nothing from the movie Real Genius?

eNtRopY

sheldon Jul2-03 01:33 AM

Re: Re: Robotics AI competition, million dollar prize
 
Quote:

Originally posted by eNtRopY
Honestly, have we learned nothing from the movie Real Genius?

eNtRopY

Funny, that was a good movie. I still think its cool and would be fun besides a big mil.

drag Jul2-03 04:06 PM

Re: Re: Re: Robotics AI competition, million dollar prize
 
Personally, I wouldn't bet on this one even 'til 2007.
If it were an aircraft then it's easy (relativly [;)]),
but a ground vehicle ?! [g)] No way ! [<:)]

schwarzchildradius Jul2-03 07:10 PM

Ahh, I donno, drag, it doesn't seem too terribly wild if a sophisticated AI is involved, s/a triangulating position based on landmarks corresponding to GPS maps. It's just negotiating the smaller obstacles that will probably be more difficult, such as boulders and streams, but remember that size is not limited -- a pretty large bank of computers could fit into a small car.
What I don't understand is why they don't seem to want to develop remote control tanks, which would seemingly be much easier to navigate by humans. The error rate for remotes is pretty high, so far as I know.

drag Jul2-03 07:17 PM

Quote:

Originally posted by schwarzchildradius
Ahh, I donno, drag, it doesn't seem too terribly wild if a sophisticated AI is involved, s/a triangulating position based on landmarks corresponding to GPS maps. It's just negotiating the smaller obstacles that will probably be more difficult, such as boulders and streams, but remember that size is not limited -- a pretty large bank of computers could fit into a small car.
How about canyons and rivers and steps and mountains ?
It's easy to build a robot that'll just lock on to the
stripe at the side of the road and keep going, but I
don't think that that's what they want...[;)]

enigma Jul2-03 07:45 PM

I'm sure the things will be able to use GPS, and computerized maps are getting very sophisticated. With the right people, my money is that it will be done by 2007.

The Grimmus Jul5-03 10:22 PM

Re: Re: Robotics AI competition, million dollar prize
 
Quote:

Originally posted by eNtRopY
Honestly, have we learned nothing from the movie Real Genius?

eNtRopY

yes... but could it also be like that movie toys with robin wilams be controlled by kids who dont know about it. Perhaps the robots will get to the area and the children will then control them and kill thousands then the robots return home...in the guise of wee wee pants wendy

Thomas1980 Jul18-03 06:12 AM

From a technological point of view it is posible, ain't it? Very expensive indeed, but quite posible...
In some modern cars there is of course GPS to guide the direction, and when to turn, there is cruisecontrols to adjust speed, radars to meassure distances to other cars ( praking- as well as highway-aid ) servosteering (powersteering?) is mostly standard, so technologically it should be posible with relatively few hardware modifications.
So the big issue would be making a system with cameras and picturerecognition... It must see red lights and then stop... And be able to follow the road when it makes a turn, as well as stay away from roadwork and so forth... So the main issue here is actually a programming challenge, which indeed is a killer. But posible, right?

schwarzchildradius Jul19-03 07:31 AM

I really dont think it'd be a prohibitive programming problem, in any case, for 3-d space can be mapped from 2 2-d images, to the limit of the resolution of the medium and distance between the cameras etc. If good programmers could work with roboticists, it might be possible to map a 3-d space to an overhead image (from high altitude photos or GPS topo maps).
From there the A.I. sets the several destinations toward a final goal and calculates a few routes to each destination, and records a small scale map of each route as it tries each.
Sounds like fun.

drag Jul20-03 11:36 PM

Quote:

Originally posted by Thomas1980
From a technological point of view it is posible, ain't it? Very expensive indeed, but quite posible...
In some modern cars there is of course GPS to guide the direction, and when to turn, there is cruisecontrols to adjust speed, radars to meassure distances to other cars ( praking- as well as highway-aid ) servosteering (powersteering?) is mostly standard, so technologically it should be posible with relatively few hardware modifications.
That's not what it seems that they wan'na do. Making a simple
car that would just go on the road is simple - VERY simple
by comparison. The challenge is to make a surface vehicle that'll
be able to make the trip off the road, through desert/mountains/
canyons and more. Personally, I don't think that'll be achieved
until 2007.

Live long and prosper.

bleh Jul30-03 01:57 PM

i dont think it would be so hard for a team of really good programers to make a program that would read gps and sensors and such i think the main problem is having it make it their alive their are so many people who would just hit it with their cars or try to destroy it just to be jerks about the whole thing. i think if you programed a very small ATV with a high enough clearence you could make it their in a straight line.

Thomas1980 Jul31-03 05:09 AM

Hmm... If the task is to make a vehicle go from coast to coast, the task seem a bit more simple... I mean, most kids have build robots in lego that can drive and avoid obstacles... The challenge here is just in a lillte bit of a larger scale... A lego robot will eventually make it, just need some kind of energy source that is sufficient... and then patience of course.
On the realistic note, I think one of those balloony wheeled cars that Nasa has been experimenting with for Mars exploration should be able to drive over most terrain here on earth... Combined with a bit of lego programming and some giant batteries and a compass or gps, that would do it, me thinks...
Dont you?
( simplified yes, but for a million $ you should be able to figure out the rest... Money is never free... )


Best regards

Thomas Hansen

Isotope Aug9-03 12:07 AM

Quote:

Originally posted by bleh
i dont think it would be so hard for a team of really good programers to make a program that would read gps and sensors and such i think the main problem is having it make it their alive their are so many people who would just hit it with their cars or try to destroy it just to be jerks about the whole thing. i think if you programed a very small ATV with a high enough clearence you could make it their in a straight line.
I don't beleive it would hit people, if you stand still, or walk torward someone, they usually swoop around you. But the rest seems accurate.

Claude Bile Aug21-03 09:15 PM

Airplanes etc are a nice idea, but the cost of building one would likely be more than the prizemoney. I think it is definately possible in theory for such a robot to be built using current technology, it is just a matter whether it can be done for less than a million beans.

BiologyForums Aug21-03 09:34 PM

Just to let you know there are plenty of robots that can already travel from one location to another - no matter the distance, in this manner.

The first time I saw one that did this was literally in 1993.

No big deal, so this must be a catch or bogus...

enigma Aug22-03 08:46 AM

Quote:

Just to let you know there are plenty of robots that can already travel from one location to another - no matter the distance, in this manner.

The first time I saw one that did this was literally in 1993.

No big deal, so this must be a catch or bogus
Actually they don't. These robots must go from point A to point B - several hundred miles - under their own power and completely autonomous (aside from GPS, etc.) - in only 10 hours. The course goes up and down mountains, on-road and off, over, under, around, and through unknown obstacles. The robots must also be able to recognise their competitors and know when to go around them or follow behind them.

The sophistication of the object and obstacle recognition AI required does not exist yet, and that is what this competition is really going to be about.


All times are GMT -5. The time now is 02:34 AM.

Powered by vBulletin Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
© 2014 Physics Forums