Grand Challenge: Technical papers about the AI used in the various robots

My advice to you is to ignore what the grandchallengers did. I think that (with the exception of Team Phantasm's Ladibug, which couldn't start because a PC failed and they had no money for a new PC) they all took the wrong path by using complex AI. Cockroaches and ants navigate over uneven terrain using very limited processing power. There is no reason why a vehiclle cannot use the same techniques thay use rather than AI.

It would be nice if this could work, but there are at least a few reasons why insect navigation methods are inappropriate. The limited range of insect sensors works well in the insect environment because they can sense far enough ahead to alert the insect to a catastrophic situation.

So if the insect is heading towards a brick wall, the length of an antenna is enough distance for it to slow down without a fatal collision. The energies involved in insect travel speeds are low enough that even a full speed collision is not fatal. This relationship between energy, sensing distance and lethaity does not scale. It is not the case with faster animals or vehicles.

It is at least suggestive that almost all reasonable speedy critters larger than a certain size use vision and some kind of vision processing.

Mitch Berkson

Timo,

you have to look for technical papers in places like:

etc.

As far as I know, none of the teams published in-depth paper about their robot. I would be surprised, if they do it in the future.

Paul.

Track down the websites of the team themselves to see what info they offer for public view.

- dan michaels

=========================

You make a good argument, but there is no reason why one cannot extend the range of insect-style sensors to accomodate higher speeds. And of course ants and cockroaches use vision and some kind of vision processing, yet still use insect-style navigation.

Given the results of the actual race, it would have been interesting to watch an insect-style vehicle slowly compete the course after the AI-style vehicles failed. (of course we dn't know whether they all would have failed; many entrants failed because of mechanivcal problems)

Imitating insect brains is a form of AI. Rodney Brooks made this famous with his rather primitive distributed AI in six legged bots.

Monarch butterflys navigate from Canada to Mexico to breed. I would not call that such a limited range.

Some insects such as dragon flies can fly quite fast. They also have two sets of wings to allow hovering and amazingly nimble flight manuvers. They have the most complex eyes and optic nerve ganglia in the insect world and have remarkable perception. I doubt if most of the robots in the DARPA challange could keep up with them.

At least one of the DARPA robot contestants used insect like AI including vision. They had a small budget and a small team and had programmed their entry in Forth. I might be able to pull up a URL with a little effort. But as you know none of the bots were up to the challange this year. Maybe next year.

Best Wishes

ok, thanks!

Ok, me too. So up to now no one has published in-depth paper - That's what I have already nearly thought :-( Nevertheless many thanks for your support on this newsgroup!

Bye,

Timo

Bats are blind, fly very fast, make turns on a dime, and I have never heard of a bat crashing into a wall... They use sonar.

It is ironic however, that our best AI, robotics, etc, cannot compare to the instincts of the lowest critters.

Saw a show on Army ants tonight... un fricken believable... They use hooks in their legs to bind together to bridge large gaps, create ladders, etc for the rest of the ants to traverse. All done using chemical communications....

I love this field!

However, i agree that the increase in speed of travel necessitates an increase of sensor range, sensor processing, and computing power to handle obstacle avoidance.

Btw, theres a lot at stake here.... DARPA is upping the award to $2million for completing their course, not to mention a HUGE contract to the winner for their autonomous vehicle.

Good luck Mark

I would look elsewhere

Try MIT , everything they do is online :)

Also keep in mind that AI is more than obstacle avoidance... some obsticals can be driven over.... a hill may appear to be impassible at first [horizon recognition]... man that stuff can get complicated quickly

Imagine

Max course length: 300 miles [LA to Las Vegas] Max time: 10 hours Min average speed: 30 miles per hour

Terrain: sand, rock, bushes, plants, animals possibly, and other vehicles

No human control.

Payoff: $2 million [2005 challenge]

wow, sounds like fun.

Bats are not blind, but your argunment is still valid. They fly very fast, make turns on a dime, and don't crash into walls while inside totally dark caves.

In three dimensions.

While thousands of other bats are echlolocating inside the same reverberant cave.

Alas, the only entrant that (as far as I know) even tried insect style vavigation (Ladibug by team phantasm) was a non-starter because the couldn't afford to replace a standard PC. :(

I disagree. I think that the increase in speed of travel necessitates an increase of sensor range alone.

Let's see if they unfairly exclude worthy applicants again this time.

here's some great information i just found, wanted to pass on to you:

each of the links shows a summary of some article or what not, the actual article is a small link at the top says "Adobe acrobat format (pdf)"

I had to really look to find that link to the actual article text :)

Regards, Mark