Via the Puppy Blender we see the Slashdot article linking to this page that's an open access summary of an experiment into robot evolution and this page that discusses the first page and has vids of the experiments.
Very cool. Read the whole thing, you'll enjoy it.

I don't really understand how the researchers recombine the robots' "genomes" to make "generations", but they do so I'll let it slide.
What they did was let robots run around doing stuff, the ones who did what they wanted best "reproduced", they kept going until the populations were stable, in other words, didn't learn anymore (except once).

For instance, they told robots to run around longer than their batteries could handle with the instructions on how to recharge, so they had to go home periodically.
After 100 generations they all went back with 10% power left, that didn't change at 200 generations. They obviously developed a Murphy's law.

The exception was the predator/prey scenario.
The predator had to touch the prey, the prey had to run away. The prey was twice as fast but the predator could "see" the prey from 200 times the distance.
The prey figured out the best way to evade the predator and then the predator figured out how to overcome that stragety so the prey figured that out so the predator changed....
Quote (from the first link)
Overall, these experiments revealed that a large variety of sophisticated behavioural strategies could evolve, but none of them were stable over time because of the coevolutionary dynamics.

It never stabilized.
Which makes them uniquely like people I think, these guys must be on the right track.

It seems to me that animals stabilize how they get their prey and how they avoid their predators, that's why new predators or prey can destroy an eco-system.
Man is the animal that best evolves its strategery in killing. It seems to stabilize, but then someone figures out a better way to do it an off we go. That's how empires are founded, like the Romans or the Greeks before them. Not many animals do that and none do it nearly as well.

Interesting and, I also think, important if they can evolve an intelligent robot.

We know lots of stuff, but not a lot of people know a lot about a lot of stuff and nobody can know everything we know all at once.
A brilliant physicist doesn't know all that a brilliant astronomer does and neither of them know what a brilliant mathematician knows.

Asimov, I think, wrote science-fictionally about how a lot of advances were made very quickly simply because a thinking machine could know lots of stuff and correlate it to do unexpected stuff.
That would be cool. If we could make reasoning computers they could figure out important stuff, like how to shut up a global warmmonger or  make the perfect s'mores.

Related, excellent sci-fi books, Code of the Lifemaker and The Two Faces of Tomorrow, both by James Hogan, he's hit and miss, but those two are excellent about evolutionary robotics.
Related Video below the fold, because I'm supposed to.

Posted by: Veeshir at 02:29 PM | Comments (13) | Add Comment
Post contains 526 words, total size 4 kb.