Slashdot Mirror

We've had Neural Nets for quite awhile. I would have taken the course in 4th year Computer Science but it was taught by a horrible professor I'd already suffered through enough times not to bother.

For the most part they're used for software recognition.

http://nerogame.org/ is a game that uses artificial neural nets.

There's also a fantastic daily wtf article about a company that insisted on using them for something mundane like hydro switching. Someone else can dig of the link.

Speaking of evolutionary behavior in a simulation/game environment...

I liked NERO a lot. The original idea was to have the bots run a capture the flag type game. However I found training mode more fun and interesting by making mazes in the map editor and seeing how AI behavior developed. "Training" my bots to run a maze was a simple yet timely processes of adjusting the reward value sliders for certain behaviors and manually smiting the bots that would get stuck or cling to the walls too much. Usually what would happen once they started getting good at running mazes quickly while showing more interesting behavior than running in circles was that the computer would crash and the last saved "DNA" neural algorithm file would swell from under a megabyte to well over a gig.

Unfortunately nobody has done jack shit in regards to further development since 2007. (At least anything apparent on the public side.) Seems a shame, since it was good up to a point. There should be potential to do more with it. I guess either computers aren't powerful enough to handle the simulation past a certain level or that things just get too complicated on their own which would make debugging nearly impossible.

I figure the NERO bots could also be made to play soccer too, provided there was the right LUA script written. (It used the Torque game engine.) But being that the simulation brickwalls at some point by simply doing maze runs, that seems unlikely at this point.

... the RTS-like game of Neuro Evolving Robotic Operatives. Instead of controlling your units directly, you had bots that you'd develop with a genetic alogrithm in a training mode. (And you'd train them by killing off "family-lines" of the bots that sucked, and using bots with interesting behavior as seeds for sucessive generations.) Then once trained, you'd give them a fairly general waypoint and they'd do the rest on their own. Which was a very cool concept, but in practice it would hang up the CPU/RAM once some really interesting emergent behavior stated to develop. (By the time you "taught" your bots to do something "useful", the game would freeze. And the data files that recorded the behavior would be HUGE!)

Only other thing to be said about NERO is that it's a shame that jack-shit hasn't been done with it since 2007. I think it would have some interesting potential if somebody on the development side could actually figure out how to make things more efficient and not have it freeze up.

There's also supposed to be some OpenNERO derivative, but it appears to be stillborn. :( Looks like it would be an interesting project though.

Here are two great examples of using evolving neural networks to drive game AI:

Nero:
http://nerogame.org/

Galactic Arms Race
http://gar.eecs.ucf.edu/

They're both the brainchild of Kenneth Stanley.
His current research can be seen here:
http://eplex.cs.ucf.edu/

There was this neat game where robots "wanted" to get to a target position on a maze. And by adjusting the other factors (avoidance, aggressiveness, or attraction) you could affect how they developed in relation to their maze solving or goal seeking ability. The neat thing was that the robots adjusted their attributes by successive generations and some kind of evolutionary scheme. So you could punish a behavior by killing robots that exhibited it or reward a behavior by spawning more robots based on that behavior model. The result is that you could "teach" the robots to some modest extent by moving the goal around the maze and watchful "breeding". The funny thing was, robots that got "smart" also had associated files that grew geometrically in size. (Not sure how relevant that is, but might provide some kind of insight.)

Unfortunately the project seems to have frozen at the 2.0 stage of development. But it doesn't mean you still can't check it out.

Anyone ever play NERO? The teaching aspect applied to a simulated neural AI in order to get it to do something is rather true. Likewise it's not always predictable. It's neat teaching game bots to do something, even if it takes time. To think that instead of some developer coming up with some carefully balanced alogrithm to induce a behavior, it's your "reward" and "punishment" of the simulated robots during training mode that makes their eventual behavior in the game.

I just hope the developer of this project can keep the ball rolling. Definitely cool stuff, but it still crashes on my machine too much to really get into it just yet.

I'm not sure if I'm getting it... But now these ideas are making me wonder what would happen if the manifold dynamic qualities of SodaPlay and manifold spatial qualities of Wings3D could be combined... Maybe throw in some other node type stuff ala Nero and you could have a really interesting chaotic and emergent behavior toy. (Spore? Who needs spore?)

It'll probably never happen, but that's my crazy imagination for ya!

I think computer controlled opponents should provide a certain level of challenge without falling into the impossible; and this http://www.nerogame.org/ looks to me like a step in the right direction.

http://www.nerogame.org/
it would be nice to have these ai's trained for, say, ioquake3

You mean, applications of neuroevolution for games? Seems the hot topic du jour. Finally, bots with complex, non-deterministic, evolved behaviour...!

The University of Texas did this: http://nn.cs.utexas.edu/keyword?neuroevolution , game at http://nerogame.org/ . First you train your bots, then you test them in battle.

...the same criticism ID'ers make about evolution can be made of a ton of other scientific theories (in all sciences, not just biology), so why aren't those theories criticized as well?

In a word, technology.

Most other well known fields of science have technological manifestations. Physics: engineering. Biology: medicine. Evolution of complex forms is not used in much technology yet. But that will soon change. Once it does the opponents to evolution will have about as much credibility as flat earthers have today.