AI Researchers Produce New Kind of PC Game

Ken Stanley writes "In an unusual demonstration of video game innovation with limited funding and resources, a mostly volunteer team of over 30 student programmers, artists, and researchers at the University of Texas at Austin has produced a new game genre in which the player interacively trains robotic soldiers for combat. Unlike most games today that use scripting for the AI, non-player-characters in NERO learn new tactics in real-time using advanced machine learning techniques. Perhaps projects such as this one will encourage the video game industry to begin to seek alternatives to simple scripted AI."

Comments

Coral Cache

[Anonymous Coward]

Slashdotted before it even went live. Here is a working link. Downloads are currently at 511, I hope their counter has more than 9 bits...

Not at all new

[Digital Avatar]

This isn't entirely a new idea. CROBOTS, for example, put one in the position of designing AIs that control tanks and then pits them against one another in an arena.

Re:Not at all new

[greg_barton]

The difference between this approach and those previous approaches is the way the underlying neural networks are constructed. NEAT (NeuroEvolution of Augmenting Topologies) constructs new network structures, whereas the old approaches used existing networks and tried to train them with user input. The NEAT approach is far more sophisticated.

Re:Not at all new

[Kiryat Malachi]

God, I loved Omega. I should track that down and play it again; its gotta be abandonware.

(answer, after a quick google: http://www.toadstool.net/games/omega/ has the DOS and Amiga versions, as well as tanks and more)

Why train them to fight?

[AFairlyNormalPerson]

It would be a lot easier to train a robot to train the other robots to fight (in the long run)...Wouldn't it?

Black and White

[Anonymous Coward]

It's been done before.

Never mind!

[XanC]

I got to wondering if the correct usage was codified anywhere, and found this style guide.

The correct reference is to use "The University of Texas at Austin" the first time you refer to the title of the university in text. Upon second reference and thereafter, use "university." When writing for internal audiences familiar with the university, it is acceptable to refer to the university as UT Austin.

Apologies to the submitter!

Torrent

[TaxSlave]

Only for the purposes of helping distribution, and for a limited time, torrent available at nerogame.exe.torrent

Re:AI? on a video card?

[adam31]

but floating point operations aren't exactly optimal for things like AI.

False.

FLOPs are not generally useful for things like scripted AI which are very branch heavy with a lot of indirection, and many possible branch targets and data requirements.

The techniques described in this game are highly mathematical in nature with a small memory foot-print, (adaptive neural networks and genetic programming via Kenneth Stanley's NEAT algorithm) and would benefit hugely from parallel vector proccessing.

Additionally, at the end of the day, the AI decision making is not nearly as expensive as the proximity-query and pathfinding routines that affect the decisions. These routines also benefit hugely from vector processors and high bus-bandwidth.

So fittingly, the AI will only suffer if the human intelligence can't adapt and make the fairly obvious decision to move toward more mathematical AI routines.

Re:Or perhaps...

[Anonymous Coward]

This will seem even more like a flame but I can't think how else to word it.

In Iraq, the uk lost more troops to US "friendly fire" than to the Iraqis.

Unfortunately I'm not taking the piss.

Re: What is old is new again

[Black Parrot]

> Call me short sighted, but isn't it at least possible that training soldiers is different to training tic-tac-toe players?

Yes. Tic-tac-toe has a manageable decision tree, and all MENACE did was prune branches that led to losing. It still required many playings, because it always pruned at the last decision that led to the loss. (Thus it trimmed the decision tree from back to front.) It would be completely untractable for chess, let alone for continuous-state games or simulations.

Still, MENACE was a brilliant insight for the time. IIRC it was done way back in the 50's -- practically the beginning of time so far as computer science is concerned -- and brought to public attention when Martin Gardner covered it in his Scientific American column in the early 60's.

Finally!! Geeky Gameplayer Employment!

[smchris]

Will these things be marketable? "Ma, I'm not playing games, I'm training my robo-warrior!"