Awaken, foul beast! (Game AI)

[MickeX]

If anyone's seriously thinking about tackling this monster, I would suggest starting by seeing if you can successfully program a simpler subset. See if you can program a one-off match of orc vs. humans with premade rookie teams. If this is successfully pulled off then it can later be expanded with other teams, skill selection, etc.

Maybe something like playing goblin vs goblin with no Big Guys, passing or hand-offs?

If you succeed, you'd get a team that can actually form some kind of advancing cage and have a chance to score with it. It'd have to predict and defend against the opponents tries to stop the cage, break it up and blitz the ball-carrier. That sounds near impossible to me, but I'm no AI programmer.

Micke

[bitMonkey]

Thanks everyone for your response! It's really fun to read what you all think about this hypothetical computer opponent. Has anyone decided to give a serious go at building it?

Personally I am actively thinking about it. My approach is very similar to that of TD-Gammon. Meaning that at its heart there will be a function approximation, V, of the true value function of each state. So at every state I can ask V what the chances are of scoring a TD. The policy for choosing actions would then be easy enough to implement I think. This would be where one would have to weigh risk against possible gain. Knocking down Griff with a (standard) halfling is great, but the odds of actually pulling it off are bad.

I plan to use either a linear function approximation, or a multilayer backpropagation neural network to do this. Tesauro used a neural network, but he's an expert at tweaking NNs.

Furthermore I plan to implement only the basic rules with the two teams that comes with the box. To begin with, anyway. If that goes well, who knows. Anyway, knock on wood and all that.