The reverse could be said for lost combats.

Or you can win very decisively losing very few units. You can win by a tiny edge over your opponent, losing lost of units and leaving your position weak for counterattack. There are many ways a combat can be won or lost. I agree with you that the combat uncertainty aspect could be 'averaged out' to somehow reduce the branching factor.īut I think that reducing a combat to a simple 'win/loss' split is a too much aggressive simplification. So I was wondering what would be your approach to solve this problem ? I'm still scratching my head at the moment, convinced I could do better than existing AI but still unsure of the right approach to take. On the negative, I see that the branching factor (the number of possible moves on a given turn) is very high, and a small amount of nondeterminism stems from the combat system that involves some dice rolling. On the good side, The game's environment is accessible (information about the game state is fully disclosed to all players), static and discrete (turn-based). I'd like to convince myself that this is just a consequence of sloppy programming and that it is indeed possible to make a computer-controlled player match human skill levels for this particular game.

I have yet to find a game where the computer AI matches a good experimented player. I was playing Axis & Allies board game on my PC the other day against the computer, and, well, the AI of this game is very weak to say the least.

Best axis and allies version