Alright im creating a squad based first person shooter type game and i have an idea for AI but i want to make sure im on the right track. My Idea is that overall strategy is fuzzlogic: i would divide the arena (its a sphere type shape) into 5 zones middle, top, bottom, left, right. Then depending on whats happening issue overall commands...ex. left is being routed reinforce from middle or something like that. I would probably decide what commands are needed by checking the states of the men, ex 6 out of 8 dead might need reinforcments Then for indivdual men i would do something like a finite state machine (i think). ex. being shot at...look for cover/shoot back does anyone have improvments/better ideas, im open to anything

You might want to use Influence Mapping to check relative strength of attack and defence. You could also have a ''lookahead'' where you estimate future strength
(i.e. 3 more enemy soldiers heading that way).
You could then use this to plan your reinforcement/withdrawal etc.

quote:Original post by tonyg
You might want to use Influence Mapping to check relative strength of attack and defence. You could also have a ''lookahead'' where you estimate future strength
(i.e. 3 more enemy soldiers heading that way).
You could then use this to plan your reinforcement/withdrawal etc.

I dont think influence mapping would be useful because all the guys have the same attack/defense (think clones). Or am i wrong?

Would this lookahead require a "cheating" ai, as in knowing where all the enemy guys were even if i friendly men cant see them. Im trying to create a non "cheating" ai if i can and make it reasonably difficult.

Basically the AI is broken down into 2 layers, the strategic and tactical element. The strategic element can use a simple rule book approach, where given a set of inputs find a matching set of good outputs and choose from them. That will allow the AI to be somewhat responsive and cohesive on the strateigc front. However it won''t be creative, inspired nor original, as it lacks the capacity to learn and rememebr. High level strategic AIs of this typed are used RTS like Age of Mythology and their ilk.

On the tactical side, using a fuzzy state machine would be the best bet. They are adapatable to a variety of sitautuions the individual agents will encounter and can repsond reasonably well. A few tactical games use this, I think Close Combat uses this as well as the new Worms 3D game, just feed it the various external and internal states.

Because there is so much data an AI has to process, some games assit the AI out by placing markers in the world. These are invisble to the player but notify the AI of things like cover, pathing, or sensory information. I''m sure this will also be a key component of your AI.

Good Luck!

-ddn

If you don''t want to cheat then you should only add the strengths and heading of known enemies. This is only a theory so far. However, if you have, for simplicity, 2 layers of influence map. First layer is ''known'' influence and the second is ''expected'' influence you can not only check which sectors are winning, losing or holding out but also which are likely to be
overrun in the near future.
Imagine the left sector was losing and 5 more enemies are heading for the middle but have yet to arrive. Without the lookahead you might decide to take people from the middle to reinforce the left when, in a couple of mins, the middle is to be completely overrun.
This might be used to add feints, counter-attacks etc.
The accuracy of your lookahead might reflect the experience of the commander or the difficulty level.