AI Stuff

[CRourke]

I'm curious about the approach you guys are taking to the AI.. I've heard mention of a "strategic AI", which I guess implies the existence of a tactical AI. Is the AI being layered like that? How many layers? Is national doctrine going to be represented for its own sake or will it just be like "If your american, don't go charging german inf with yer shermans"?

I'd like to see the AI's thought process taken directly from national doctrines and training manuals. Not so much for the sake of realism, but because it seems like the instruction given to future plt and co leaders was kept very simple. No one expected them to execute grand involved manuevers. ("Sgt, take 1st and 2nd squads and execute a counter march to gain a double envelopement around their refused flanks. Have Cpl Poole and 3rd squad feign a withdrawal, then wheel left and deploy skirmishers en echelon")

But I digress.. in any case, I think the nicest thing about it is that the simplified tactics taught are most readily adapted to an AI. And, when they don't work out, its easiest to dismiss it and say, "Ya, but a real plt could have made that mistake too".

Also, if things are nicely compartmentalized, one can hardly blame a plt for failing to act in perfect coordination with its neighbors. Its just a fog of war. I think alot of the reason wargamer's complain about AI is that its easy to see what the right thing was from their god's eye view of the battlefield. Real war isn't a series of chess-like moves planned out for 3 hrs in order to get the best odds attack on the krauts killer stack.

And while I'm rambling, its seems like the first thing an AI would want to do would be to sit down (figurtively) and spend some time analyzing the map. Especially with the non hexed based approached, I imagine its pretty important for the computer to simplify the map into "this region is a strong point" "this hill has great observation" "this open ground is death". Maybe this kind of analysis could be really exhaustive, but only need be done once for a map and saved.

Anyway, I'm just curious as to how you guys will get this AI thing to work. With so many factors in both the units and the terrain, I'll be really amazed if you can pull of even a mediocre AI.

Chris Rourke

[Guest Big Time Software]

There are basically three layers of AI. At present, we have only put in one (the lowest level). Here they are:

Tactical - think of this as the reactive behavior of a unit. It is the logic that gets your units out of tight spots, determines when they shoot, etc. Both players' units, human or computer, utilize this AI.

Operational - Gets a unit from A to B in the best way possible for the particular unit, mission type, terrain, weather, etc.

Strategic - This is where the real planning goes on. Units are told what tasks to do, when, and where. Coordination of asset movement, fire, direction, etc. is key. Most importantly, planning to obtain short and long term goals is done here.

There are a whole lot of variables here, but honestly we know we can do a good job with them. Charles has a long history of creating very good AI for very complex game systems. Not only that, but both of us have a lot of practical experience with fuzzy logic. The key to good AI is to make it flexible, and fuzzy logic is the means to that end. The fantastic results produced from CM's Tactical AI lead us to suspect that combination of the three AIs will kick ass. Will it be perfect? Will it never be defeated? No to both. But will it challenge you, surprise you, punish you for your mistakes, minimize its own goofs, alter strategy to fit current conditions, and do it all without cheating? Yup. And that is a whole lot more than what most, if not all, other games do.

Steve

[This message has been edited by Big Time Software (edited 04-29-99).]