ai engine research project HELP!!

aiResearch · April 3, 2002, 14:18

hello all!!

I'm doing a small cs research project due in next week (aagh!) entitled: "The ai engine in Sid Meier's Civilization 3" to be exact.

Basically I'm finding it hard to find concrete information about the actual structures used in creating the ai itself and how it was created/written etc.

I've been trawling through sites for about a week now and have found lots of information on playing the game, creating new units etc but not alot on the actual core ai engine itself.

Could anyone lead me in the right direction??

many thanks
p

dunk · April 3, 2002, 14:29

I think you picked a tough topic for your project. All you'll find on these boards is players' theories on how the AI works. There will be some evidence to back up these theories, but nothing past circumstantial. Firaxis is not going to give you any specific details on how it's put together because of copyright concerns.

You should have picked something like AI in Computer Games in general... then you could probably get some source code from a freeware game and disect that for how it works.

DaveV · April 3, 2002, 14:41

Try this link

aiResearch · April 3, 2002, 14:56

Quote:

I think you picked a tough topic for your project

yeah, would you believe you're not the first person to say that!

Quote:

There will be some evidence to back up these theories, but nothing past circumstantial. Firaxis is not going to give you any specific details on how it's put together because of copyright concerns.

sure, obviously Firaxis aren't going to give away their secrets, so turning to boards like this where serious gamers really try to understand and analyze all ai behaviour, can lead to some very interesting "theories"!!

Quote:

You should have picked something like AI in Computer Games in general

the original title was "The history of AI in computer games", indeed a bit more generic, however speaking to my professors they thought this more "selective" title(!) could be (and is proving to be) more of a challenge!!!

Thanks for the feedback Dunk!

would you believe DaveV, that interview was one of the reasons for choosing the more selective topic, however thats about as much info from the firaxis boys as I seem to be able to get! (yet)

p

Inverse Icarus · April 3, 2002, 15:15

make sure to mention that the AI cheats more and hates you more rather than get better as the difficulty level increases

aiResearch · April 3, 2002, 15:41

Quote:

make sure to mention that the AI cheats more and hates you more rather than get better as the difficulty level increases

from a programmers point of view, how exactly would you define "get better"??

"cheats" and "hates" aren't easily programmed functions, more the ai is trying to gain a desired advantage using the information it has, ie its pre-written functions, to "get better" as you progress through the difficulty levels.

"Ambiguity, is not the friend of a good engineer"

Sorry, but this is a serious thread in search of more concrete information/direction/"theories", not usual "whimperings"!!!

p

Inverse Icarus · April 3, 2002, 15:51

basically, one simple way to impliment an AI's "getting better" as you raise difficulty levels, would be quasi-coordinated battle plans.

if you were to play on Diety, the AI should examine the terrain around the area to be attacked and use it to it's advantage (stack some infantry and artillery on a mountain to bombard the hell out of a city). As it stands now, I have never seen the AI use artillery (excpet on defence) and i definately haven't seen the AI use terrain effectively. I've seen enemy units move onto jungle squares near my cities rather than the mountain square, just as easily accesible.

if you think about the difficulty levels in terms of varying strategy, it makes perfect sense. if you made the chieftan AI charge randomly across grasslands (as they do all to often on Diety) and the Diety AI actually use the terrain to make it difficult for the human to blow them away.

Keep in mind, i'm an 18 year old kid, going to Rensselear Polytechnic next year. The only code i write is for enjoyment, nothing professional or extremely in depth. I don't know exactly how such a program could be implimented with Civ3's AI, but give me 4 years of college and i'll figure it out

also note that this was just a war improvement, as i am a warmonger. i'm fairly sure some of the "builders" on this forum can point out ways the AI's domestic game could be revamped.

ALPHA WOLF 64 · April 3, 2002, 17:48

have the AI remember its last few attacks. attacking a tank over and over with swordsman isnt likely win (except those dreading tank killing spearmen). Therefore, the AI should wait to attack until all its units are ready. Attacking 6 swordsmen in one turn might actually have a chance wheres the one/turn for 6 turns isnt likely to ever win.

I've seen discussions on how people setup killing zones between forts. it would be nice if the AI "remembered" those zones and avoided them.

Ethelred · April 3, 2002, 17:50

Quote:

i'm fairly sure some of the "builders" on this forum can point out ways the AI's domestic game could be revamped.

Well for one they could quit irrigating half the grasslands and winding up with entertainers because the population grew faster than they could build happiness improvements. When they get railroads they should replace excess irrigation with mines.

Then again I like it this way. If the AI did those things the game would be a lot harder. I wouldn't be able to out produce with less cities on high levels.

Inverse Icarus · April 3, 2002, 17:52

i've noticed the AI just has a "checkerboard" pattern of irrigation / mines that it hardly ever seems to break.

TechWins · April 3, 2002, 18:31

aiResearch, I think you would be better off posting this in the Strategy forum. Not because this (General) is the wrong place to post this, yet, because you'll be provided with better info in the Strategy area. Possibly reading Vel's strategy guide could help you out too (you'll be able to find that in the strategy section), and talking to Vel himself might be a good idea too. BtW, Vel is short for his real name, which I don't remember the exact spelling of.

Jack_www · April 3, 2002, 19:44

Anther game you could use for your project is freeciv. This game is based on the games Civ2 and Civ 1. I am pretty sure you would have no problems getting the source code for the game and maybe even documentation as well. The website for this game is www.freeciv.org.

I agree with TechWins, the Strategy forum would be a better place to go, also firaxis people post there more often there then anywere else. Also there is a thread in there on the cheats the AI usess againts the human player, but the info in there is not all that accurate.

Ninot · April 3, 2002, 21:30

Quote:

Originally posted by ALPHA WOLF 64
have the AI remember its last few attacks. attacking a tank over and over with swordsman isnt likely win (except those dreading tank killing spearmen). Therefore, the AI should wait to attack until all its units are ready. Attacking 6 swordsmen in one turn might actually have a chance wheres the one/turn for 6 turns isnt likely to ever win.

um... maybe im just as stupid as the AI, but in Industrial to modern age wars, if a wave of my tanks just cant possibly kill all the infantry in a city.. and i have Cavalry close enough (or any other unit that can't/wasn't upgraded for some reason)... i just rush them with everything i have.

However, this usually only happens when the unit is down to 1 hp, and i dont feel like fighting the city again after it has healed and found a few drafted buddies to fight with

but if it gets knocked down to that 1hp... i send EVERYTHING i have after it, regardless how good the unit is. If that next cavalry might be the one who is gonna get the lucky hit on the hurt mech infantry, i find its worth the chance...

However, the AI rushing with nothing but swordmen in an already lost battle... i cant explain it, but i dont see it that often either. I find in my games, the AI usually sends the best its got, right up untill the end...

Inverse Icarus · April 3, 2002, 23:03

the best thing you could teach the AI to do is how to reload and re-try that turn