DESIGN: AI Improvement discussion

[Locutus]

FSMs for tactical manouvering? Sounds sensible, that could certainly work. And fairly easy to program.
I really haven't looked into what the current AI looks like yet, do we know what kinds of algorithms it currently uses for this purpose (if anything)?

AI Game Programming Wisdom, you say? I guess I should look into those (if my bank account allows for it )...

[child of Thor]

sounds reasonable

He he - lets face it any info on helping in improving the CTP2 AI is going to be a good thing. I think its quite sophisticated and was made to be - it just somehow never really got the polishing it needed, or at least it feels that way.

I know Pyraray when he is around might have more to say on it, there was some good stuff in those interviews Locutus ran a while back as well - about how the game was not rushed out, which seems a bit odd cause compared to CTP1, there are definatly more erm 'un-finished' bits to CTP2, and the vanilla AI did seem fairly uniterested the second time around?

Still if we can even just fix the bits that are in place(like all diplomacy issues/co-ordinated attacks/AI personalities etc) we will have accomplished good things.

[MrBaggins]

There are two of them.... and each have slightly different stuff. They are $70 US retail each, but I've got them from amazon for a bit more than $40.

When I've got them I'll let you know which is the more applicable.

[MrBaggins]

Originally posted by Locutus
*SNIP*

I really haven't looked into what the current AI looks like yet, do we know what kinds of algorithms it currently uses for this purpose (if anything)?

*SNIP*

Yes... it uses "agents". There is a goal system, which is used by the AI scheduler to task squads (groups of agents.)

The essential difference is that the agents are essentially dumb. They are "told" what to do, pretty much directly from above. Its possible for you to have smart behavior using this system, but typically much harder. You have to deal with many more exceptions, and basically abstract multiple state machine behavior in one pass... not easy.

A FSM is smart in that it listens and talks to other FSM's, and acts appropriately. The individual parts aren't that complicated, but when they work in concert, they are moreso.

[Locutus]

http://www.aiwisdom.com

Interesting, Soren Johnson (of Civ3 AI fame) wrote an aritcle in one of those books about TBS AI as well.

Abstract: Writing a turn-based AI presents a number of unique programming and game design challenges. The common thread uniting these challenges is the user's complete control over the game's speed. Players willing to invest extreme amounts of time into micro-management and players looking to streamline their gaming experience via automated decision-making present two very different problems for the AI to handle. Further, the ability to micro-analyze turn-based games makes predictability, cheating, and competitive balance extremely important issues. This article outlines how the Civilization III development team dealt with these challenges, using specific examples to illuminate some practical solutions useful to a programmer tasked with creating an AI for a turn-based game.

Browsing through the rest of the articles, sounds like 2 fascinating books (another Apolyton-related game that's covered is MoO3: an article about terrain analysis). For the right price I would buy them even if I didn't have any intention of using any of the knowledge (particularly part 2). Just reading about how the AI of some (types of) games works alone is fascinating enough.

[Locutus]

Hmm, my personal experience with agents is somewhat limited, but from what I understand about them that certainly seems like an awkward way of managing tactical manoevering. Based on the knowledge that I have, FSMs seem much more useful in this context.

[MrBaggins]

Chad Dawson (Stainless Steel Studios) of Empire Earth fame wrote most of the stuff that stood out to me, as being most applicable to what I'd *like* to be able to do.

My favorite title of any article wasn't by him, however... it would be "How Not To Implement a Basic Scripting Language" by Mark Brockington (BioWare), Mark Darrah (BioWare) regarding NeverwinterNights. Quite the interesting tale of woe.

[Locutus]

There's lots of stuff there that sounds great. The Strategy articles about terrain and maps and the like stood out the most for me: the Steven Woodcock (Wyrd Wyrks) article about strategic dispositions, the MoO3 article about terrain awareness, the Shawn Shoemaker (Stainless Steel Studios) aritlces about map generation and transport AI.

I know a thing or two about writing scripting/programming languages already, so those articles have slightly less appear to me (though I'd still devour them).

Damn, why did you have to tell me about these books when my birthday is still 6 months away?

[Fromafar]

Originally posted by st_swithin
A* is the WORST pathfinding algorithm EVER!!!

Actually, A* should have been [0] in the array of 'required bug fixes.'

She may have been (partly) right after all. I just found out that the CTP2 implementation of the pathfinding algorithm does not do A* justice. When a node is in the open node list, it is not updated when a better path is found.

So, with the Activision* algorithm, you may get something like:



While with a correct A* implementation, you could get:



Looks much better, doesn't it?

[Flinx]

Does this mean you (think you) fixed it? Yea!

[Maquiladora]

On a different note, can someone look at making the AI place all its tile improvements around the INNER most ring/s before it starts placing on any outer rings? It seems like a dodgy leftover from CtP1 or something. It would help the AI alot.

[Gilgamensch]

Actually I think this would need to be tied with the discussion about Governor. The AI shall build useful, so in other words if surrounded by hills, convert forest to grassland and stuff like this.

[Locutus]

Ehm, the governor and the AI are one and the same thing, Gilg...

[Gilgamensch]

Maybe a bit bad expressed: For sure it is the same, otherwise we would have a schizophrenic AI.

Where I wanted to point at, is that the AI shall build according to the cities surrounding and global goal. And that is something which isn't really happening. It is 'only' building for growth even as it can't grow further. (simplified )

[Ekmek]

bump
for calvitix to use (if he needs it)