Better Single Player AI

posted on January 30th, 2011, 3:05 pm

good luck in accomplishing all of that

you make it sound a lot easier than it actually is.

posted on January 30th, 2011, 3:08 pm

good luck in accomplishing all of that

you make it sound a lot easier than it actually is.

Which is why I (in the first post of this thread) asked people to test my numbers.

Because it will take lot of tinkering and testing to reach the best compromise juggling the numbers.

posted on January 30th, 2011, 6:14 pm

Dim With all due respect to your Knowledge of FO you don't know what you are talking about. You really need to go talk to him and get the full 411 before you go running around declaring what he has and has not be able to accomplish.

While Freyr's buildlists won't direct port to FO They do provide a means by which you can program a FO AI. The behavior perameters should port over however as the Devs have not riped the old AI out yet. That wont happen until they redo it. I know it is a short term solution but it is better then stabing in the dark for answers.

Feel free to explain to me why I don't understand it - but having played the Klingon Merciless one he released on Filefront and seen how it works and plays, I do feel I have a fairly valid understanding of the way it operates .

I stand by it being impossible to make an A1/A2 AI that, without using cheating, can compete with a decent player.

posted on January 30th, 2011, 6:24 pm

Who cares.

Thats not relevent to the thread.

posted on January 30th, 2011, 6:45 pm

Please be more respectful Tok'ra - this is a forum, not your home.

posted on January 30th, 2011, 9:41 pm

As said in earlier posts, we do not consider the current AI to be a competitive gameplay format. It's for mission-like "lets whoop massive fleets!" games.

We are working on features enabling an advanced AI that may compete with a human player and still be interessting (without cheating). Will probably lead to a complete replace of the AI system sometime.

If you are looking for intelligence, try human intelligence! we managed to assemble you a human intelligence pool over there in the tunngle channel.

posted on January 30th, 2011, 9:49 pm

"lets whoop massive fleets!"

shoop da whoop?

posted on January 30th, 2011, 10:08 pm

As said in earlier posts, we do not consider the current AI to be a competitive gameplay format. It's for mission-like "lets whoop massive fleets!" games.

We are working on features enabling an advanced AI that may compete with a human player and still be interessting (without cheating). Will probably lead to a complete replace of the AI system sometime.

If you are looking for intelligence, try human intelligence! we managed to assemble you a human intelligence pool over there in the tunngle channel.

You keep saying the same thing.

This isnt intended to be a major AI overhaul, it's just fiddling with the build prce/time modifiers until the AI is more of a challenge on easy, but less spammy at medium/hard.

Yes, I know theres MP servers (I dont use them myself) and I dont care. This is for SP flops, not MP.

posted on January 30th, 2011, 10:43 pm

yep thats fine Everybody is free to mod the AI to his or her taste

I just responded to the point that a competitive Fleet Operations AI could be done with the Armada 2 AI system.

posted on January 31st, 2011, 1:37 am

You keep saying the same thing.

This isnt intended to be a major AI overhaul, it's just fiddling with the build prce/time modifiers until the AI is more of a challenge on easy, but less spammy at medium/hard.

Yes, I know theres MP servers (I dont use them myself) and I dont care. This is for SP flops, not MP.

How will tweaking the AI make them less spammy but just as hard?

posted on January 31st, 2011, 5:08 am

Feel free to explain to me why I don't understand it - but having played the Klingon Merciless one he released on Filefront and seen how it works and plays, I do feel I have a fairly valid understanding of the way it operates  .

This AI is at least 2 generations out of date, thats why.

I stand by it being impossible to make an A1/A2 AI that, without using cheating, can compete with a decent player.

I do not agree.

Dim

Sorry for the spelling error

I just responded to the point that a competitive Fleet Operations AI could be done with the Armada 2 AI system.

Never said it could I just said that an accetable placeholder could be fashioned. But  knowing how to code the stock AI right couldn't hurt when the time come to fix and improve it.

[Tok`ra]

posted on January 31st, 2011, 6:20 am

Last edited by Tok`ra on January 31st, 2011, 6:30 am, edited 1 time in total.

How will tweaking the AI make them less spammy but just as hard?

edit: answer removed.

You're not worth answering.

Read the bloody thread.

posted on January 31st, 2011, 9:27 am

Never said it could I just said that an accetable placeholder could be fashioned. But  knowing how to code the stock AI right couldn't hurt when the time come to fix and improve it.

thanks for the help offered, but we already know how the A2 Ai works We are happy with the current AI as a long-term-placeholder, but feel free to make a mod with one thats more to your taste. We are always for community input.

posted on January 31st, 2011, 10:33 am

Why not analyze the gameplay of the good players and implement it into the AI?

Sure, then you have about 500 additional scripts running, that compare many other factors together and works like a chess-computer. This is serious work and will halt the rest of the project for a year or more.

And when a new patch comes out - the AI must be rewritten or you code it in a dynamic way - then its a hell of work. Either way, the normal AI should get 125% of the players ressources, the hard 175% and the merciless 250%. That would be more fun to play I think.

posted on January 31st, 2011, 9:33 pm

Since you've basically just gotten the answers straight from the horse's mouth, I'll just say these few things Starfox1701

Given that I can't discuss what hasn't been released, I'll start off with two base assumptions from a gross simplification of RTS gameplay:

1. Players optimize their build orders based on unit composition, cost and efficiency, time to construct, how quickly they can multitask, and what the opponent's build order is (among other things). Logical so far?

2. Players optimize their tactics by knowing when to engage, where to engage, and when to retreat - and also when to be able to guess at those things.

Consequently, in order to make an AI that can compete with a player, it has to be dynamic - it must be able to determine enemy force content (that includes size and strength), probably locations of attack (which, if the AI only can rely on sensor data like a human player, then it must have a predictive system).

It must also be able to change its build order according to what the enemy has done, accommodate resource falloffs from destruction, and know when to tech up, or stay down in the tech tree, amongst a great many more variables that make a poor player a better, or even a great player (like knowing when to cloak and run, knowing when to use special weapons or save them, knowing when to sweep through mining instead of staying and killing all of it).

Given this enormous quantity of variables and possibilities of strategies and tactics and the fact that there hasn't yet been an AI in an RTS which can reliably replicate a decent percentage of the ones necessary to good gameplay tactics... well, I leave you to the logic

In A2 you can program the build list, you can even tell the AI to change a build list when encountering certain large scale variables (did you attack with this many units etc), and you can change how many units the AI defends with, attacks with, what it guards, and other force-composition variables.

However, you cannot get it to predict attack routes, nor split up forces as a human would, nor even mine like a human player would.

In essence, you cannot get the AI to do most of what a human player can do. The one upside is that you can have the AI in A2 reassess its position a good deal faster than a human player can, but you cannot control for individual battles - only the entire game (aka, you cannot cause the AI to only fire a few special weapons in one instance, but all of them in another dependent on what units are there, etc). Even if it were possible to program with A2's primitive system, you'd need thousands of additional variables, because that's how many possibilities are being used. Do I use Phase Plates when there is only a single B'rel to my 3 Rhienns? How about a single S-2? Or Negh'var? Or do I just cloak? Or if there are 2 B'rels? Do I cloak then, or do I use Phase Plates? Or do I keep some units cloaked to cause psychological issues for my opponent?