Attack AI improved in CMBB?

[BeWary]

Hi all,

I'm guessing this has been discussed already but a quick forum search was fruitless, so... will the A.I. for the attacking computer player be enhanced in CMBB? It could be quite, shall we say, lacking in CMBO. For instance, I recently played All or Nothing as the Germans against the CPU, and 75(!) turns later the computer had barely moved 1/2 way across the map. It had a ton of tanks but it moved them so slowly and without direction that they never got anywhere. It was kind of anti-climactic.

Thanks,

BeWary

[Schrullenhaft]

Unfortunately such leaps in AI 'intelligence' come at excessive costs in development time. While there are some minor improvements to the AI, it's still a bit subjective to say how much of an improvement these tweaks are. I don't think any of them will be to the level of improvement that many people are hoping for. In the past it's been said that it takes a lot of coding time for Charles to make even small behavioral changes in the AI.

I don't know what the engine rewrite will hold for us in terms of an improved AI. There are so many things on the plate for improvement already that I have no idea how much time will be allotted for AI coding beyond the necessary changes for the new engine.

[Tarquelne]

In case you haven't seen it: There's a thread that discusses how a scenario can be designed to help the AI attack.

Or maybe it was just a tip on flag placement. Try another search, if you wish.

[Commissar]

Actually BTS posted something on this a long way back. If I remember correctly, they said that the AI would not be greatly improved. The reason was that they could spent a whole lot of time on the AI but only get very minimal improvement. They felt that time was better spent on other parts of the game. I believe the substantially improved AI will most likely have to wait for the engine rewrite. Anyway, if you want a real challenge, you need a human opponent. No matter how great the AI, you really can't beat an unpredictable, human opponent.

[ July 24, 2002, 01:36 AM: Message edited by: Commissar ]

[chris talpas]

I still think Battlefront should consider contracting out the AI for the rewrite. I nominate Strategic Studies Group because: 1) they have a proven AI track record. 2) they know wargames.

Chris

[Bruce70]

OK I appreciate all the difficulties that the average developer has with AI and I perosnally beleive that the AI for CMBO is better than average. But I am sick of reading comments like this:

Anyway, if you want a real challenge, you need a human opponent. No matter how great the AI, you really can't beat an unpredictable, human opponent.

(not having a go at you in particular Commissar)

As an AI researcher I have to say that this simply isn't true. To a certain extent it is true of typical rule based AI but it certainly isn't true of AI in general. Rule based AI has been terrfic for relatively simple games like chess and so game developers have assumed, incorrectly, that rule-based must be the way-to-go in general. However you simply can't cover all the possible states of a complex game like CM, you need an AI that can generalise and improvise. AI's do exist that can do this and they do not require a lot of development time or even too much CPU time (for a turn based game like CM). The sooner we stop saying "you can't expect too much from the AI" the sooner we will start seeing some decent AI.

Sorry for the rant.

[Thomm]

Well put, Bruce70!

I enjoy watching wargames evolve, because they are very challenging testbeds for AI development!

With regard to the CMBB AI, nobody has yet answered whether the AI supports covered arcs on its own, and, if not, whether they can be pre-set by the scenario designer.

Also, does the AI use the Hull Down and Scoot and Shoot commandos; both of which imply better AI tactics per se.

Regards,

Thomm

[CMplayer]

Originally posted by Rollstoy:

With regard to the CMBB AI, nobody has yet answered whether the AI supports covered arcs on its own, and, if not, whether they can be pre-set by the scenario designer.

Bump this question because I'd like to hear an answer as well.

[KNac]

The AI.

Probably one of the most challenger things in computer´s world is to program any kind of AI, along with that programming real physics and collision models, or complex 3d engines.

I´m not a programmer (at least a pro programmer) but I´ve been in computer´s world for a bunch of years, I know a lot of things about it and even if I don´t know a lot about hardcore programming, I know a lot of other areas (some involved in the gaming world). I know that the 98-99% of the population just would get insane trying to code relative easy things. Hardcore programming requieres a mathematical knowledge that it´s not reachable for most of the population

In a logical way, you should deduce how difficult is to "recreat" the human mind. You know that computer´s try to emulate human mind, something that even us (humans) don´t figure how runs, even for the most simple things. Just imagine.

I LMAO when I read some stupid reviews saying that "the AI is poor" (should I say SC review in GameSpot?) Probably if they know... however I LMAO in all the crtiqs about all things, those guys should get a real job

Now returning to the original question. You musn´t know a lot of programming to know that programming atack/agressive AI is much more difficult than defending AI. Just because the level of complexity and variables is much higher. Just counting "where should I move my troops?" increase the level of difficult a lot.

As other have said I prefer the working more on other 100 features than in the AI which is a very fine time-comsumer

For the future, IMHO they should contract an other programmer or as other said, contract the AI with other company.

[Commissar]

Originally posted by Bruce70:

OK I appreciate all the difficulties that the average developer has with AI and I perosnally beleive that the AI for CMBO is better than average. But I am sick of reading comments like this:

Anyway, if you want a real challenge, you need a human opponent. No matter how great the AI, you really can't beat an unpredictable, human opponent.

(not having a go at you in particular Commissar)

As an AI researcher I have to say that this simply isn't true. To a certain extent it is true of typical rule based AI but it certainly isn't true of AI in general. Rule based AI has been terrfic for relatively simple games like chess and so game developers have assumed, incorrectly, that rule-based must be the way-to-go in general. However you simply can't cover all the possible states of a complex game like CM, you need an AI that can generalise and improvise. AI's do exist that can do this and they do not require a lot of development time or even too much CPU time (for a turn based game like CM). The sooner we stop saying "you can't expect too much from the AI" the sooner we will start seeing some decent AI.

Sorry for the rant.

Can you name a game that really had the AI you talk of? I've been a gamer for a long time and have yet to see any AI give me the challenge that a competent human opponent can give me. Even really good AI pales in comparison. If getting this AI that can "generalize and improvise" is as easy as you say, why haven't AI game programmers not done so? I just have a hard time believing that the lack of really challenging AI is a result of a mere misunderstanding or misperception by the world's programmers. If you are really in the know, perhaps you can let Charles know how such an AI can be easily programmed. An AI in CM that actually knows how to attack would be very welcomed.

Another thing is that it's much more satisfying for me to beat a human opponent than it is to simply beat the computer. When I beat the computer I just beat a machine, a bunch of mathematical formulas. It's not the same as crushing the hopes and dreams of a human player (game-wise speaking). But that's probably because current AI in games is either much too easy or too hard and always predictable.

[ July 24, 2002, 10:24 AM: Message edited by: Commissar ]

[CMplayer]

Originally posted by Commissar:

Another thing is that it's much more satisfying for me to beat a human opponent than it is to simply beat the computer.

On the other hand, a well-made single player scenario can feel like a challenge, since you know that the designer honed and balanced it, and optimized it to how the AI functions.

In that way it becomes like having a human opponent.

[WWB]

Damn skippy. The AI does not handle QBs so well--the target is way too broad. But one can use a trick or three to manipulate the AI in scenarios, making it a much more effective opponent.

And, when on the defensive and given a proper setup, the AI can have a can of whup ass.

WWB

[BeWary]

Originally posted by KNac:

<snip> Now returning to the original question. You musn´t know a lot of programming to know that programming atack/agressive AI is much more difficult than defending AI. Just because the level of complexity and variables is much higher. Just counting "where should I move my troops?" increase the level of difficult a lot.

<snip>

Well sir, I actually have a B.S. in Computer Science and took an Artificial Intelligence course as part of it, so I am not as dumb as you propose. I simply stated a fact, that the A.I. can be lacking in certain circumstances. I never said it would be easy to fix.

Back to my point... yes, human opponents are indeed fun and challenging, but my style of play doesn't often work too well with a human opponent. The choices are either PBEM, which gives a quick few minutes of play, and then it's over for the day (most likely), or TCP/IP, and I rarely have hours to spare at a time to play. So the A.I. is my opponent 99% of the time, and sometimes I'd like to be able to defend instead of always attack.

As for the scenario I played, All or Nothing, it was included with the game and seemed to have good flag placement that would encourage the A.I. to advance (i.e., the BIG flags were in town, with only one little flag on a bridge in the middle of the map). In was raining in that scenario, but in the AAR only about 10% of the computer's tanks were bogged/immobile, so that wasn't the main problem.

A couple of quick suggestions if I may. I noticed that the A.I. seemed to cancel the movement orders of most tanks between turns. This caused further delay as they stopped and waited for the command delay to pass. Something like that could probably be tweaked for CMBB.

The computer's tanks kept running into each other (i.e., bunching up), which added a lot of delay as the computer tried to figure out how to resolve the traffic jam.

The A.I. also liked to go the hard way. It decided to go across wet, hilly fields, and even up steep slopes scattered with trees instead of sticking mainly to the roads, which would have gotten them to the town MUCH faster. I think out of 30+ vehicles the A.I. had, I only saw 2 ever use a road, and then not for long. Keep in mind that most of the roads were concrete and therefore un-mineable (except for daisy-chains), so the A.I. should have used them more.

I'm not asking for perfect attack A.I. In fact, I have had a number of rather good times defending against the A.I., but there's definitely room for some tweaking.

BeWary

[Vergeltungswaffe]

I won't even begin to try to claim that a human opponent isn't a bigger challenge than the AI, in this, or any other, game.

However, for this particular, most wonderful, game, I will tell you that if you play by the hardcore Franko's True Combat rules, the AI can definitely present a very entertaining game for those times that playing a human isn't feasible.

Look for any post of his, a link to it is in his signature, plus some other places, none of which I happen to recall off the top of my head as I am already intimately familiar with them.

[TSword]

BeWary,

My impression of your described behaviour of AI tanks has to do, beside other factors, that they are trying to get LOS or avoid LOS on already identified targets (AI knows very well the sweet LOS spots).

I'm also more than convinced that a lot of "bad" AI behaviour in the attack role could be atleast partially be overcome by more options for the Scenario Designer (Avenues of Attack, triggers and the like..).

This would be available for a predictable price, while investing in improved "AI" is a very risky and expensive business with only minor results.(As long as AI is not able of Learning (Improving in doing the same over and over again) -> Possible today but very expensive, not to mention higher forms of intelligence/learning taking (multiple) contexts into account -> No way in current AI-science). CM-Example: Sometimes Hills are good to possess, other times not, the context or surrounding defines wether it's a worth objective or not -> No AI capable to do today.

So BTS will no option be left, then to improve control for Scenario Designers.

Greets

Daniel

Greets

Daniel

[BeWary]

Tsword,

I initially thought the same thing about the tanks trying to stay out of LOS of my units, however, after losing a couple bunkers I had no units that the computer could see. In fact, I couldn't see his tanks for most of the game, because I had no guys near him. All my guys were in a town on the other side of some big hills. The A.I. did a good job of pushing forward its infantry, but it left the tanks behind so they got chewed up.

As far as A.I. not being able to take into account various possibilities, I would have to disagree. That's what A.I. does. It analzes all of the possibilites and chooses the best one for what it wants to accomplish. There are various approaches (algorithms) to accomplish that. I've actually been very impressed on many occasions at what the A.I. has done in various situations. I think Charles did a great job with it. I was just hoping for a little improvement in CMBB.

BeWary

[Scipio]

Originally posted by Bruce70:

Anyway, if you want a real challenge, you need a human opponent. No matter how great the AI, you really can't beat an unpredictable, human opponent.

I can only second that! And this will not change until AI is not just a term, but really artificial intelligence.

The CM AI is okay, it's good to train the basics, and it can even beat you. But a human opponent is the only real challenge.

[xerxes]

If scenario designers had some control over the a.i. it would be an entirely different story.

[Sheff]

quote:

--------------------------------------------------------------------------------

The A.I. also liked to go the hard way. It decided to go across wet, hilly fields, and even up steep slopes scattered with trees instead of sticking mainly to the roads, which would have gotten them to the town MUCH faster. I think out of 30+ vehicles the A.I. had, I only saw 2 ever use a road, and then not for long. Keep in mind that most of the roads were concrete and therefore un-mineable (except for daisy-chains), so the A.I. should have used them more.

--------------------------------------------------------------------------------

Actually, this is an issue that infringes on path-finding, which is another algorithm issue all together. Pathfinding, like AI development, has it's own development. Programmers can, depending on the project, spend a lot of time developing a new algorithm; some time tweaking an existing algorithm; or simply transplant one that has already been developed somewhere else. They can require a wealth of mathematical knowledge and insight and can be VERY time-consuming. I'm not trying to cut anybody down, but hopefully to increase the appreciation that all of you already have for all the hard work these guys do to bring you good games to play and how gifted they are (even if you believe their AI to be "okay").

[Scipio]

Originally posted by xerxes:

If scenario designers had some control over the a.i. it would be an entirely different story.

Yes, scripting would be very helpfull.

[BDW]

Hey guys - I've found a way to make the AI very challenging on either defense OR attack. All I do is give it 175% force strength. It makes it pretty fun. You pretend like you are up against a drunk with a huge force at his disposal! (and I get my ass kicked about half the time. especially when defending against the computer attacker who has 175% force on top of the bonus for attacking)

[Grisha]

Bruce70,

You say you're an AI researcher. Does that mean you do code? If so, how far is AI coding these days? To what extent are neural networks used and is there anything else out there these days? Finally, how is AI coding not as taxing as before on a system, and is this due to more processer power or better code algorithms? Basically, I'm just curious as AI code has always fascinated me (in college was seriously considering robotics for a time).

[ July 24, 2002, 03:13 PM: Message edited by: Grisha ]

[Doug Beman]

Originally posted by Bruce70:

OK I appreciate all the difficulties that the average developer has with AI and I perosnally beleive that the AI for CMBO is better than average. But I am sick of reading comments like this:

Anyway, if you want a real challenge, you need a human opponent. No matter how great the AI, you really can't beat an unpredictable, human opponent.

(not having a go at you in particular Commissar)

As an AI researcher I have to say that this simply isn't true. To a certain extent it is true of typical rule based AI but it certainly isn't true of AI in general. Rule based AI has been terrfic for relatively simple games like chess and so game developers have assumed, incorrectly, that rule-based must be the way-to-go in general. However you simply can't cover all the possible states of a complex game like CM, you need an AI that can generalise and improvise. AI's do exist that can do this and they do not require a lot of development time or even too much CPU time (for a turn based game like CM). The sooner we stop saying "you can't expect too much from the AI" the sooner we will start seeing some decent AI.

Sorry for the rant.

Yes, but look at how long it took, and how much computing power was used, to create a rule-based AI that could beat a human in chess.

DjB

[Scipio]

Originally posted by Doug Beman:

</font><blockquote>quote:</font><hr />Originally posted by Bruce70:

OK I appreciate all the difficulties that the average developer has with AI and I perosnally beleive that the AI for CMBO is better than average. But I am sick of reading comments like this:

Anyway, if you want a real challenge, you need a human opponent. No matter how great the AI, you really can't beat an unpredictable, human opponent.

(not having a go at you in particular Commissar)

As an AI researcher I have to say that this simply isn't true. To a certain extent it is true of typical rule based AI but it certainly isn't true of AI in general. Rule based AI has been terrfic for relatively simple games like chess and so game developers have assumed, incorrectly, that rule-based must be the way-to-go in general. However you simply can't cover all the possible states of a complex game like CM, you need an AI that can generalise and improvise. AI's do exist that can do this and they do not require a lot of development time or even too much CPU time (for a turn based game like CM). The sooner we stop saying "you can't expect too much from the AI" the sooner we will start seeing some decent AI.

Sorry for the rant.

Yes, but look at how long it took, and how much computing power was used, to create a rule-based AI that could beat a human in chess.

DjB</font>

[zukkov]

guys, you better be careful what you wish for. one day in the not too distant future, the computers may be saying the same thing about us(ie: "i wish they made smarter humans, it's just too easy to beat them. guess i'll just have to keep playing myself to get a good game!"). and if an arnold swartzeneggar look alike shows up at your door some day, then you can kiss your wargaming days goodbye. lol..