Suspect A.I. in CM 1. Will CM 2 be better ?

[-=reef=-]

I have just come from playing a river crossing game. The map was 1.5km wide with 2 bridges crossing a deep, wide river. The computer player, supposed to be defending decides to cross the river and attack me while im advancing on the bridges. But instead of heading for the bridges, he moves towards the river till he cant get his units across, then walks all his troops along the river till he gets to the bridge. Reminded me of watching lemmings. Needless to say, exposing most his force on the bank of the other side meant i just shot him to bits while he tried to cross. Shame as the game was fun till that point.

Will CM 2 A.I. be any better ? Seems a shame to waste such a good game with poor A.I. Ill stick to multiplayer i think.

[Joe Shaw]

<BLOCKQUOTE>quote:</font><HR>Will CM 2 A.I. be any better ? Seems a shame to waste such a good game with poor A.I. Ill stick to multiplayer i think.<HR></BLOCKQUOTE> I certainly can't answer for BTS, though I suspect that few changes in the AI can be expected beyond minor tweaks. I do, however, disagree with your characterization of the AI as "poor".

Any rating of the AI in CM has to be a relative one. I've yet to see ANY AI in any comparable game that can consistently match a real person. I've yet to see SOME real people always play cleverly. Anytime that you present the computer with a non-vanilla terrain mix you run the risk of goofy behavior, but a single test is hardly grounds for the wholesale indictment of the AI as "poor."

There is no question that you're going to get a better match from a real person than you will from the AI ... USUALLY! But every once in a while, and frankly more often than not, you'll get your butt handed to you if you don't play well against the AI.

In conclusion and speaking relatively, I think the AI in CM is miles ahead of the pack and I am quite satisified with it. Is it perfect? Of course not, and it never will be because the judgement of it's performance will always be subjective and it's not a person ... just a clever computer program.

Joe

[Cull]

Yes, there's always little stuff to nitpick, and occasional AI disasters.

I get beaten by the computer---and I think I'm a better-than-competent player.

The AI is lacking enough to cause me to usually give the computer a force advantage, which can make it quite challenging. And in my own scenario-design experience, I find it takes a careful design and tweaking to provide the AI with the ability to challenge a human. Granted, these are "compensation" techniques, but I agree with Joe that it is still quite impressive for a game so rich in detail and realism. There is so much beneath the surface of this game that I have yet to tire of AI games after a year of playing. I urge you not to jump off too quickly, reef!

[Schrullenhaft]

That's a bit of a pathfinding issue. I'm a bit ignorant of how CM's AI is coded up (or any other AI for that matter), but I believe that it has a hard time with such a barrier to movement (one possible reason why there are no rivers in QBs). The AI finds the river edge and then probes along it to find a crossing. Something supremely simple for a human but quite difficult to program in a pathfinding routine (especially one that could maneuver around an obstacle without coming into close proximity to it).

Charles may come up with something to solve this, but such work may have to be left until work commences on CMII and even then AI pathfinding is a difficult issue to program; so no guarantees on some problems.

[David Aitken]

This is the classic AI thread. Go and read it:

CRAP AI, Other BUGS ,Fix Or implement IPX/TCP!!

[Stacheldraht]

The AI does have quirks, blindspots, and easily exploitable propensities, but it's still amazingly good, considering all that's required of it.

[Jeff Heidman]

The AI in CM is like democracy as a form of government: It sucks, but its better than the alternative.

I can't say that I have played a game against the AI in some months. It is pretty boring. Even if it doesn't do anything blatantly stupid, it is still very precitable.

I would be MUCH more interested in seeing the Tac AI improved rather than the computer opponent AI.

Jeff Heidman

[Londoner]

I agree with both arguments here. The AI sucks, full stop. Unless you make the force sizes really silly, beating the AI, even on human designed maps is very easy. But and it is a big but, what can i expect here, I've got a cele 433 not Big Blue ffs. With the hardware to hand, I think BTS has done the best that is possible with the AI, considering the complexity of the game. We aint playing chess here.

[Keith]

I am sorry but I must take issue with those that the "AI sucks". I rank the AI, particularly the tactical AI, to be one of the best in the industry. Often funny behavior can be attributed to the inexperience of the scenario designer. For instance, if you capture an enemy objective flag the AI will try to counter attack. The scenario designer should put the flag on the far side of the bridge if the designer did not want the defenders to cross the bridge and get mowed down counter attacking.

In the end you will ALWAYS need to find a human opponent if you want a real challenge!

Saying things like "It Sucks" is really not to productive and is intended to tick the software developers off. Quite juvenile.

[ 05-09-2001: Message edited by: Keith ]

[Terence]

<BLOCKQUOTE>quote:</font><HR>Originally posted by Keith:

I am sorry but I must take issue with those that the "AI sucks". I rank the AI, particularly the tactical AI, to be one of the best in the industry. Often funny behavior can be attributed to the inexperience of the scenario designer.

[ 05-09-2001: Message edited by: Keith ]<HR></BLOCKQUOTE>

This was going to be my question. Keith, you said it well.

Its true that the AI does do some crazy things from time to time, but often that is because the scenario is set up in such a way -- with exit zones and flags that make the AI operate strangely.

REEF: What scenario were you playing?

In the scenario you describe, were I the designer, I would have put a flag in the midst of a good defensive spot within LOS of the bridge, on the opposite bank.

Then, the AI would probably have given you a much harder time.

Unless of course that WAS the case. If it was, I plead ignorance.

[russellmz]

well, the ai is impresssive without rivers:

i once made a large hedge maze and the ai got a tank to solve it....impressive!

http://mapage.cybercable.fr/deanco/hysterical.htm

[ 05-09-2001: Message edited by: russellmz ]

[Deadly Dave]

Ok so lets accept that th eAi has limitations how do we make co9mpensations for better games.(lets find a solution)

how to get a better balanced play even from axis/allied pbem games.

My gripe is this if the allied /axis sides are equal (?) why do so many people always want to play Axis.

I was playing a game last night that my buttoned tank was killed by a sharpshooter at 480m Okay i can live with a flukey shot now and again -BUT in the same turn i had a tungsten shell and 2 Ap shells bounce off a hertzer at 600m while I was staitonery and so was he, while at the same time the hertzer pinged my A. car travleing at full pace at nearly 700m away. To add insult to injury I had yet another tank blown out by AT gun from 900m while the dozy tank was trying to kill an infantry mob 400m away. Now this is all in the same turn!!!!

So if I'm a total loser let me live wth that -BUT if the AI is not up to speed how can I give the allied the extra edge (in points at set up ) to compensate. And if so what percentage any thoughts from more experienced players

[Col Deadmarsh]

<BLOCKQUOTE>quote:</font><HR>Originally posted by -=reef=-:

Will CM 2 A.I. be any better ? Seems a shame to waste such a good game with poor A.I.<HR></BLOCKQUOTE>

Obviously, you've never had the pleasure of playing the Close Combat series...

[JMcGuire]

<BLOCKQUOTE>quote:</font><HR>Originally posted by Deadly Dave:

My gripe is this if the allied /axis sides are equal (?) why do so many people always want to play Axis.<HR></BLOCKQUOTE>

Whomever told you they were equal obviously just wanted to you to agree to let them play the Axis... They are anything but equal, and rightfully so (the Germans had all the cool toys).

[Jeff Heidman]

<BLOCKQUOTE>quote:</font><HR>Originally posted by Keith:

I am sorry but I must take issue with those that the "AI sucks". I rank the AI, particularly the tactical AI, to be one of the best in the industry

[ 05-09-2001: Message edited by: Keith ]<HR></BLOCKQUOTE>

So?

That does not mean it doesn't suck. It just means it doesn't suck compared to other AIs. I would agree with that. In fact, I think I said exactly that.

As a tool to provide a challenge to a human player who actually has a pulse, it sucks.

That is not a knock on Steve or Charles, it is just a statement of the reality relating to where AI is in the industry, and not just the game industry. It is not reasonable to expect Steve and Charles to be able to accomplish what dedicated sceintists have not. If they could, the world would be much better served if they used that knowledge for something more substantial than a really good computer game.

Jeff Heidman

[Stacheldraht]

Well, I think a number of us have become jaded as we've played for months and seen our skills improve against human opponents. Personally, I remember thinking that the AI was pretty tough and convincing when I first started playing, and I'll admit to losing my share of games to it back then. But in time, particularly after lots of PBEM games, I've come to see the AI as a pretty darn poor opponent, and I beat it handily every time.

It's still a superb achievement compared to most games, though game AI has been showing marked improvements of late with titles like B&W.

[MikeyD]

There's criticism and then there's constructive criticism. Is there anything we users can suggest to help make the AI less predicatable/more intelligent?

My idea is to give the AI a 'left/right/center-dominant' function. The AI could randomly(?) select which side of the board to focus on, then choose the best attack route for that side. That would make for some unexpected (and probably occassionally strange) tactics to defend against.

Just a thought.

[-=reef=-]

:eek: Wow, so many responses. The information provided has been very useful. I understand that rivers are the main cause of the AI problems, and will now be able to create much better mini campaigns from what i have read here.

For those that asked it was my own campaign. After spending hours making it and then seeing the AI do foolish things, i was a bit sour for sure, hence the post.

The scenario basically led through bocage to a river with 2 bridges and a ford. On the far bank was a town. Some bocage, the river and 50% of the town were on one of the maps. The German AI then decided to leave the very defensible town and launch an attack across the crossing points with the effect of my origional post. This is mainly what i found annoying.

However it seems there is not alot i can do to get the outnumbered Germans to hold instead of launching futile attacks. Another Campaign for the scrap heap.

Ive been building mini campaigns of D-Day and the push out from the beacheads. All Historic and maps as near as i can recreate them. But i have never been able to get the balance just right so they can be played succesfully from either side. MAybe i should just build multiplayer Campaigns.

Thanks for all your information, much appreciated. Regards, reef.

[WWB]

Well, there are a few, realatively simple things which would help the AI greatly.

1) Give it some ability to keep formation. If you have ever looked at the AI deploy you know how random it is. Keeping platoons together would do it a world of good.

2) Give scenario designers the ability to place some sort of faux flags, which would behave as if real flags for the AI but would be inivisible to players. It would greatly help the AI in operations to have some sort of guidance. Possibly even add the ability for scenario designers to divide forces and map into sectors.

3) I have never, ever seen an AI that can pull of a river crossing op, so lay off of that one. It is tough enough for a human to accomplish, much less a PO.

WWB

[Stacheldraht]

Teach the AI to cover its movements, particularly attacks/assaults, with smoke screens, and not just use smoke to temporarily silence tanks or guns (though it usually resorts to HE arty for the latter, happily).

[Schrullenhaft]

I don't think that there is anything we can suggest to "fix" the AI. Unless you have extensive knowledge of AI programming and you've discerned exactly what type of environment has been created for the AI to operate in. All most of us can do is create logical constructions of what we think the AI should do, as if it were a human just making simple decisions on tactics and deployment. There are so many other details to the AI's development that most of the suggestions we make would probably be useless because they would hinge on certain particulars that the AI isn't employing. I'm sure BTS doesn't mind our suggestions, but I'd like to warn those that think that it would take just a little effort to have the AI performing much better due to their certain approach to the logic of the battlefield.

[Keith]

I think the biggest trick in making a scenario enjoyable against an AI oppenent is knowing how to place the victory flags. The flags are the primary decision factor used by the AI. If you want to have the AI defender hold defensible positons and not get wasted in fruitless counter attacks you have to place the flags in sufficient depth behind the main line of resistance.

When designing a scenario for the AI as the attacker the most important thing is learning how to synchronize the timing of forces and reinforcements such that the AI can pull off coordinated attacks. This is really a black art and requires lots of patience tweaking scenario parameters. Having a good road net to the objectives with non-constricting terrain is essential making the AI look smart. Also placing the objectives in a coherent fashion, not willy nilly all over the map, will help the AI. Think of the AI as a rat in a maze and you have to place the cheese (flags) at the appropriate places along the way to help him out.

The bottom line is the AI is very respectible and I think carefull scenario design can make the AI fight quite well. Of course the AI will always be better defending than attacking, so keep this in mind.

[MikeyD]

I know someone who's tried to construct several Villers Bocage scenarios and has always been dissappointed. Basically, no matter where you place the flags you'll never recreate Whitmann's movements with AI logic.

BUT... I've detected the occassional unpublished tweak from one CM upgrade to the next. I'd bet money that there's bound to be SOME tinkering with the AI in CM2, whether BTS announces it or not. These guys are such perfectionists, how could they resist!

[James]

I recently started a topic called: Case for new type of chain of command AI. Those dreaming of a better AI may find find it interesting.