Another stupid demo AI question

[adrake]

Before you say anything, I already performed multiple searchs looking for this info and couldn't find a definite answer...

I recently downloaded the latest demo of CM (Gold Demo I think). From my forum searches, it seems that this demo version is (just about) up to date with the retail version in terms of AI. Is this true?

I played Chance Encounter for about 10-15 turns to learn the interface and basics. I then proceeded to play both maps from both sides winning each the first time I tried (i.e. I didn't know the senerio before playing). I'll give you that I didn't do it as pretty as possible, but still...

I do like the game alot, but before I buy it I would like to know if the retail version AI is any better then the Gold Demo?

Thanks!

[Guest Ace]

Originally posted by adrake:

I then proceeded to play both maps from both sides winning each the first time I tried (i.e. I didn't know the senerio before playing). I'll give you that I didn't do it as pretty as possible, but still...

I'm a little confused by this statement. If you are playing both sides, then you HAVE seen the scenario before at least for one of the games. Playing two games at once from each side is even worse, since you are fully aware of the force composition of your enemy.

BTW, are you using the full fog of war option? This makes a big difference.

Also, in an attack scenario, it is best not to play as the defender first. It gives you an unfair advantage when playing as the attacker the next time. Try the attacking side a few times before taking a peek at the defensive setup.

If it's still too easy, you can give the AI an advantage, moving the level to +1, +2, or +3. Many players do this.

Admittedly, Chance Encouter is not the hardest scenario, although it is a great intro to the game. The full game comes with a lot more scenarios (about 50, I think), plus there are a bunch of great designers out there that keep putting out great scenarios and operations, infinitely expanding the available missions.

The full version also has a great Quick Battle option which is great for playing human opponents since maps are randomly generated (that way no one has seen the terrain prior to unit purchase).

For a more challenging game, try Valley of Trouble as the Americans (Hint: try leading with your Shermans - they'll scare off all the germans)

[This message has been edited by Ace (edited 02-28-2001).]

[L.Tankersley]

There have been several relatively minor AI tweaks made to the retail CM code; I don't know how many of these have migrated over to the demo. That being said, the demo gives you a pretty fair indication of what the strategic and tactical AI is like.

Now, THAT being said, a few comments: while far from perfect, the CM AI is quite good. Particularly when you consider that it has to develop a complete plan of action with no input from the human scenario designer, and that it doesn't cheat (if you play with fog-of-war on, the computer only knows about the forces it sees). It's not as good as a good human player, but it can occasionally administer a good whupping to a good player even without the benefit of amazing luck. If you're concerned about a lack of challenge, there are a couple things you can do. First, play against another person. CM really shines here, and you can now play both via email or in real time via TCP/IP. Second, many scenarios are specifically designed to be challenging, particularly when played from one side or the other. Finally, there are myriad options in CM to make things harder on you. You can give the computer player an experience bonus (essentially giving them higher quality troops), you can handicap the force sizes, you can play certain challenging kinds of scenarios (try defending in an assault for example), and you can play quick battles in which the computer automatically purchases your forces and end up having to improvise a defense with whatever's at hand.

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

Leland J. Tankersley

[Schrullenhaft]

As always... it took me too long to post this now redundant info...

With the exception of a few tweaks here and there... NO the AI is the same for both the Gold DEMO and the full retail versions.

It's extrememly hard to code a competent AI. Large scale changes in the AI take a long time to code. CM has the best, commercially available tactical-level AI available. There's always room for improvement and there will always be niggles about one situation or another not being realistic. But CM has the best AI available in a wargame of this level. It doesn't "cheat" (which is how some games "code" a tougher opponent).

Playing the same scenario against the AI gives you an advantage that the AI doesn't have - you've remembered (in general) what you're facing and now you have more options to choose in preparing for the engagement. The AI doesn't have this capability.

If you still feel the AI is no challenge modify the "Play Balance" at the beginning of the scenario to give the AI larger forces, etc.

[This message has been edited by Schrullenhaft (edited 02-28-2001).]

[Guest Madmatt]

Actually....That's not really true. The Gold Demo was based on v1.02 release of the code. The current game is now up to v1.12 and there were several pretty significant changes to the game, TacAI and other internal factors that make the retail game much harder to win against human and artifical opponents than what you get in the Demo.

Also understand that the demo is only two rather standard fair battles with not a whole lot of finesse required to fight them efficiently. Once you try your hand at some of the larger or more tactically challenging battles that the game ships with, or any of the myriad of user created offerings then you will see a huge difference in how the AI conducts itself as compared to the demo.

The demo is a great introduction to the game, but to say that the demo AI plays identically and with the same skill level as in the retail game is just not true at all.

Madmatt

[adrake]

Originally posted by Ace:

First, thanks for the candid and insightfull responces. I realize that this is going over old stuff again, but I couldn't find a good thread about it using the search feature.

ACE: BTW, I did play as the attaer first, with full fog of war. By saying I played both sides, I ment that I played as attacker first then replayed as defender.

Increasing the experience or reducing my forces may make the game harder but doesn't make the AI smarter. Although I did like the comment that there are larger, more tactically complex scenarios in the full version.

I'm still not sure if I'm going to buy, but thanks for the info!

[armornut]

I would do it man. I would say that it is the best buy for what you are going to spend. Even if you only play the AI you will be busy for a long time, but factor in the IP games and the Play By Email games you can play, and you will have a game that will challenge you for years.

Just a thought

Armornut

[The_Capt]

I would recommend CM for the multi-player alone.

I will admit, I don't play the AI anymore but PBEMs have provided more enjoyment than any game has in a long time.

[Mannheim Tanker]

In my opinion, CM has the best AI out there to date. If you decide to pass by CM based on its AI, you might as well do so with all of the other wargames out there

Also, if you get a chance, play a PBEM. It changes the face of the game considerably.

[SybotCB]

Let me just tell you that when I downloaded the demo I played both of the scenario's several times. I liked the game although I did beat both of the scenario's as both sides with full fog of war on the first times that I played them. I thought hey this is a pretty fun game and I'm actually pretty good at it. I'll get it and play all the grog's out there and show them how it's done.

This was a completely naive and idiotic state of mind. I received my game in the mail (finally!) on February 16th. (Yes only one day after they emailed me saying it was back in stock, I am so lucky). I immediately loaded up the AAchen scenario against the AI playing as the allies.

Cool I've got some new units here that weren't in the demo so I was playing around with them. Anyway I started walking my troops down the street toward the hotel. Using the same basic tactics I used to win the Demo's scenario's.

Next thing I now an AT gun takes my tank out in turn 4 and my Halftrack in the next turn. I have 300 German troops in buildings all around me raining bullets on my mens heads. My men are running everywhere just trying to get out of the way. My Company HQ sent a letter of resignation to me saying he could not work for someone as incompatant as I am. And the scenario ended with an Axis Total Victory.

Well OK I need to be more careful so I try it again. Still got beat although not quite so quickly. At this point I felt that I needed to try the tutorial. So I went through that several times before finally being able to pull out a tactical victory over the germans. (Just in the tutorial).

Anyway, Let's put it this way. BTS seriously mislead's you. The scenario's included with the demo are really easy compared to all the other ones. Also I haven't even played a Human AI as I'm still working on tactics against something as lowely as the computer. I can't imagine playing against a person yet.

This game is excellent. The AI is pretty competant. And if you can beat it routinely there's always TCP/IP and PBEM after that. I think BTS has made a lifelong fan. (Only one other company has that kind of devotion from me where I'll by anything they make sight unseen (Blizzard)).

Get this game. I'm pretty sure you won't regret it.

I still liked the concept of the game and I bought it

[jasoncawley@ameritech.net]

You should buy CM. You are right that the AI is not a chess program; if you want a truly challenging opponent you will want to play humans. But that is true of every strategy game I know of except chess. (There may be decent AIs for checkers and bridge I suppose, but I don't know it from personal experience). CM has more depth of gameplay than any computer game I've seen, at least since "Stars!". I doubt you will regret it.

All of that said, there are improvements possible in the CM AI, and I have some concrete suggestions to make about it. I know such changes are hard to code and I do not expect the moon. But I am also aware that many AI programming problems are related to the difficulty of reducing to rules of thumb, the tactics of good human players. And there is some danger, when such rules of thumb can be found, that following them too strictly may lead to exploitable predictability.

Right now, I think the CM avoids the latter horn of the bull rather well. Its problems are not "doing the same thing" every time (some things, sure, but not many), but instead, not using some rules "rigidly enough", as it were. Or not having such rules available to adhere to or not. So I suggest some tactical rules of thumb, meant more to avoid mistakes than to achieve brilliance.

#1 vehicles should avoid locations where other friendly vehicles have already been knocked out. These are often "kill sacks", under LOS from many capable enemy AT assets.

#2 off-board artillery should be called for much sooner. Once a target is engaged, however, at most 1/2 the rounds in the module should be expended (1/3rd to 1/2 is a good target range), before shifting to a new target (either "adjust fire" or "cancel" if no adjust target is available). (Rockets are an exception - just fire the whole thing since the spread is too wide to "aim" anyway). The reason is that most of the enemy have left the area, or have already been hurt as much as the arty can hurt them. These days it is too common to see AI FOs delay fire until very late in the game, or to expend the whole module on a single gun position, or both.

#3 The AI needs to make better use of initial pauses to maintain "station" and formations. HQs should pause once when their platoon moves out, for instance, lest their lower delay times push them out to a "point" position.

Similarly, when attacking, a lead platoon can create a "point and overwatch" effect simply by pausing everyone but the front-center squad, once. The lead unit only will tend to draw fire on contact, while the rest can return it unsuppressed. One light-armored vehicle can do the same for a pair of tanks or heavy TDs.

#4 platoons should be kept together (the AI mostly does a good job of this), but also used in support of each other. Vehicles should likewise be paired in support of each other. Mix it up as to whether both remain stationary to fire, both advance, or one does either to cover the other, fire and movement style.

The key thing is to have them make these determinations but to keep them reasonable close to each other, but deployed side to side. Thus, if one in a pairing is farther forward, it should be more inclined to sit and fire - and vice versa for one farther back. The result should be more fire and movement "walking", and fewer piecemeal attacks in sequence along a single avenue of advance.

I realize that one is hard to code perfectly, but it does not need to be perfect.

#5 infantry under artillery fire, unless in heavy buildings, should withdraw-run to the nearest cover toward the rear, rather than trying to "ride out" a bombardment in place, or advance into the open with greater delays. Return after the bombardment. This one must have some "fuzzy" in it, to avoid becoming too predictable. But sitting under bombardment is both predictable and dumb. It would be better to be heavily weighted toward "smart", but not all the way (to leave some difficulty in prediction).

#6 the location of enemy units should have more weight in choosing directions to attack, and the location of objectives less. It is a bit too easy to lure the AI onto flags without securing its flanks first.

#7 when making unit purchases, pairs of vehicles of the same type are often more useful than singles of multiple types. They have a more pronounced battlefield effect in whatever particular direction. One Sherman 76mm, an M8, a halftrack, and an M-20, is a poor mix of vehicles. 2 Shermans and 2 halftracks, or 2 M-10s and 4 M-20s, or 2 Stuarts and 2 M-8s, are better mixes.

Why is the first mix bad? One AP round destroys the heavy AT capability. The force can't carry enough infantry for the light armor to produce an "assault" effect. The main effect is multiple light-armored MGs, but way too much is paid to get this effect.

Compare the last force - 4 light AT guns and 10 MGs on 4 speedy armored platforms. Or the middle one - 2 strong AT guns and 6 .50 cals. The first (Shermans and 'tracks) can carry an infantry platoon in assault, and has strong HE and MG firepower. The costs are the same, ~325 for each mix.

In other words, don't take a little of everything, but pick one or two definite somethings and take enough of it to matter. Also, notice that realistic weapons mixes work better in the above examples. U.S. armor used Shermans and 'tracks; Tank Destroyers used TDs and light scout vehicles (M-20s, Jeep MGs etc); cavalry used light tanks and ACs (and jeeps BTW).

#8 The AI needs training in setting ambush markers, using "hide", and deciding when to open fire. This is not easy to program, I know. Some rules of thumb may help, though -

If you can't see an enemy but aren't going to move, then set an ambush marker if you can. Set it to a place where you have LOS and closest to any reports of enemy, or in default of those, near his edge of the map. Don't worry about it being in the wrong place. Just set a new one when you get a new sighting report. Weight toward existing ambush markers from other units, and set yours near them but not exactly on them. Treat TRPs the same.

If you have an ambush marker set and aren't going to move, then count the number of friendly units that can see enemies, and the number of enemies that can be seen by somebody. As this number goes up, increase the propensity to drop the "hide" and fire. When one does so, every unit that can currently see an enemy should drop its "hide", so that all fire at once. If others are already firing (e.g. because of triggered ambush markers), drop yours too and fire, if you can see enemy.

The multiple ambush markers, and placing them "dynamically" = anew as newer spotting reports come in, have a good chance of putting one of them in a spot that will be triggered. The linkage of firing to dropping of "hide" by others, will allow a "mass ambush", at least in the second minute. And the rule about many shooters seeing many enemies, leading to dropped "hide"s regardless of ambush markers, should prevent screwed up "hide" orders and ambush marker placement, from preventing the AI from firing.

#9 dismounted teams slower than "fast" movement type, should ride rather than walk when vehicles are available and enemies have not yet been spotted. Vehicles carry such teams should avoid the center of wide open areas, and instead try to stay within 20-40 yards of some kind of cover, in case the teams have to dismount.

The teams themselves should also avoid moving over wide areas of open ground, unless friendly units are in cover on both sides of the route. If you can't see a friendly unit in the cover you are moving toward (or near it), it is too soon to try to cross a wide area of open ground, if you are not "fast". Gaps of <50 meters you can ignore.

This will produce an "overwatch" effect, as the heavy teams pause and cover faster units moving into cover farther ahead. And it will protect the teams themselves from being pinned or panicked in the open, where they can't reach cover quickly because of their heavy gear. It will also prevent dumb decisions like charging with an HMG 42 team, instead of a squad, increasing the life expectancy of flamethrower crews, etc.

#10 Light guns (20mm FLAK, 81mm mortars, 75mm infantry guns, 40mm AA, 57mm ATGs, etc) should be set to fire at will on targets they can see. But heavy AT guns (75mm PAK and up) should be set to "hide" until they see enemy armor, and should not fire on enemy infantry (unless the range is so close the guns are threatened/discovered, or) unless they have already revealed themselves firing at other target types.

In other words, heavy ATGs only drop "hide" to shoot armor. Let the lighter guns suppress infantry, kill light armor, and absorb artillery "counterbattery" fire. These days, an 88mm FLAK battery will too often and too easily reveal itself to a scouting infantry platoon.

Some of these will be harder to try to impliment that others. (I haven't written code in a long time, so I am not the best judge of such things). But I think they would markedly improve the existing CM AI. Which incidentally, already does some things quite well, lest I leave the wrong impression.

Those are my concrete AI suggestions, for what they are worth...

[jasoncawley@ameritech.net]

I should have said something about ambush marker ranges in that section. It depends on the weapon, but these are good rules of thumb -

The ambush range for Panzerfausts is 25 meters, using the Platoon HQ to set the marker. Same for flamethrowers.

The ambush range for Screcks, Bazooka, and PIATs is 50 meters.

The ambush range for anti-infantry fire is 100 meters, again using Platoon HQs.

The ambush range for light guns, heavy guns, mortars, and artillery FOs (TRPs) is anyplace they can see. For HMGs as well, using a weapons Plt HQ, for example. (Collectively, these are "heavy weapons" in infantry terms).

Another place where the AI could use some rules of thumb about distance scales, is minefields. When mines are encountered, go around them not through them. The first try, go around 25-50 meters. But if you hit two, then do not keep trying little "hooks" around each 20x20 tile and running into the next one. Go wide around, 100-200 meters. Notice, the scales must vary by a factor of two - any predictable distance would lead to gapped enemy minefields seperated that far apart.

I hope these are helpful.

[adrake]

Hey Jason, how are you doing? I should have known it was you once I saw the length of your post!

I'm sure you don't remember me, but I used to post as carcassonne on rec.games.computer.stars back 1997-98. I'm still waiting for Supernova.

Well, thanks to you all, and especially SybotCB, I have decided to buy the CM! I love WEGO strategy games (like Stars! mentioned by Jason) and I've never really gotten into a hard core tactics game before so it should be some good experience. Plus who knows when Stars: Supernova will come out.

So once I get the single player down, I'll be back for some PBEM.

By the way, are the PBEM games usually one turn per day or do players expect a 2-5 hour block of time that will be dedicated to playing a game? I am very busy all the time so would be more interested in the 1-2 turns per day type of game.

Thanks all!

EDIT: Typo fix.

[This message has been edited by adrake (edited 03-02-2001).]