Why i the AI so stupid

[Strat]

I have tried time and time again to make a battle where the AI has a chance. To no avaial the AI always chooses a route to self destruction. If it is not doing a suicide charge toward the flag it is doing some other suicide strategy. Once it starts a strategy it doesn't deviate. The programmer should make the game somewhat more challenging for the private gamer and design the AI with enough intellegence so scenarios can be tested. The current AI is useless.

Two cents,

Strat

[edgars]

AI sucks at attacking.

If you want a challange, let the AI defend and give it +3 expirience setting and maybe some force bonuses.

[KiwiJoe]

Show me a war startegy game where the AI isnt stupid. Play humans, thats were the game and the challenge is. The AI is for getting the "feel" of the game/commands/units etc and play testing, nothing more.

[Tanaka]

If you ask me, I don't know why they do an AI... All we need is a TacAi, I myself only played the game against the AI 2 or 3 times... :cool:

I think in your state of frustration (no other explanation for your post) the best thing to do is to play the game against an human opponent.

In complex 3D tactical games like CM, you will never find (at least on the near future) a non "stupid" Ai...

So my advice is, don't waist your time playing the Ai, because this game excels in TCPIP and PBEM mode.

PS- If your problem with the Ai is much more specific, scenario testing, do like at chess, play against yourself...

[Stixx]

As mentioned above CM's AI isn't brilliant but i've never seem a wargame with brilliant AI.

If you thinks CM's AI is bad have a go at Close Combat. Then you can really see some bad AI.

[KiwiJoe]

close combat muhaha! yeah its real fun watching your tank spin around 180 degrees and charge the enemy ass 1st!

[Jarmo]

Yeah the AI is pretty dumb. But still about the best one I've seen in a computer game. I've played against her a lot and manage to lose now and then.

I think there was a better one in "Strategic conquest" at some point, but that was way simpler game.

Simple game = easy to do a good AI.

I've never met a chess game that couldn't beat me at easiest settings. CM is not a simple game, that considered the AI is pretty amazing.

Although I could do without last minute spotter charges..

If you want a challenge, don't play at equal odds. Try something like doing an attack -50%. And use random purchasing for both.

[Stacheldraht]

<BLOCKQUOTE>quote:</font><HR>If you want a challenge, don't play at equal odds. Try something like doing an attack -50%. And use random purchasing for both. <HR></BLOCKQUOTE>

That's just a challenge of odds, though, not strategy or tactics, which isn't much fun. The TacAI is decent, but the strategic AI is generally pitiful. People should remember, too, that not every wants or has the opportunity or time to play against humans. The game should provide a solid solo experience, too.

[Thomm]

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

If you thinks CM's AI is bad have a go at Close Combat. Then you can really see some bad AI.<HR></BLOCKQUOTE>The AI of Close Combat is not bad.

[CrapGame]

<BLOCKQUOTE>quote:</font><HR> The AI of Close Combat is not bad. <HR></BLOCKQUOTE>

Well, then how do you explain the Close Combat AFV Dance of Death???

The CC-series AI is nowhere near to the same caliber of the AI in CM. Not even close.

Crapgame

[Thomm]

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

Well, then how do you explain the Close CombatAFV Dance of Death???[/QB]<HR></BLOCKQUOTE>A problem of map design/terrain coding. <BLOCKQUOTE>quote:</font><HR>The CC-series AI is nowhere near to the same caliber of the AI in CM. Not even close.<HR></BLOCKQUOTE>Well, once, when I still had time for such follies, I rebuilt a CC scenario in CM. Same map, same units. The battles and the results were very similar. Human victory, either side, either game.

The point is that in CC you cannot create force ratios of 1:3 because of the 15 unit slots limit. This always puts the attacker in a difficult situation.

And for that stereotyp "Dance of Death", I encountered it rarely enough to take away from the game, especially in CC5, which had appropriate map design to avoid this pathfinding problem in most cases.

The AI of CC is not bad, in fact it is quite amazing considering the COMPLEXITY of the environment and the soldier model.

Regards, Thomm

[Jarmo]

Close Combat AI.

First, it cheats.

In a same situation, the AI always gets more units. (CC1 and CC2, haven't played the rest)

Second, the complexity of the terrain is nowhere near the complexity of CM. What you see is a painting slapped over the "real" battlefield. You get the same basic components in both games, but CM is 3D. A lot, lot more difficult.

[dalem]

1) The CM AI is pretty good, for an AI these days.

2) The CM AI will punish you if you make blatant mistakes, like leaving flanks unguarded, advancing AFV without infantry support, etc. So at the least you can use it as a reminder of what not to do.

3) The CM AI is fairly decent when placed on the defensive. Giving it force and/or morale bonuses definitely helps but is not strictly necessary.

4) 'Realistic' force picks will help the AI as well - let the AI pick forces for both sides and again I think you'll see your game experience improve.

-dale

[tiborhead]

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

close combat muhaha! yeah its real fun watching your tank spin around 180 degrees and charge the enemy ass 1st!<HR></BLOCKQUOTE>

And let us not forget waiting around for up to a half an hour before the AI actually does anything. I can't count the number of times, when defending, I had to wait like 20 minutes just for the AI to send 1 infantry squad of a flamethrower or something. The exception being CC2. I recall that to be a challenging

game.

[Thomm]

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

Second, the complexity of the terrain is nowhere near the complexity of CM. What you see is a painting slapped over the "real" battlefield. You get the same basic components in both games, but CM is 3D. A lot, lot more difficult.<HR></BLOCKQUOTE>Terrain data in Close Combat is defined at a resolution of 2mx2m, in Combat Mission it is 20mx20m. That means that for every terrain tile in Combat Mission there are 100 terrain tiles in Close Combat. Whether this is an advantage is arguable. It certainly puts a lot of strain on the planning and pathfinding AI.

Also, terrain data in Close Combat has a height value for every 2m x 2m tile. Therefore, the terrain data (LOS) is 3D, but the display is 2D (large bitmap instead of 3D polygons).

So what you say with regard to complexity may be wrong: both engines use 3D terrain, yet Close Combat works with tenfold spatial resolution.

Regards,

Thomm

[Louie the Toad]

Strat,

Are you taking advantage of the AI by doing things you should not be able to do?

Such as:

viewing from a higher elevation than you realistically should; zipping around the battlefield with your mouse to view places and terrain which you should not be able to see; zooming in to view the enemy when your units do not have that capability; selecting enemy units with the mouse thereby letting the AI tell you more info than you should be able to know(see example below).

Try using some or all of Franko's Iron Man Rules and see the difference. Oh, did I mention playing at the Realistic setting and using Level 1 except when your guys are in a two story building (level 2) and turning detailed hits off?

Example: My Cromwell is waiting hull down and sees a PzIVj (not completely sure since my guys did not completely identify it) back up between two houses. The Cromwell shoots and gets what appears to be a hit. BUT, the enemy tank continues to back up and stops almost completely out of sight. The turn ends. Did I hit the PzIV, did I kill it? The enemy gun didn't appear to slew down as so often we see. There was no explosion. My tank crew didn't yell "we got him" which they have in the past. Some of the tank can still be seen. Do I click on the enemy unit to see what the AI says? NO. Absolutely not. Besides that, I have switched off Detailed Armor hits during Set Up. Do I use the Line of Sight tool to see what my guys can see. YES. But that is all. What I can see and hear is the only information I should be able to get. What happened? Fog of War happened.

I could have flown with my mouse over to the enemy unit, examined it from all sides, toggled on the Detailed Armor hits, selected the enemy unit with the cursor to find out information. BUT --

"That would be telling" -- Number 6

[Maximus]

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

Terrain data in Close Combat is defined at a resolution of 2mx2m, in Combat Mission it is 20mx20m. That means that for every terrain tile in Combat Mission there are 100 terrain tiles in Close Combat. Whether this is an advantage is arguable. It certainly puts a lot of strain on the planning and pathfinding AI.<HR></BLOCKQUOTE>

Well in this case it should be easier for the AI to plot around an obstacle. More tiles to use to plot a more defined path.

However, your anology is wrong anyhow. CM does not use tile to tile movement ploting. It uses meter to meter plotting. I guess you have never seen the AI plot a tank to move inbetween the narrow corners of impassable terrain. If CM uses tile to tile movements, then you would see very unrealistic movements.

<BLOCKQUOTE>quote:</font><HR>Also, terrain data in Close Combat has a height value for every 2m x 2m tile. Therefore, the terrain data (LOS) is 3D, but the display is 2D (large bitmap instead of 3D polygons).<HR></BLOCKQUOTE>

Wrong for the same reason above. In fact this is the beauty of the 3D maps in CM. Granted they could be even more varied if the tiles were smaller, say like 10m x 10m, but the engine dows calculate the 3D LOS down to the meter. Why do you think that each AFV has a silloutte rating? Do AFVs in CC have such ratings?

<BLOCKQUOTE>quote:</font><HR>So what you say with regard to complexity may be wrong: both engines use 3D terrain, yet Close Combat works with tenfold spatial resolution.<HR></BLOCKQUOTE>

I can't comment on this as I have no idea what you mean by "tenfold spatial resolution". You mean that since CC's "tiles" and "elevations" are 2x2m as opposed to 20x20m?

Now as for AI differences, there's no comparison. I for one never got into the CC crowd. Didn't like the top-down, real-time aspect of the whole thing. I liked the isometric look, turn-based thing with Talonsoft's Campaign Series.

CM took the best of both worlds and made them better. Isometric to full 3D and turned-based/real-time to WE-GO (turn-based with real time action). Now with as complex of a engine that CM has, the AI still does manage to whip my butt from time to time even on the attack. And I've been around since the Beta Demo days. Try a scenario called "Tiger Trap", play as the defender. I'll be the attacking French will whip your butt.

[ 07-10-2001: Message edited by: Maximus ]

[Thomm]

Maximus, I really admire the confidence with which you put forward your "down-to-the-meter" theory!

Let me just ask you one question: if there was this geometric precision you claim, why do those LOS issues involving the corners of houses and ridges of hills exist?

I am convinced that movement and LOS on Combat Mission "Tiles" is resolved on a sub-grid of smaller tiles, not with geometric operations as you state. This is because everybody tries to use integer arithmetic for this task, and tile-based methods are the method of choice for this.

Regards,

Thomm

[Moon]

Don't confuse the map editor "tiles" with the in-game "resolution" of what's happening around you. After all, your tanks are not moving from tile to tile, are they? The internal resolution of CM has been mentioned somewhere in public by Steve or Charles in the past, and I believe it is much less then 2mx2m. In fact, I seem to recall that it's not tied into an overlying grid at all, but rather uses real point-to-point resolution. Makes sense, since it's 3D. But someone more knowledgeable will have to chime in here.

[StellarRat]

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

Yeah the AI is pretty dumb. But still about the best one I've seen in a computer game.<HR></BLOCKQUOTE>

Actually there are some very challenging chess programs out there, but I'm sure you meant computer strategy game (other than chess).

[Dittohead]

<BLOCKQUOTE>quote:</font><HR> 4) 'Realistic' force picks will help the AI as well - let the AI pick forces for both sides and again I think you'll see your game experience improve. <HR></BLOCKQUOTE>

Actually I have found that the AI gives a much better game if you select its forces and let the computer pick your forces. That way you can insure the AI has a maximum of combat units.

Don't pick any FO's, the computer always puts them where they are easily killed and fires smoke more often than not, same goes for mortars.

Tony

[L.Tankersley]

<BLOCKQUOTE>quote:</font><HR>Originally posted by Louie the Toad:

"That would be telling" -- Number 6<HR></BLOCKQUOTE>

Actually, it's Number 2 (all of them) that says that.

[David Aitken]

Here we go again. I would direct anyone who hasn't yet seen it to this thread:

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

[russellmz]

<BLOCKQUOTE>quote:</font><HR>Originally posted by David Aitken:

Here we go again. I would direct anyone who hasn't yet seen it to this thread:

CRAP AI, Other BUGS ,Fix Or implement IPX/TCP!!<HR></BLOCKQUOTE>

Explanation of CM's Tile System

the above link is...an explanation of...well...YOU KNOW

LOS is the big factor why we can't have pixel by pixel terrain. For even a small map the CPU would crawl to a halt if we went with an absolute "pixel" based system. There are all sorts of other reasons, but this single one kills it dead all on its own.

So what does CM use for terrain and how does it work?

Basically, the terrain is done in tiles (squares), with each being 20x20 meters. This tile is then subdivided into 2x2 meter squares for determining unit position and special terrain placement (like a wall or hedge). Elevation differences are averaged between each 2x2 square as well. So a large map of 3000x3000 meters has 2,500 tiles and 225,000 subtitles (provided I did the math correct ). This is a HUGE amount of data and takes up about 7MB of RAM to store. Just imagine what would happen if we increased the resolution any!

There can be multiple types of terrain in one tile (road and Heavy Woods for example), but all are predefined "building blocks". The base terrain contained in each tile is an approximation of whatever it is supposed to be. For example, a Woods tile does not have individual trees, instead it is an approximation of large, dense tree growth. However, something like walls and buildings are defined using the 2x2m internal grid. Therefore, if your unit looks like it is in a building, and the display panel says it is in a building, it is in a building (unlike some others ). Windows, doors, interior walls of buildings, etc. are all approximated and are not represented directly.

A unit does not move from one 20x20m tile to another one like a unit would move from one hex to another. The unit actually moves smoothly along using the 2x2 meter grid. Also remember that a squad is has a "footprint" but no actual tracking of where each man is at any given moment. The end result at CM's level of simulation is identical, or so close that it isn't interesting to note differences. However, the difference makes it possible to make CM now instead of waiting several years for CPUs to improve.

The KEY point to take away from this is that having both units and terrain be approximated makes the game work, not only from a technical standpoint, but also from a SIMULATION standpoint. The problem with something like CC is that the units are "exact" but the terrain is full of approximations (apparently larger ones than CM too). This means that a soldier standing in the open might be behind a wall, or one behind a wall in the open. Trying to get fair results from incompatible treatments like this is very difficult.

In the end Combat Mission's tile/unit system is infinitely more flexible and realistic than something like West Front's traditional 2D hex design, but it also lacks the problems associated with CC. So while it isn't pixel perfect, it is the best system possible for today's computers.

Steve

(steve made a math error above somewhere with a square)

I think we need to clear up a few misconceptions.

First, remember that CM does not literally simulate the positions of individual men on the map. Reasons should be faily obvious, but the biggest one is CPU power. Consider that your average game of CC has maybe 100 men involved. CM can go as high as 3000 or even higher. Thirty-to-one! Now figure in that most algorithms that depend on number of soldiers, like line of sight, are what we call "n-squared" algorithms (which means that every man on one side must 'look at' every man on the other side) this increases the CM:CC CPU load ratio from 30:1 to NINE HUNDRED to one!

Ouch. 900:1. Now you can see why we simulate squads as a "whole", not as individual men. NOTE: We do track individual casualties and weapons within a squad, however, just not multiple locations. Note: For most purposes, squads "occupy" a single point in the map, not an amorphous footprint. So a squad centered on the edge of woods occupies entirely woods not half-in, half-out. The only time the game considers dispersal directly is when incoming fire takes casualties. For example, a direct artillery hit on the squad's central location point does not stupidly wipe out the entire squad. They're not literally all standing on the head of a pin.

Geek time! Now a little math. Each point on the subgrid (2m x 2m) requires two bytes: 6 bits for terrain type and 10 bits for elevation. A typical CM battlefield (1000m x 2000m) requires a 501x1001 subgrid, or 501501 points, or 979K. A large battlefield (3000m x 3000m) requires a 1501x1501 subgrid, or 2253001 points, or 4400K. Because of the need for dynamic playback, at times a copy of this information must be held internally, so we're up to 8800K for a large map. Shew!

Tanks are also considered to also occupy one "point" on the map, not a "bounding box" (again, due to CPU limitations). So you either see the tank or you don't. You don't see the front fender but nothing else, for example. It would be nice to add this detail, but we'll need faster hardware first, so perhaps in the future. Another reason we don't allow spotting "just the fender" is that CM allows (optional) scaling-up of unit graphics so you can see your tanks and men better. 1:1 scale is like fighting with armies of ants. So, because your tanks are often drawn larger than real life (though you can turn this feature off!) it would get strange when you, as a player, can "see" the fender because the tank is enlarged, but the program tells you that you can't, because it's using the real size of the tank to make its calculations. So it's an issue that has user-interface implications as well as CPU limitations.

All in all, the idea of a grid is hardly one we can claim to have invented. But as far as I know, no wargame of CM's scale exists that goes to anywhere near the lengths that CM does to provide precise LOS calculations. Most are just based on hexes - i.e. if the center of hex A sees the center of hex B, the LOS is clear. Obviously that's very crude by CM's standards. The only game I know of that even comes close is CC, and as I've described it's at a lower scale with far fewer men, and yet its LOS is still less precise than CM's.

Charles

[Los]

BTW how one designs the scenario, the terrain, placement and types of victory points and other scenario/operational variables all go into determining how well the strat AI performs on attack and defense.

Los