AI : unable to attack?

[Scipio]

I created another small scenarios,Allied attacks. I tested it vs the AI. It was okay when the AI played the defense, but when I set the AI to attack...

There were two large VLs in the scenario, no river to cross or another impassable terraine. I waited til turn 20 of thirty, but not a single attacker showed up. I surrendered at this point, just to see what the attacker is doing: nothing! They were all still hidden in their start up positions. What am I doing wrong?

[ May 22, 2002, 05:02 PM: Message edited by: Scipio ]

[JasonC]

Give the AI intermediate small flag objectives on the way to the main objectives. Give the AI LOS to things. Provide some defenders close enough to make contact early. The AI uses enemy sighting reports to orient itself. If the whole enemy force is hidden, it can easily get disoriented.

Fights in terrain with practically no LOS involve guessing where the enemy is likely to be. You and I have no great difficulty doing that; we readily just pick something arbitrarily if we lack information, and expect to revise our plans as new info comes in. The AI hasn't a clue where to start. Without a deterministic reason to do this rather than that, it can't decide and does nothing.

The AI is fairly weak overall, anyway. And weaker on the attack than on defense. But its inability in situations described above, completely hidden enemies etc, are beyond "bad at". They are just "can't do at all". With intermediate objectives and enemies it can locate, it will move and fight. But that doesn't mean the attack will be delivered well.

[Cameroon]

In addition to Jason's points, I believe that if the AI is on attack and its units are hidden, they tend to stay that way. Same with units on transports (like trucks). Moral of the story, don't hide the AI's units if it is attacking and don't put 'em on trucks.

[WWB]

Another thing about the AI: it really has trouble when it cannot see enemy units. Start the battle during the skirmish at the outpost line and it does much better.

WWB

[russellmz]

Originally posted by wwb_99:

Another thing about the AI: it really has trouble when it cannot see enemy units. Start the battle during the skirmish at the outpost line and it does much better.

WWB

there a couple ai tips in the faq: split squads to reduce global morale, switch friendly edges, etc.

[Mord]

This discussion brings up a thought I have been tinkering with for a while. Why no allow trigger points or way points to help the AI along in premade scenarios. Something along the lines of objective flags but they'd be placed in the editor and would not show up in game. Maybe something to just get in going in the right direction when attacking. It would give the battle designer a little more control.

I played a game last night and basically waited on the AI for most of the scenario. Near the end it really pulled some wicked moves on me using a ton of smoke to get in very close and personal. But alas just when things got really interesting and kinda scary the game ended. I wish it would play that way all the time. But in all fairness I must say it still suprises me sometimes.

Anyway trigger point/way points might be a good way to get the AI to act alittle more human. Just a thought.

Mord.

[JasonC]

The Battleground Series had a nice way of handling this. They allowed "scripts" for the AI in particular scenarios, layouts of plans for moving the various formations making up its force. The scenario designer would set them beforehand, and the AI would pick one of the plans. It is much easier for a human designer to see a reasonable way to get the ball rolling and drive events. And a lot easier to build an AI that can handle the smaller tactical stuff at least competently, than one that thinks clearly about the bigger picture moves.

[TSword]

Can only agree

Much more control for the scenario designer is necessary to create really interesting Singleplayer scenarios and even more so "Historic battles". This addit. control should come as a surplus, so it's free to the designer to use them or not (For Instance scripts).

Greets

Daniel

[Lawyer]

I agree this is a problem. I am designing a scenario based on history where a convoy of trucks needs to move along a paved road to trigger battle events. I can't get the AI to make the trucks move along the road, even though I have put some small VF's along the road to entice it, and also made the map edge eligible for points if they exit.

On the German side, a line of tanks and halftracks controlled by the AI just rotate and muddle around each other like a bunch of beetles trying to mate. A clear road with VF's in front helps a little, but not enough. Reinforcements really jam up the works.

I think the designer needs the ability to script AI moves in advance, at least for the first turn or two to get things moving and LOS. Otherwise, I don't see how to make scenarios like mine work unless they are PBEM only. That would be a shame.

CM scenarios become repetitive. Making interesting maps is a great advance in gameplay, but I now think giving the designer more control over unit movements is even more important.

[Panzer76]

This has been discussed many things, as most topics on this BBS

Anyway, yes, the AI, especially the StratAI very weak. I only hope BTS have done something with it in CMBB. Anybody know?

[Panzer Leader]

Originally posted by Panzer76:

This has been discussed many things, as most topics on this BBS

Anyway, yes, the AI, especially the StratAI very weak. I only hope BTS have done something with it in CMBB. Anybody know?

About the only thing they've mentioned is "minor tweaks" to the AI quite a while ago, so I'm thinking there will bo no major changes.

[Scipio]

Thanks for the help. The 'unhidding' helped, however, the AI guided attack was a desaster. I guess I better create scenarios for H2H.

[WWB]

As a PO goes, the CM PO is very good. But it is still definitely a PO and not up to human snuff. That said, what one needs to do is help it along as well as understand its limitations. When I design a battle, early on I decide which side should be playable--usually PBEM only. After I get it working for that aspect, I will look to see how it works as single player vs. each side. Then I put in strong warnings on how the battle should be played in the briefing. Getting a battle which works as PBEM and single player from both sides is nearly impossible, and should not be attempted from anyone short of a CM Scenario Designing God.

For example, in your truck scenario Lawyer, we know that the AI will not move the convoy. So make it playable as a human convoy leader. Scipio, your case is odd, take a good look at all the parameters and you will probably find the error. Have you, for instance, set the attacker? What about the proper map-edge settings? A good place to start is McAufflie's Belgian CM Homepage (link is at home--any help?) which has an excellent scenario editing FAQ.

While scripting is to some extent cool, I have found it to be more trouble than it is worth. If one has ever tried to design a TOAW scenario, you know what I am talking about. Placing the forces, objectives and making the map is relatively easy. Then you get into this primitive branching logic language to attempt to setup your battle, generally leading to frustration. I think going with scripting would be a fundamentally bad idea, in terms of ease of generation not to mention PO quality and suprises.

The idea of 'Faux Flags' is alot better. It would work with the current AI, and would definitely help to guide the PO without doing weird things to the scoring. BTS, Fix or Do Somefink!

WWB

[Mr. Johnson--]

Set up those small flags at nice jump off points, use a few more big flags. Then put in the briefing that the human player turn flags off when playing.

If the AI is attacking they will almost always need massive reinforcement, thereby making the game uneven if 2 humans decide to play it, so plan ahead for what kind of scenario you want.

Also try giving the AI a few elite battalion commanders with full bonuses. They will help keep those infantry under command better, and will help prevent a short 81mm barrage scaring half the attackers.

here is McAullife's website

http://users.pandora.be/aneric/tipsindex.htm

[ May 23, 2002, 01:01 PM: Message edited by: Mr. Johnson-- ]

[JasonC]

I agree with wwb_99's comments about the difficulty of designing TOAW scenarios, but I don't think that difficulty carries over to all use of scripts to direct AIs. It was much simpler to use in the Battleground Series, and the resulting performance was much better. (Although the AI was still too weak against good humans in the BG series, in general).

Part of the reason is undoubtedly scale. Grand tactical scale works with simple overall movements of large units. TOAW is not grand tactical scale, and trying to treat its arbitrarily layered formations that way just doesn't work. The level of coordination between the movements of this small group and that small group is too high for scripts to realistically handle.

But in the grand tactical case in BG, the scripting system let you set movements as low as brigade, but it was often enough to just direct entire corps, or at the most divisions, towards sensible intermediate objectives. The result was far lower "combinatorial" complexity. Which corps goes first, and toward this obvious point or that one, could result in a reasonable plan of battle, out of a managable four or five possibilities.

I suspect in CM it would be possible to get the same sort of grand tactical effect, by only scripting rough orders for whole companies, or even for the entire force. At most, one might use seperate plans for armor platoons and infantry companies - still a handful of forces. And nothing facing in the sense of "if-thens", just "go here (prominent terrain), then here (an objective)".

It really is the broadest strokes that we humans do so much better than the AI. Our minor placements and coordinations of units are undoubtedly somewhat better, but it is the larger scale stuff that we do effortlessly and that AIs routinely stuff up completely. They do not transition between nested levels of planning and command easily, as we do. If they get higher level planning direction - even roughly - from a pre-planned set of reasonable options, and therefore only have to figure out the lowest level stuff all on their lonesome, they are much more successful.

Otherwise put, AI 2nd lieus are fair, AI captains are poor, and field grade AI officers are vegetables.

[ May 23, 2002, 02:56 PM: Message edited by: JasonC ]

[steve_lara]

Years ago I played Typhoon of Steel, a tank battle game like a 2D Combat Mission. I liked the way AI controlled light tanks in that game would NOT try to attack heavy tanks or bunkers frontally at long range but instead would charge heavy tanks to hit the enemy in the side or rear at point blank range some times even under cover of smoke. I remember one of my Panther tanks almost being taken out by a Staurt light tank that way. I wish Combat Mission would do this so that AI controled light armored vehicles would not be so useless.

[Scipio]

Some kind of scripting would be very helpful. When I make a scenario, I have a raw idea how the attack should run. That does not mean that it is the one and only possibility - but with scripting I could give the AI at least a senseful 'standart attack'.

BTW, if someone want to test the scenarios, both can be downloaded at www.warfarehq.com