AI on the Offense

[Childress]

What's the point, really? No dig whatsoever on CM or the authors but it's just not possible doing it right given the complexity of sims like this one and the current state of the art in AI algorithms. It's too hard. The number of computer games in which the Artificial Intelligence can mount a coherent attack are few and far between. As a matter of fact I can't think of a single one. Maybe the pricey Panther series? Panzer General (if one considers that a 'wargame')?

So arriving at the next Scenario in this campaign we find ourselves in defense mode. I plotted some artillery, re-arranged some units then pressed the GO button until the end. Result:

Believe me, I'm no great shakes as a player. Maybe if and when 'triggers' arrive this will change. The computer opp on defense can prove challenging in CM given good positioning. But for now, if you enjoy playing the defending side, only take on another human. Unless you want to role play an Aztec god ravenous for blood.

[General Jack Ripper]

Personally, I think the A.I. shouldn't be in an attacking role unless it outnumbers the player at least 2 to 1. In your example Childress, the AI is at a 1.1 to 1 ratio, and that situation always favors the defense.

The only other idea, is that maybe the AI is supposed to lose, but is just meant to inflict some casualties to reduce your forces for the next scenario?

I'm not saying the A.I. can conduct an effective attack though, just that if the AI has enough combat power, it can win simply through weight of numbers.

[Childress]

In your example Childress, the AI is at a 1.1 to 1 ratio, and that situation always favors the defense.

But not in combat power. Note the destroyed tanks. Yes, the AI needs overwhelming, no, make that ridiculous numbers to pull off an attack in single player.

[MikeyD]

The AI does not respond well to sloppy programming. The scenario designer needs to be rigorous, pay close attention to the timing, composition and coordination of forces, to the terrain, to any likely player reaction. This is not something you're going to see in a generic '1 size fits all' QB offense. The best offensive AI scenarios can kick your butt three ways to Sunday first playthrough. It should also be noted, quitting out and restarting a scenario if the battle's not going well then finally succeeding does not really count as 'winning against the AI'.

[sburke]

The person who programmed it though could probably kick my ass every game while drunk, stoned and having a houseful of crazy teenagers and babies screaming.

[John Kettler]

Childress,

Have you played Barkmann's Corner as the Germans? You might find that far less easy to handle. The scenario is, in my view, wrong to label the Germans as Attacker, when they are clearly playing the Defender.

Regards,

John Kettler

[Childress]

I played Barkmann 2-3 times. I've won and lost crushingly. There's a considerable element of luck present in the battle re the survival of the Panther and/or its components.

Yes, it's really a defensive situation. It works well because it's small and the Sherman wave attack is not without plausibility. The AI runs into trouble coordinating masses of armor, infantry and artillery in an offensive posture. CM shines, like any wargame, in player vs defending AI setups or H2H (where it rules) .

This is not something you're going to see in a generic '1 size fits all' QB offense.

This was not a QB, but a scenario within one of the CMFI campaigns. And not the only one of its ilk.

[noxnoctum]

The AI in the Command Ops/Airborne Assault series is excellent. Other than that I can't think of any game with really good AI.

The addition of triggers though will definitely make a huge difference if they get implemented.

[Childress]

The AI in the Command Ops/Airborne Assault series is excellent. Other than that I can't think of any game with really good AI.

So I heard. Never owned one of their games but downloaded the demos. But comparing them to the CM series is like comparing apples to Buicks. In CO the player directs his square counters from a god-like elevation over a grand-tactical battlefield. Terrain is simplified, units flow like water over enemy positions. It's all very abstracted but convincing in its way; good command and control is essential, the player experiences realistic delays, etc. But it doesn't require the precision and careful positioning of individual vehicles and squads within a detailed environment that CM imposes. It's simpler to mount an AI attack in the former.

[ZPB II]

Buy fanatic troops for the AI.

[Rokko]

The problem with offensive AI is mainly, that it is not able to keep its troops within C2 range, that it can't keep any formations and that it is unable to react to anything going on on the battlefield due to lack of triggers. For instance, in reality if company of tanks attacked and lost maybe 5 tanks, it would often call of the attack. In Combat Mission the AI will keep sending that company forward regardless of what happens to it.

Triggers are really what Combat Mission needs (more than fire or hit decals or anything).

Also AI programming should be less unintuitive. The system with AI groups is less than optimal and very uncomfortable to work with. The best would be if any platoon or maybe even squad/vehicle could be given an own "plan" with the possibility to add as many exceptions as possible.

Some people think, triggers be used to create unrealistic effects, but I believe people would just not play such scenarios.

[Badger73]

The AI does not respond well to sloppy programming. The scenario designer needs to be rigorous, pay close attention to the timing, composition and coordination of forces, to the terrain, to any likely player reaction. This is not something you're going to see in a generic '1 size fits all' QB offense. The best offensive AI scenarios can kick your butt three ways to Sunday first playthrough. It should also be noted, quitting out and restarting a scenario if the battle's not going well then finally succeeding does not really count as 'winning against the AI'.

Agreed. Would love to see a primer on this for CMx2 scenario design. Is there such documentation somewhere? Or a thread that shows and tells?

[jep]

The AI in the Command Ops/Airborne Assault series is excellent. Other than that I can't think of any game with really good AI.

The addition of triggers though will definitely make a huge difference if they get implemented.

Only if scenario designer is willing to sacrifice days to implement AI.

[Childress]

Triggers are really what Combat Mission needs (more than fire or hit decals or anything).

Wouldn't that render scenario making more complex? There hasn't been an avalanche of user created battles in CM 2.0 despite the vaunted overlay feature.

[Rokko]

Maybe, but it could be an optional addition to the system now.

Besides. Considering guestemating very precise timing for each AI Plans for every AI group for EVERY group's subobjective can be VERY time consuming, considering you have to test a lot and change back and forth.

So actually triggers might make scenario creating less time-consuming.

Triggers could be like:

if casualties > 20%

{

go for Objective C (continue attacking)

}

else

{

go objective D (abort attack, retreat)

}

or maybe AI Group 1st Platoon/ A Company gets a trigger to wait for 2nd and 3rd platoons before continuing with its next objective.

That way you wouldn't have test when the other platoons arrive by the clock.

[womble]

Wouldn't that render scenario making more complex? There hasn't been an avalanche of user created battles in CM 2.0 despite the vaunted overlay feature.

It might make it more complex, but it would provide tools for wannabe designers to do a better job, which might make them more inclined to attempt it. At the moment, it seems, perhaps, like trying to build a working automobile with a leatherman tool and a pair of mole grips: not a task many would relish, or be able to accomplish, despite the sterling examples of the McGyvers of AI plans...

[Childress]

Or simply make 'AI on defense' scenarios which already work pretty decently. But trying to make credible battles starring the Artificial Intelligence on the attack is a lost cause. Which several of the battles from the, otherwise excellent, GL campaigns attempt to pull off.

That was a very intricate response, Womble.