Propensity for AI to hit HQ tanks

[sgtabell]

A question to the grogs...

When playing in the game, I notice the AI killing or shooting at the command tanks before others, with what seems to be an unrealistic, (at first glance) ability.

Were tanks during real war able to spot command tanks easier?

Is there a visual difference in the game between a command/hq unit and others?

Is it a unit type by unit type difference? (I remember hearing that there Russian tank(s) which stood out due to a radio transmitter antennae sticking out.

Just wondering,

SgtAbell

[Zimorodok]

For sure the AI targets your command tank first. How it knows (besides cheating) I couldn't tell you, but I have noticed this time and again. For a while, I was doing the same thing (the collar tabs on the Tank? icon change...look for the sub-leader pips or whatever denotes an HQ) but I felt that was more a defect in the layout of the screen than actual battlefield intelligence, so I don't do that anymore. I think I read that it was going to be 'fixed', but I haven't checked to see if v1.03 does that or not.

(i have also noticed, incidentally, that my command tanks are usually the worst shots, despite their point value...but that could just be my units)

Zimorodok

[c3k]

I, too, have noticed this behavior. I brought it up several months ago. I got dog-piled upon by various self-nominated grogs who told me I was wrong. However, I've seen my HQ tank get targeted first too often. Even if it's the last tank to come into LOS.

In real life, early war tanks would have to remove their main gun to make room for the radio sets. Other vehicles sported large antenna farms. In the game, these visual clues don't exist. Perhaps they're modelled in the code. I don't know.

Ken

[manchildstein (ii)]

it seems to me that the antennas were on the company command vehicles and up... not on platoon commanders...

from what i've seen the ai definitely likes to take out platoon commanders first though...

[MikeyD]

Back in the 'dim time' before CMBB was even in Beta there was a long debate about whether or not we'd be able to visually pick out a HQ tank during a battle. There was a great clamour to include this feature in the game.

Obviously we didn't get that feature, BUT it seems the game engine is able to do what the player can't. And everything's equal -- the AI for YOUR side has the same abilities as the opponent AI. You simply have to learn to let the AI pick your anti-tank gun's targets for you. It's cleverer than you are at it!

[tar]

IIRC the resolution of this was that after a while, you could figure out which tank was the HQ tank. It is indicated to the player by subtle changes in the picture of the soldier associated with the tank. When he gets the rank insignia, then you have identified the HQ tank.

[Lady Roxanne]

On a similar note, send Green, Regular, Vet, and Crack tanks of the same type over a rise into enemy AT assets. Guess which tank will get targetted first? Second? Last?

Do the same thing again but send the tanks over the rise spaced about 5 seconds apart with Mr. Green first into the firestorm. Watch the enemy gunners switch targets as the pros come into sight.

[Doodlebug]

Originally posted by CrankyKris:

On a similar note, send Green, Regular, Vet, and Crack tanks of the same type over a rise into enemy AT assets. Guess which tank will get targetted first? Second? Last?

Do the same thing again but send the tanks over the rise spaced about 5 seconds apart with Mr. Green first into the firestorm. Watch the enemy gunners switch targets as the pros come into sight.

Oooh. If that's true then it smells like a problem to me. That can't be right. Can it?

[c3k]

A bump and an additional comment or two.

I haven't noticed the "pip" thing in the manual. If, indeed, the command tank is denoted by a VERY subtle change in the unit info screen, a mention of that in the manual would be good. Also, a picture to show EXACTLY what the change is. Obviously, it would be better to have a "?HQ?" marker to make it as obvious as a 2x4 to the forehead.

How DOES the AI discover a command tank? Especially at start?

Has anyone tested the targeting issue vs. better quality crews? (Crack, vet, reg, conscript issue.)

Let's get this to BFC's attention before the final patch is released.

Ken

[Wicky]

In the first version of CMBB, HQ units esp. Tanks would be spotted after a couple of minutes even in Xtreme FOW in the demo(see pic below)- The ease and speed of spotting was sorted in the first release/patch

I've made some bmps (based upon demo) which more easily show the HQ units - Available from here:

HQUnits.zip

[PeterX]

Watch the enemy gunners switch targets as the pros come into sight.

I set up a scenario like this and didn't notice a propensity on the past of the AI to target the aces. It seemed random. There was a marked tendancy for Fighter Bombers to take out HQ and more experienced AFVs in earlier versions of CMBB. This doesn't seem to happen anymore- based on non-scientific observation. The targetting algorythms may, as far as I know, incorporate purchase cost which is higher for the better crewed units.

[ May 17, 2003, 01:50 PM: Message edited by: PeterX ]

[Lady Roxanne]

I have a scenario using 1.02. All things being equal, except experience, the crackerjack will die first. Last to die will be your Greens. This is true even if the Cracker is the last spotted and at the greatest range. I did see one time where a Crack was hit before an Elite. I suspect a small bug there. BTW, the enemy knows which of your squads carry the most AT assets. Given a choice of several squads to fire at, in equal cover and at the same range, your guys with the most AT assets will be targetted first. I will admit I haven't tested this in quite some time however.

All this is really no big deal, IMO. The AI needs the help, and it works for you too if you let the AI do the targetting for you.

[ May 17, 2003, 08:44 PM: Message edited by: CrankyKris ]

[Joachim]

If you think the AI "cheats", let it work for you. If the AI targets crack and HQ even if the hit probability is lower, then it will be those doing the overwatch in hull-down positions.

OTOH I remember two '41 QBs (v1.02 or below)where I had green/regular Romanian tanks (those slow ones with 2 man crew, 37L21 gun but thick armour) in the first line with a veteran/crack PzIIc, IIIg and IVe in the second line about 50-100 metres back. The poor Romanians took the incoming - but usually survived 45mm and 76mm fire with their thicker armor - while the German tanks run out of ammo doing the kills on ranges above 500m. This experience suggests that the AI is not cheating - it acted pretty dumb on this occassions . Even a SdKfz231/8 scored some BTs from that "cover". Had I done the targetting for the Soviet guns, I know what I would have shot at

Gruß

Joachim

[Andreas]

Ahem - the TacAI can not 'cheat', since it works the same for you, as it does for the computer, or your opponent. So if there is a propensity to target HQ tanks (there certainly seemed to be one in the early builds, as MikeyD says), then it will work just the same way for you as against you, unless you override the TacAI targetting orders.

[Joachim]

Originally posted by Andreas:

Ahem - the TacAI can not 'cheat', since it works the same for you, as it does for the computer, or your opponent. So if there is a propensity to target HQ tanks (there certainly seemed to be one in the early builds, as MikeyD says), then it will work just the same way for you as against you, unless you override the TacAI targetting orders.

That's why I wrote "cheats" instead of cheats. Hehe, maybe the AI acts gamey then . But using micromanagmenet ain't gamey, so even this ain't true. Hey, then I am cheating, as I can use that feature and abuse the AI's use of that feature. :cool:

But I seriously doubt the AI will shoot on targets who have a significantly lower hit prob. Nevertheless I don't put HQs or the best tanks in the first row.

Gruß

Joachim

[Soddball]

Well, I wanted to see this in action for myself, so I ran a test.

I went into the Scenario Editor, and placed a patch of woods on an otherwise entirely bare map. I then placed a Russian 37mm AA gun in the woods - I wanted something that was fast firing.

I then picked a platoon of Marders (nice and vulnerable to the AA gun), and removed all their ammunition.

I left the HQ as Regular, but I set one Marder to Conscript, one to Regular, and one to Crack.

Then I proceeded to line the Marders up, with the Conscript marder 75m away from the hidden AA gun, the Regular (non-HQ) 175m away, the Crack Marder 275m away, and the HQ Marder 375m away.

I then hotseated the game, letting the AI do the targetting.

First to get busted was the Crack Marder (275m), then the HQ(375m), then the Regular(175m), then the Conscript Marder(75m).

I ran the test 10 times, and the AI followed this pattern every single time.

Whilst this might not be 'gamey' because it works both for and against you, I don't like it and would like to see it changed.

Edited to add: This test was carried out under Extreme Fog of War, and the AA gun was able to see all four Marder tanks at the beginning of turn 1.

[ May 19, 2003, 01:04 PM: Message edited by: Soddball ]

[Guest PondScum]

I bet that this behavior is a side-effect of the TacAI trying to micro-optimize survival chances by evaluating all threats. So in Soddball's excellent example, the TacAI looks at the opponents and computes their to-hit chances: "Ok, the Crack Marder has a 20% chance of hitting me, the Regular has an 18% chance, ..." etc. Then it tries to destroy the most threatening opponent first, where "most threatening" is based on their to-hit chance, with some additional factor thrown in for being the HQ tank. But the distance is irrelevant, UNLESS it affects the kill chances (i.e. penetration), which it won't in this particular scenario.

Soddball, did the 37mm gun ever see the experience level of the Marders? If it saw them all from the start, then this targeting behavior is probably "working as designed" from BFC's standpoint (although you could argue that they should add a "damn, that tank is REALLY NEAR, let's shoot THAT ONE" factor for the next engine ). But if it was just identifying them as generic Marders all the way through, then it looks like information leakage is taking place - the game engine should be using "regular Marder" hit chances, which would make it take out the nearest one first. I'm wondering if the to-hit-chance routine has perfect information, even though this breaks EFOW...

[ May 19, 2003, 03:03 PM: Message edited by: PondScum ]

[MikeyD]

I'm interested in seeing the piping change on the corner portraits of enemy HQ tanks as I select them. That's something I've never noticed!

Another thing about AI targeting HQ. I believe it breaks down if you've got two or more guns firing on the enemy. Single gun may be more consistent at targeting HQ units.

[Soddball]

Originally posted by PondScum:

I bet that this behavior is a side-effect of the TacAI trying to micro-optimize survival chances by evaluating all threats. So in Soddball's excellent example, the TacAI looks at the opponents and computes their to-hit chances: "Ok, the Crack Marder has a 20% chance of hitting me, the Regular has an 18% chance, ..." etc. Then it tries to destroy the most threatening opponent first, where "most threatening" is based on their to-hit chance, with some additional factor thrown in for being the HQ tank. But the distance is irrelevant, UNLESS it affects the kill chances (i.e. penetration), which it won't in this particular scenario.

Soddball, did the 37mm gun ever see the experience level of the Marders? If it saw them all from the start, then this targeting behavior is probably "working as designed" from BFC's standpoint (although you could argue that they should add a "damn, that tank is REALLY NEAR, let's shoot THAT ONE" factor for the next engine ). But if it was just identifying them as generic Marders all the way through, then it looks like information leakage is taking place - the game engine should be using "regular Marder" hit chances, which would make it take out the nearest one first. I'm wondering if the to-hit-chance routine has perfect information, even though this breaks EFOW...

Good question. I reran the test and at no point did the Marder's experience pop up - as one would expect in EFOW.

If you'd like to try this test scenario yourselves, please email me and I will pass a copy over to you.

It does seem possible that the computer is evaluating its targets on the basis of risk - but if that's the case, then why would it be targetting HQ vehicles?

I would like to hear commentary from someone in BFC on this matter (if they have time).

[Guest PondScum]

I would bet that the targeting algorithm takes multiple factors into account. Risk is one factor, because you want to stay alive. Another factor is the chance to take out enemy HQ units, and weaken their command-and-control. Pure unit cost could be another, trying to kill whatever is most expensive. Of course, we can't differentiate the effects of unit cost from those of risk in this micro-scenario - the crack Marder is the most dangerous opponent, but also the most expensive.

Either way, it looks like you've definitely found an EFOW bug - the TacAI's targeting routine is getting full information about the experience level and/or cost of units, when it shouldn't have this information. Without a full ID, it should be treating them as generic regular Marders, and thus taking them on closest-first (because if they're all the same experience level, the closest one will be the easiest one to hit *and* the one with the greatest chance to hit the gun).

[illo]

This enables a ridicilous gamey tactic which i just tried out.

I bought 4 PSW 232, 3 greens 1 elite for jerries and regular 76,2mm ATG for ruskies. Map was flat and open and i took command of germans.

When battle started I ordered my elite SPW to drive back and forth at fast speed on the map edge. I gave orders for green PSWs to drive directly towards enemy positions trusting the enemy would notice their incompetence and ignore them totally.

The 76mm atg was spotted by my green attack team from some 150m away. As soon as russian gunners noticed that they were spotted they took aim at my elite PSW some 800-1000m away...and suprise suprise. They couldn't hit this speedster at all. Meanwhile my other PSWs just sat there 150m in front (and in full view) of ATG having a lunch break (i issued them covered arc command so they woudnt supress the poor ATG).

Ivan has sharp eyes.

[ May 19, 2003, 11:22 PM: Message edited by: illo ]

[AJSL]

*bump*

[ May 19, 2003, 07:22 PM: Message edited by: AJSL ]

[Wolfe]

Originally posted by illo:

Ivan has sharp eyes.

:eek: Somebody is actually using my green interface mod?!? I'm flattered, but I assume you haven't seen Pakfan's mod?

A bit on-topic. I thought I had seen a propensity for the AI to target the more expensive units first, even ones further away as far back as some of my first CMBO games. But I don't think it's always the case, especially concerning infantry. The game won't necessarily target a more expensive Elite Flamethrower over a nearer infantry squad, but it does seem to pick out the more expensive tanks rather often to me. No hard proof, though.

- Chris

[ May 19, 2003, 11:05 PM: Message edited by: Wolfe ]

[Brigadier]

Gentlemen,

In light of the rather damning evidence presented by the Prosecution (the grogs) and before the Jury retires to consider it's verdict , perhaps BFC should present it's defense on the matter.

BFC , your witness :confused:

[Wolfe]

Originally posted by Wolfe:

a propensity for the AI to target the more expensive units first

Well this doesn't seem to be the case. Vs a Rifle platoon (4 squads and HQ) along with a support unit which is always further away from my MG than the squads are, the MG seems to be targeting based on threat. I think. :confused:

Using the 'N' key the flamethrower is targeted 6th (last), an 81mm mortar is 3rd (after 2 close squads, but before two other squads that are further from my MG, but closer than the mortar), a MG42 HMG is 5th (after all suqads, but before the HQ), and a 105mm FO is 6th in targeting preference.

Replacing my MG with a Soviet 76mm gun, the 105mm FO is targeted last, but if you replace it with an 81mm mortar, the mortar is targeted 1st.

All enemy units are identified only as "Infantry?".

Anecdotal, but interesting.

- Chris

[ May 19, 2003, 11:32 PM: Message edited by: Wolfe ]