Seriously flawed ammo selection algorithm?

[NON]

In a ongoing QB ME my M18 Hellcat encountered a "Assault Gun?". The distance to the target was around 600-700m and since my M18 was camouflaged and would get the first shot at it's flank I ordered it to sit tight and let it rip. For almost two entire rounds the two gunned it out at a distance. The M18 managed to get 6 rounds off, all which connected and promptly ricochetted of the unknown assault gun, while the assault gun, having to turn and face the M18 and having a slower ROF missed twice, and then connected on the third shot,making a lovely bonfire out of the M18.

Now here's the thing, the M18 was carrying no less than 6 tungsten rounds, the same amount it went to AFV heaven with. It did not once switch to tungsten, even after having witness 6 AP rounds just bounce off.

Is this not somewhat ridiculous? Since CM strive first and foremost for realism, would you expect a real tank crew to sit there and bounce round after round when they knew that 1)They had 6 rounds of something that would almost certainly penetrate sitting around and 2) The opposing tank has just fired 3 shots and has us bracketed, and it's certain that whatever gun it mounts will blow us to high heaven ?

The M18 falls apart if you throw bits of ripe Brie at it, and the tank crew had to know that whatever kind of assault gun they were facing they wouldn't stand a chance once the opposing gunner got his act together.

I've noticed this kind of silly behavior before, but on those occasions the vehicle didn't have as many as 6 tungsten rounds and didn't have time to bounce 6 AP rounds so I let it go as just a minor annoyance.

[ April 13, 2004, 09:22 AM: Message edited by: NON ]

[SpitfireXI]

I have noticed that too, I can't remember the last time a tungsten round was fire in CMAK. I really loved it when my Russian tankers would fire those off with no hesitation to the disapointment of the Germans. But in a related note, I have notcied that the StugIII has some crazy tough armour as compared to it in CMBB. In both games a Sherman in less then 500 meters was firing a total of 5 to 6 rounds at the flank of the thing and it kept going.

[Michael Emrys]

This is an old complaint going back to the days of CMBO. I don't recall what BFC has said in explanation/justification of the existing mechanism though.

NON, realistically speaking, what the M18 would have done historically is to get out of the LOF after it was observed that its shots were having no effect and it was about to be brought under fire. That's one reason it had so much speed.

Michael

[Redwolf]

In principle I agree.

But you have to take into account that the tungesteen rounds lose much of their extra penetrarion against angled armor or when shooting at armor at an angle.

If the assault gun is turning and at the same time is one of those with highly angled front plate you would get no benefit out of using tungsteen.

The after-armor effect of the tungsteen round should also be less but I thing CM doesn't model this for the 76mm US AT and HVAP shots.

[c3k]

NON,

Just a dim recollection here, but if the enemy AFV was not identified, then your M18 would not fire the tungsten. It needs to KNOW what it's firing at, ISTR, for it to use the high-value rounds.

Or, it could be that the crew realized that whoever gets to AFV heaven with the best penetrating rounds gets the oft-mentioned virgins of paradise? I could be wrong on that....

Regards,

Ken

[YankeeDog]

I think you're right that there's room for improvement here, but I don't think the problem is with the ammo selection algorithm *per se*.

If you put an M18 up against a target it *knows* it's going to have a hard time penetrating with regular AP, it will usually (1) switch to tungsten, possibly after a ranging shot or two, (2) pop and/or fire smoke, (3) run like hell, or (4) some combination of the preceeding options. While still not perfect, IMHO the TacAI for situations like this has actually been substantially improved since CMBO.

In your example, though, the target was only partially IDed. When the TacAI has a partial ID like "Assault Gun?", it appears to assume that the enemy vehicle the most common type within that general class for the purposes of determining TacAI behavior. To put it in terms of your example, an M18 up against an "Assault Gun?" will generally behave like it's up against a StuG; if it sees "Tank?", it probably assumes it's fighting a PzIV. If this "Assumed ID" is more or less correct, this works fine and the TacAI generally behaves as it should; it usually stays and fights if it has a good chance of KOing the enemy before getting KOed itself (using T rounds if available and necessary), and otherwise it often runs and/or pops smoke.

The problem comes in then the "Assault Gun?" or whatever is actually of a rarer type, and especially a rare, more heavily armored type that the unit in question is unlikely to be able to KO. In this situation, no matter how many rounds of regular AP your AFV bounces off the target, it won't "learn" that the enemy is actually made of tougher stuff tha it first assumed.

IOW, what is really needed is some sort of rudimentary "learning" AI routine for behaviors like ammo selection, target ID, fight/flight, etc. IRL, an M18 crew that bounced 6 rounds off what it initially thought was relatively easily penetrated StuG would probably conclude that it needed to change tactics in some way. I can believe that the TD crew might attribute as many as 3 bounced rounds to bad luck (or simply not seen the hits - it's not true that a gun crew always see their hits), but 6 hits should be enough for the TD crew to know that something funky is going on.

At present, I don't think the TacAI contains these kind of "learn from experience" routines, and this affects other areas of the game as well. For example, unless the player specifically tells them not to, TCs will stupidly unbutton themselves in an area where they have just been forced to button by heavy small arms fire. IRL, if the small arms fire was heavy enough, the TC would probably decide that he was better off inside the fighting compartment so long as the AFV was in that particular area. Again what is needed is some kind of dynamic memory routine that allows the TacAI to conclude "This area is dangerous; it's not worth the risk to unbutton," not changes to the button/unbutton algorithms *per se*.

Charles and Steve have already made some comments about improving the TacAI's ability to "learn" and "remember" in the new engine, and I am hopeful that these changes will make unit behaviour in situations like the one you bring up more realistic. For now, whenever possible try to get Full IDs on enemy AFVs before engaging; this increases the chance of your units using Tungsten if it's needed.

Cheers,

YD

[NON]

But you have to take into account that the tungesteen rounds lose much of their extra penetrarion against angled armor or when shooting at armor at an angle.

Maybe so, but the tank crew didn't know exactly what they were shooting at. And after the 6 rounds bounced, you think they'd be encouraged to at least TRY something else.

Also, I think the armor has to be pretty steep (greater than 30 degrees) for tungsten not to have an advantage over AP at this distance.

NON, realistically speaking, what the M18 would have done historically is to get out of the LOF after it was observed that its shots were having no effect and it was about to be brought under fire

It wouldn't even try tungsten first? In the case in question, it didn't back away either (they had a round and a half to do so) so either way something is smelly

Since it's been said that this is an old complaint, can anyone dig up BFC's explanation for this odd behavior?

[NON]

The problem comes in then the "Assault Gun?" or whatever is actually of a rarer type, and especially a rare, more heavily armored type that the unit in question is unlikely to be able to KO. In this situation, no matter how many rounds of regular AP your AFV bounces off the target, it won't "learn" that the enemy is actually made of tougher stuff tha it first assumed.

I think you've hit the nail on the head. Since the shooting went on for more than one turn, I checked the M18's target and it said that chance to kill was good, and from the stats it looks as if they assumed it was a StuG. It was in fact a Jagdpanther.

So it seems the TC just goes with the guessed stats, and completely ignore what they actually see happening (the shells bouncing)

[NON]

For now, whenever possible try to get Full IDs on enemy AFVs before engaging; this increases the chance of your units using Tungsten if it's needed.

Unfortunately that is next to impossible at ranges over 500m when playing with normal or extreme FOV

[Redwolf]

Originally posted by NON:

</font><blockquote>quote:</font><hr />But you have to take into account that the tungesteen rounds lose much of their extra penetrarion against angled armor or when shooting at armor at an angle.

Maybe so, but the tank crew didn't know exactly what they were shooting at. And after the 6 rounds bounced, you think they'd be encouraged to at least TRY something else.

Also, I think the armor has to be pretty steep (greater than 30 degrees) for tungsten not to have an advantage over AP at this distance.

</font>

[Dittohead]

At that range it should easily tell the difference between a stug and jagdpanther. doh Unless the crewmen are blind in 1 eye and blond in the other.

[Michael Emrys]

Originally posted by NON:

</font><blockquote>quote:</font><hr /> NON, realistically speaking, what the M18 would have done historically is to get out of the LOF after it was observed that its shots were having no effect and it was about to be brought under fire

It wouldn't even try tungsten first?</font>

[Michael Emrys]

Originally posted by redwolf:

...armor angle (always vertical in CM)...

Eh? How's that again?

:confused:

Michael

[Redwolf]

Originally posted by Michael Emrys:

</font><blockquote>quote:</font><hr />Originally posted by redwolf:

...armor angle (always vertical in CM)...

Eh? How's that again?

:confused:

Michael </font>

[NON]

I said the combination of armor angle (always vertical in CM) and facing angle. if the target is turning and the 45 degree border for the side shot is passed, then you have a pretty big facing angle and in combination with the armor angle a very high effective angle

I see what you mean, but apart from the first shot made while the opposing JagdPanther was turning to face my M18, all shots were fired directly at the front and the elevation of the two vehicles were identical or close to as well

[Panzer76]

Originally posted by redwolf:

The after-armor effect of the tungsteen round should also be less but I thing CM doesn't model this for the 76mm US AT and HVAP shots.

BFC members (MadMatt) has told me it does.

[Panzer76]

As others have mentioned, it acted like that because

1. It though it was a Stug, which it would be able to penetrate with AP.

2. It cannot learn from the bouncing shells.

On another subject, Im pretty tires of seeing ALL the assualt guns IDed as Stugs, just lik eth eT34/85 was always IDed as T34/76s. The spotter must ID the bastard right away once in a while !

(long range)