Retreating Monster Tanks revealed (monster pictures)

[aka_tom_w]

Steve writes:

"Tom,

quote:

The question is can the CMBB engine accomodate a tweak or change to the Tac AI to acount

for the "caught with the pants down" factor?

The TacAI already has this available to it. See my comments about the Stuart test. A tank looks at the

chances it can kill another tank based on its CURRENT situation. If the situation is that an otherwise

crappy tank is facing the rear armor of a big brute, it will likely take a shot or two (or five or six in my

example). This is even after the big baddy puts a target line on it. In fact, I would argue that the TacAI

should have moved my Stuarts before they all got brewed up."

Ok Thanks Steve

Since I tested it with Redwolf's example I found the Russian tanks it not back away, at all, from rear shots at Mk IV's and were not scared of them.

Thanks for the clarification.

Here's hoping you actually get some "time off" on the weekend sometime

-tom w

[ December 01, 2002, 12:31 PM: Message edited by: aka_tom_w ]

[Licensed Fool]

Maybe Battlefront should publish a list of ALL factors that the A.I. takes into account when making some AFVs back off from a "threat" and with what weighting,so we could see the formula for ourselves? E.g.-Respective crew quality/armour thickness/gun capabilities/cover terrain/hull-down/moving or stationary/in or out of command/buttoned or unbuttoned etc.(Anything I've missed?) The bottom line is,some AFVs in the game retreat when they shouldn't,hence the game seriously lacks realism until its put right.

[Harv]

Originally posted by Licensed Fool:

The bottom line is,some AFVs in the game retreat when they shouldn't,hence the game seriously lacks realism until its put right.

And which ones would those be? I've been trying to get this to happen myself but apparently haven't been using the correct AFVs. Please send a list to me as soon as you can.

Thanks,

Harv

[Redwolf]

OK, folks, I really really hope this leads somewhere this time. I

evaluated how big the probablity is that my savegame shows the bug. As

I said, it did for 2 of the 3 tests I ran.

I re-ran my last savegame a couple of times to see how often it

reproduces the offending behaviour. Note again that it is not my

fault that the savgames don't store the random number seed for the

next turn so that the movie is different each time.

Here is how it is done: download

http://65.96.131.208/tmp/isu-coward/isuretreat6.cme

Open hotseat.

Do not give or change any orders. The orders to reproduce the problem

are already in the savegame.

Just hit "GO", two times (one for Axis and the for Soviets). Select

the ISU to see what it does.

Here are my tests:

1: kills Pz IV fine

2: nails Pz IV

3: misses, get a ricochet fromt he Pz IV and retreats. That is OK

since it cannot fire during reload

4: same

5: retreats before firing, fires on the backward move (bug)

6: nails Pz IV

7: shoots and reverses after being hit (like 3)

8: same

9: same

10: same

11: same

[restart CMBB]

12: retreats before firing (like 5) (bug)

13: shoots and retreats afterwards

14: shoots and retreats after the Mk IV fires the second time retreats

15: nails the Pz IV

16: starts reversing before firing (bug)

17: reverse after firing

18: misses and retreats after second shot from Pz IV

[restart cmbb]

19: retreats before shooting (bug)

20: retreats after missing

So I have 4 out of 20 times that the bug is shown, or 20%. In these 20

I had no case where the ISU didn't shoot at all, which was the case in

the very first movie I got when playing this and which my screenshots

are made of.

I don't see how anybody could come up with a better savegame given

that the movies are not reproducable. If there is a desire to debug

the retreat-before-shooting condition then it is pretty obvious that

re-runnign this savegame an average of 5 times will show the problem

faster than random isolated testing which never shows the bug.

[update: ran two more times. One retreat after fire (fine), one retreat without ever shooting, neither before retreat nor while moving backwards (bug)]

[ December 01, 2002, 02:22 PM: Message edited by: redwolf ]

[Redwolf]

Originally posted by Harv:

</font><blockquote>quote:</font><hr />Originally posted by Licensed Fool:

The bottom line is,some AFVs in the game retreat when they shouldn't,hence the game seriously lacks realism until its put right.

And which ones would those be? I've been trying to get this to happen myself but apparently haven't been using the correct AFVs. Please send a list to me as soon as you can.

</font>

[aka_tom_w]

Hi Refwolf

Is it really a bug if %80 of the time the TAC AI behaviour if acceptable? But by random choice %20 of the time the tac AI behaviour favours self preservation, but is irrational from your point of view?

Is that really a bug in your opinion?

I'm just askin'

-tom w

[Harv]

Oh, I read "some AFVs" as meaning certain types when I first read it. But it could also mean all AFVs will do it sometimes in certain circumstances I suppose.

Either way, it's a pretty sweeping statement, and I was hoping to get some more detail from him before I continued with this.

Harv

[Redwolf]

Originally posted by aka_tom_w:

Is it really a bug if %80 of the time the TAC AI behaviour if acceptable? But by random choice %20 of the time the tac AI behaviour favours self preservation, but is irrational from your point of view?

These 20% (or whatever it really is) represent a bug. They see a late-war heavy tank destroyer flee before a mid-war upgraded early-war, clearly weaker medium tank, and by doing so losing many advantages.

I do not believe that this is CMBB's normal TacAI self-preservation code in action, it is a bug that looks like it is the self-preservation code.

In reality it is a probably some temporary screwup with vehicle data. That a certain sequence of spotting and targetting has the wrong armor and penetration data inserted for the calculation. Crazy example would be for this test case that somehow for a given sequence of spotting for this ISU and the Pz IV the other vehicles on the map influence it, in a computational-wise broken way, like accidentially inserting a SU-76's data when doing the ISU calculations. Just a crazy example - if Steve could mail me the sourcecode already I could have a btter look

[Redwolf]

Originally posted by Harv:

Oh, I read "some AFVs" as meaning certain types when I first read it. But it could also mean all AFVs will do it sometimes in certain circumstances I suppose.

Either way, it's a pretty sweeping statement, and I was hoping to get some more detail from him before I continued with this.

Nobody ever in this thread or in the several previous once had any serious test data that indicate that one vehicle combination always shows this bug. We only observe that Soviets show the problem more often.

I recommend you concentrate on the testcase we have here instead, which is a spotting/targetting sequence or influence by other happenings on the map.

[Harv]

OK, I just ran your savegame 50 times and I didn't see the ISU retreat a single time unless it was gun damaged or reloading.

I didn't do anything except hit the Go button twice and watch the movie.

The ISU also was killed 6 times, was gun damaged 4 times and was penetrated 5 times killing a crew member so I'm not sure about your statement that the Pz IV is a clearly weaker medium tank.

I'll run it another 50 times just because it's better than watching hockey, but what I'm finding so far is that there isn't a problem.

Harv

[Redwolf]

Here is a PBEM movie file which always shows the behaviour:

http://65.96.131.208/tmp/isu-retreat/isuretreat8.txt

It will show what I mean everytime your run it.

It took, BTW two tries to produce it, overall I am at 27 tries and 9 exposures of the bug now, which is a 33% chance.

[ December 01, 2002, 02:33 PM: Message edited by: redwolf ]

[Redwolf]

Originally posted by Harv:

OK, I just ran your savegame 50 times and I didn't see the ISU retreat a single time unless it was gun damaged or reloading.

You're on 1.01?

I wonder how you can possibly have tried it 50 times in 21 minutes. How long does it take your machine to load the 3D graphics?

[ December 01, 2002, 02:37 PM: Message edited by: redwolf ]

[Redwolf]

A single try takes me exactly 1:05 minutes, and that is with skipping the first 10 seconds of the movie. And I have a fairly fast machine.

[Harv]

Forget it redwolf. Your stopwatch doesn't have any idea how many tests I've run with all of the savegames you had listed on your first post here.

If you wish to doubt me, that's fine...I don't care either way now. The onus is on you to prove that it's broken, not on me to show that it works as intended.

Have fun,

Harv

[Battlefront.com]

Redwolf,

I feel directly insulted by the following items. I know in my head

that most of them are caused by too low an attention span

I am directly insulted!! I am leaving this BBS for a month

My ISU is hulldown. There is proof in my screenshots. The proof is

that I gave a "seek hulldown" command to the PZ IV position and it

stopped on that hill. I posted a screenshot why said stopped. It

will only stop when it reached hulldown. The ISU is hulldown.

If you reply to this stating it is not hulldown, then please explain

the standing ISU.

I found this to be an irrelevant issue in and of itself and therefore disregarded it as such.

Other irrelevant stuff snipped.

At the distances of my test the kill chances are "excellent" for the

ISU and "ok" to "fair" for the Pz IV. The hit probablities are equal.

I would not assume this. The hit probabilities are probably a little in favor of the PzIV. The ISU-122 is slightly bigger. But on the whole, they are likely similar at this range.

It it bejond me how nobody goes into the editor for two minutes and

checks this. Worse, in all humbility(?), it is bejond me how anybody

can assume that I post stuff like this without checking such simple

facts first - I don't.

First of all... remember this thread is not just a discussion between you and me. There are others posting here and some of them stated quite definitely (in the previous thread) that the PzIV had no chance against the ISU-122. Go back and check it if you doubt me. Who has the short attention span?

I never claimed that the ISU always retreats.

See above comment.

I only claim that there is a bug in the CMBB TacAI that for some specific sequence of spotting and targetting the mechanism is messed up.

Yet I still have yet to see it. Plus, others in this thread (reminder again about above comment) have said this is a sort of epademic with SU heavies.

It doesn't happen always (and it certainly didn't happen in the

isolated tests I ran before making these test savegames), but it is

not uncommon enough to ignore either.

It is easy to ignore something that I haven't seen any evidence of. Sorry, but after the last big discussion (Sneak) with you I do not trust your observations nor your reporting of them. Therefore, I require save files to see it with my own eyes.

And at the very least my tests of your file proved, without any doubt, that some are trying to make a mountain out of a molehill (at best).

I tried to do it better than in the auto-sneak-exhaustion thread

Providing a file is the best way. Posting screenshots is not. But your effort to improve over your last "big issue" is noted and appreciated.

For me, it showed the offending behaviour in two of the first three

runs, starting from the last savegame with changing no commands in the

savegame, just hit "go".

All I got out of this was taunting.

Where? I saw me blasting a bunch of people's convictions out of the water with rather quick and easy tests, but where were you taunted? Ah... yes, I know. I asked you to explain what it was that the TacAI was doing wrong. Well, I think that was more or less a fair remark to make.

It is neither my fault that CMBB doesn't have a "save all autosaves"

functionality

Not my fault that you are too lazy to rename the Autosave file after each turn. I just played a PBEM game that was 26mb. I shudder to think what my Save Games folder would look like if we saved EVERY TURN automatically.

not is it my fault that hotseat savegames don't store

the random number generator seed and hence come up with a different

movie each time. My last savegame showed the offending behaviour 2 out

of the 3 times I tried it.

Actually, this is a good thing. If something is an issue it should be easy to replicate with continued playing out of the turn. Sure, it might take more than one try, but if it is 2 out of 3 as you claim, that isn't such a big deal now is it? If it is... who has the short attention span?

I stated very clearly that isolated tests did never show the behaviour for me, that is why I sat down and reimported the old terrain. Still, Steve, you go on and on and state results from other artificial situations. This will not show the bug, and I pointed it out. Now you blame me for wasting your time in these tests I rated as useless in first place.

A reminder again that you are not the only one in this thread making claims that need to be challenged. Sorry that you can't see that.

The bug is worth fixing

It isn't a "bug" until we say it is. Until then it is user conjecture. And as the Sneak thread showed quite clearly, as well as this one, conjecture isn't worth a damned.

Steve, you ask for people sending you savegames which show the behaviour. What can possibly be better with the savegames I provided?

One that showed the problem. I tested your file isuretreat6.cme and results are posted in next post.

If you place yourself in my point of view, would you agree that I provided the maximum of savegames and descriptive screenshots short of doing it in PBEM (to replay the same movies)?

Sure, you did just fine. Again, I used your files to knock down lots of other stuff in this thread that was, without question, incorrect.

Steve

[Battlefront.com]

Using Redwolf's isuretreat6.cme file I found...

1. ISU-122 advanced, fired, missed, was struck twice by PzIV (1st shot killed ISU)

2. ISU-122 advanced, fired, killed PzIV first shot.

3. ISU-122 advanced, fired and shot hit small rise above it. Proximity to shot + incoming from PzIV caused to back away a few meters out of LOS. Next turn I advanced it forward again. It got into LOS of two PzIVs, took a hit that killed a crew member. Shocked. Took two more hits, pulled back in Panic. Penetrating kill, bailed out.

4. Nearly identical to #3. Difference is that on the next turn it got a shot off before being hit and losing a crew member and being Shocked. The other PzIV was not in LOS this time either. ISU took another hit, Panic. Did not retreat. Took two more hits, killing another two crew members. Bailed out.

5. ISU-122 advanced, fired one shot and was struck with partial penetration. ISU was Shaken and pulled back.

6. ISU-122 advanced, buttoned up do to MG fire, started to pull back. No shots exchanged. Next turn I moved it forward, again it came under two PzIV's fire, started pulling back and got partial penetration knocking out gun and crew member. Several more hits followed making the unit Routed.

7. ISU-122 advanced, retreated without anything happening. Next turn moved forward, exchanged one shot each, retreated. Next turn repeat of first. Next turn advanced again, both tanks exchanged shots and neither flinched. ISU actually burried one shot in front of it but didn't move back like earlier examples. End of turn PzIV scored one partial penetration hit, no damaged but changed status to Shaken. Didn't retreat. Next turn they traded shots and ISU was hit 4 more times. After 4th hit it pulled back.

I'll post my conclusions, as such, in next post.

Steve

[Battlefront.com]

Redwolf,

These 20% (or whatever it really is) represent a bug.

No, there is no bug. Of that I am as convinced as ever.

They see a late-war heavy tank destroyer flee before a mid-war upgraded early-war, clearly weaker medium tank, and by doing so losing many advantages.

Yet another sweeping generalization that is easily disproven. My earlier tests that you so easily cast aside prove that quite easily. My more recent seven tests show something else...

The ISU-122 was put into a fairly crappy firing position vs. the intended target. This is clearly evident to me. 1 in 7 times it scored a hit. 6 out of 7 times it was struck and either died or otherwise on the losing end of the engagement.

So... with this and the other tests done by people, even your own tests, it is clear to me that the TacAI is behaving quite correctly. It is not predictable so sometimes it is a bit more brave than other times. That is also correct.

The bottom line is that the ISU correctly is skittish about being in this specific location vs the specific target. The results clearly show that it has very, VERY good reason to be unhappy about this position. Therefore, pulling back does not show a "bug" but very sensible assessment of the situation.

The only thing I would change here, if I could, would be for a vehicle to ignore the impact (firepower) of its own shell nearby. I think that caused at least two of the pullbacks. But the system is not sophisticated enough to figure out "oops, that was my bad shot, not someone elses good shot".

In reality it is a probably some temporary screwup with vehicle data.

You don't know what you are talking about. And you wonder why the threads you start go nowhere slowly?

I see nothing wrong based on this test. I see nothing wrong based on my previous tests. I see nothing wrong in the games that I play. In short... I don't see anything wrong at all that can't be explained reasonably and rationally. There might be something wrong somewhere, but until it is shown to me... there is no "bug".

Steve

[aka_tom_w]

Steve Says:

" The bottom line is that the ISU correctly is skittish about being in this specific location vs the specific target. The results clearly show that it has very, VERY good reason to be unhappy about this position. Therefore, pulling back does not show a "bug" but very sensible assessment of the situation."

Now that sounds like the Tac AI is erring (if it is in fact, in error??) on the side of Self Preservation to me. :confused:

Redwolf,

Why can't you admit that the Russian ISU in your example is in an untenable situation and KNOWS it, , sometimes it wins and sometimes it loses. Somehow it knows it can be killed and wants to back out of that situation, sometimes it is braver than other times BUT (this is the BIG one) it would seem that most of the tests by you and Steve indicate that when it is brave it gets hit and penetrated thus (logically, ala Mr Spock) implying that in that situation it should retreat because when it does not it usually loses the dual it would seem. Thus the %20-%30 of the behavior you call a bug when it retreats is in fact its best move because it could stay and get knocked out if it is hit in that situation.

Am I confused by the logic that suggests that there is a bug because a unit that could get knocked out if it is brave and stays to dual instead chooses to invoke its self preseravtion sub routine and instead retreats to save its hide???? :confused: :confused:

-tom w

[ December 01, 2002, 03:43 PM: Message edited by: aka_tom_w ]

[Battlefront.com]

Tom,

Now that sounds like the Tac AI is erring (if it is in fact, in error??) on the side of Self Preservation to me.

Yup. For the most part, this is exactly what the TacAI does and should do. In real life AFVs did not like to engage each other when they felt there was a decent chance of not making it. Tanks are not robots. They are manned by Humans and, unlike the tank itself, the Humans care if they survive.

However, it doesn't always mean the vehicle pulls back. The ISU tests and an overly unconcerned armored car in a battle I recently played showed. I wish my AC had been a bit more biased towards self preservation

Steve

[aka_tom_w]

Originally posted by Battlefront.com:

Tom,

</font><blockquote>quote:</font><hr />Now that sounds like the Tac AI is erring (if it is in fact, in error??) on the side of Self Preservation to me.

Yup. For the most part, this is exactly what the TacAI does and should do. In real life AFVs did not like to engage each other when they felt there was a decent chance of not making it. Tanks are not robots. They are manned by Humans and, unlike the tank itself, the Humans care if they survive.

However, it doesn't always mean the vehicle pulls back. The ISU tests and an overly unconcerned armored car in a battle I recently played showed. I wish my AC had been a bit more biased towards self preservation

Steve</font>

[Redwolf]

[Tarquelne]

So I have 4 out of 20 times that the bug is shown, or 20%.

Isn't it unusual for a "bug" to show up only occasionally - _excatly_ the same situation, just some different random numbers.

I was able to coax the ISU into a hull-down position at the end of the turn (took quite a few tries, what with the frequent retreats), so I could check the hit and kill chances at that point.

PzIV vrs. ISU - 29% hit, Fair kill

ISU vrs. PzIV - 27% hit, Excellent kill

That looks pretty iffy for the ISU to me, given that the odds of a 1 shot kill are close to 1/4 (treating "excellent" as "100 % kill", which is of course optimistic) and the odds of the PzIV connecting with it's second shot before the ISU gets a second shot are good.

I really doesn't look like a "bug" to me, redwolf. At most, I think you could argue that BFC made the self-preservation tropism too strong.

Not a bug, at least based on this one test. Do you have a scenario you could easily edit to try and give the ISU a slightly less worthy opponent?

BTW - Characterizing the matchup as a "late war tank destroyer vrs. a mid-war tank" might be throwing you off a bit. I think it's more accurate to think of it as a "unit with good gun and armor vrs. a unit with a better gun and poor armor." The 75mm gun and good optics on the PzIV make it something more than a generic "mid war tank". (And I wonder if similar thinking ("huge Russian beasts backing away from puny PzIVs")isn't behind the thinking that the Soviet tanks are "cowardly.")

I'll see if I can get an ISU-122 to retreat from a clearly inferior foe...

[Redwolf]

Originally posted by Battlefront.com:

6. ISU-122 advanced, buttoned up do to MG fire, started to pull back. No shots exchanged. Next turn I moved it forward, again it came under two PzIV's fire, started pulling back and got partial penetration knocking out gun and crew member. Several more hits followed making the unit Routed.

7. ISU-122 advanced, retreated without anything happening. Next turn moved forward, exchanged one shot each, retreated. Next turn repeat of first. Next turn advanced again, both tanks exchanged shots and neither flinched. ISU actually burried one shot in front of it but didn't move back like earlier examples. End of turn PzIV scored one partial penetration hit, no damaged but changed status to Shaken. Didn't retreat. Next turn they traded shots and ISU was hit 4 more times. After 4th hit it pulled back.

Good, at least you could reproduce the retreats without shots.

So, given the hit probablities and kill chances I just screenshotted about, tell me again why is it so desireable for the the ISU to pull back about 1/3rd of the times and why never for the Pz IV?

[aka_tom_w]

Originally posted by redwolf:

[QB]

That is an interesting question, perhaps warrenting further discussion. Should the Pz IV in that example not have the same self preseravation tac AI behaviour invoked sometimes as well?

Has anyone seen the PZ IV back off in that situation?

I don't think I have seen the German tanks back off or retreat very often at all, BUT I admit I have not been looking and I don't play the German side very often

:confused:

-tom w

[ December 01, 2002, 04:38 PM: Message edited by: aka_tom_w ]

[Tarquelne]

So, given the hit probablities and kill chances I just screenshotted about, tell me again why is it so desireable for the the ISU to pull back about 1/3rd of the times

Looks like you posted the same time I did.... If _I_ were in the ISU I'd consider retreating. Hmm... I have something to try...

and why never for the Pz IV?

It it didn't have to rotate in order to move away, it might. And I wonder if the AI is more likely to retreat a unit that was just in motion.

[ December 01, 2002, 04:38 PM: Message edited by: Tarqulene ]