Monster Tanks Retreating #2 And Steve's List of Why the Tac AI tells tanks to retreat

[aka_tom_w]

Battlefront.com

Administrator

Member # 42

posted December 02, 2002 12:29 PM

Folks,

I will try and sum up what is going on within the game.

1. The TacAI's Prime Directive is to prevent harm from coming to the vehicle under its command.

However, the TacAI fully understands that it is in command of a war machine which in a setting that is

not risk free. Therefore, the TacAI often errs on the side of slugging it out instead of buggering off if

the risk is more or less marginally equal between killing something and dying in the process of

attempting such a kill. The TacAI is also modified heavily by Experience and random "luck". Sometimes a

crappy unit will do something "brave", other times a really good unit will do something "cowardly". But

in all cases the TacAI is making a logically based assessment of the situation without the bais of what

the player expects of it.

2. The TacAI is less concerned with what damage it can cause vs. a particular target than it is what kind

of damage can be done against it by that target. For example, a Nashorn can eliminate practically any

vehicle on the map. But at 200m even small arms can take it out of action. Should the TacAI ignore this

fact and just drive around the map wherever the player dictates, thinking "I'm king of the world baby!"?

No. It is more concerned with its own survival, as it rightly should be. Otherwise the player probably

wouldn't have any AFVs left after the first turn of enemy contact.

3. It is the player's "responsibility" to understand and respect the TacAI's behavior. You don't need to

agree with it, any more than a battlefield commander would, but understand that it is far more informed

than the player is. A good player, however, understands this and LEARNS from the TacAI instead of

fighting it. A good player needs no more feedback from the game beside what is already provided.

4. The TacAI's primary directive is at least initially overridable by the Human player. This means the

Human can put a vehicle into a near suicidal position without the TacAI having a say about it. This is

mostly becaue the TacAI lacks situational awareness and therefore is ignorant of potential threats the

Human is exposing it to until they actually materialize. Even if we could program in some sort of

situational awareness, we wouldn't because that would pretty much remove the player from the game.

In short, the player is allowed to screw things up to a large degree. However, once the TacAI is

presented with a direct "what do I do now?" situation, it will ignore the player's intentions (actually, it

most often doesn't even know what they are to begin with) and follow its Prime DIrective. If the Prime

Directive decides that a shoot out has a decent chance of a positive outcome, it stays and shoots. If

the TacAI instead decides that the situation is unfavorable, then it scoots.

5. The chance of actually causing damage to another vehicle has a lot to do with RoF. If at X range it

takes roughly Y shots to score a hit, and I shoot 3 times as fast as my opponant, who is more likely to

score a hit first? Simple math shows that I am. The other tank is better off not engaging at all in that

specific situation. Instead it should seek an alternative means of causing me harm. The TacAI gives the

player that chance by retreating instead of being brewed up on the spot.

6. If there is criticism to be made about the TacAI's decisions, it isn't that they are wrong (or buggy as

Redwolf would claim), but that they are too correct. It could be that the TacAI is too smart compared to

a real life crew in the same situation. However, this is something that we can argue about forever.

Perhaps more randomness needs to be injected, but I can promise you all that if we did this we would

see far more threads complaining about "the stupid AI didn't back up" than we have seen of "the stupid

AI backed up". But this is a whole 'nother discussion to have and I don't want to get it confused with

this one.

Well... that is about all I can think of to say at this point. I guess I could comb through this thread and

point out (for the 100th time) why Redwolf's example is functioning correctly, but since he is pretty

much the only one that needs to be convinced of this... I'll pass. He hasn't listened so far.

Steve

[aka_tom_w]

this is the new home of the Monster Tank Retreating thread

-tom

[Foxbat]

Starting out with the conclusion and a complete debunk of the 'Monster Tank-Retreat "Bug"' won't help starting a debate

[Terrapin]

Originally posted by Foxbat:

Starting out with the conclusion and a complete debunk of the 'Monster Tank-Retreat "Bug"' won't help starting a debate

Was the original thread a debate? I must have missed that. All I saw was one poster saying "This is a bug! Fix it!". Not "Is this a bug? If it is, what should we do?"

[aka_tom_w]

Now,

where were we......

Lets review

Redwolf

Member

Member # 3665

posted December 03, 2002 10:56 AM

May I kindly suggest we wait until I answer Steve's detailed question when I go home?

This spinning out of control serves no purpose.

And yes, shoot and scoot is not the correct command to use here. In the sitiation we have here a

stop at contact is required, and the seek-hulldown command does that, the shoot-and-scoot will

continue to the ordered point, and -as documented- it will wait for a few seconds at that point. This

is no good here. The only good use of the ISU, if you choose to take on the Pz IV is getting into LOS

in hulldown, stop, get a shot off and then retreat. This works perfectly fine with the "seek hulldown"

command, except that the retreat happens, in 30% of the cases in my test scenario, before the shot.

Andreas, can you elaborate on why you think it is a good idea to have an optional automatic retreat

in seek hulldown, but not in shoot-and-shoot? Is this really in the game? Why should seek-hulldown

have code in it to override player's orders with a retreat and shoot-and-shoot not?

[ December 03, 2002, 10:59 AM: Message edited by: redwolf ]

IP: Logged

Licensed Fool

Junior Member

Member # 11273

posted December 03, 2002 11:04 AM

For my part,I apologised in a post way back for sounding disrespectful,and I also said I'd taken on

board Steve's excellent explanation of A.I. routines and would be testing his and other peoples

suggestions re the backing off aspect,and that I was already drifting from Redwolf towards Steve on

this. This seems a logical course of action for me,but if I'd have dug in and refused to bend or see

another guys point of view,I'd fully deserve to be called a fool!

--------------------

"A British anti-tank unit blew up my Tiger in Normandy and killed my crew but I was treated well after

capture" (Waffen-SS tanker in letter to L.Fool)

IP: Logged

Andreas

Member

Member # 1091

posted December 03, 2002 11:11 AM

quote:

Originally posted by redwolf:

Andreas, can you elaborate on why you think it is a good idea to have an optional

automatic retreat in seek hulldown, but not in shoot-and-shoot? Is this really in

the game? Why should seek-hulldown have code in it to override player's orders

with a retreat and shoot-and-shoot not?

It is my understanding there is no specific code in it. I'll elaborate tonight or tomorrow on my thinking

about this (it may well be flawed).

--------------------

Is it a joke? IS2 was DESIGNED to fight Tigers! - Cpt. Kloss

Of course what I write is worthless... - Cpt Kloss

QED

Der Kessel - Say 'Git stoofed' to crappy scenarios

German artillery observation practice, photos from AG North - at Der Kessel

IP: Logged

Vanir Ausf B

Member

Member # 5261

posted December 03, 2002 11:38 AM

quote:

Originally posted by redwolf:

And yes, shoot and scoot is not the correct command to use here. In the sitiation we

have here a stop at contact is required, and the seek-hulldown command does that, the

shoot-and-scoot will continue to the ordered point, and -as documented- it will wait for a

few seconds at that point. This is no good here. The only good use of the ISU, if you

choose to take on the Pz IV is getting into LOS in hulldown, stop, get a shot off and then

retreat. This works perfectly fine with the "seek hulldown" command, except that the

retreat happens, in 30% of the cases in my test scenario, before the shot.

I completely disagree.

You say "The only good use of the ISU, if you choose to take on the Pz IV is getting into LOS in

hulldown, stop, get a shot off and then retreat", which is exactly what the Shoot & Scoot command

does. This situation is what Shoot & Scoot was made for. If it's not appropriate here, it's not

appropriate anywhere.

You also ignore the fact that if you use Seek Hulldown the ISU-122 is more likely to stand and shoot

it out to the death than retreat after one shot, which has been demonstrated to result in a dead ISU

more often than a dead Mk IV.

It would be nice if the first move in a Shoot & Scoot order was Hunt instead of Fast, but even as-is it

gives the ISU it's best chance of beating the Mk IV. I would use it without hesitation.

--------------------

ye gods, Slapdragon, have you gone nuts? You can't possibly expect me to read all that. -Lars

In my defense I spent so long typing my post the rational debate started without me. -ScouseJedi

DUDES!!! -Grog Dorosh

[Redwolf]

Just for note, Steve had these questions for me in the other thread and I'll get to them, but not from work. Hasty arguments serve no purpose here.

As for seek hulldown and shoot-and-shoot, can anybody explain to me whether and if so why one or the other has more tendency to either stop for a shot, or retreat, or to proceed to the ordered point? The manual says nothing about them being different in these respects. If they are not different in these respects, then there is no point in telling me I have to use shoot-and-scoot instead of seek hulldown.

[ December 03, 2002, 01:54 PM: Message edited by: redwolf ]

[Foxbat]

Originally posted by redwolf:

As for seek hulldown and shoot-and-shoot, can anybody explain to me whether and if so why one or the other has more tendency to either stop for a shot, or retreat, or to proceed to the ordered point?

Shoot 'n Scoot tells the tank to do what it would naturally want to do anyway so it is more likely to succeed in doing that

[Tarquelne]

As for seek hulldown and shoot-and-shoot, can anybody explain to me whether and if so why one or the other has more tendency to either stop for a shot, or retreat, or to proceed to the ordered point? The manual says nothing about them being different in these respects.

I think the difference is that shoot-and-scoot more or less gaurantees a "retreat" - that's what you've told the unit to do, while Seek Hulldown leaves further movement to the TacAI. Once it's found it's hulldown position you're orders are over.

Furthermore, and possibly more importantly for your purposes, from a few trials this moring Shoot-and-Scoot always results in the ISU firing before retreating. I didn't run lots of tests, so it might still do it sometimes, but I'm confident it retreats-before-firing less often.

So it looks like there's a way to sorta force a unit to advance to a hull down position, fire, and then retreat - Use Shoot-and-Scoot. You lose.. sorry "loose" the convenience of having the vehicle seek hulldown for you, but (in my experience) you do take the shoot/scoot decisions out of the TacAI's hands.

[Battlefront.com]

Gentlemen...

Let's keep this one less heated than the last three or so pages of the previous thread became, OK? I think most of us realize that there really isn't a problem. So I don't think it is wise for any of us, especially me, to invest much more time into this. However, if people want to continue a discussion about how to work WITH the TacAI instead of AGAINST it, I am all in favor of that. The manual is purposefully designed to not get into such details. For better or worse, we want the players to learn for themselves what works and what doesn't.

So, along that line of discussion:

Tarqulene,

I think the difference is that shoot-and-scoot more or less gaurantees a "retreat" - that's what you've told the unit to do, while Seek Hulldown leaves further movement to the TacAI. Once it's found it's hulldown position you're orders are over.

Totally correct. Seek Hull Down is not supposed to produce a reverse order, Scoot and Shoot is. Seek Hull Down only produces a reverse order if the player really shouldn't have put that vehicle in that spot in the first place. As was seen in the previous thread, the TacAI basically has four possible reactions to an unfavorable Seek Hull Down order:

1. Scoot imediately and not fire a shot at all.

2. Scoot after firing one shot or while firing a shot.

3. Scoot after firing several shots and (possibly) taking some hits.

4. No Scoot at all. Just sits there until victory or defeat is determined the hard way.

On the other hand, Shoot and Scoot is basically telling the TacAI "go to this spot, identify target, and fire. Don't hang around because you probably won't like the attention you receive. So as soon as your shot is fired, kick in reverse and get out of LOS."

Hopefully one can see the difference inherent in each.

Also, Tarqulene, I thought I prefaced my comments about your test better than I apparently did. I think what you were attempting to sample (which TacAI behavior had better success) was a really good idea. The problem I had with your test was that it only gathered some of the information I felt was necessary to make such conclusions.

Therefore, I suggested a way to basically cover all the bases with one set of tests. That way we would have all the relevant data in case someone (like me) said "yes, but how is that bit of data compared to this bit of data". I had such questions about your test but you didn't track the data I felt was necessary to answer my questions. Not that your tests were flawed, just the information logged incomplete.

Statistics and scientific sampling is something I was not trained to do in school. Math is also not my strongest subject. Probably why I majored in History instead of the sciences However, after 5 years of doing CM I have become, I think, pretty good at seeing how data interacts with other data. But I have also learned that having too MUCH data is better than having too little. If a test has too little, and the question is legit, then the test can not answer that question. If all the data is logged, it can answer more questions than was originally thought of by the guy setting up the test. This is what I meant about "bias".

Anyhoo... on with the show

Steve

[Battlefront.com]

Oh, and one other note directed at Redwolf specifically here.

In the previous thread I posted some questions for you. They were issues that I saw you either ignore or continue to repeat even though they were not true. I asked you to answer these questions, not to compare genitle sizes , but to make sure you understand how off base you were *and*, much more importantly, why yet another thread you were in turned into a long and (variably) unproductive discussion.

However, I will instead take a simple admission from you that your assessment of the ISU and the TacAI's treatment of it were offbase. That at best there is a tiny disagreement with the frequency of the TacAI using Behavior #1 (noted above) and are prepared to back up that claim with some facts. I think after all that has gone on in that thread, you owe me at least this direct and simple statement.

If you do not wish to either answer the original questions or take my more simplified offer above, then I am sorry to say that I don't hold out a lot of hope for future discussions with you. And if this is your choice, I feel obligated to be "on altert" to shut down any thread I see you moving in the same direction as the Sneak and Monster Tank threads. There was something like 30 pages between the two topics (three threads), and that is simply way too much over way too little. I'd rather close it up and start a smaller thread to discuss how the game works so it won't get bogged down with all the junk that went into the other three threads.

Steve

[aka_tom_w]

Ok

Just hold on Steve

Redwolf posted this this morning...

"Redwolf

Member

Member # 3665

posted December 03, 2002 10:56 AM

May I kindly suggest we wait until I answer Steve's detailed question when I go home?

This spinning out of control serves no purpose. "

So he is interested in replying and we can read his thoughts this evening

-tom w

[ December 03, 2002, 02:53 PM: Message edited by: aka_tom_w ]

[Tarquelne]

I had such questions about your test but you didn't track the data I felt was necessary to answer my questions...

Ah, OK. That's why you should go do your own tests.

[ December 03, 2002, 03:38 PM: Message edited by: Tarqulene ]

[Redwolf]

I think I should make an easier question out of the scoot-and-shoot versus seek-hull down issue:

If the TacAI overrides the command anyway before reaching the waypoint and retreats, what difference does it make which command it originally was? I mean very specific in CMBB terms, not in real-world terms. What Steve said above is vague and mixes up situations where the command actually gets executed to the end and where it is not.

If there is a different probablity of the TacAI overriding the player command while executing the two different commands, then I would like to know why that is.

[Licensed Fool]

Originally posted by Battlefront.com:

Gentlemen...

(snip) The manual is purposefully designed to not get into such details. For better or worse, we want the players to learn for themselves what works and what doesn't.

Steve

AHA ALL IS REVEALED! CMBB is one of those puzzle games! Never was much good at them myself,I've been stuck in The Hobbit on the Spectrum for the past 20 years!Hope I can figger out CMBB a lot quicker than that!

[Tarquelne]

Originally posted by redwolf:

[QB]I think I should make an easier question out of the scoot-and-shoot versus seek-hull down issue:

Hmm... I had a reply, but the Internet Monsters seem to have eaten it.

If the TacAI overrides the command anyway before reaching the waypoint and retreats, what difference does it make which command it originally was?

In short, it was: What command is being overriden, and when? AFAIK, there are no commands being overriden with either S-a-S or SkHldn in the tests, and I've never noticed a S-a-S command overidden in a game.

What Steve said above is vague and mixes up situations where the command actually gets executed to the end and where it is not.

I, at least, am still confused about just what your question is. It's the term "override."

Err... actually, maybe it's just the fact that I keep trying to fit your question into the framework of the ISU retreat issue and tests. If you're just asking about the chance that a unit will override the various sorts of movement orders due to self perservation, then I certainly don't know. I've never noticed a S-a-S order overriden, but I've never given a unit a long forward "scoot" order either.

I havn't noticed much of a difference between various move orders and a unit's chance to bug out, but I've watched for it, and it's fairly rare in my experience. Besides, when a PzIII starts frantically reversing because it's seen a T-34 come around the corner, I'm generally too busy cursing to remember what move order I'd given the panzer.

Babble babble... anyway, assuming I'm right about what your question is, maybe Steve was confused in the same manner, and can now focus on it.

[THumpre]

Originally posted by Licensed Fool:

</font><blockquote>quote:</font><hr />Originally posted by Battlefront.com:

Gentlemen...

(snip) The manual is purposefully designed to not get into such details. For better or worse, we want the players to learn for themselves what works and what doesn't.

Steve

AHA ALL IS REVEALED! CMBB is one of those puzzle games! Never was much good at them myself,I've been stuck in The Hobbit on the Spectrum for the past 20 years!Hope I can figger out CMBB a lot quicker than that!</font>

[Guest PondScum]

Originally posted by redwolf:

If there is a different probablity of the TacAI overriding the player command while executing the two different commands, then I would like to know why that is.

TacAI appears more likely to override "move-to-hull-down" commands where that would leave the tank with its ass hanging out in the open.

TacAI appears less likely to override "scoot-and-shoot" commands where that would leave the tank in cover after the command is over.

As to WHY, I hypothesize that it's because it makes sense from a real-world standpoint. Imagine you're a tanker, and you're ordered to go hull-down on a ridge against enemy tanks. For all you know, you might get no orders for the rest of the battle, i.e. you HAVE to stay there, in view of the enemy. That would have a high pucker factor, and crews might be more likely to disobey the order and bug out.

Now, imagine you've been given a scoot-and-shoot command that leaves you back in cover. Again, you have no knowledge of future orders, BUT if you carry out this one, you're only going to be in view of the enemy for a short time. Much lower pucker factor, and crews might be less likely to disobey the order.

Does that answer your question of "why"?

[Licensed Fool]

YES STEVE! I fully agree with you on point 6 of your TacAI posting where you say-"If there is criticism to be made about the TacAI decisions,it isn't that they're wrong...but that they are too correct.It could be that the TacAI is too smart compared to a real life crew in the same situation." Now at last we're getting to the nitty-gritty of this long debate! No disrespect to our Russian tank crews,but historically their level of education was not too hot,and I can't therefore see them being able to assess an enemy tanks threat capability as quickly as they do in the game.In an earlier posting I presented my "credentials" that I think show I'm something of an authority on the Eastern Front 1941-45,and was called a liar by at least one immature subsequent poster.I hoped that my postings would give Battlefront and the more sensible fellow gamers some valuable and interesting feedback on how I see CMBB, but if I'm to be howled down every time,why should I bother?

[Marlow]

Originally posted by Licensed Fool:

and was called a liar by at least one immature subsequent poster.I hoped that my postings would give Battlefront and the more sensible fellow gamers some valuable and interesting feedback on how I see CMBB, but if I'm to be howled down every time,why should I bother?

Now that you have toned down your posting, maybe you can contribute (remains to be seen); however, you were treated in accord with your original several posts.

[Michael Dorosh]

Respect on the forum doesn't come from "credentials" it comes from what you post.

So post something we can respect.

[Berlichtingen]

Originally posted by Licensed Fool:

</font><blockquote>quote:</font><hr />Originally posted by Battlefront.com:

Gentlemen...

(snip) The manual is purposefully designed to not get into such details. For better or worse, we want the players to learn for themselves what works and what doesn't.

Steve

AHA ALL IS REVEALED! CMBB is one of those puzzle games! Never was much good at them myself,I've been stuck in The Hobbit on the Spectrum for the past 20 years!Hope I can figger out CMBB a lot quicker than that!</font>

[Redwolf]

Originally posted by Battlefront.com:

[...], but to make sure you understand how off base you were *and*[...]

I dont think that this kind of "tuning in the audience" serves any purpose. It only strengthens my position because it enables me to claim your questions are not even meant to clean anything up. And there's nobody here who needs any tuning towards one side anymore anyway.

[Redwolf]

1. Do you now admit that the ISU-122 is a vehicle designed not as a

Tank Destroyer but as an Assault Gun (i.e. infantry support, not AT

duty)? If you say "no", please cite historical refferences to refute

Zaloga and Valera's websites (the two sources used here).

Yes.

As far as I understand the CM system, there is no special coding for a vehicle "you are an assault gun and now behave like one", correct? Everything the TacAI comes up with is purely based on real data, like vehicle characteristics, range and terrain, correct?

2. Do you now admit that the ISU-122 has some serious shortcomings vs.

a PzIV at short range (600m or less)? Before answering this recheck

all the various test files, including your own, and add up how many

times the ISU-122 triumphs and how many times it is defeated. If you

say "no", please show us something, ANYTHING, that shows that the

ISU-122 should not fear a PzIV at this short range.

It has shortcomings. However the TacAI you programmed makes them worse, not better. It ruins its own chances to kill the opponent with the one shot it likely has by often wasting it by firing while retreating, or getting out of LOS without shooting at all, which obviously wasted the already invested "danger time" and opens the problem that *somebody* has to kill that Pz IV sooner or later.

Obviously the right thing for the ISU is to always wait until the first shot gets off while standing and then start doing whatever the TacAI wants. As for a model of crew cowardice, see below.

3. Do you now admit that there is not rampant running away problem of

"über tanks"?

I don't use the phrase Uebertanks for Soviets, but otherwise yes, it is not rampant running away. It is an annoying detail and I wasn't the first to notice at.

Before you answer this, make sure you fully understand

and grasp the various tests conducted using various different

vehicles, ranges, and conditions. For example, my tests on Page 1, and

plenty of others after. If you answer "no" to this, please provide

files which show various problems that are inexplicable according to

what you have read here.

We have plenty of files of the TacAI retreating before getting the first shot off. This is not desireable to have in a tactical wargame. It is a "soft factor", as as people noted it can very well model a "screwup" on part of the crew.

However, while it might desireable to have crew screwups in the game mechanics in theory, in practice CMBB only has them for a very limited set of units, so far we have only seen slow-firing assault guns.

If you model "crew screwup", then you would have to have the Panzer crew screw up occasionally as well. If you don't model "crew screwup", then don't ever let them do a moronic thing like shooting one second after beginning to retreat.

4. Do you now admit that there is no "bug" with the TacAI pulling back

without firing a shot? [reminder skipped]

You condived me that it shouldn't be labeld bug. It is an inappropriate model of soft factors. The model is inappropriate because this kind of soft factors happens for a too minited set of units.

5. Do you now admit that the TacAI is behaving in the best interest of

the player by moving a friendly vehicle out of harm's way *before* it

is *likely* to be killed? Again, if you answer "no" please so some

sort of statistical sampling from a provided file that supports your

claim that the TacAI is robbing the player of victory.

No. The TacAI should always wait for the first shot to come off and then start reversing, if at all. That is in the best interest of the player.

We are talking about 1-3 seconds more exposure here, that is the time that would be required to wait for the shot to get off. The ISU is driving around the enemy LOS for 10 seconds anyway, adding the 1-3 seconds that it needs to wait longer to get one shot off while standing is the right thing to do.

[General Panic]

Originally posted by redwolf:

It has shortcomings. However the TacAI you programmed makes them worse, not better. It ruins its own chances to kill the opponent with the one shot it likely has by often wasting it by firing while retreating, or getting out of LOS without shooting at all, which obviously wasted the already invested "danger time" and opens the problem that *somebody* has to kill that Pz IV sooner or later.

Obviously the right thing for the ISU is to always wait until the first shot gets off while standing and then start doing whatever the TacAI wants. As for a model of crew cowardice, see below.

The ISU-122 isn't likely to hit the PzKwIV with one shot, because it doesn't have the chance to bracket the target before it either retreats or gets clobbered by the tank with the faster ROF.

Thus, squeezing off a quick shot before retreating is a just a waste of ammo and the AI's behavior is correct.

[Vanir Ausf B]

Again: If you want a tank to fire a single shot while stationary and then retreat, use Shoot & Scoot. This is what it's there for. The tool you need to do what you are asking for is there. You just have to use it. I don't know why this is so hard to understand.