Odd AI Behavior
 

Since it seems to be coming up, I figured I would start a thread on AI actions that seem out of the ordinary or illogical:

1. Running into their own minefields, before visible contact established between opposing sides.

2. Dashing West out of prepared positions during a defend, before victory point hexes have been captured. The time this happened, my units had been spotted and engaged, but the AI units were actually heading away from those units. I was in the enemy rear at the time and do have a nice screen print of this.

3. AI dropping artillery 9 hexes in front of my LD before I have even moved on turn 1. Impossible for me to be there yet.

4. Auto pivot to face frontal armor toward firing unit. I understand what the intent was here, but this gets more tanks killed than anything. Normally, the firing unit is off at an angle to the target and this angle creates an artificial armor slope which the code takes into account when determine shot effectiveness. That artificial slope causes shots to deflect the vast proportion of the time. I know often I won't get a kill on a target at an angle when firing a gun marginally capable of killing the target. I just keep firing until the pivot and then the kill comes. When I fire, I always angle my armor off full frontal so that I get benefit from that when the target fires back, if it doesn't die. I have to re-angle my tank after each responding shot. It renders British 2pdrs very ineffective except at very close range.

Re: Odd AI Behavior
 

1, 2 and 3..... when you get a save game everyone can look at, do post it

4. That's your experience. Other will say without the pivot their tank would have taken a side shot. This is not going to change.

Don

Re: Odd AI Behavior
 

I have attached a .ZIP file with a save in it. Background information is listed below to try to help understand what was occurring at the time. The game was a save for an ongoing campaign.

With respect to item #2, at locations (70,70), (70,71), (70,72) and (70,73), the AI had four British Valentine III tanks positioned. Those bolted out heading West. One can still be seen heading West at location (59,71). At locations (70,49), (70,51) and (72,49), the AI had three British Grant I tanks positioned. They started out briefly Southwest and then also turned West. One of those Grant I tanks can be see heading West at location (64,54). Numerous infantry positions also appear to have been abandoned by units heading west.

This all started after my units engaged two 40mm Bofors in the enemy rear area. The Valentines started moving when I attacked the first AA gun. Then the Grants started to move.

Additionally, it might be possible to validate item #1. As mentioned, the Valentine tanks headed West. The one tank still visible to me might have passed through a mine field to get to it's present location at (59,71). It is West of a mine hex at (61,72). I can't tell if the line of mines continued North and was consequently in the path of the Valentines, but you might be able to.

It probably varies with tactics employed. I try to be respectful of the AI guns and engage at a distance, which makes most shots against some frontal aspect of the target. With stronger tanks mixing with the enemy at close range, flank shots are more likely. Since it depends on tactics and equipment used, I can understand not changing it.

Re: Odd AI Behavior
 

Just to see what would happen, I pretty much holed up my force where they were in the enemy rear and zipped through the remaining turns of the battle.

In the post battle review, I found the most of the AI tanks milling around in what would have been my rear area if I had any units back there. Some were literally three hexes from the back edge of my side. Most of the AI infantry moved out as well. Only ATGs(4), AA(1), mortars(1) and bunkers(3) were around the objectives. Everything that moved abandoned the objectives. I had my whole battalion close by but could have taken the objectives with a platoon.

As a follow-up on item #1, the that mine field hex I mentioned was the Northern most in that line, so the Valentines don't appear to have passed through them.

Re: Odd AI Behavior
 

Regarding auto pivot
How about teaching the AI not to present 12 o'clock but 11 or 01?
Just like the drivers of a Tiger were taught, show the corner, not the front (for Tigers it was 10:30 and 01:30).
This adds 15% to the front armor and doubles the side armor. It protects the front and all hits to the side should glance off.

Maybe a check could be implemented if the front armor is a lot stronger than the side armor (e.g. Panther) and then decided if it's better to present the front or the corner.

Re: Odd AI Behavior
 

Would any tank crew under fire have time to adjust to optimum angle?I think panic and self preservation would be the first reaction(knowing the 2nd or 3rd could be a kill shot) and they would quick move to frontal position, and pray!:hide:

Re: Odd AI Behavior
 

From a programming perspective, it opens a can of worms to try to code specifically for one type of tank. As Don mentioned, what I was seeing was based on my experience. I'm playing a German campaign and am currently in North Africa. I did some test firing against of a British 2pdr against a Pzkw IIIe. At a 30 degree angle, 61% of hits deflected, while at a 0 degree angle 26% of hits deflected. Hits that didn't deflect penetrated doing damage or killing the target. So what I was seeing is true.

But lets look at the tank I tested. It has hull armor of 3 all around and the turret front is 5 and sides and rear 3. With a tank like that, it benefits from the artificial slope my 30 degree angle created. That's because the hull armor on the front and side are the same. Odds are, the turret will be turned toward the enemy so that's not a factor.

Now, lets look at a tank like the Panther A. It has front hull armor of 14 with sides and rear of 5. Without doing the math, I have to believe that my 30 degree angle is not going to generate a ballistic thickness for the side that occurs with the frontal armor and it's normal slope at a 0 degree angle.

With that, there is one instance where the angle helps and one where it doesn't. More than likely, considering tank design focused on making frontal armor significantly thicker than side and rear armor as the war went on, the auto pivot helps more than it hurts. It's a matter of knowing your equipment and doing what you can with it. As long as I have tanks with armor on the front, sides and rear that are close, I will impart the angle before I take my shots. When the frontal armor starts being significantly thicker than my side and rear armor, then I will face my enemy straight on before I fire. And all of this only matters if you are trading shots with one enemy tank. When there are dozens returning shots at you, auto pivot or manually setting the angle won't matter. Someone is going to get a sweet shot at you.

Re: Odd AI Behavior
 

At least that's what written in the manual of the Tiger and the Panther (Tigerfibel, page 84 and following, Pantherfibel, page 5 and following, readily available on the net).

The manual of the Tiger teaches the driver to present the corners (10:30 or 01:30). This makes the hull invulnerable even against 15,2cm (says the manual).

The manual of the Panther teaches the driver to present the front (12). Actually any angle between 11 and 01 makes the Panther invulnerable (says the manual).

If the enemy is located more than 30° (Panther) or 45° (Tiger) off, then it's faster to present a corner. Most German tanks were able to turn on the spot (all the late ones I think).

So we have the Tiger with an almost even armor distribution and the Panther as example for a very uneven distribution. The Tiger hugely benefits, the Panther is not more vulnerable from showing the corner (hits on the Panther side should ricochet and the line of sight thickness of the armor is doubled).

That's why I proposed 11 and 01 (Panther) and not 10:30 and 01:30 (Tiger) as a simple solution, it should be better for all kinds of armor distribution.

The actual armor distribution is in the database, so it should be no big problem to calculate the "perfect" angle for each tank. That angle should be taught to the (real life) crew during training, in my opinion.

Re: Odd AI Behavior
 

A 30° angle (towards enemy) adds 2 to the front armor of 14 and doubles the side armor from 5 to 10. Side hits come in at an angle of 60° and should ricochet very often, as you experienced (just like me).

I would (and I do in the game) angle my tank and pray that the auto pivot does not kick in, especially when the range is such that the enemy round is just barely able to penetrate the armor. On the other hand I'm praying for my opponent to present the full front, which is the stupidest thing to do for that crew...

The probability of a side hit should be very small compared to a frontal hit as there is only a little bit of the side visible, so I'm taking the chance (and counting on ricochets). I hope the game takes this into account when determining if it's a front hit or a side hit.

Re: Odd AI Behavior
 

Do you have a save of this game made at least 3 turns earlier ?. Even one made near the start would be helpful

Don


Ignore this. I found the info I was looking for

Re: Odd AI Behavior
 

The shot coming from marginal penetration range is what makes the angling work for the tanks I'm using. If the ranges is too close or the enemy's gun is so big it doesn't matter, the angle isn't going to help and in reality would hurt. In reality, an angled tank presents a larger visible target to be hit and should be hit more often. A 0 degree angle presents the smallest possible target to the firing unit.

Hit probability on the side is going to depend on the angle you set. With the Pzkw IIIe tanks I tested, assuming 0 degree armor sloping (not sure what it really is), a 45 degree angle to target should be best as it will create a ballistic armor balance between the side and the front since both are 3. In that case, the side should be hit more often than the front simply because there is more side to a tank than front.

As far where hits actually landed, I didn't get into those kind of details when I ran my test. I set the range and angle of the target and fired. Tracked were if the shots deflected or penetrated. On the angled tank, I didn't check to see where the deflected shots actually hit (front or side, turret or hull). Also, I didn't check to see how many shots missed as that can be influenced by the gun firing and the crew experience. It would be possible to check to see if an angled tank gets hit more often, but that was more than I was looking to see. In the case of the Pzkw IIIe, even if shots hit twice as often, it still would have had a better survival rate than taking shots straight on. That's just this one tank, however. It will vary with different tanks as will the optimum angle. That's where it becomes a programming nightmare to figure out and I wouldn't bother with it.

One thing that did seem to occur and this may be more perception than reality was, during a firing sequence, kills seemed to be more frequent after a damage penetrating hit occurred. In reality, if a tank is penetrated, the overall armor integrity would be compromised. Armor fractures, weakening it. If it's welded as opposed to cast, seams break. A kill would be more likely on subsequent hits. I didn't track it, however. It just seemed that way and I don't know if that aspect is incorporated into the game. If it's not, I'm not asking for it. It was just my perception.

Re: Odd AI Behavior
 

Sorry, no :(

I didn't even realise I had the one I sent. It was just something I stumbled across.

I do have another assault getting ready to start so what I'll do is save the game at the start and turn by turn, just in case something useful comes up. If the AI acts as it has been, it should move units out before I get to the objectives. Also, I used my full complement of support points this time, so the I should expect to see the minefield belt be complete from the top to bottom of the map. That way, if the AI units do move out, they should cross the minefield to get to me.

Re: Odd AI Behavior
 

No worries. I don't think I'll need them now.

Don

Re: Odd AI Behavior
 

That's where training comes into play. Under pressure, trained troops will generally react instinctively to their training, suppressing self presevation. Generally militia units tended to run when trained units would hold fast when faced with the same risks.

As for angling Tigers when engaging the enemy, I had heard that before. That's why I started looking at it in the game since I knew it did take into account engagement angles when calculating armor penetration.

Re: Odd AI Behavior
 

What you are seeing is the normal reaction turns build into the game for all battle types kicking in. It's simple enough to see, just set up a normal battle. Set player 1 to all computer and set player two ( the AI ) to all computer except deploy and set that to human. Press continue then autodeploy and click on the HQ button

Click on each formation and take a look at the "reaction turn" That's the turn they will react. In most cases how they react will be OK within the game setting up counterattacks that keep people on their toes but occasionally it won't and your save showed the perfect "it won't"

Andy and I will discuss how this is handled.

Don

Re: Odd AI Behavior
 

One if these threads I did mention where the reaction threw a monkey wrench into my plans, so it wouldn't be good to eliminate it.

Re: Odd AI Behavior
 

1) No idea - never seen that one, since if the defender it will generally be stationary. If 2 below - then it might decide the quickest path is via the mines, if there are no nearby gaps, of course.

2) The AI will sometimes do a mini counter attack even before you have taken any objectives. Usually with armour, but sometimes not. This should be quite rare - in 2 recent WW2 long campaigns the defenders have come at me maybe twice with a tank platoon before I was near the objectives. (Being near the objectives if spotted by the AI can sometimes trigger counter attacks too, though generally you need to flip one or more).

3) If you mean dropping arty ahead of your deployment zone, it will sometimes aim at the front of the deployment line and then there is the usual spread of fires which may take the aim point off and away into the neutral zone. But fires landing ahead of your deployment zone can drop on the next turn as well thus splatting your advancing troops from turn 1 if you decided to risk running into the shelled zone. And if heavy arty, then it can crater approach roads in front of your advance which also may slow your approach down.

4) Before that was introduced stugs, marders etc were simply dead meat if engaged from a flank as only turrets would turn to engage. It also solved problems with Char Bs and Lee/Grants which would almost never use the hull guns, simply engaging with the sub-turret unless a human player manually hull-turned them - which the AI rarely did.

Cheers
Andy

Re: Odd AI Behavior
 

:)Seems like an reconnaissance probe. Sometimes it works and sometimes it doesn't. In the case where they hit the minefield, the units were East of it before they started and there were no gaps.

Except in my case, the AI didn't run to the sound of the guns. At the end of the battle, it had possession of my deployment zone, but I didn't have anything back there. My entire force was in AI deployment zone.

Desert battle in this case, without roads. I've seen the AI hammer roads before so I don't deploy on them. The fire came in too far ahead of my deployment zone to be a risk to anything and I definitely wasn't going to go through the impact zone with artillery still coming in. If I'm receiving incoming mail, I get out of the impact zone and if I have to go where it is coming in, I wait. I suspect this is true for most people. The only exception to this would probably be the last turn or two of the battle and the impact zone is on victory hexes not controlled by the player.

For fixed or sponson mounted guns, it makes sense. They are going to turn to fire. For other units, auto pivot's usefulness varies from tank to tank. Some it helps, some it doesn't.

Coding for each situation is not something that I'd want to tackle. You would have to run a numeric calculation for each unit to determine the optimum angle each time it ends it's turn or receives fire. Can't do it by unit name because names can be changed. Since armor values also can change, it would have to be a run-time calculation unless you calculated it in advance and maintained the values in a table. Then you have to consider multiple threats to the tank. With a 50mm L42 to the right 30 degrees off at 16 hexes and a 50mm L60 to the left at 20 degrees, 22 hexes off, which is the bigger threat? Has one of the guns been suppressed and really isn't a threat? Could it possibly recover and become a threat before the next turn starts? You see where all of that goes. Coding it would be a project in and of itself.

Auto pivot at worst puts the player and AI on the same playing field. If the player chooses to angle a unit before firing or at the end of their turn to optimize armor, that gives them a leg up on the AI. No reason to complain there. When I first brought it up, I saw the advantages to manually setting tank angles. As Don pointed out, it was based on my experience. When I get to Tigers and Panthers rather than various versions of the mark IIIs, I'll worry less about tank angles. Things are fine the way they are :)

Re: Odd AI Behavior
 

Counter attackers should head off towards any flipped V-hexes eventually in order to retake them. Or, go for nearby located enemy. This results in AI units appearing up your rear, as has happened to me on woods maps with T34s appearing up the kilt of my assault group..

They will go deeper into your rear while those conditions are not met. Sometimes that will result in a force running over your artillery park.

However if the AI counter-attackers never saw anyone nearby (e.g. visibility is way low) it wont have something to attack, and if you only flipped over the V hexes right near the end of the battle then it may not have time to go back from its rear-area mission and attempt to retake them.

Cheers
Andy

Re: Odd AI Behavior
 

You pretty much described the situation. I had not flipped any victory hexes at the time, visibility was low (9) and the units that ended up in my deployment area did not have eyes on any of my units. My scouts had eyes on the objectives and the units that moved out allowing me observe what was occurring. When the AI units got to my deployment area, I'm sure they were pretty lonely. There was nothing back there but the enemy :)

Re: Odd AI Behavior
 

As Andy noted, the AI will sometimes do a mini counter attack even before you have taken any objectives. That was the intent of the reaction units but something may have gone slightly astray over time as now in a typical assault / defend situation the AI is allowing roughly half of their formations to react on a wide variety of turns. The only way I can know if this has crept up a bit is to run some tests with an older version of the game which I may do this weekend.

However, under normal circumstances these counter attacks are not seen as odd. In the save game you offered the whole situation was radically different than most games with the way you attacked and the visibility in the game. Moderate pointed games on larger maps will have minefields more dispersed than they would be in high pointed games and if a player tends to play those on medium to small maps the incidences of AI tanks running into their own mines would rise so , like many things in this game, what players see and what strikes them as "odd" depends a great deal on their preferences, points and map size

Don

Re: Odd AI Behavior
 

Another aspect is, being a game, I don't have to stay "honest" so to speak. In reality, if an attack is launched, defensive lines by the assaulting side aren't abandoned as I did in my attack. Beyond the 100x100 battlefield I was on, there would be supply and ammo dumps, hospitals, communications, headquarters, etc. that would have to still be protected. If those things had been back there at in my game, the AI would have had a field day. In the game, I packed up all my "stuff" and took it with me. Also, a narrow front attack isn't typical. Most real assaults are broad based. In the Battle of the Bulge, the Germans didn't punch a hole one mile wide and send everyone through it. It was an assault on an 80 mile front.

Oh well, enough on this. I think it's been pretty much beat to death.

Re: Odd AI Behavior
 

5. AI clearing it's own obstacles.

Just started an assault and the AI must have a rifle section next to it, because after the pre-turn 1 artillery bombardment, a message popped up saying "rifle section CLEARS 1 STRAND OF WIRE" and showed me the wire at the bottom edge of the map.

This has actually been an issue since the first SP version came out back in the stone ages, but I don't know if anyone has tried to do anything with it.

As a follow-up thought, since the AI tends to target locations with artillery when an obstacle is cleared, will it nail its own troops as a result?

Neither one hurts the player so it doesn't need to be fixed, but I was curious if any attempt has been made to correct this. I have my turn 1 save file and attached a screen print of the message.

Re: Odd AI Behavior
 

Nope - that one has been there since SP 1 days.

SP 3(?) had a manual flag for engineers to not clear mines, but that is of little use for the AI.

In SP there are just mines or other field defences - there is no ownership marker. So the AI engineers/ordinary troops cannot be coded not to clear "own" mines, since all field engineering items belong to the same pool. Scenario designers can do tricks to give both sides field defences so a "defender does not clear mines" global is not a good idea either, since the defender sometimes does want to clear mines (e.g. to open a lane for a counter attack) or both sides to deal with scatter mines in MBT.

The deployment code is supposed to keep troops clear of engineer obstacles, but has never been able to fully keep it so.

So basically there is no real "sane" way to stop the AI troops clearing mines they are deployed on or in front of, but still clear any they should clear (such as when reacting to a V-hex flip by moving through own minefields).

As to targeting own side - the targeting is done for the opposite side (if AI controlled) to the troops during the clearance phase of the owning player. So that should not happen.

Cheers
Andy

Re: Odd AI Behavior
 

Yes, we all know the AI will clear it's own wire ( and it's own mines, DT's etc etc and "YOUR" wire if you are defending if you put an engineer unit beside one)

As has been stated many times these ( wire, mines. DT's ) have no "friends". They affect and cause damage equally to both sides. One of the "jobs" of an engineer is to clear these obstacles and there is no way to tell them if they "own" the obstacles or not. If we removed the ability to clear mines , DT's wire etc from the defender to clear these obstacles you KNOW someone will howl who "needs" to be able to clear some of them so he can mount a counterattack and we took that ability away.

So yes, this minor little game quirk has been discussed many times and we are all aware of it

Don

Re: Odd AI Behavior
 

The only reason I mentioned it was I had played the other version by another company who's name I won't mention here and the mine clearing ability for each unit could be toggled on and off.

There are other problems with the version from that unnamed company which is why I play WinSPWW2 instead of it. Overall, WinSPWW2 is a much better product IMHO :)

Re: Odd AI Behavior
 

********Which, as Andy noted, is of no use to the AI which was the basis of your observation that the AI was clearing it's own wire.


********Yeah well, we think so to FWIW.:) That "version from that unnamed company" is SP3 based and a spin off of SPww2. There were "disagreements" as to the direction the games developement should take and that's why there are now two versions each with different things people like and don't like proving once again it's impossible to make everyone happy. ( especially "wargamers" )

We've been "fixing" the game for 10 years now and people still find things about it they don't like and I have no doubt we could "fix" it for ten more and we'd still be in the same boat we are now.


Don

Re: Odd AI Behavior
 

I have the benefit in coding in an environment where things are more black and white. If it costs more to leave the problem unfixed than to fix it, we fix it. Otherwise, we live with problem. It's not a matter of making people happy.