Jump to content

Broken tacAI


OlafP

Recommended Posts

4 hours ago, OlafP said:

Here is another video of a team suddenly changing to an entirely different action spot when there is a junction and a Face command involved:

This seems to be related to repositioning and junctions. The AI seemingly gets confused when there is a junction involved and sends all or part of the team to the wrong action spot.

Another reason why I don't think TacAI self-preservation is involved, is because TacAI self-preservation always issues either a Fast- or Slow-command, but the team in my example seem to be entirely orderless.

In this video it appears that the TacAI is performing the new "cornering" mechanic added in v4. That is, the team was given a face command outward along the building they are behind, which tells them to peak around the corner. The one soldier who moved to the other side of the building was doing this as well, he just had to run around a wall to get there. Personally, I do not see a problem with behavior in this second video.

On 2/18/2019 at 4:55 AM, OlafP said:

The tacAI seems to be very dysfunctional when it comes to infantry moving around corners of buildings, walls, and the like. In fact, entirely broken in some cases, I would say. I've found various bugs related to pathfinding in particular. Don't know if there is any point in posting this now, since I understand that there is a patch underway, but here is a typical example:

Btw. is this the proper place to post bugs?

The scenario in the first video you posted is a lot more complex. First off, do you have a save file of the replay of this turn to share?

I've watched the clip a few times and I can't really figure it out. At first, I thought the team had taken fire first, causing the team to displace. However, it appears that this is not the case. The team spots one enemy team, and without taking fire appear to displace, which is when they run into the open and get gunned down. If I were to guess, it appears that when your team spotted the enemy, it triggered the TacAI to displace. This is odd, as normally that behavior is only triggered after taking fire. It's possible that a combination of the 4.0 TacAI, along with the specific move command you used (Hunt) and the nature of the immediate terrain all conspired against you to produce the result you recorded. 

I understand that this is frustrating. No one likes losing units, especially when it appears that they were killed by a glitch/bug. My advice would be to first, send/post the save file of the replay (if you have it) here so that beta testers and those curious can load it up and see for themselves what happened. Second, right now it is best to wait for the upcoming engine patches to be released. I know for a fact that the patches will be addressing a number of TacAI behaviors in 4.0 and fixing them. It is very possible that the behavior you saw in the first video will be mitigated/eliminated from happening again after the patches are released. 

It sucks that we have to wait for the patches, but the good news is that they should be released any day now. In the meantime I would recommend replacing the 'Hunt' command with the 'Slow' command in situations like your first video. Added with a 'Hide' command, this would have caused your team to crawl up to that hedge and remain prone. Then, the following turn you could turn off the 'Hide' command to allow the team to sit up and look out over the yard to their front. Also, a liberal application of the 'Pause' command helps to mitigate the current 4.0 behavior.

Edited by IICptMillerII
Link to comment
Share on other sites

I´ve encountered many similar situations when infantry moves into confined spaces, incl. 8x8m buildings, hedges, walls and quite particularly cross point AS. It not just takes considerable longe for the AI to deploy properly in such AS´s, even for small teams, also a lot of weird things happen, like single pixeltroopers running away and then return when they find their desired place within the team can finally be taken. I found this to be particularly bothersome in a running game with Panzergrenadier squads having 2 lMG and the assistant MG gunners totally confused about finding their place beside the lMG gunners. While one can be more sloppy with infantry squads just having one lMG, I now settled with always split squads (with 2 lMGs) for maneuvering and deployment in target AS´s.

Off course things are worse for quick and fast movement as the only purpose of these modes is to get every pixeltrooper as fast as possible to the target AS "simultaneously". Sort of all move there first and sort out later. Normal, hunt and slow provide better results as the TacAI appears to consider final deployment while still on the move and final deployment then works somewhat smoother. While this appears quite logical actually, I´d still prefer some real SOP for squad/team deployment and not the oftentimes total confusion at the targeted deployment AS, no matter if unter fire or totally unmolested.

Link to comment
Share on other sites

3 hours ago, Xorg_Xalargsky said:

Just tested a simple scenario (JTAC team vs. transport taxi) and it appears that the Hunt command does not cancel a Hide "at waypoint" command if a unit is spotted. You learn something every day...

I stand corrected. This is good to know. I suspect I had hide or cover arc commands on subsequent hunt order segments which were cancelled when the soldiers were engaged while executing an earlier segment.

I tested this in the released version in a cooked up scenario and found that once they cancelled their movement (because they were hunting) they reacted as if they needed to find cover. This leads them to automatically plot a withdrawal order. As we all know this process is not perfect with the Tac AI - in my test case they were on a street and the sought cover in a near by house - perfect. In the OPs case they made a poor choice. Statistically, if we were to recalculate the turn each of us could have seen different results but I suspect given the terrain I would mostly see good choices and the OP would see more poor ones. Such is the imperfect world we play in.

The good news is that the patch coming for the v4 engine reduces the tendency of the tac AI to decide to withdraw from a fire fight. I'm not saying it's suddenly going to be perfect but you will see less instances of the Tac AI deciding to create its own orders. With less auto generation of withdrawal orders there will be less poor choices made by the Tac AI. We should all be happier.

I am forbidden from confirming that the patched version of FB behaves better - so don't ask. :)

Link to comment
Share on other sites

Some of these examples are at the extreme margins of AI capability. You place the units in a no-win situation in a maze-like environment where movement choices are restricted then are surprised when there's a negative result. Its like claiming a car is defective if it doesn't stop you from deliberately driving off the road into a ditch.

That last film clip, players had demanded that units have the capability to 'peek around corners' when placed behind a building. That  clip is an example of that. Most times there's not an intervening wall to scamper around first, though.

Link to comment
Share on other sites

17 hours ago, Bulletpoint said:

The manual is wrong here. The unit only stops when actively spotting an enemy.

(or when taking enough fire to fill the suppression bar, which generally cancels any move order)

Oh, they stop and hunker down when they take incoming fire. I think you might be confused by the fact that squads sometimes do not realize that they are taking fire in CM, even though sound and graphical effects indicate to you as a viewer that they are.

17 hours ago, 37mm said:

You are confusing 'hunt' with the (in my opinion) superior "move to contact" order of CMx1.

No, I'm not confusing anything. Look, test this out, it's very simple. CM has had this feature since BtB. It behaves exactly as it did back then.

16 hours ago, IICptMillerII said:

In this video it appears that the TacAI is performing the new "cornering" mechanic added in v4. That is, the team was given a face command outward along the building they are behind, which tells them to peak around the corner. The one soldier who moved to the other side of the building was doing this as well, he just had to run around a wall to get there. Personally, I do not see a problem with behavior in this second video.

Yes, the cornering mechanic, but they're supposed to position at the corner indicated by the face command, not the corner which the face command points away from. From the manual: "If you only wish to man one corner and not the other, give the team or squad a facing command angled slightly towards the corner you want, and away from the one you don't.". That's not really the issue I had in mind though. The problem is that they are repositioning to an action spot on the other side of the wall. It didn't have any bad consequences in this example, but in the first one it did. I do suspect that there might be some overlap of issues here.

Quote

The scenario in the first video you posted is a lot more complex. First off, do you have a save file of the replay of this turn to share?

Yes: https://ufile.io/tx7n7

Quote

I've watched the clip a few times and I can't really figure it out. At first, I thought the team had taken fire first, causing the team to displace. However, it appears that this is not the case. The team spots one enemy team, and without taking fire appear to displace, which is when they run into the open and get gunned down. If I were to guess, it appears that when your team spotted the enemy, it triggered the TacAI to displace. This is odd, as normally that behavior is only triggered after taking fire. It's possible that a combination of the 4.0 TacAI, along with the specific move command you used (Hunt) and the nature of the immediate terrain all conspired against you to produce the result you recorded. 

Well, like I said, the Hunt command normally does not behave like that and is not supposed to behave like that. They're supposed to hide, not open fire. Hunker down, not reposition. I don't think they're displaced in the sense of self-preservation. They are reevaluating cover based on either the face command being reached at the end of the command chain or new information popping up. Or both. That causes the unit to erroneously think that it needs to move to the other side of the wall, just as in the second example I gave.

Quote

I understand that this is frustrating. No one likes losing units, especially when it appears that they were killed by a glitch/bug. My advice would be to first, send/post the save file of the replay (if you have it) here so that beta testers and those curious can load it up and see for themselves what happened. Second, right now it is best to wait for the upcoming engine patches to be released. I know for a fact that the patches will be addressing a number of TacAI behaviors in 4.0 and fixing them. It is very possible that the behavior you saw in the first video will be mitigated/eliminated from happening again after the patches are released. 

No worries, thanks for acknowledging the issue. I will hold back further reports until I have tested the new patch.

Quote

In the meantime I would recommend replacing the 'Hunt' command with the 'Slow' command in situations like your first video. Added with a 'Hide' command, this would have caused your team to crawl up to that hedge and remain prone. Then, the following turn you could turn off the 'Hide' command to allow the team to sit up and look out over the yard to their front. Also, a liberal application of the 'Pause' command helps to mitigate the current 4.0 behavior.

Well, according to the manual Crawling soldiers tend to pause and return fire at nearby/ exposed enemy troops often, then resume moving. That's not really what I wanted here, especially since there are tanks in the background and infantry often open fire on tanks. I've found that a crawling soldier is often a soon-to-be dead soldier in this game. I rarely use it. I guess what I could have done was to issue a Target Arc command, but it seems that if they ignore the Hide command, they will ignore the Target Arc command as well.

10 hours ago, MikeyD said:

You place the units in a no-win situation in a maze-like environment where movement choices are restricted then are surprised when there's a negative result. Its like claiming a car is defective if it doesn't stop you from deliberately driving off the road into a ditch.

You can't be serious?

Link to comment
Share on other sites

30 minutes ago, OlafP said:
19 hours ago, Bulletpoint said:

The manual is wrong here. The unit only stops when actively spotting an enemy.

(or when taking enough fire to fill the suppression bar, which generally cancels any move order)

Oh, they stop and hunker down when they take incoming fire. I think you might be confused by the fact that squads sometimes do not realize that they are taking fire in CM, even though sound and graphical effects indicate to you as a viewer that they are.

No, I'm quite sure they don't stop because of incoming fire as such.

They stop if 1) they spot an enemy, 2) they get suppressed to a certain (quite high) level. Just as any other movement order gets stopped if the moving team gets suppressed enough.

I've seen scout teams take repeated MG42 bursts and still hunt on until one of them got killed or their suppression meter filled up, whatever came first.

Edited by Bulletpoint
Link to comment
Share on other sites

19 minutes ago, OlafP said:

The problem is that they are repositioning to an action spot on the other side of the wall. 

Straight walls always run through the centre of the action spots, dividing them in two, so I believe that pixeltrooper intended to stay in the same square, but had to run around the wall to reposition on the other side...

19 minutes ago, OlafP said:

Yes, the cornering mechanic, but they're supposed to position at the corner indicated by the face command, not the corner which the face command points away from. From the manual: "If you only wish to man one corner and not the other, give the team or squad a facing command angled slightly towards the corner you want, and away from the one you don't.".

I think you're supposed to face them more away from the building. You're clicking basically on the building itself - try to use the facing option and click on the small shed to the right instead. Does that help?

That being said, I also have many problems making the corner-peeking mechanic work consistently, depending on what kind/size/angle of building it is.

Edited by Bulletpoint
Link to comment
Share on other sites

1 minute ago, Bulletpoint said:

I've seen scout teams take repeated MG42 bursts and still hunt on until one of them got killed or their suppression metre filled up, whatever came first.

Yeah, I've seen that as well. But I think it's just the game going a bit too far in modelling the difference between actual incoming fire and perceived incoming fire. Usually spotting occurs before the unit perceives incoming fire, which is a bit strange. In cases where spotting is hard but incoming fire is easy to perceive (such as a sniper far away), they do stop when taking sufficient incoming fire.

The saved game I provided can actually be used to test this easily. On the right side, I have a platoon hunkering down in the open because they took several casualties from an unspotted marksman.

Link to comment
Share on other sites

7 minutes ago, OlafP said:

But I think it's just the game going a bit too far in modelling the difference between actual incoming fire and perceived incoming fire. Usually spotting occurs before the unit perceives incoming fire, which is a bit strange. In cases where spotting is hard but incoming fire is easy to perceive (such as a sniper far away), they do stop when taking sufficient incoming fire.

The saved game I provided can actually be used to test this easily. On the right side, I have a platoon hunkering down in the open because they took several casualties from an unspotted marksman.

The team does notice taking incoming fire - this is shown by the suppression meter rising. When the meter becomes full, any movement order is cancelled and the team drops down. This happens both on hunt and on the other movement modes, so it's a different discussion.

And if the team also takes a casualty, the suppression meter very quickly goes to max. I believe this is what you're seeing in the case of the unspotted marksman.

Edited by Bulletpoint
Link to comment
Share on other sites

25 minutes ago, Bulletpoint said:

I think you're supposed to face them more away from the building. You're clicking basically on the building itself - try to use the facing option and click on the small shed to the right instead. Does that help?

That being said, I also have many problems making the corner-peeking mechanic work consistently, depending on what kind/size/angle of building it is.

No, that just made them stay away from both corners. It's very finicky. Sometimes they corner post without any input, sometimes you have to spend several turns to find the right angle, and sometimes they don't seem to be able to do it at all. 

Edited by OlafP
Link to comment
Share on other sites

15 minutes ago, Bulletpoint said:

Straight walls always run through the centre of the action spots, dividing them in two, so I believe that pixeltrooper intended to stay in the same square, but had to run around the wall to reposition on the other side...

No, they don't. In this case the centre of the action spot was the corner of the building.

Link to comment
Share on other sites

4 minutes ago, OlafP said:
20 minutes ago, Bulletpoint said:

Straight walls always run through the centre of the action spots, dividing them in two, so I believe that pixeltrooper intended to stay in the same square, but had to run around the wall to reposition on the other side...

No, they don't. In this case the centre of the action spot was the corner of the building.

Ah, ok, it's a diagonal wall :) Hadn't thought of that. Yeah that's true. He does run to a different square then. It seems the corner peeking mechanic takes over and makes the troopers stick to the corners of the building in this case, instead of staying in their square. The game engine mistakenly believes that you want a guy posted at each corner, and tries to do that.

Link to comment
Share on other sites

17 minutes ago, Bulletpoint said:

The team does notice taking incoming fire - this is shown by the suppression meter rising. When the meter becomes full, any movement order is cancelled and the team drops down. This happens both on hunt and on the other movement modes, so it's a different discussion.

And if the team also takes a casualty, the suppression meter very quickly goes to max. I believe this is what you're seeing in the case of the unspotted marksman.

Obviously, when the unit notices incoming fire, the suppression bar is going to change. So saying that the suppression bar is changing and that the unit notices incoming fire is the same thing. What seems to be your point is that the suppression bar needs to be at MAX in order for units with the Hunt command to stop moving, and therefore that there is no difference between the Hunt command and other commands in this regard. This is wrong. Here is an example I just recorded from my savegame:

The interface got cut off for some reason, but you can see that the suppression bar is not at max. In fact, it is only slightly elevated, and there are no spotted units, only sound contacts.

Link to comment
Share on other sites

6 minutes ago, OlafP said:

Obviously, when the unit notices incoming fire, the suppression bar is going to change. So saying that the suppression bar is changing and that the unit notices incoming fire is the same thing. What seems to be your point is that the suppression bar needs to be at MAX in order for units with the Hunt command to stop moving, and therefore that there is no difference between the Hunt command and other commands in this regard. This is wrong. Here is an example I just recorded from my savegame:

The interface got cut off for some reason, but you can see that the suppression bar is not at max. In fact, it is only slightly elevated, and there are no spotted units, only sound contacts.

Your video doesn't show the suppression meter, but it does show your troops are green, nervous and under bad leadership. That means they will drop down much faster under suppression. I still believe it's the suppression caused by the incoming fire that makes them react - not because the way the hunt command works.

If we assume that the hunt command makes troops stop and drop on receiving fire, then well led, highly motivated veteran infantry should do the same.

By the way, they are not sound contacts - just contacts :)

Edited by Bulletpoint
Link to comment
Share on other sites

1 hour ago, OlafP said:

No, I'm not confusing anything. Look, test this out, it's very simple. CM has had this feature since BtB. It behaves exactly as it did back then. 

I just tested this out on CMA & you are indeed correct. I have no idea if the feature was put back in at some point or has always been present (here are players complaining about its absence in 2011, for instance) but it's good to know, thank you.

However my overall point (which is a tactical one) remains... "hunt" is a poor command choice for close quarters urban combat.

Have your men maximize their stealth (by crawling, with a target arc) or their speed (use quick, again with a target arc)... having them slowly walk around urban death traps is just asking for trouble.

Don't believe me, believe the US Army...

 

Edited by 37mm
Link to comment
Share on other sites

3 minutes ago, Bulletpoint said:

Your video doesn't show the suppression meter, but it does show your troops are green, nervous and under bad leadership. That means they will drop down much faster under suppression. I still believe it's the suppression caused by the incoming fire that makes them react - not because the way the hunt command works.

If we assume that the hunt command makes troops stop and drop on receiving fire, then well led, highly motivated veteran infantry should do the same.

No, the suppression meter is a reflection of subjective suppression, in other words suppression relative to the current variables of the unit (experience, moral etc.). It is not a measure of objective, actual incoming fire. That's why you see the suppression meter going bananas when Green troops take fire, while it changes only slightly with Veterans in the same situation.

Link to comment
Share on other sites

11 minutes ago, OlafP said:
20 minutes ago, Bulletpoint said:

Your video doesn't show the suppression meter, but it does show your troops are green, nervous and under bad leadership. That means they will drop down much faster under suppression. I still believe it's the suppression caused by the incoming fire that makes them react - not because the way the hunt command works.

If we assume that the hunt command makes troops stop and drop on receiving fire, then well led, highly motivated veteran infantry should do the same.

No, the suppression meter is a reflection of subjective suppression, in other words suppression relative to the current variables of the unit (experience, moral etc.). It is not a measure of objective, actual incoming fire. That's why you see the suppression meter going bananas when Green troops take fire, while it changes only slightly with Veterans in the same situation.

I always believed the suppression meter showed an indication of the objective amount of incoming fire a team is taking. And I believe veterancy just makes the meter empty quicker, as well as possibly making return fire more accurate despite suppression. But it's possible I could be wrong here.

Did you try to set up the same situation as in your video, but to use a good condition veteran squad?

Edited by Bulletpoint
Link to comment
Share on other sites

1 minute ago, Bulletpoint said:

I always believed the suppression meter showed an indication of the objective amount of incoming fire a team is taking. And I believe veterancy just makes the meter empty quicker, as well as possibly making return fire more accurate despite suppression. But it's possible I could be wrong here.

Well, it could be tested by changing that scenario to only have Elite troops. I think you'll find the exact same behaviour, but it might be more difficult to find a case, since Elite troops are much more likely to spot the unit first.

Link to comment
Share on other sites

2 hours ago, Bulletpoint said:

I always believed the suppression meter showed an indication of the objective amount of incoming fire a team is taking. And I believe veterancy just makes the meter empty quicker, as well as possibly making return fire more accurate despite suppression. But it's possible I could be wrong here.

Did you try to set up the same situation as in your video, but to use a good condition veteran squad?

I´d support your assumption. To get things straight for any such test one needs standardized parameters which IMO should be to test with regulars, leader 0 and normal morale, as well as C2 connection to superior HQ (having equal parameters). Based on that one can deviate to check out other parameters to achieve anything conclusive.

Link to comment
Share on other sites

5 hours ago, OlafP said:

Yes, here it is: https://ufile.io/tx7n7

Thanks for this. This helps a lot. I ran your saved game using my FB 2.0 installation and saw the same behavior you did, just slightly different end results but pretty much the team running around panicked.

Then I loaded it in the patched beta version of FB and the behavior was markedly improved. In fact, I'd say, exactly what you would want. The team stopped at that row of bushes (not really a hedgerow), either kneeling or prone, took a couple shots at an enemy soldier in the stone courtyard in front of them, who then appeared to run off, then took a couple rifle shots at a Sherman that suddenly appeared far off through the trees - presumably to make it button up. Then they hunkered down. All four guys in the same place, no running in circles, screaming and shouting. Proper behavior.

That should cut through the remainder of the above discussion.

Link to comment
Share on other sites

4 hours ago, OlafP said:

No, the suppression meter is a reflection of subjective suppression, in other words suppression relative to the current variables of the unit (experience, moral etc.). It is not a measure of objective, actual incoming fire. That's why you see the suppression meter going bananas when Green troops take fire, while it changes only slightly with Veterans in the same situation.

 

3 hours ago, Bulletpoint said:

I always believed the suppression meter showed an indication of the objective amount of incoming fire a team is taking. And I believe veterancy just makes the meter empty quicker, as well as possibly making return fire more accurate despite suppression.  <Snip> 

First, thank you both for having this interesting, informative conversation and keeping it civil while disagreeing.  +1 up vote to you both.    

I also thought the duration a unit remained suppressed depended on the Experience & Leadership of the unit.  However, the dynamics and effects of soft factors (unit attributes) in combination with other factors and the resulting overall effect on unit behavior can be a little confusing.  @Josey Wales is IMO the resident expert on soft factors so I tagged him to get his opinion on the soft factor / suppression part of this conversation.   

Link to comment
Share on other sites

4 hours ago, Ultradave said:

Thanks for this. This helps a lot. I ran your saved game using my FB 2.0 installation and saw the same behavior you did, just slightly different end results but pretty much the team running around panicked.

Then I loaded it in the patched beta version of FB and the behavior was markedly improved. In fact, I'd say, exactly what you would want. The team stopped at that row of bushes (not really a hedgerow), either kneeling or prone, took a couple shots at an enemy soldier in the stone courtyard in front of them, who then appeared to run off, then took a couple rifle shots at a Sherman that suddenly appeared far off through the trees - presumably to make it button up. Then they hunkered down. All four guys in the same place, no running in circles, screaming and shouting. Proper behavior.

That should cut through the remainder of the above discussion.

This is really good to hear. Add another reason why everyone should be excited for the engine patches to be released. Speaking of that, get back to testing so the patches can come out already! 😁

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Unfortunately, your content contains terms that we do not allow. Please edit your content to remove the highlighted words below.
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...