Concerned over rare pathfinding problems in SF2 demo

[HerrTom]

1 hour ago, Battlefront.com said:

FWIW, there was no explicit changes to the TacAI's pathing in Engine 4 that I'm aware of.  Perhaps tweaking here or there for a corner case, but nothing more than that. Sometimes fixing a corner case produces a new corner case.  Which is why corner cases in games are often left alone by game developers.  Trying to fix a 1 in 1000 event could introduce a 1 in 100 event.

Steve

Could this be an auxiliary effect to the fleeing implementation in engine 4?  It's possible that we never really saw this happen because fleeing (without losing command, perhaps?) is more common?

Anecdotally, I've found myself disappointed in the choices the TacAI makes when it does the "self preservation" pathing mid-turn (as opposed to broken/transparent icon running).  It occurs to me that I don't really pay close attention to what the broken units are doing...

[Battlefront.com]

7 minutes ago, HerrTom said:

Could this be an auxiliary effect to the fleeing implementation in engine 4?  It's possible that we never really saw this happen because fleeing (without losing command, perhaps?) is more common?

Anything is possible as the TacAI is extremely sensitive and can easily produce "unintended consequences".  It's one reason we're often reluctant to try and solve corner case problems.  Often times it creates more problems which then have to be corrected.  That's the downside of AI behavior modeling for something this complicated.  GO! is much easier

7 minutes ago, HerrTom said:

Anecdotally, I've found myself disappointed in the choices the TacAI makes when it does the "self preservation" pathing mid-turn (as opposed to broken/transparent icon running).  It occurs to me that I don't really pay close attention to what the broken units are doing...

Self preservation pathing is prone to making decisions the player doesn't agree with because the player has a broader view of the battlefield, intentions, history of activities, etc.  The unit doing the evasion is thinking only of itself based on the knowledge it has available to it.  Often a unit is like teenagers at an abandoned summer camp on Halloween night.  As a member of the audience you know they shouldn't go into that room because you know the killer is in there, but to the teens they're just looking for a place to shag

Steve

[sburke]

1 hour ago, Pericles said:

sburke - I didn't realize TacAI problem identification was so complicated. I thought that identifying TacAI problems in the most current engine version 4.0 was all that was relevant in improving the engine. The developers would take your ticket submission regarding an identified TacAI problem and make changes to the most current engine version 4.0. 

If you are correct, and it is actually relevant to know whether or not TacAI problems identified in the current engine version were also present in the previous engine versions, then my belief that your question is not relevant to the discussion is wrong. 

That being said, it would have been constructive if you would have given your opinion on the topic of the post, which is TacAI problems with the Shock Force 2 demo. You've had a chance to look at the videos and to experiment with the demo. What do you think? 

again you are jumping to a conclusion. 

1. So far all I have done is read through the thread.  I have a full time job that is in hyper mode at the moment and I have been traveling so I can't even load the game to check the saves, the map etc.  therefore I haven't an opinion yet.  I try not to form one without first looking through the data and making sure I understand the case and try to reproduce it.  Forming an opinion too soon creates bias, bias leads to misinterpreting data.  What I can say is I haven't seen anything in my playing CMSF2 pretty extensively for a number of months testing scenarios etc that would lead me to think there is a lot of odd TAC AI behavior.  I do see things occasionally and if I can reproduce it, I report it.  Based on that I find it plausible you may have found an issue.  What exactly that issue is, potential frequency and my ability to duplicate it are all things I haven't the means to answer yet.

2. TAC AI issues are very difficult items with a lot of possible influences for example there was an issue a while back (unresolved as far as I know) where individual pixeltruppen get separated from their team possibly because of terrain features. I know the first time I experienced it was a CMBN scenario I was playing PBEM back in 1.0 (scenario was the mace - I can't believe I actually remember that ) Steve's post above regarding corner cases though is probably far more relevant  We all want CM to be better  However it is software.  Software tends to occasionally produce crap and the conditions that got it to produce crap can be really hard to determine.   Hell I just wrapped up a several hour call with a vendor on an issue that we fixed but we have no idea what was the initial cause (and I doubt we will get an actual cause out of them).

[ncc1701e]

4 hours ago, IanL said:

Now that part sounds suspecious. Can you share the scenario - I would appreciate it.

PM sent.

[Pericles]

6 hours ago, Battlefront.com said:

Self preservation pathing is prone to making decisions the player doesn't agree with because the player has a broader view of the battlefield, intentions, history of activities, etc.  The unit doing the evasion is thinking only of itself based on the knowledge it has available to it.

In the particular case of the video below, I don't see how the unit is thinking only of itself based on the knowledge it has available to it when it runs back toward the enemy a second time instead of down the road or behind the wall. It simply doesn't make any sense given the in-game situation (e.g. the unit is aware of enemy contacts North). Would you agree, or am I missing something? 

Battlefront wrote: "However, we have been at a point now, for a while, where it's getting harder and harder to find reproducible, addressable corner cases.  That's good."

Agreed. And the discussions in this thread may have contributed to the identification of a reproducible corner case in the Wilcox scenario. 

If it's not sufficiently addressable, then that's not good, but only mildly not good. I realize that expecting a video game like this to be completely free of errors is not realistic. As a customer and proponent of the CM franchise I will continue to poke and prod with evidence, reason, and an open mind in an attempt to contribute to the improvement of the game experience. 

 

[c3k]

^^^

Great vid. The team should NOT have gone back around that corner into the beaten zone.

Of course, one big issue is the lack of "memory" for our pixeltruppen. However, I'd think a "?" or hard-contact would be something to leverage to force them not to run that way. Meaning, if that team "knows" there is an enemy there, the team should avoid that enemy (increase distance or keep LOS from occurring or get out of LOS by the quickest means).

...and then prioritize moving towards the friendly edge.

[MarkEzra]

A Great Line from SPR:  "Where's the Rally Point?" … "Anywhere BUT HERE!"

And so it goes in CMSF2.... as the title of this thread mentions: Rarely.

Sometimes all we can do is patch up our differences rather than the game, and hope for a New CM Tomorrow.

I VOTE for that

 

[A Canadian Cat]

1 hour ago, c3k said:

Great vid. The team should NOT have gone back around that corner into the beaten zone.

And when I tried to reproduce that situation they *never* did. They always went the smart way.

[Josey Wales]

Video 1 looks like the Indirect Fire bug that @HerrTomhas alluded to. The unit takes 1 casualty and suffers about 70% suppression and  drops to the morale state of 'Nervous'. The unit leader is the casualty, but the assistant takes over who also has +1 Leadership. Can a hand grenade cause the Indirect Fire bug?

Video 2  looks like a typical auto evade when troops in the 'Rattled' morale state become Pinned.

I cannot comment on video 3 as I don't have any information on the unit status.

When I see my troops in CM behave in a way different to what I expect, I can typically 'roleplay out' the situation so as to be less dissatisfied with the result. 

It is hard to make that case here in either of the first 2 vids.

In vid 1, the unit does not panic, nor does it suffer from being led by an incompetent. The unit is merely dropped to 'Nervous' and still has a capable leader fronting it up. The unit is also aware of 2 enemy contacts to the north. I would therefore find it difficult to roleplay out the situation that developed and would come to the conclusion that something was wrong with the TacAI.

In vid 2, the unit is Rattled and (probably) Pinned whilst moving. The auto evade kicks in but again, I find it difficult to roleplay out that they would run back into the fire that was coming at them. Again I have to conclude that something here is wrong.

Having said that I have never witnessed a situation in any of my own games that I have not been able to roleplay out, but I've also never witnessed the Indirect Fire bug demonstrated by @IICptMillerII. That's not to deny it exists, just that I've not seen it in a game I've played. I do not have the Shock Force 2 demo.

 

 

[General Jack Ripper]

On ‎10‎/‎25‎/‎2018 at 5:32 PM, ncc1701e said:

This is why I always split my squads into teams, especially in tight spaces.

[General Jack Ripper]

12 hours ago, Pericles said:

 

 

Did the team withdraw on it's own, or did you order them to withdraw?

[LongLeftFlank]

1. Guys, Pericles is *not* making stuff up; everyone who's played CM has had WTF! moments like this. So no need to jump on him so hard. He's trying to help. OTOH, Ian et al are also trying to help, so do let's keep it civil.

2.  In CMSF, where BLUEFOR is (or should be) casualty-sensitive (to offset their massive FP advantages), losing half a squad to sporadic kamikaze behaviour may mean losing a game on vp. So this matters more than ever.

3.  But it doesn't sound like a straightforward AI 'bug' fix either. The cover-seeking AI has always been a bit of a black box, but clearly:

a. We *do* want unpinned squaddies to get out of a killzone, stat, not go to ground in place by default to get hosed down more.

b. But we also *don't* want them overdoing it by dashing pell mell for the nearest 'safe' place that happens to be 100 yards away across a beaten zone.

4. My personal view (fwiw) is that abstracted hard 'micro' cover seems undermodeled in general, even though that hunch is tough to confirm empirically.

In RL, it's unusual (and a memorable horror, think Omaha) when an infantryman doesn't have anything at all solid within a few paces that he can put between himself and bullets. That's why modern battlefields look 'empty.'  Incoming fire should normally encounter sharply diminishing returns after the first few shots, even in 'open ground'.

By boosting the number of perceived 'refuges' available to pixeljoes, it seems you could then reduce the chances of the TacAI choosing to run for a safe spot far away.

Edited October 27, 2018 by LongLeftFlank

[ncc1701e]

10 hours ago, General Jack Ripper said:

This is why I always split my squads into teams, especially in tight spaces.

Well, the flanking move is risky because it is done on the enemy side. If an enemy is there hidden, that is three dead. But, the TacAI behavior is the same whatever the Friendly side parameter is. What is more surprising is the soldier fleeing in the wrong direction.

But, thank you, you gave me the idea of a new test. What if this squad had the knowledge of a spotted enemy unit on its right (North), will it try the same flanking move to enter in this house? Back to the tests...

[Pericles]

11 hours ago, General Jack Ripper said:

Did the team withdraw on it's own, or did you order them to withdraw?

The team withdrew on its own. There were no orders for that team on that turn. 

[General Jack Ripper]

On ‎10‎/‎27‎/‎2018 at 5:31 AM, ncc1701e said:

Well, the flanking move is risky because it is done on the enemy side.

All I meant was trying to pack a full squad through a small gap always leads to trouble.

[General Jack Ripper]

On ‎10‎/‎27‎/‎2018 at 7:20 AM, Pericles said:

The team withdrew on its own. There were no orders for that team on that turn. 

Then, yeah. I'll agree the teams behavior was strange.

[ncc1701e]

On 10/27/2018 at 11:31 AM, ncc1701e said:

But, thank you, you gave me the idea of a new test. What if this squad had the knowledge of a spotted enemy unit on its right (North), will it try the same flanking move to enter in this house? Back to the tests...

All right, I have added one enemy unit on the North of the squad location. The enemy unit is well spotted by the squad at the beginning of its turn. Still, the squad is doing its suicidal flanking maneuver like if it is ignoring this unit or has no knowledge of this threat. This is reproducible but TacAI can not think for you after all. I will stop there since anyway I have not the patch for the Indirect Fire bug. I will let @IanL try my first test scenario.

5 hours ago, General Jack Ripper said:

All I meant was trying to pack a full squad through a small gap always leads to trouble.

Not only, use waypoints far from each other is also a source of problem for TacAI. I have find that if a team has started moving, it ignores breaches done in walls during its movement. Doing small intervals between waypoints is the way to go.

[Battlefront.com]

3 minutes ago, ncc1701e said:

All right, I have added one enemy unit on the North of the squad location. The enemy unit is well spotted by the squad at the beginning of its turn. Still, the squad is doing its suicidal flanking maneuver like if it is ignoring this unit or has no knowledge of this threat. This is reproducible but TacAI can not think for you after all. I will stop there since anyway I have not the patch for the Indirect Fire bug. I will let @IanL try my first test scenario.
 

Thanks, we'll take a look at it for sure.

3 minutes ago, ncc1701e said:

Not only, use waypoints far from each other is also a source of problem for TacAI. I have find that if a team has started moving, it ignores breaches done in walls during its movement. Doing small intervals between waypoints is the way to go.

Correct.  Paths can not be recalculated on the fly, so the unit continues to use whatever path it decided to use from its previous waypoint.  When something like enemy fire hits a unit there's a decision to either continue with its assigned path or to end it right there and recalculate.  Meaning, the unit is not recalculating its path on the fly, but having it's original path prematurely ended and a new one calculated.  Terrain based changes do not, nor should not, interfere with pathing in that sort of way.  It's not realistic.

Steve

[General Jack Ripper]

3 hours ago, ncc1701e said:

Not only, use waypoints far from each other is also a source of problem for TacAI. I have find that if a team has started moving, it ignores breaches done in walls during its movement. Doing small intervals between waypoints is the way to go.

Yes, because the unit doesn't "know" that you're going to blow up that wall. I think people think the TacAI is smarter than it is, to their detriment.

I came up with a maxim a long time ago, "When playing with AI, treat them with kid gloves." Meaning, you have to prod your guys with a stick every once in a while to keep them in line, and be forgiving when they mess up occasionally.

Or as Badger73 says, "It's like herding cats," or something.

[ncc1701e]

54 minutes ago, General Jack Ripper said:

Yes, because the unit doesn't "know" that you're going to blow up that wall. I think people think the TacAI is smarter than it is, to their detriment.

In fact, I discovered this by chance. A team has destroyed a wall using a demolition charge. The squad that was just behind and that could see the wall has started moving about 10 seconds before the explosion. So, instead of using the breach in the wall, the squad went around it to be catch by enemy fire.

WTF I said  and I try to understand the problem. Next time, pause and patience. Patience is my problem you know.

[Battlefront.com]

I always put a waypoint just before the expected breach location, just in case.  As long as the wall is blown by the time the guys get there, you can launch the assault team prior to demolition.

Steve

[MikeyD]

Fitting a full squad through a narrow wall gap is like trying to squeeze dough through a pastry tube. Three teams converging together on a single point at the same time. Its not a big surprise that one team will consider the opening blocked and seek an alternate route.

Edited October 28, 2018 by MikeyD

[Battlefront.com]

Yes, the TacAI has logic to try very hard to be patient with bottlenecks, but sometimes the results aren't optimal.  Squads are particularly prone to this because the TacAI might decide "OK, this Team is busy getting through the space, so why not move the other Team elsewhere?".  Meaning, the TacAI is more flexible when it comes to diverting whole Teams more than individual Soldiers.

Steve

[aleader]

5 hours ago, ncc1701e said:

All right, I have added one enemy unit on the North of the squad location. The enemy unit is well spotted by the squad at the beginning of its turn. Still, the squad is doing its suicidal flanking maneuver like if it is ignoring this unit or has no knowledge of this threat. This is reproducible but TacAI can not think for you after all. I will stop there since anyway I have not the patch for the Indirect Fire bug. I will let @IanL try my first test scenario.

Not only, use waypoints far from each other is also a source of problem for TacAI. I have find that if a team has started moving, it ignores breaches done in walls during its movement. Doing small intervals between waypoints is the way to go.

I've run into this 5 or 6 times now in the demo in different battles, but mainly in Breaking the Bank when attempting to enter buildings across the street.  I split my squads, give them a quick order just outside the door, and then another inside the building, but one will often run around to the front of the building (depending on your perspective I guess), run right past the side door and will then try to enter from the front, getting mowed down in the process by enemy on the opposite side of the next street. 

I catch it during the replay when I find myself saying "WTF are you dummies doing running around the building!"  I don't know if it's because they're thinking the rear door is blocked or what, but they should have then tried the side door at the very least.  I've resorted to playing the demo in real-time so I can catch these before I lose my squads.  I have to admit I'm really enjoying playing it in real-time, something I never do.

They also seem to be running away a lot when only taking suppressing fire or taking one casualty, even when they have overwhelming fire superiority.

[domfluff]

I had a frustrating - but explicable - case earlier today in CMRT.

A team came under SMG fire from a nearby building, but they did not gain a spotting contact (I could see where the fire was coming from, and overwatching units *did* gain a contact icon, but the specific team under fire did not).

The team then ran for the nearest cover, which was the building that the fire was coming from. They ended up charging into the SMG fire, losing a man in the process.

I don't think this is an example of bad AI at all - just something that's an unfortunate consequence of the excessive information that we gain as commanders in Combat Mission - it's "clearly" the wrong course of action, but "seek cover when under fire from an unknown source" was an understandable decision to make in the moment.