hummm patche 4, I need your opinion

[domfluff]

The above, with grid lines approximating action spots:

[MarkEzra]

1 hour ago, domfluff said:

I still suspect that in Roadblock map that due to map curvature and obstacles like the low wall, there's a position out-of-LOS just in front of the hedge, which is close or closer than the alternatives.

OK   than we may make an assumption moving on that Bocage terrain is not, of itself, buggy.  I have not looked at the Roadblock map but will do so.  Your Road Block Pics very helpful.  I did similar but will try that exactly and see what's happening.  But won't happen until later tonight

 

51 minutes ago, The Steppenwulf said:

No I observed the AI behaviour in 4.0 - a long long time ago and not played BN for two years as a result. You are all playing catch up 

 

Thank You.  I will keep your experience in mind.

Edited May 15, 2019 by MarkEzra
add info

[Howler]

24 minutes ago, domfluff said:

This spot has LOS with the Target tool to that MG team. It doesn't have LOS to the action spot the LMG is on, since the low wall is in the way. I still think that's the problem. It's trying to find the nearest solution out of LOS, and it chooses this one.

Wouldn't the nearest spot out of LOS simply be a move back  and away from the hedge/incoming fire? Heck, I'd accept them running back 50m if that would make them feel safer. From Deville this LMG team was stationary and expected to provide covering fire... Instead, they have decided to run forward rather than back. I got lucky here as they 'break' as the playback ends. I am able to adjust the waypoint...

 

From CW 18 Platoon this team was also stationary and expected to provide covering fire. It occurs early during playback and by the time the commands phase arrives - they've already surrendered. The started in the red ellipses and received fire from 11 o'clock position (upper left from the red ellipses).

In either case, all one needs to do is keep hitting the red button until the units do take 'flight'. The waypoint will always be placed in the same spot whether it happens now or 4 minutes (turns) later.

I have saves of both. @IanL has one of the saves but I don't know if he's had a chance to load or if it was  helpful...

[domfluff]

6 minutes ago, Howler said:

Wouldn't the nearest spot out of LOS simply be a move back  and away from the hedge/incoming fire?

So, that seems logical, but (taking my very broad guess as true for the moment) it might depend on geometry, the way that movement orders are handled from units that cover three spots, and exactly how the LOS tool sees bocage gaps.

If you imagine that the bocage gap is empty space, then it might just be that there's a shorter path to one of the three action spots the unit has to take up, rather than the fairly large arc that LOS could be traced through the gap. Might be extremely tricky to pin down without proper debugging tools though, and I'm certainly not confident that's the solution, but it seems like a fairly likely one to me - fundamentally, the AI sees this as a safer spot than the one that it's in, and dies on the way to it.

It's interesting that in your example it's the same setup though - one space to the left of a bocage gap, with fire coming from 11 o'clock or so.

[Howler]

7 minutes ago, domfluff said:

It's interesting that in your example it's the same setup though - one space to the left of a bocage gap, with fire coming from 11 o'clock or so.

Indeed, interestingly enough, loading the save and selecting the Evade places the waypoint back and away from known contacts. So, the evade routine during the commands phase differs from the evade routine that we will see during playback.

[snarre]

10 hours ago, rocketman said:

I think, a bit confusing, is that it is called Low Bocage, but isn't as you can cross through it with infantry. Same thing in FB.

yah it is call whit same name but only on Normandy it work different . Like that tanks canot go ower it .

[Warts 'n' all]

11 hours ago, The Steppenwulf said:

No I observed the AI behaviour in 4.0 - a long long time ago and not played BN for two years as a result. You are all playing catch up 

 

No we NOT "all playing catch up". The behaviour with the patch is different. With engine 4 as in with CMFB units that had taken fire were breaking either to the side, or at a diagonal backwards out of, and away from cover sometimes leading them into the line of fire of the enemy.

What we are seeing with the patch is units charging towards the enemy before they have sustained any casualties.

[Warts 'n' all]

@MarkEzra I've tried the map test a few times as both Allies and Axis. No gap charging to report so far.

[Wicky]

Playing test range - As US noticed on left hand field Germans had come forward into the field through hedge gap and duly got shot up

Edited May 16, 2019 by Wicky

[Wicky]

[Warts 'n' all]

Ah, that's bad news. 

[MarkEzra]

@Wicky The German Units have no movement orders...just a setup/ambush 1000 meter order.  What you are observing is only the setup phase.  The AI has chosen to setup on the "wrong side" of the bocage.  Check in the editor (AI) and you will see the yellow paint falls on both sides of the terrain tile.  My understanding is:  The AI generally makes it's setup placement  based of which side IS the Defender setup side...in this case North, then, LOS, then concealment.  You will note that the setup was near passable Bocage tile....less concealment... so perhaps LOS trumps all.  But no movement to destruction. Not a bug.  Just code

[Sgt.Squarehead]

Both sides continue to be rock solid.....My men shot their rifles dry, killed a lot of Germans, but nobody on either side did anything dumb. 

[MarkEzra]

@domfluff  First... Thank you for posting the movement pics and caring enough about CM to question and seek to improve it's game play.

I replicated your movement order and attached a save game file to watch the fireworks.  I think you will agree a bit different than yours.  Why this is so may well be,

as the say in the Real Estate game, Location, location location.  Please note my quick order goes up to but not through the passable Bocage tile.  The Infantry avoids the hole and sets up on either side.  The Save game file starts with the infantry reaching their setup.  play the game for awhile and let me know what you find.

 

Road block pic 2.bmp Roadblock Inf runs to open Bocage Tile.zip

[domfluff]

Will give this a poke, but it's worth pointing out that whenever I made a movement order similar to the above - with the central squad action spot on the bocage gap - they behaved reasonably (staying in place, or sometimes retreating).

The way to reliably recreate the suicidal forwards-evading behaviour was to place the order one square to the left of where you have it in the picture above. That way, the rightmost action spot of the squad is in front of the bocage gap, rather than the central square.That's what produces the forwards-running behaviour.

[MarkEzra]

25 minutes ago, domfluff said:

Will give this a poke, but it's worth pointing out that whenever I made a movement order similar to the above - with the central squad action spot on the bocage gap - they behaved reasonably (staying in place, or sometimes retreating).

The way to reliably recreate the suicidal forwards-evading behaviour was to place the order one square to the left of where you have it in the picture above. That way, the rightmost action spot of the squad is in front of the bocage gap, rather than the central square.That's what produces the forwards-running behaviour.

OK.. I will check that out.

Results:  I found no problems when setup left or very slightly left

Edited May 16, 2019 by MarkEzra
add info and pics

[domfluff]

Okay, will recreate, and get a save on the turn before they start running, since I've just run through the same thing again and got the same result.

The exact move to get into the situation was:

[domfluff]

Zip file here, containing two saves. One before the behaviour (005), and one immediately after (006). The only user-action taken was a pair of quick moves to the above action spot (I have that save as well if you'd like it, but I don't imagine it's needed).

https://uploadfiles.io/pnl1430q

The trigger seems to be the HE fire from the AT guns, which isn't surprising, and the retreat direction matches the screenshots above. Some of the AI plans move the AT gun positions, so the squad doesn't come under the same amount of HE fire in all tests.

I have seen a little variety in that - mostly they head to exactly the spot they're moving in save 006 (and in the previous shot), but sometimes they've headed to one square closer in the same direction.

Start of 006 looks like this: 

That waypoint has been generated entirely by the squad, after taking HE fire from the opposite AT gun.
 

[domfluff]

Having reloaded that save a few times, it's definitely the HE avoidance behaviour that's triggering it (which wasn't unexpected). The unit doesn't always start moving immediately, but it will when it starts taking HE. Not necessarily straight away, but within a couple of turns.

[MarkEzra]

19 minutes ago, domfluff said:

Having reloaded that save a few times, it's definitely the HE avoidance behaviour that's triggering it (which wasn't unexpected). The unit doesn't always start moving immediately, but it will when it starts taking HE. Not necessarily straight away, but within a couple of turns.

which becomes reasonable in the short run.  I checked the elevation tiles: West of Bocage 16 meters to lower ground and concealment vs 32 meters of observable terrain East.  Player input may well have led to a better outcome. 

 

 

[domfluff]

That kind of thing is still my expectation here.

It doesn't mean that it's not a behaviour that needs tweaking in some manner, but it's a normally-reasonable response to stimuli, which produces unreasonable results in an edge case (or even a few edge cases).

Edited May 16, 2019 by domfluff

[MarkEzra]

34 minutes ago, domfluff said:

That kind of thing is still my expectation here.

It doesn't mean that it's not a behaviour that needs tweaking in some manner, but it's a normally-reasonable response to stimuli, which produces unreasonable results in an edge case (or even a few edge cases).

The major thrust of my testing has been to clarify if bocage terrain tiles are creating unrealistic or unreasonable AI reaction when under fire.  As best as I can see, bocage style QB maps work as intended. The Training scen is not my creation, but I have looked at that as well and feel confident in saying that it , too, works as intended.  There are two areas of concern that I do not offer opinion:  Further altering AI behavior and Player technique.   

[domfluff]

Oh quite, but I'm a lot less restricted, and free to speculate wildly

• I think it's clear that this is not problem with bocage tiles themselves, since it doesn't occur on the flat test map. I suspect the bocage gap tiles are illustrating the problem more immediately than anything else, but other circumstances could cause this.
• I do not believe this is a question of player technique. Certainly putting units in the tiles adjacent to bocage gaps is a bad idea, but the punishment for that shouldn't be the pixeltruppen committing suicide. This isn't as simple as putting waypoints to select the correct entry point to a building - this evasion behaviour is out of player control. Certainly "don't put them next to gaps" is good advice in general, and that's the immediate player-fix, but it's still unexpected, unpredictable and uncontrolled behaviour, which you really don't want.
• I also don't believe it's a problem with the Roadblock map as such (or any other map with a similar situation) - more that the AI behaviour is interpreting the situation, and choosing that depression in Roadblock as the best available spot to retreat to.

I do therefore believe we've shown that it's a consequence of the Engine 4 HE avoidance changes. It's worth pointing out that this behaviour is significantly better than it was in 4.00, but it looks like there are still tweaks that can be made in 4.01.

Without knowing how the AI is set up (again, I'm free to speculate wildly here, since I'm just a punter), this may be a question of messing with weighting, and giving greater priority to action squares which are closer to the friendly map edge. I also don't imagine that's a simple thing to "just" change, since any AI changes will have knock-on effects.

If the exact same behaviour occurred but resulted in the squad retreating backwards, I don't think anyone would notice this as a problem. It's that they're moving forwards, and specifically in the case of bocage gaps that means exposing the entire squad to fire, rather than one third of it, with disastrous results.

I think that the bocage is just bringing attention to the AI limitations in a not-elegant way.

[Howler]

While both my saves involved units that were stationary, the captured image below without mods this time, could be caused by nearby HE falling. There were a couple of impacts 50m to their 5 o'clock (lower right of center). It's the only difference I can find between replays (regenerating the turn) that show them evading from those that show them remaining in place.

The LMG team is unordered and already in position to overwatch. I'll I do is load + red button...

The other example (I have a save of) shows the team evading forward and away from the designated friendly map edge solely as a result of small arms fire.

I'm still running cycles of your (MarkEzra) provided QB; I can say that I have yet to see anything unusual other than a tendency for teams to snake through the two gaps in the center hedgerow splitting the Allied position. I"m also running a variation of your file that has lower (by one) Allied morale grade in order to force more evasions.

Edited May 16, 2019 by Howler

[rocketman]

I haven't followed this thread extremely closely but it reminds me of something that players complained about when SF2 was released (or demo) in that when units inside walled compounds panicked they ran in what was seemingly the wrong direction. Since walls and bocage are impassible and the bocage gaps act like the opening in walls - maybe the problems are related. Just a thought...