ArmA 2 OA Beta Build 83181

[Dwarden 1125]

http://www.arma2.com/beta-patch.php

[83156] Fixed: AI units no longer firing at empty enemy vehicles (http://dev-heaven.net/issues/5183)

[83137] Fixed: Walking no longer causes prone-style recoil to be used. (http://dev-heaven.net/issues/22828)

[83122] Fixed: AI no longer considering a car horn as a threat.

[83118] Fixed: AI no longer using AA against ground targets or guided missiles against soft targets unless desperate (issues/7578)

While the fix itself can be considered minor, it was done in quite sensitive area of code

(it can affect what targets does or does not AI engage anywhere in the game),

if the fix is to stay in the game, extensive testing is needed now, especially in campaign missions and other complex missions,

to confirm some key aspect of the AI target selection is not broken by the fix.

The fix consists of:

- changed conditions when an enemy vehicle is considered as "finished"

- changed criteria when AI engages a target

- changed criteria when AI stops firing at a target

and some more fixes

Edited July 28, 2011 by Dwarden

[metalcraze 290]

I like seeing "AI" in every beta's changelog.

[NoRailgunner 0]

Another bug bites the dust... :cool:

[suma 8]

[83156] Fixed: AI units no longer firing at empty enemy vehicles (http://dev-heaven.net/issues/5183)

While the fix itself can be considered minor, it was done in quite sensitive area of code (it can affect what targets does or does not AI engage anywhere in the game), if the fix is to stay in the game, extensive testing is needed now, especially in campaign missions and other complex missions, to confirm some key aspect of the AI target selection is not broken by the fix.

The fix consists of:

- changed conditions when an enemy vehicle is considered as "finished"

- changed criteria when AI engages a target

- changed criteria when AI stops firing at a target

[[frl]myke 14]

[83122] Fixed: AI no longer considering a car horn as a threat.

Nice. Just as little input: is there a way to make AI react on car horns in a way, they are standing in the middle of the road, there is not enough place to drive around them so when using the horn they would check the orientation of the vehicle that used the horn and goes out of the way?

Basically the same way as you would if a car horns behind you. ;)

[gossamersolid 155]

[83118] Fixed: AI no longer using AA against ground targets or guided missiles against soft targets unless desperate (http://dev-heaven.net/issues/7578)

Thank you so much Suma, great fix!

[NoRailgunner 0]

It would be great if AI + animals would just go/sprint away from roads where vehicles drive - at least somekind of reaction on vehicle speed + horn.

Fleeing animals and civilians from combat/battle areas would be great too...

[LondonLad 13]

Myke;1992003']Nice. Just as little input: is there a way to make AI react on car horns in a way' date=' they are standing in the middle of the road, there is not enough place to drive around them so when using the horn they would check the orientation of the vehicle that used the horn and goes out of the way?

Basically the same way as you would if a car horns behind you. ;)[/quote']

+1 on that.

One of those smaller priority issues/feature in the grand scheme of things; but also one that is an extreme annoyance to me ingame. The horn (vehicle) is almost pointless in game due to the nature that the AI ignore it.

I suspect you'll have to do a little extra coding to get this one implemented but if its something you DEV's can find time for in your schedule I'd definately appreciate it :)

[pauliesss 2]

Just a simple and honest thank you, will of course report if I find any issues. :)

[fabrizio_t 58]

Hi,

Good job Suma and thanks.

Any planning on further working on AI target selection/prioritization?

I see the AI still having problems in CQB, meaning moving units usually prefer keep moving instead of stop and fire on target, when threat are very close.

In these cases Neartargets info and knowsabout tell the threat is well known, usually.

I bet this problem can be somewhat related to the (low) frequency of visibility checks ?

Can we eventually have some higher "shooting" prioritization against "movement" when threat is very close and in line of sight ?

[OMAC 254]

It would be great if AI + animals would just go/sprint away from roads where vehicles drive - at least somekind of reaction on vehicle speed + horn.

Fleeing animals and civilians from combat/battle areas would be great too...

+2 on that. Excellent idea.

---------- Post added at 07:20 AM ---------- Previous post was at 07:16 AM ----------

... if the fix is to stay in the game, extensive testing is needed now, especially in campaign missions and other complex missions, ...

Any suggestions as to which campaign mission or scenario would be most useful to test given the changes?

[dmarkwick 261]

While the fix itself can be considered minor, it was done in quite sensitive area of code (it can affect what targets does or does not AI engage anywhere in the game), if the fix is to stay in the game, extensive testing is needed now, especially in campaign missions and other complex missions, to confirm some key aspect of the AI target selection is not broken by the fix.

The fix consists of:

- changed conditions when an enemy vehicle is considered as "finished"

- changed criteria when AI engages a target

- changed criteria when AI stops firing at a target

IMO units firing on enemy vehicles can be considered legitimate, whether empty or not. There was a thread about it not so long ago where the various aspects were talked over....

[Muahaha 10]

Hi,

Good job Suma and thanks.

Any planning on further working on AI target selection/prioritization?

I see the AI still having problems in CQB, meaning moving units usually prefer keep moving instead of stop and fire on target, when threat are very close.

In these cases Neartargets info and knowsabout tell the threat is well known, usually.

I bet this problem can be somewhat related to the (low) frequency of visibility checks ?

Can we eventually have some higher "shooting" prioritization against "movement" when threat is very close and in line of sight ?

I think getting a repro sample and a CIT ticket might be able to get it enhanced/fixed faster.

[suma 8]

Any planning on further working on AI target selection/prioritization?

I see the AI still having problems in CQB, meaning moving units usually prefer keep moving instead of stop and fire on target, when threat are very close.

In these cases Neartargets info and knowsabout tell the threat is well known, usually.

I bet this problem can be somewhat related to the (low) frequency of visibility checks ?

Near targets visibility is checked very frequently, therefore I doubt this is the cause. Most likely there is some reaction to the situation missing or handled improperly. There are reaction like this in Danger FSM for events "EnemyNear" or "CanFire", perhaps some tuning for them would be enough.

I think getting a repro sample and a CIT ticket might be able to get it enhanced/fixed faster.

Exactly. If you are able to create a good repro (esp. if the repro is short, located on Utes or Desert, and full automated, not requiring any interactions from the player), I can check the cause and we shell then see if it is fixable easily or not. Posting in a CIT is preferred, but if you feel that too clumsy, you can also post it here and somebody else can create the ticket later.

[.kju 3245]

@ Suma

Are you talking about the danger.FSM in \ca\characters\scripts\danger.fsm used for:

	class CAManBase: Man
{
	fsmDanger = "Ca\characters\scripts\danger.fsm";

To me it looks like there is no reaction defined in the "EnemyNear" or "CanFire" event:

Or is the handling not exposed?

Edit: Fixed links to the pictures.

Edited July 28, 2011 by .kju [PvPscene]

[neokika 62]

Have you checked the Condition Action?

It would make sense to have the reaction there since it would run on a non-scheduled environment. I could be wrong though. :)

_neo_

[Big Dawg KS 6]

Myke;1992003']Nice. Just as little input: is there a way to make AI react on car horns in a way' date=' they are standing in the middle of the road, there is not enough place to drive around them so when using the horn they would check the orientation of the vehicle that used the horn and goes out of the way?

Basically the same way as you would if a car horns behind you. ;)[/quote']

Couldn't you script this?

[[frl]myke 14]

Couldn't you script this?

If they can't fix it i probably can script it. But i would prefer to see it in AI's usual behaviour.

:EDITH:

But more likely i will keep my usual behaviour in such situations: i just drive over them. :p

[suma 8]

@ Suma

Are you talking about the danger.FSM in \ca\characters\scripts\danger.fsm used for:

	class CAManBase: Man
{
	fsmDanger = "Ca\characters\scripts\danger.fsm";

To me it looks like there is no reaction defined in the "EnemyNear" or "CanFire" event:

Is see a following reaction on canFire event, in the Stop to Fire state, which should make the soldier to stop for 4-8 seconds (or shorter when target is destroyed) on a fire opportunity (duration controlled by the Timeout and Destroyed condition):

_stopUntil = time + 4 + random 4;
if (vehicle _this==_this) then {_this forceSpeed 0;}

[Rough Knight 9]

wow....thanks for the hard work...loads of fixes coming out here. I'm off to do some testing : )

[fabrizio_t 58]

Near targets visibility is checked very frequently, therefore I doubt this is the cause. Most likely there is some reaction to the situation missing or handled improperly. There are reaction like this in Danger FSM for events "EnemyNear" or "CanFire", perhaps some tuning for them would be enough.

Ok, thanks, i already fiddled with danger.fsm in the last weeks.

From what i've experienced, "CanFire" event is reported extremely rarely, maybe something is missing there.

I was expecting to see it triggered quite a lot of times in a gunfight, but even if the AI is taking thousands of shots, only a handful of "CanFire" events are queued.

This was checked by me by continually dumping to .rpt all queued .fsm events;

I already tried maximizing "CanFire" priority (to check whether it's eventually overriden by higher priority events in queue), but saw no increase in event reporting.

I'll put together a Repro.

Finally a different question: neartargets target position accuracy is shared for all the units in a group right?

I was wondering how to check the line of sight for a single unit ...

Edited July 28, 2011 by fabrizio_T

[suma 8]

From what i've experienced, "CanFire" event is reported extremely rarely, maybe something is missing there.

That is quite possible. Once I see some repro, I will be probably able to tell more quickly. Note: this intention of this event is to handle a situation when you can fire on a target which could not be fired on before.

Finally a different question: neartargets target position accuracy is shared for all the units in a group right?

Yes, it is shared in a group. I can imagine it would be possible (and not very hard) to simulate reduced the accuracy when we can be sure some particular unit cannot see the target, but nothing like this is implemented now (and even as simple as it sounds, it would require to proceed very carefully with implementation, and it would require lots and lots of testing once implemented).

[fabrizio_t 58]

That is quite possible. Once I see some repro, I will be probably able to tell more quickly. Note: this intention of this event is to handle a situation when you can fire on a target which could not be fired on before.

I see.

Please clarify the clause: "on a target which could not be fired on before":

if for whatever reason a unit times out when performing the "Can Fire" state code in .fsm ( = does not kill threat or does not even open fire on it), the same threat would potentially cause a new "Can Fire" event to be issued later or not?

Thanks.

Yes, it is shared in a group. I can imagine it would be possible (and not very hard) to simulate reduced the accuracy when we can be sure some particular unit cannot see the target, but nothing like this is implemented now (and even as simple as it sounds, it would require to proceed very carefully with implementation, and it would require lots and lots of testing once implemented).

It would be very useful in my opinion! Maybe it would be possible to add a hidden or visible key (value = true|false) to the output of neartargets?

Edited July 28, 2011 by fabrizio_T

[metalcraze 290]

Yes, it is shared in a group. I can imagine it would be possible (and not very hard) to simulate reduced the accuracy when we can be sure some particular unit cannot see the target, but nothing like this is implemented now (and even as simple as it sounds, it would require to proceed very carefully with implementation, and it would require lots and lots of testing once implemented).

Would be really nice to have this. Will make AI more humanlike when it comes to engagements.

[KeyCat 131]

Here is a repro for the MG/HMG issue where bullets hits ground, looking close on their barrels it looks like the recoil is inverted when prone?

http:\\keycat.dyndns.org\files\A2OA_B83181.zip

/KC

Edited August 6, 2011 by KeyCat