TPWC AI suppression system

[tpw 2315]

...(might itself be a problem if you run debug on a 1000+ unit battle and have 3-4000 civs generated at start up-haven't tested this yet, but may do today).

So are have you booked the time on a NASA supercomputer? :)

I'm still plugging away on a replacement for these bloody civilian balls.

[ollem 4]

tadanobu, ChrisB, are you running 'latest version of ArmA' (1.60) or the (latest?) post 1.60 Beta?

(ArmA2, ArmA2 OA, ArmA2 CO?)

(Difficulty setting? (e.g. Recruit, Veteran))

(map and type/side of AI used?)

[tpw 2315]

ah I just ran some quick tests of 2.04, cba and no other mods. Not looking good. As soon as they become suppressed everyone starts shooting at sky except machine gunners who now spin around firing erratically mowing down their entire squad. This was using both pbo and script versions. Units used were plain us army rifle squad and takistani militia group. Default skills in editor. Playing veteran with both friendlies and enemies on normal. Tried with and without debugging enabled. Same problem occurs. I ran arma without tpwc as comparison and no such behaviour occurs. I ran 2.03 script again which runs fine. No shooting at sky but I did notice that when units become surpressed they shake their heads at the sky every so often. Have to go out for awhile but will do some further testing later.

I have to say I'm very suprised at this, it's behaviour that I simply don't see to this degree - and as you'd imagine, I do test pretty intensively. I'll reiterate what I've said before, not to try and absolve myself of culpability, just to give an insight of what might potentially be going on

• TPWCAS changes unit stance
  
• TPWCAS changes unit aimingshake, aiming accuracy and courage
  
• TPWCAS reveals shooter to the suppressed unit
  
• TPWCAS sets a few custom variables and fired/killed eventhandlers

Apart from this, there really is no other major behaviour modification going on.

So I'm going to request that before you post any reports of erratic AI behaviour, please do the following

• Reset your computer!
  
• Try the exact same mission without TPWCAS
  
• Try the same mission with no other mods except TPWCAS
  
• Try with tpwcas_skillsup = 0 (stance only modifications, no skills modification)
  
• Try with tpwcas_reveal = 0 (shooter is not revealed to unit)
  
• Try with tpwcas_debug = 0 (no pesky civilian balls to confuse the units)
  
• Tell me what difficulty settings you are using
  
• Tell me what OPFOR and BLUFOR you are using, and what side you are playing as
  
• Tell me if you are using the addon or script version of TPWCAS
  
• Tell me what version of TPWCAS you are using
  
• Tell me what version of Arma you are using (esp if beta)

I'm hopeful that we can get to the bottom of this soon, and get the hell out of beta city.

Edited July 5, 2012 by tpw

[CameronMcDonald 146]

I wonder if it is (like suggested earlier) to do with the debug spheres.

[tpw 2315]

I wonder if it is (like suggested earlier) to do with the debug spheres.

tadanobu's above report suggests otherwise.

[fabrizio_t 58]

ah I just ran some quick tests of 2.04, cba and no other mods. Not looking good. As soon as they become suppressed everyone starts shooting at sky except machine gunners who now spin around firing erratically mowing down their entire squad. This was using both pbo and script versions. Units used were plain us army rifle squad and takistani militia group. Default skills in editor. Playing veteran with both friendlies and enemies on normal. Tried with and without debugging enabled. Same problem occurs. I ran arma without tpwc as comparison and no such behaviour occurs. I ran 2.03 script again which runs fine. No shooting at sky but I did notice that when units become surpressed they shake their heads at the sky every so often. Have to go out for awhile but will do some further testing later.

Also check ArmA .rpt file, see if there are errors.

Check there whether there is anytihing more being loaded, apart from CBA and TPWCAS.

@tpw: did you make some check on reveal? For example make unit A reveal unit B to unit C only if unit A knowsabout B is > C knowsabout B.

Otherwise you may end up "fibrillating" AI, since everybody is cross-revealing anybody (with geometrical progression, so it's likely to put anything out of control).

Also if you use lookAt with reveal consider it momentarily stops unit movement, so in a huge cross-revealing situation it can eventually make units appearing to go forth and back, as they simply move->turn for looking->move ...

Better than that, if you don't do it already then reveal anything just from leader of A to leader of C, since knowsabout is shared through group.

This should eliminate quite some load.

Sorry, i can't directly look at code, maybe it's already fine.

Edited July 5, 2012 by fabrizio_T

[tpw 2315]

Also check ArmA .rpt file, see if there are errors.

Check there whether there is anytihing more being loaded, apart from CBA and TPWCAS.

@tpw: did you make some check on reveal? For example make unit A reveal unit B to unit C only if unit A knowsabout B is > C knowsabout B.

Otherwise you may end up "fibrillating" AI, since everybody is cross-revealing anybody (with geometrical progression, so it's likely to put anything out of control).

Also if you use lookAt consider its momentarily stops movement, so in a huge cross-revealing situation it can meke units appearing to go forth and back, as they simply move->turn for looking->move ...

Better than that, reveal just from leader of A to leader of C, since knowsabout is shared through group.

This should eliminate quite some load.

Sorry, i can't directly look at code, maybe it's already fine.

The more I think about it, the more I just want to get rid of reveal altogether and let other more capable AI mods handle the behaviour.

[ollem 4]

Agree - but keep in mind the 'bird watching' was already spotted before 'reveal' was added to the code.

However, did anyone experience AI behavior improvement using 'reveal'?

Because that was the purpose in the first place.

[tpw 2315]

OK so I think I might have tracked down at least part of the problem.

In what must be one of the biggest regressions in history, units are failing to have their skills readjusted once they are unsuppressed. Which means they stay very low once suppressed, and which is probably leading to sky shooting.

I'm frantically trying to work out how to solve this problem.

EDIT:

Solved!

• Skill reduction code was calling a nil variable that had changed once bDetect was introduced, so all skills immediately hit 0 once suppressed
  
• Skills restoration was also calling a nil variable that I had somehow forgot to reintroduce while implementing the bDetect version

I'm very sorry if this has caused you anywhere near as much stress as it has caused me.

TPWCAS 2.04 beta (regression fixed): http://filesonly.com/download.php?file=249TPWC_AI_SUPPRESS_204.zip

I also introduced a tiny if statement, so that if tpwcas_reveal = 0 then the reveal function is actually disabled.

Links updated

Edited July 5, 2012 by tpw

[tadanobu 1]

Quick update. I just ran 2.03 and 2.04 and what I noticed was that in 2.03 the balls are instantly deleted, while in 2.04 they remain after unit is killed.

Forgot to mention in earlier posts that the only adjustments I have made to tpwc is enabling/disabling debug. Haven't touched other settings because I've tried to make tests as consistent as possible. Regularly reset computer. I have steam version, so auto-updated to 1.60.87580 and latest (I believe) cba 0.8.6.182.

I can run a 1000 unit battle at 4 fps lol. Just tried it and yes approximately 3000 civs were created which was to be expected with debug running. But I just comfortably ran a 600 unit battle on OA's desert island using 2.03 and you'll hopefully be encouraged to know that tpwc does not really add too much overhead. Lost a couple of FPS but was running around 23 FPS when the fighting kicked off.This is what I'm used to any way. I have to say again that 2.03 script seems the most reliable. I saw no unusual behaviour even in such a large battle. With debug turned on I eventually went down to around 15 FPS. Running on a toshiba with i7 quad and geforce gt 630M.

Edited July 5, 2012 by tadanobu

[tpw 2315]

Quick update. I just ran 2.03 and 2.04 and what I noticed was that in 2.03 the balls are instantly deleted, while in 2.04 they remain after unit is killed.

Forgot to mention in earlier posts that the only adjustments I have made to tpwc is enabling/disabling debug. Haven't touched other settings because I've tried to make tests as consistent as possible. Regularly reset computer. I have steam version, so auto-updated to 1.60.87580 and latest (I believe) cba 0.8.6.182.

I can run a 1000 unit battle at 4 fps lol. Just tried it and yes approximately 3000 civs were created which was to be expected with debug running. But I just comfortably ran a 600 unit battle on OA's desert island using 2.03 and you'll hopefully be encouraged to know that tpwc does not really add too much overhead. Lost a couple of FPS but was running around 23 FPS when the fighting kicked off.This is what I'm used to any way. I have to say again that 2.03 script seems the most reliable. I saw no unusual behaviour even in such a large battle. With debug turned on I eventually went down to around 15 FPS. Running on a toshiba with i7 quad and geforce gt 630M.

You are right, I'd commented out the dead ball removal code and forgot to uncomment it.

Just to be sure, here's the absolute latest version with that particular fuckup removed.

http://filesonly.com/download.php?file=249TPWC_AI_SUPPRESS_204.zip

EDIT: If you guys haven't noticed, I'm starting to get stressed out here! Feels like 3 steps forward, 2 steps back with every release.

[ollem 4]

EDIT: If you guys haven't noticed, I'm starting to get stressed out here! Feels like 3 steps forward, 2 steps back with every release.

You shouldn't get stressed :cool: It's already quite amazing how much time you can spend to work on this.

So you just released an updated v2.04, or wre you stressed again and should it be called 2.05? :p

[tpw 2315]

You shouldn't get stressed :cool: It's already quite amazing how much time you can spend to work on this.

So you just released an updated v2.04, or wre you stressed again and should it be called 2.05? :p

Just updated 2.04. I'm sorta hoping that the time difference between me and everyone else means that not too many people will notice that I snuck a new version over the top. I will update to 2.05 tomorrow with bDetect 0.67 (or whatever it will be by then).

I do get a bit stressed because poor code reflects badly on the coder. I try to put in a couple of hours coding each day, in between work and being single dad, so I'd prefer that I was releasing versions without stupid omissions, regressions and basic mistakes, and instead moving the whole thing forward to make Coulum and Fabrizio's ideas really shine.

[fabrizio_t 58]

Just updated 2.04. I'm sorta hoping that the time difference between me and everyone else means that not too many people will notice that I snuck a new version over the top. I will update to 2.05 tomorrow with bDetect 0.67 (or whatever it will be by then).

:) Don't worry, 2 more days and i'm out of vacancy, so things will definetely slow down on bDetect side ...

On a more serious note, v0.68 is in the works, but fo far the additions / fixes are so tiny i think i'll hold it.

Glad you fixed it, don't be stressed, in 10 years the one thing i really learnt is that OFP/ArmA/ArmA2 editing is a matter of trial and error.

Edited July 5, 2012 by fabrizio_T

[-Coulum- 35]

I do get a bit stressed because poor code reflects badly on the coder. I try to put in a couple of hours coding each day, in between work and being single dad, so I'd prefer that I was releasing versions without stupid omissions, regressions and basic mistakes, and instead moving the whole thing forward to make Coulum and Fabrizio's ideas really shine.

tpw don't sweat it, your doing a great job. And I am sorry I didn't pick up on the skills not regaining earlier. Honestly I just haven't gotten to really work on this as of late - and I am not nearly as busy as you sound to be. Your dedication to this mod is great and trust me, everyone appreciates it a ton.

[Robalo 465]

Hey tpw,

About the ASR skill reset, you may find this tip useful if you want more control: instead of setting the "configured" variable and depending on ASR_AI to run it's cycle and catch that, you can alternatively

call the skill reset directly like this:

_unit call asr_ai_sys_aiskill_fnc_SetUnitSkill;

This will restore whatever skills the unit would receive initially. Should be harmless to add it with the extra check like this so it works with or without my mod:

if (!isNil "asr_ai_sys_aiskill_fnc_SetUnitSkill") then {_unit call asr_ai_sys_aiskill_fnc_SetUnitSkill};

Maybe this will help you get rid of that extra loop.

[chrisb 196]

Quick easy standard 'clear town' mission: here

Opposing ’ai’ side was USSR standard small 5 man groups x 11, placed around town with no waypoints just synced to a ‘Defend’ module (GL4), i.e. they use their own initiative to take up positions in town, usually always in building on balconies or rooftops, wherever there is a good position and occasionally moving positions to other buildings etc.

My side USMC standard infantry section (13man section) plus 3 man BAF squad (just to see if they did it as well, they did) I was a BAF unit. All ai apart from me. Multi ‘Move’ waypoints through the town to clear, on ‘combat’ setting, ‘open fire, fire at will’, other no changes.

All set to 1.00 skill, both friendly and enemy in difficulty, and standard mid slider skill in the ’Editor’, playing Arrowhead inc baf/pmc so reinforcements 1.60 one of the latest Beta's 94444.

I disabled your ballons, debug = 0.. Although it still does it with =1.

The machine gunners didn’t do as ‘tadanobu’ i.e. killing friendly’s, everything as normal other than the sky thing. They don’t all fire into the sky, around 70% do, the rest are normal..

As said, ‘All ai’ on the map, apart from me, so ‘70 ai’ plus 1 me, 71. No civies..

They tend to do what you see occasionally when ai throw smoke grenades, they follow the grenades course with their weapon i.e. slowly dropping the weapon towards level position from skyward..

Map: Takistan. Town: Chardarakht.

In-game settings = here

Playing on ‘Veteran’

Playing 1920x1080. FXAA used in configs to beef up AA to VH effect, thats why the AA is set to low..

Ignore view distance in screen shot as I use ’IVD’ set at 2000vd in-game for ground assault.

I have a lot of mods at work, but the principle two ai mods are SLX & GL4.

I think its definitely a conflict with another mod, just which..

Going to do the same mission, with all other mods disabled.. Will post result..

Edit: yeah I think its a mod conflict somewhere..

Edit again: o.k maybe its not a mod issue unless its CBA. Anyhow, quick mission two opposing sides (standard infantry sections 23 men in total, I was one, rest ai) USMC v USSR, only mod CBA. The same thing happens, the sky thing. No other mods running. All settings as above concerning ai skill etc. I was a USMC rifleman.

Edited July 5, 2012 by ChrisB

[jiltedjock 10]

Just a quick query on the ASR AI reset - presumably this would reset all the skills back to their initial ASR starting point.

Could this have negative effects if any of those skills have been degraded during a mission by other factors? I am thinking particularly of courage. Do the AIs' skills take a permanent reduction during missions in normal play?

[ollem 4]

-----

Thanks for detailed description.

Could you perhaps share your mission file, so me/others could run the exact same test?

[chrisb 196]

Thanks for detailed description.

Could you perhaps share your mission file, so me/others could run the exact same test?

Looking at it again, it seems to be the ai with grenade launcher weapons i.e. M16A4. They may be in grenade launching mode ? maybe worth a look..

[tadanobu 1]

I did very quick test of the amended 2.04 pbo and script with 600 unit battle. Just tpwc and cba. I ran my civ player character(allowdamage false) along the us line several times and saw just one us unit, an assistant auto gunner raise and not lower his weapon.But only this individual unit. His skill in editor seems to have very low default. The grenadiers always raise their weapons to fire grenades and lower them again when returning to normal fire mode. Machine gunners (it was these units raising their weapons that first brought my attention to the problem of sky watching) now seem to behave normally. However, spheres did not show in debug mode with either script or pbo and the enable/disable in the userconfig did not work so using pbo I was unable to disable debug. The civ counter kept me up-to-date of the civs being generated so I know it remained enabled. Sorry not really able to devote much time to testing today and so the info might seem a little bit incomplete and rushed. But I'm keeping things consistent so I think my observations may still be useful.

[orcinus 121]

Hi

Just ran several SP test missions (Trial by Fire, bboy's "Pinned Down", a port of thomsonb's Flashpoint v1.14 to Podagorsk) 2-4 times each. Results were excellent - good long firefights, AI using buildings, firing from rooftops & windows, flanking, etc.

asr_ai v1.15.3_test5, TPWC_AIS204, TWP_AI_LOS v1.0, selected COSLX files (ai_dodge, ai_no_autoengage, ai_findcover, netcode, shout, cloud; plus ORC_slx_wounds which is slx_wounds edited to remove weapon-dropping). CBA 1.006pre.

All were PBO versions.

TPWC bugging was on. Only change to the .hpp was to up the start-delay delay to 30 (Flashpoint can take 20 seconds plus for a lot of groups).

Beta 94444 on CO1.60/1.11.

XP Pro, i5-750 @ 4GHz, 4GB RAM.

Settings: Modified regular - tags off, AI skills set to 1.0/0.8 (but would be over-ridden by asr_ai), SuperAI off. (Autoaim? WTF is that doing in this game? <g>)

I did not see any fly-shooting; the only time I saw units aiming into the air was when they were firing at choppers. In fact it seemed to work brilliantly, the best combination yet.

One RPT error - on loading a saved game, this error seems always to appear:

Error in expression <ll = _unit getvariable "tpw_losball";  
detach _ball;   
deleteVehicle _ball;  
>
 Error position: <detach _ball;   
deleteVehicle _ball;
 Error Generic error in expression

which seems to have no impact at all on the game. I suspect that it might result from a dead unit either not being saved or deleted on reloading the mission (I've no idea how the engine handles such stuff).

@tpw - this thread does indeed reflect on the coders - and none of them have any reason to feel stressed or depressed about that.

This is one of the best threads I've read here, and TPWC_AIS is already a great mod, and a credit to the coders. Thanks for all the hard work. This mod is already on my 'must-use' list.

Edited July 5, 2012 by Orcinus

[SavageCDN 231]

@tpw - this thread does indeed reflect on the coders - and none of them have any reason to feel stressed or depressed about that.

This is one of the best threads I've read here, and TPWC_AIS is already a great mod, and a credit to the coders. Thanks for all the hard work. This mod is already on my 'must-use' list.

Could. Not. Agree. More.

[Robalo 465]

Just a quick query on the ASR AI reset - presumably this would reset all the skills back to their initial ASR starting point.

Could this have negative effects if any of those skills have been degraded during a mission by other factors? I am thinking particularly of courage. Do the AIs' skills take a permanent reduction during missions in normal play?

I did not experiment with this, but I highly doubt the skill levels would ever change dynamically during a mission, unless they're modified on purpose with scripting commands (like the mod here does). You can run a simple test. Without any mods, have some units fight and observe their courage level with this in their init:

this spawn {while {alive _this} do {_this globalChat format ["Unit %1 now has courage level %2",name _this,_this skill "courage"]; sleep 5}}

Edited July 5, 2012 by Robalo
code corrected

[jubxxxjub 10]

Just played it on a MP server with 40 peeps, and UPSMON. It seems either it's conflicting with UPS, or it just doesn't work in MP.

Other than that noice mod.

Edited July 5, 2012 by jubxxxjub