[WIP] bCombat infantry AI Mod [SP]

Is no one else getting bomb spammed with this message? http://i1133.photobucket.com/albums/m594/froggyluv/arma32013-11-0515-52-04-138_zps9a7359c2.jpg

It literally flickers the entire mission and if you can't read it: Undefined Variable: BCombat_Enable

I get these kinds of Undefined Variable errors in literally every AI mod: Tpw/WW etc.. and have no idea why as no other mod is running and I've verified cache via Steam multiple times. Ah well, sorry I can't help test :/

@froggyluv, no i have not seen that error, honestly i dont get any scirpt errors from bcombat, Its a nice thing to see.

If your getting a bunch of undefined variable errors from all mods then you may be right, and have a problem with some core files or something. the last time i had a problem similar, I deleted a few files and then verified my cashe with steam. some times that helps sometimes its not related at all. Goodluck!!

nothing of the like here either..

have you tried but on some other mission? Or that happens no matter the mission at hand?

Happens in all missions I create -yeah must be some corrupt or outdated core file that's not being caught by Steam verify and it's been going on forever. I'll try a full uninstall -reinstall so I don't have to miss out anymore. Cheers!

Happens in all missions I create -yeah must be some corrupt or outdated core file that's not being caught by Steam verify and it's been going on forever. I'll try a full uninstall -reinstall so I don't have to miss out anymore. Cheers!

Check the CBA version you have installed ;)

Currently at work, i'll post later.

I have been testing and the guys have covered everything "wrong" that I have seen so far, I like the faster movement and overall the AI seem to be much more effective than i have experienced either vanilla or with other mods, hopefully I find a glitch or two because I am feeling kind of useless compared to some of the other guys here!

Same here RAINF trying to find time this week to do some testing. Does anyone have a test mission they've created that they would be willing to share?

Same here RAINF trying to find time this week to do some testing. Does anyone have a test mission they've created that they would be willing to share?

+1 for this one, if someone can share his pbo

Check the CBA version you have installed ;)

Currently at work, i'll post later.

Indeed. That did the trick :)

Ill be testing some really high number Ai vs Ai soonish this week.

Good luck with the testing guys, hopefully this can get to full release and MP compatible :D

Thanks!

I just had a test comparing bCombat AI and vanila AI.

I found that in clear open terrain, bCombat AI tend to disperse more and become lonel unit whereas vanilla AI they move too slowly but within range of each other

Not good. Please tell the file names for those .SQF files, i'm failing to reproduce the problem.

Put my RPT in your inbox.

---------- Post added at 22:58 ---------- Previous post was at 22:57 ----------

Also, I finished the last campaign episode mission "Tipping Point" today without a single team casualty, with bCombat providing all of the Ai survivability. I feel like that was how that mission was designed to be played.

Big difference from my first vanilla run-through, when I was the only one to make it to the ridge, let alone through the mortars and paratroopers.

Would it be still possible to test this ai mod ...I have been testing asr,twp,wwai and also with those mods combined I used kronzy urban patrol srcipt...Then I heard about this and it sounds great and would love to give it a go if possible. Thanks

Would it be still possible to test this ai mod ...I have been testing asr,twp,wwai and also with those mods combined I used kronzy urban patrol srcipt...Then I heard about this and it sounds great and would love to give it a go if possible. Thanks

Ok, queued for next release.

Put my RPT in your inbox.

---------- Post added at 22:58 ---------- Previous post was at 22:57 ----------

Also, I finished the last campaign episode mission "Tipping Point" today without a single team casualty, with bCombat providing all of the Ai survivability. I feel like that was how that mission was designed to be played.

Big difference from my first vanilla run-through, when I was the only one to make it to the ridge, let alone through the mortars and paratroopers.

I saw your .rpt. Indeed there are some problems.

Did you enable the mod within configure / extensions menu?

I just had a test comparing bCombat AI and vanila AI.

I found that in clear open terrain, bCombat AI tend to disperse more and become lonel unit whereas vanilla AI they move too slowly but within range of each other

Thx for feedback, this allows me to give some clarifications.

Let's speak of vanilla AI.

There are 2 main reasons for units spreading, while in combat mode:

1) moving to some nearby cover:

2) being ordered to attack something

Actually 1) is just hindered by some cheap stop-and-go behaviour coded within danger.fsm.

This behaviour makes units simply stop on spot for 4 to 8 seconds whenever an enemy is in line-of-sight.

This whatever their position and the situation.

I find this behaviour undesirable for many reasons, but i'll list just two:

1) this way units stop their movement towards safety/cover. They end up firing from potentially vulnerable position (let me say on vanilla they do it all the way).

2) group movement is severely hindered, since group leader will wait for any stopped units to fall back.

For sake of clarity i find this behaviour somewhat acceptable in CQB.

Now, speaking of bCombat, it's not making anything different from vanilla, except for much more careful evaulation of whether to stop a unit or not when some enemy comes in sight.

So, basically, bCombat unleashes the ability for units to move into cover, as well as group mobility.

I think you all have observed this.

This is done under some circumstances, not all the way.

CQB is an exception, in this case bCombat closely mimics vanilla behaviour.

The outcome of "unleashed" behaviour is what you observed.

On open terrain AI groups spread more (scarse cover -> more distance to travel), at least on a same timeframe.

This is actually good in my opinion.

About units moving out of range, this is not coded within bCombat, it's some pathfinding issue arising.

That said, bCombat addresses this problem with an optional enhancement (formation tightening).

It just forces units to fall back when in formation and too far from leader.

This behaviour is disabled by default because i found it collides with the ADVANCE vanilla command and i'm trying to find a way around.

---

Now i'll take a deep breath and will answer to the extensive Lordprimate feedback (http://forums.bistudio.com/showthread.php?167288-bCombat-test-candidates&p=2550396&viewfull=1#post2550396) , which will allow me for further internals disclosure.

That was an interesting insight, thanks.

Here we go.

First thing i would like to thanks Lordprimate for his feedback.

pretty sure were still doing closed tests.

ok, time to report some of my observations.

for starters i launched everything default style, with most differences being noticed in the 50v50AI mission. Honestly it was hard to notice the difference. After a few missions i could see that the AI were Not flanking as effectively, they often hung out too long(getting them killed) at corners facing the wrong dir( facing the wall or object). They were not readily engaging targets. their rules of engagement seriously limited their fire. You could almost see the gears grinding when they had clear line of site, but wouldnt shoot. Often times getting themselves killed because, "they couldnt take the shot" (morale speaking, they had LOS). This would also happen a lot getting them killed.

I won't comment on this now, since feedback is too generic to draw some conclusions.

Assuming all testing was done on default settings, as you said, let me ask some more information:

1) what is the average distance from which units engage within your mission? It makes quite some difference, since AI behaviour does change alot on distance.

2) What kind of waypoynts are given to the groups clashing ? This does matter

3) It would be much important to know the bullets fired / kills ratio for your mission. Please uncomment "call bdetect_fnc_benchmark" at the bottom of bcombat.sqf to gain access to some critical data.

4) what are your average FPS, since bDetect is auto-adaptive and bCombat relies on it.

5) what is the game difficulty you play on

Including what others have said about the team/squad lead going rambo. He(SL/TL) often hangs out in the open to long on first contact, getting himself killed almost every mission almost every time.

Yes, this is being addressed / reverted to stock behaviour.

so then i enabled AI_hearing. Honestly didnt notice a thing. I have seen an enemy kill a team mate of an AI, while they were close.. the surviving friendly team mate never once moved/ looked to where the shooter was, again, often resulting in the death of that unit... The killer would be no more then 20m away gunshots are very audible, however, the AI seems completely unaware that his buddy was just shot by someone directly left/right/front/rear, based on sound. only until the friendly survivor has LOS does he engage(per moral/engagement rules) however, it would seem that if you mis-fire at a distance they will hear that.... kinda wierd..

Strange indeed, care to share some repro?

Hearing does work by revealing some information about shooter to nearby units and switching their behaviour to combat.

Currently they also glance towards the perceived shooting position, if they weren't already in combat mode.

You can check bcombat_fnc_soundalert function within common.sqf, it's pretty straightforward.

Alerted units are not ordered to move anywhere, it would be a bad design choice (potentially mission breaking).

That said if you give a guard waypoint to groups supposed to be alerted by sound, they may look for the shooter.

then i enabled fast_turn. because of above i thought it might have been a "turning to enguage" issue. Enabled, I noticed that the AI would Twitch, tward their target. and by twitch i mean. they would turn, then suddenly snap back to their previous pos(+ a few deg), then turn again and snap back(again close to their original angle, plus a few degrees). Once disabled this behavior dissipered. something is buggy in your turning function.

Fast turning is just in its prototypical form, hence it was disabled by default.

I will look into this.

then i enabled fast_move for rushing. AI were borderline suicidal, turned it back of.

"Fast move" is a test-feature designed to make units move individually and groups spreading more.

Its benefits are for assault on exposed terrain or under artillery fire.

It's not an all-around enhancement, the option to enable or disable it should be left to mission maker (who will be able to override bCombat config).

It has to be fine-tuned, as well as most mod features.

That said, try mission Assault on Charkia and you'll hopefully see how this behaviour may be useful already.

I then enabled tight_formation. AI would be way to close to each other. as if they were constantly put in "FILE" formation. They would all get mowed down. turned it off

All that formation tightening does is commanding far units back into formation.

It does not affect formation itself nor makes units any closer than vanilla settings.

It wasnt until i enabled debug balls that i saw something to play with. The supression settings, needed to be tweekd. IMHO

I ended up lowering most of the values. What i noticed was that a small firefight would completely drain their moral. I would see a lot of red debug balls, sometimes they would turn purple , the ai would run away (away from the enemy ) and then after a few hundred meters their debug ball would turn white again. As if they were in a short tactical retreat to regain moral. then they would come back to the fight ( i have fleeing and surrender Disabled).

most of my supression values are now:

// Suppression-related params

bcombat_penalty_bullet = 0.25

bcombat_penalty_flanking = 1

bcombat_penalty_scream = 0.25

bcombat_penalty_enemy_unknown = 5

bcombat_penalty_enemy_contact = 10

bcombat_penalty_enemy_close = -10

bcombat_penalty_casualty = 25

bcombat_penalty_wounded = 5

bcombat_penalty_fire = .25

bcombat_penalty_recovery = 1

Sure settings have to be tweaked and that will be next phase.

I've introduced configurable suppression settings in order to make anybody comfortable with them.

It's probably too soon to give insight on these (i avoided doing that on purpose till now), but since you are already fiddling with this please just consider the following head-ups:

* Some settings are being rounded. For example there's no actual difference in putting bcombat_penalty_bullet=0.25 or 1. Same applies to bcombat_penalty_recovery. It definitely may be debatable whether to round them or not, but currently that's how it works.

* bcombat_penalty_flanking adds to bcombat_penalty_bullet. By putting bcombat_penalty_flanking = 1 and bcombat_penalty_bullet = 0.25 you make a bullet fired from behind add 4x the original morale hit.

* bcombat_penalty_enemy_unknown adds to bcombat_penalty_bullet as well. Given your config a bullet fired from an unknown shooter add 20x the original morale hit.

* bcombat_penalty_scream is to be removed, as it looks useless (the unit screaming is the one being wounded).

* bcombat_penalty_enemy_close = -10 sounds wrong to me, are you really supposing units to gain any morale from seeing any nearby enemies?

so far i see that this has GREAT potential!!!!! In general i can see the rules of enguagement hindering AI, in some situations. I have seen many times the AI had Clear LOS. but didnt take the shot. I imagine that it was a moral issue. but after enableing the debug i could see that some of them were a %100 percent moral and they wouldnt shoot to save themselves.. and often died or got a team mate killed by not killing the enemy.

I can see how this might "slow ai down" as they tend to hang back instead of keep moving (as soon as their in contact)

Sure finding the sweet spot between returning fire and moving is the biggest challenge, in terms of balancing.

That said i'm not overly concerned about a AI unit failing to open fired onto an enemy in LOS while it has other options (move to destination, move to cover, flee ...).

May i spend some final lines on this mod philosophy ?

Players usually consider opening fire on sight as the best option, but in my opinion that's a legacy of their gamers' background.

Humans do not usually react this way when threatened. It's called self-preservation and makes them unpredictable.

Despite this being the apparent rule in mods, i hardly think making AI more deadly would be, by itself, an enhancement.

Infact i'm putting the same efforts in making AI better as well as worse, depending on some criteria.

Vanilla AI is (still) robots with guns.

You can't really apply tactics against robots.

Think about it, you end up exploiting their (programming) weaknesses or trying brute force.

I'm bored playing this way.

I would like to have those robots sense fear.

Fear leads to mistakes.

Mistakes make into greater tactical opportunity for the opponent.

Cheers.

"The supreme art of war is to subdue the enemy without fighting" - Sun Tzu

May i spend some final lines on this mod philosophy ?

Players usually consider opening fire on sight as the best option, but in my opinion that's a legacy of their gamers' background.

Humans do not usually react this way when threatened. It's called self-preservation and makes them unpredictable.

Despite this being the apparent rule in mods, i hardly think making AI more deadly would be, by itself, an enhancement.

Infact i'm putting the same efforts in making AI better as well as worse, depending on some criteria.

Vanilla AI is (still) robots with guns.

You can't really apply tactics against robots.

Think about it, you end up exploiting their (programming) weaknesses or trying brute force.

I'm bored playing this way.

I would like to have those robots sense fear.

Fear leads to mistakes.

Mistakes make into greater tactical opportunity for the opponent.

Cheers.

"The supreme art of war is to subdue the enemy without fighting" - Sun Tzu

I like your philosophy a lot AND I'm bored playing the way you described too :)

Thanks for your mod !!!

Thanks for the opportunity Fab!

I can answer a few things now and implement the others and get back.

1) In the City average distance is under 300m. Most open field enguagements happen further apart could be about 500 between them.

2) I started with "MOVE", with no other waypoint mods. I noticed a few ai would reach the waypoint (after the main battle) and declair the area all clear. only to get ambushed. It was funny. Then i changed to "SearchAndDestroy". so the ai would look around for a little while once they reached the way point. so instead of a few ai from each team standing around at each endpoint. there would only be 1-2 left from each side evading and searching respectively.

3) ok, will do.

4) around 20 with the 50v50.

5) when making/testing missions i use veteran for 3rd person when sitting down and playing a mission its Elite.

6) I basically have 5 groups of 10 ai with one waypoint that is in the same place as the opposing forces waypoint. I move them around the map changing the distance between them. IE. in an open field ill spread them apart more because theres nothing to block LOS. In a city setting or hilly area i may have the closer. Most of the time this happens in cities. as the the group moves the ai cover in the direction of movement. An AI friendly may move out of the flow and around and make contact. He dies, the friendly at the wall continues to look in the direction the group is moving. If by design they just glance "move their head" then it may very well work fine. I just thought they would turn their bodies at least. An assumption on my part. And i dont know about this common.sqf, ill look into it.

7) shared my results, your looking into it.

8) they were spartans what can i say. :) in the city they would rush around buildings , threw buildins expose theselves and get killed. in the open they would rush toward the enemy running into bullets getting mowed down. basically whoever opened fire first slaughtered the enemy in my tests. ill try that mission out with it enabled. this is purely feedback, i know there is fine tweaking to be done. I will try that mission out with bcombat and fast move enabled.

9) i wouldnt lie i swear this happened!! the stacked up like they were put into file formation. when moveing they would move like they were in file. really close to eachother.

10) Thank you for the explaination!! i thought it was the max % that was subtracted from 100 and nothing was mutlipied by each other. and i do believe they are being rounded because well a enemy bulled didnt cause that much reduction in moral. I just wanted to share that turning them down from default seemed to smooth things out a lil bit. i had increased the settings a little bit. that mission was a dissaster.. one team member was killed the whole group was just "scared", all their debugl balls were yellow and red. some turned purple they tried to run but were mowed down. It sounds like this is going to be a part of a userconfig file, I hope so. once again just wanted to give feedback i know that were just testing. :)

I most certainly hope that i did not convey that i dont know the philosophy behind the mod. Again i can see the potential and know it comes down to tuning , scripting and tweeking. sometimes i blather on in my explanations and blurr the focal point of my feedback. Thanks again for explaining the details of the suppression settings. Im tooling around with tpwcas right now after that ill go and implement those changes and play with bcombat some more.

I would like to get / test your mission, is that possible?

Then just a few more explanations, below (sorry, another wall of text).

1) In the City average distance is under 300m. Most open field enguagements happen further apart could be about 500 between them.

Then we get 2 different scenarios there, 150m. distance being the actual threshold between them.

150m-0m:

As of v0.11:

unit behave live vanilla, they prioritize fire against movement.

Any targets of opportunity are engaged, there's a probability they skip them only if suffering a morale hit.

They do return fire, with probability proportional to skill and inversely proportional to morale hit.

Many other criteria act on return fire logic, i won't go much in depth now.

>=500m-150m:

units prioritize movement against fire.

Units do return fire, as above.

Units don't engage targets of opportunity unless:

As of v0.11:

* their current command != "MOVE" (e.g. ATTACK)

* their combat mode is RED (fire at will)

Added in v.012:

* they're group leader (as of v0.12)

Planned /evaluated:

* they're NOT moving and onto their planned destination (in cover / hidden)

* their vector of movement is incident with line of fire ( trigger bening probabilistic and / or angle threshold )

----

Here is the full picture:

in CQB units act like vanilla ( except they lock targets more efficiently, hence they should open fire faster ).

On medium to long range mobility is prioritized and they usually avoid attacking targets of opportunity, since lowered firing accuracy makes movement advantageous.

Accuracy is a key factor here.

As it gets too high, no kind of movement can be risked on the battlefield, hence no tactics (this plagued ArmA2).

bCombat itself reduces accuracy:

* statically: capping its max value within CfgAISkill to about 70%

* dynamically: morale hits reduce firing accuracy

So basically we do reduce accuracy in order to make units able to maneuver at range (better stance handling helps achieving the same).

Maneuvering is crucial in many cases, such as:

* ASSAULTING: Closing onto enemy faster, reducing the time being exposed to fire while traversing terrain.

* KEEPING ON ROUTE: moving towards destination waypoint as needed, without being kept in undesired skirmishes (Vanilla ArmA3 fails here, engaging at any times).

* FLANKING

* WITHDRAWING

* FLEEING

* ...

I think the overall concept is correct.

Sure it has to be balanced, for instance we will have to tweak the 150m. threshold.

At the same time we also SHOULD weight carefully the effect of firing accuracy on movement.

This is the reason i asked your difficulty settings as well as bullets / kill ratio.

By experience, i found that 20/1 to 30/1 ratio is normally healthy for medium range up to 300m., with mobility being possible.

Lower ratios would progressively disrupt mobility.

2) I started with "MOVE", with no other waypoint mods. I noticed a few ai would reach the waypoint (after the main battle) and declair the area all clear. only to get ambushed. It was funny. Then i changed to "SearchAndDestroy". so the ai would look around for a little while once they reached the way point. so instead of a few ai from each team standing around at each endpoint. there would only be 1-2 left from each side evading and searching respectively.

In my opinion MOVE should really mean MOVE: reaching (safe) destination being priority.

Groups with a MOVE waypoint should possibly avoid derailing into off-route clashes, unless you give them RED combat mode.

S&D is probably the best way to setup attacking groups.

4) around 20 with the 50v50.

20 FPS average is not far from bDetect min. setting (15FPS).

Any low spiking would potentially trigger features degradation.

If by design they just glance "move their head" then it may very well work fine. I just thought they would turn their bodies at least. An assumption on my part.

They use glanceAt all the way, it does quite some wonders. I think it's easy to spot them doing this, they look around a lot.

Turning body leads to glitches when units are prone, so normally i avoid using LookAt / DoWatch whenever possible (the former especially).

---------- Post added at 16:42 ---------- Previous post was at 16:04 ----------

+1 for this one, if someone can share his pbo

Why not running Assault on Charkia?

It's a simple yet very well designed mission you can get on Steam workshop.

It's my favourite testing mission at the moment, since it features both open terrain as well as CQB and many AI groups (about 80 units).

It's quite rewarding to play it as squad leader, with bCombat (please enable fast move).

Personally i play it this way and i don't have to micromanage my squaddies anymore.

The only glitches i've seen so far are pathfinding issues within the city.

That means sometimes your units fail to regroup.

It's not bCombat fault.

Why not running Assault on Charkia?

It's a simple yet very well designed mission you can get on Steam workshop.

It's my favourite testing mission at the moment, since it features both open terrain as well as CQB and many AI groups (about 80 units).

It's quite rewarding to play it as squad leader, with bCombat (please enable fast move).

Personally i play it this way and i don't have to micromanage my squaddies anymore.

The only glitches i've seen so far are pathfinding issues within the city.

That means sometimes your units fail to regroup.

It's not bCombat fault.

I'll give it a try tonight.

Thanks

Interesting DEV branch update today:

http://forums.bistudio.com/showthread.php?149636-Development-Branch-Changelog&p=2552265&viewfull=1#post2552265

Is anyone using exThreads param?

If so BIS fixed today a bug causing command scriptDone to fail.

That bug was affecting bCombat as well ... so you'd better update.

If anymore testers are required, I'm sure some of our guys would be glad to help. Currently we're experimenting with various AI scripts on our public server, but there are still many features I would like to see implemented (which hopefully bCombat has).

You can contact me via PM ;)

If anymore testers are required, I'm sure some of our guys would be glad to help. Currently we're experimenting with various AI scripts on our public server, but there are still many features I would like to see implemented (which hopefully bCombat has).

You can contact me via PM ;)

GDSN you understand this is for SP only, currently?

---------- Post added at 18:39 ---------- Previous post was at 18:37 ----------

Testing v0.12. So far so good.

The reverted behaviour for group leaders changed overall behaviour quite a bit.

Groups do move much slower.

Early to say whether that's good or bad.

Updated (candidate) changelog:

v0.12
-----

Core:

* Better filtering of BIS danger.fsm events. 
* Distance limit of 500m. from source for danger.fsm events collection. 
* AI units may open fire upon on targets of opportunity from distance, if:

	* target is near unit's planned destination ( < 50m )
	* target is near group's current waypoint position ( < 50m )
	* units are "hidden" and not moving

* Disabled vanilla fleeing behavior (creating mayhem) when bcombat_allow_fleeing = false
* Minor tweaks and bugfixes.

Balancing:

* Lessening of morale hits, about 15-20% in average.

Gameplay:

* AI Group leaders reverted to default fire vs. movement prioritization. They move as in vanilla, except for morale effects. This was done to avoid them rushing. No changes to sub-formation leaders.
* AI Units positioned over terrain level (e.g. guard posts, some buildings) should prefer crouching instead of going prone, when fired upon.
* "Fast move" enabled by default on Player led groups. Should make units under player command move faster while in combat mode.

Performance:

* Code optimization for functions (only)
* Faster / lighter line-of-sight checks

I didnt know you can use higher numbers in exThreads then 7 ... It mentions 8. whats the point?

GDSN you understand this is for SP only, currently?

Yes fabrizio, that will not be a problem.