Jump to content
nkenny

The effect of weapon on AI accuracy

Recommended Posts

The effect of weapon on AI accuracy

by nkenny

Abstract*
Testing the effect of weapon on AI accuracy. 

Introduction
The weapon an AI soldier is equipped with greatly affects killing power.  Even when the class of weapon is nominally the same, weapon configuration plays enormous part. This is especially evident when similar weapons from different mods are used together.  To explore the effect of weapon on AI killing power I built an experimental suite. 

Experiment
The test was conducted while running only CBA, CUP Weapons, RHS Russian Armed Forces and NIArms. The player profile was newly created and no AI skill tweaks were made. 

Description
I placed a single shooter on flat terrain, back turned towards a four man group of enemies. The four man team is concealed from the shooter (his back is turned). Each soldier is identically equipped with no body armour or helmet or weapon. 

After five seconds the shooter is given a weapon and given perfect information about the enemy. This results in him turning and engaging the enemy. The targets are spaced at two meter intervals and the test was run on a 100m and 200m range (normal combat engagement ranges).  Shooter and targets were forced to remain static and remain in a standing position. My point of comparison was the AKM rifle. Ubiquitous in enemy hands and mod packs alike.  I provide data for the modern MX rifle with a HAMR RCO as a frame of reference. 

The exercise is over once three minutes has passed, the shooter has expended all ammunition, or all targets are dead.  Measurements are printed on screen and recorded on the clipboard. I paste that information into a document. 

Measurements 
I ran the exercise six times with each weapon. While not enough for statistical evaluation, I nonetheless deemed it sufficient to get a sense of the performance characteristics of each weapon. 
Hits, each bullet that impacts a target.  
Shots fired, maximum is 6 magazines: 180 rounds. 
Accuracy, hits divided by shots fired. 
Time, time in minutes and seconds. 

Record
After each round I recorded weapon class name, Accuracy, (hits / shots fired) and time. Each measurement is averaged (in red). 

Expectations
Mod configuration will vary somewhat. With CUP weapons being more accurate than RHS counterparts. 

Data @ 100 Meters

Spoiler

// 100M Vanilla
wpn: arifle_AKM_F - acc: 50 percent (5/10) - time: 00:05
wpn: arifle_AKM_F - acc: 57 percent (4/7) - time: 00:04
wpn: arifle_AKM_F - acc: 57 percent (4/7) - time: 00:05
wpn: arifle_AKM_F - acc: 46 percent (6/13) - time: 00:07
wpn: arifle_AKM_F - acc: 38 percent (5/13) - time: 00:07
wpn: arifle_AKM_F - acc: 61 percent (8/13) - time: 00:07
  51% accuracy -- 10.5 shots -- 6 seconds

// 100M CUP
wpn: CUP_arifle_AKM - acc: 50 percent (4/8) - time: 00:07
wpn: CUP_arifle_AKM - acc: 38 percent (5/13) - time: 00:10
wpn: CUP_arifle_AKM - acc: 50 percent (4/8) - time: 00:06
wpn: CUP_arifle_AKM - acc: 50 percent (4/8) - time: 00:07
wpn: CUP_arifle_AKM - acc: 30 percent (4/13) - time: 00:09
wpn: CUP_arifle_AKM - acc: 41 percent (5/12) - time: 00:08
  43% accuracy -- 10.3 shots -- 8 seconds  

// 100M RHS 
wpn: rhs_weap_akm - acc: 27 percent (6/22) - time: 00:11
wpn: rhs_weap_akm - acc: 19 percent (9/47) - time: 00:28
wpn: rhs_weap_akm - acc: 12 percent (6/48) - time: 00:29
wpn: rhs_weap_akm - acc: 14 percent (7/47) - time: 00:29
wpn: rhs_weap_akm - acc: 25 percent (10/40) - time: 00:26
wpn: rhs_weap_akm - acc: 12 percent (4/32) - time: 00:21
  18% accuracy -- 39.3 shots -- 24 seconds

// 100M NIArms
wpn: hlc_rifle_akm - acc: 40 percent (4/10) - time: 00:06 
wpn: hlc_rifle_akm - acc: 16 percent (4/24) - time: 00:09
wpn: hlc_rifle_akm - acc: 40 percent (4/10) - time: 00:06
wpn: hlc_rifle_akm - acc: 50 percent (4/8) - time: 00:05
wpn: hlc_rifle_akm - acc: 36 percent (4/11) - time: 00:05
wpn: hlc_rifle_akm - acc: 23 percent (4/17) - time: 00:08
  34% accuracy -- 13.3 shots -- 6.5 seconds




Data @ 200 Meters

Spoiler

// 200M VANILLA
wpn: arifle_AKM_F - acc: 9 percent (4/41) - time: 00:47
wpn: arifle_AKM_F - acc: 27 percent (6/22) - time: 00:23
wpn: arifle_AKM_F - acc: 15 percent (7/45) - time: 00:49
wpn: arifle_AKM_F - acc: 14 percent (8/56) - time: 00:59
wpn: arifle_AKM_F - acc: 14 percent (4/27) - time: 00:28
wpn: arifle_AKM_F - acc: 10 percent (7/69) - time: 01:17
  15% accuracy -- 13.3 shots -- 47 seconds

// 200M CUP
wpn: CUP_arifle_AKM - acc: 13 percent (4/30) - time: 00:31
wpn: CUP_arifle_AKM - acc: 33 percent (4/12) - time: 00:15
wpn: CUP_arifle_AKM - acc: 50 percent (4/8) - time: 00:09
wpn: CUP_arifle_AKM - acc: 23 percent (5/21) - time: 00:22
wpn: CUP_arifle_AKM - acc: 18 percent (4/22) - time: 00:25
wpn: CUP_arifle_AKM - acc: 28 percent (7/25) - time: 00:26
  27% accuracy -- 19.6 shots -- 21 seconds

// 200M RHS
wpn: rhs_weap_akm - acc: 4 percent (6/131) - time: 01:57  
wpn: rhs_weap_akm - acc: 3 percent (5/156) - time: 02:05
wpn: rhs_weap_akm - acc: 8 percent (5/62) - time: 00:50
wpn: rhs_weap_akm - acc: 5 percent (8/158) - time: 01:51
wpn: rhs_weap_akm - acc: 5 percent (7/130) - time: 01:46
wpn: rhs_weap_akm - acc: 2 percent (5/180) - time: 02:36  ** RAN OUT OF AMMO ** 
  4.5% accuracy -- 136 shots -- 111 seconds

// 200M NIArms
wpn: hlc_rifle_akm - acc: 60 percent (6/10) - time: 00:12    
wpn: hlc_rifle_akm - acc: 30 percent (4/13) - time: 00:15
wpn: hlc_rifle_akm - acc: 38 percent (7/18) - time: 00:22
wpn: hlc_rifle_akm - acc: 66 percent (6/9) - time: 00:12
wpn: hlc_rifle_akm - acc: 71 percent (5/7) - time: 00:09
wpn: hlc_rifle_akm - acc: 77 percent (7/9) - time: 00:12
  57% accuracy -- 11 shots -- 13.6 seconds

// 200M VANILLA MX with HAMR RCO
wpn: arifle_MX_Hamr_pointer_F - acc: 40 percent (4/10) - time: 00:12
wpn: arifle_MX_Hamr_pointer_F - acc: 36 percent (4/11) - time: 00:13
wpn: arifle_MX_Hamr_pointer_F - acc: 42 percent (6/14) - time: 00:14
wpn: arifle_MX_Hamr_pointer_F - acc: 31 percent (5/16) - time: 00:17
wpn: arifle_MX_Hamr_pointer_F - acc: 50 percent (4/8) - time: 00:09
wpn: arifle_MX_Hamr_pointer_F - acc: 70 percent (7/10) - time: 00:11
  45% accuracy -- 11.5 shots -- 12.6 seconds



Findings
@100 Meters
At close range  the vanilla AKM and CUP AKM are closely matched. With a slight edge to the vanilla gun. RHS weapons are considerably less accurate than all other counterparts and require four times the amount of time and rounds fired for similar effect. The numbers for the NIArms AKM belie their effect on target. Toadie's AK tended to fire in bursts, but with a highly accurate first shot. In effect the gun killed with the first shot, but burned off a few rounds into the air-- wasting time and bullets. 

@200 Meters
At mid range the differences between the weapons become evident. CUP weapons are almost twice as effective as their vanilla counterpart. RHS weapons are considerably less accurate than the others. Which is not to say ineffective. Rate of fire is high and spread provides greater coverage-- area of fire-- than the other counterparts. As for NIArms: the rifle comparatively is laser accurate, out shooting all other alternatives. Contrasted to RHS it provides almost eight times more killing potential. It even compares favourably to the vanilla MX with a optic. 

Conclusion
Not all AKMs are created equal. The killing potential of each gun varies wildly from mod to mod.  The most deadly close quarters AK is the vanilla, while in most combat ranges, the NIArms gun outperforms all other alternatives. 

 

One obvious take away is that there are radical differences of capability with two presumably identical weapon systems.  This is one reason why I would advice  consistency in mod sets. Another is that this vast difference in capability need not match the players performance, expectations and experience, with the same weapon system.  In the case of the AKM, I believe most players find them very similar-- aside from minor differences in sight picture and sound.  In fact to test that very thing I used the same suite to record my own times. I won't post the results here, but they were largely similar across mods. 

 

The test itself remains neutral to the intended design intention of the style of configuration each mod adopts.  The comparable lack of accuracy of RHS weapons need not be read as a flaw.  There are arguments both from the perspective of realism and game play to prefer less accurate weapons. 
Firstly, the sort of 'white room' engagement against a non-responsive, non-moving target is hardly applicable. Handling performance on sterile ranges fail to convey the stress of combat shooting. Most studies into the effect of combat on marksmanship show a massive decrease in effective accuracy. RHS weapons simulate combat shooting, rather than target plinking.
Secondly, more shots fired and fewer one-shot-kills, is generally good for game play. At 100 meters vanilla, CUP and NIArms equipped soldiers can eliminate an inattentive fire team in less than 10 seconds.  While demanding some caution is good, such a quick end also means a quick end to the session for the player in question. 

- Thirdly, RHS comes with an entire dimension of weapons. Accuracy comes with modern quality. A RHS M4A1 is more accurate than an AKM. So is a an AK74. See references.

 

Many communities depress AI accuracy values. The considerable accuracy of CUP and NIArms weapons need not be a flaw. Follow the links in the references and one can see the extreme degradation of accuracy based suppression levels. The considerable default accuracy of vanilla weapons, the presence of multiple stances, all are suggestive of a very specific type of gameplay.  Soldiers should be cautious and reveal their positions only momentary.  Suppressive fire is very effective. It is more important to suppress all dangerous enemies than it is to pursue time consuming killing shots!  When the AI is accused of laser accuracy, it will in some cases be due to failure to adequately suppress-- through fire superiority-- the enemy position. 

 

Finally, I suspect, particularly in regards to the user made modifications there are considerably differences.  Test are done in different AI setting ecologies-- tweaking AI skill is easily (if somewhat obtusely) done. If I shall brave a conjecture I expect CUP weapons, being added more along open source model will show greater variance in configuration.  NIArms weapons are developed with an eye towards delivering an uncompromising experience for the shooter. The AI comes much later.  All of this means that mod makers and mission designers must pay attention to the source of the weapons given to  the AI. 


Future tests
1. The same platform can easily be used to test other weapons or family of weapons
2. In the future I would like to test ranges out to 300 and 400 meters**

 

References

- Original thread on nopryl.no, further numbers, weapons and experiments are here

- Experimental mission, links to mission for your own testing


-k 
 

* Dear mods: Feel free to move this thread. While relevant to mission makers and mod developers equally. It seems sufficiently meta  to concern general players, but it is not pointed enough to belong to the mod discussion forum. 
** My initial run with standing RHS weapons yielded one hit across 180 rounds. In other words. If faced by a lone, standing gunman armed with a RHS AKM. At 300+ meters you might as well remain stationary and standing yourself. 

  • Like 8
  • Thanks 2

Share this post


Link to post
Share on other sites

Fantastic. This kinda testing is just so important - i had no idea different mods had different dispersion for same weapons. That was one of my turnoffs to the game a while back in which low skilled Insurgents using AK's were completely dominating high skilled spec ops (5.56). Really giving no reason not to arm everyone with 7.62 arms which was kinda depressing

Share this post


Link to post
Share on other sites

I don't know jack about scripting to build a proper test but I do have an interest in learning what makes bot kill faster/slower.

 

As far as I know even vanilla weapon has some wildly inconsistent configuration for AI accuracy. I guess aiDispersionCoef probably has the most significant effect, and they're configured to have such massive gaps between weapon classes(sometimes within nominally the same class) that it totally outweighs base weapon dispersion. Marksman rifles generally have a aiDispersionCoefX/Y value of 2*3, original vanilla rifle of 4*6(4x greater area than markman rifle!), however most APEX DLC addition has a 6*6 for reasons unknown, including SPAR-17 which you might consider a marksman rifle otherwise. Machineguns are on the otherhand in a league of its own with aiDispersionCoef on the order of 20s in both axis, in turn means they have rather limited actual killing potential under vast majority of circumstances.

 

At higher unit skill/AI aim level it starts to equalize across the board as aiDispersionCoef eventually goes to 1 in which case weapons are treated as exactly what they are. Machineguns are still hindered by their burst behavior which reduces the fire rate of actual aimed shots, since only the first shot within a burst has chance to hit(I don't observe any meaningful recoil compensation even at higher skill). But relatively speaking some become quite a bit better like SPMG with impressive dispersion.

 

The information regarding to suppression is new to me but from the interactions between AIs I've seen, all else being equal machineguns will be consistently beat by average rifles in the hands of AI, in a straight up duel type of scenario at least. Whether they could provide any benefit that outweighs the abysmal lethality in a unit is unclear to me. But if I were to play dress up doll with my AI team, I'd favor very heavily on weapons configure as marksman rifle(CMR-76/Mk14 in, SPAR-17/MXM out). Short of that weapons with favorable ballistics in the assault rifle category.

 

Just from gameplay I was kinda under the impression that optics on AI doesn't really meaningfully affect their accuracy/"time to kill", but mostly the range they're willing to engage, and their ability to maintain track on a already detected contact. Will you be able to point to me where in the config does it specify it's effect on AI accuracy?

 

And did you investigate stances? resting(don't believe its a feature AI can make use of)? I'm under the impression they don't really matter to AI anymore.

Share this post


Link to post
Share on other sites

A tip for the investigator: see if you make graphs from your quantitative data. They are easier to read, you can spot differences immediately, and hopefully draw more people to the discussion. Nobody (or is it just me?) likes reading lists of numbers.. 🙂

  • Like 1

Share this post


Link to post
Share on other sites

@mmm

Interesting. I did not delve into the dark arts that is the skill settings of Arma3.  Instead I wanted to run a baseline of sorts were ostensibly identical weapons were compared in similar situations.  Repeating the experiment with the AI skill set to 0.25, 0.5, 0.75 and 1.0 would perhaps yield interesting data. 

 

To respond to some of your questions and input. 

 

1. Suppression dramatically decreases the effective accuracy of the soldier. Follow the references for more numbers.  In regards to automatic rifles vs marksman rifles. I ran some tests where I compared a standing AKM shooter to a standing machine gunner.  While I did not perform a full experiment, the general trend was that standing at 100-200 meters, assault rifles are more effective. Machine guns will consistently overshoot the target. Of course, this is a white room experiment, and not a duel environment with cover, concealment and multiple stances. I will agree to your other assessment.  My limited tests with marksman rifles show that these, particularly paired with optics, are very deadly. 

2. The effect of weapon optics on AI is considerable. An MX rifle equipped with a HAMR optic is nearly twice as efficient as a bare bones rifle (Source). As I'm not sat before an Arma capable computer at the moment, I can't point you to the config values. But I do recall there are different shooting (or recoil?)  patterns for optics. Let me get back to you on that. 

3. I have not checked the stances! It is a very interesting area. Still this is easy to do within the testing suite.  It is my belief that weapon resting and bipod use are not considered for AI. But the prone stance in particular is.  I did run some tests at 300 meters, unrecorded, but stopped because the RHS AKM was simply unable to score more than a single hit. Setting the unit to prone resulted in two hits.  Repeating the test across a wider range of distances and stances is next on my todo list. 

Share this post


Link to post
Share on other sites

@froggyluv Thanks! I suspected seeing some data would be useful for mission and mod designers and players alike.  It isn't as bad as you might think. Within an ecology of vanilla weapons, the AKM is markedly inferior to both the MX and SPAR15 rifles, on all ranges. Add RCOs and the modern rifles pull out ahead. Note also that the AK12 is intrinsically more accurate than the AKM.   That said, vanilla weapons are very accurate overall. It may be difficult to distinguished between being one shotted in six seconds rather than four point five...

 

@joostsidy Good point. I will think about it. My reasoning for avoiding it has been this: the sample size is quite small. I therefore wanted to avoid giving the impression that this experiment is suitable for statistical numbers mathemagics

 

-k 

Edited by nkenny
clarity
  • Like 1

Share this post


Link to post
Share on other sites

@nkenny

Well according to THIS aiDispersion will only truly equalize at 1 skill. Lower than that I will assume it will use a multiplier still larger than 1 proportional to the aiDispersion value. In my test to be safe I put the AI aim in difficulty to 1 as well. Also since it's a multiplier to weapon dispersion it informs you more about how weapons are configured for AI use, rather than weapons themselves. Basically the "unfairly" configured weapon catches up at 1 skill because the artificial inaccuracy is removed.

 

One thing I have no mean to properly investigate is recoil compensation for automatic fire. Maybe you can try some tracer effect at different skill level/stances?

 

Edit:

With regards to scope, I don't mean to dispute your test result as it is more scientific than I could ever be. And I'd like to believe scopes make some difference to AI accuracy. But I wonder if it is actually in the config, maybe it's pulled number from the optics FoV and hard coded to apply bonus to AI? Again I know nothing about it just spewing things I don't understand. Just don't seem the weapon's "AI firemode" in the cfgweapons has anything indicating an accuracy change.

 

As for stance effect. I could say, from my experience in gameplay, prone definitely HAD an effect in earlier days of ARMA 3. The focused streams of tracer bullets coming your way from an AI automatic riflemen in prone position, watched in slow motion was quite unforgettable. Even assault rifles delivered more rapid semi auto fire from prone. However It may have changed at the time or sometime after the Marksmen update(along with the excessive dispersion multiplier), as of today I don't observe any tangible effect from stances.

  • Like 1

Share this post


Link to post
Share on other sites

Good work. I think the next thing would ideally be to work out exactly what config values influence these variations such that a script might be run on the whole config tree and spit out a large result set (which could be graphed as suggested above). In as much as you've identified a potential 'problem' - this might yield its solution, whether by allowing mod makers to see how their configs compare with others or mission makers to see what weapon sets inter-operate well.

Share this post


Link to post
Share on other sites

This is a great start.

 

I'm having trouble also with the distance at which AI shoot.  Have you tried AI shooting beyond 750 meters, which seems to be a hard stop?  AI with sniper rifles, DMRs, and machineguns should easily fire beyond 750 meters, but they do not.  I've been shooting AI beyond 750 meters and they never fire back, even if you're standing in plain site.  It seems the cfgWeapons files have distance settings.  These were modified for a few weapons by the "Duckhunting" mod on Armaholic.  But that mod is 3 years old and doesn't seem to work anymore.

Share this post


Link to post
Share on other sites

AK12 in various stances
So I have rerun the original experiment with the vanilla AK12 across a wide variety of parameters, stances and distances.  The number one thing demonstrated was the unsuitability of the experimental framework.  While adequate for demonstrating categorical differences between ostensibly identical weapons between mods, it has proven unhelpful in showing the nuisance of stance and weapon attachment. The sample size being small, also the main reason I do not provide data in pretty columns and graphs, simply does not present an adequate foundation for consistent comparison.  That said, the test still showed some things. Hence this post. 

 

Method and measurements

The experimental suite is unchanged from what was presented in the first post.  To make sense of the data I decided to make a measurement of  aggregate potential.  The formula I settled on was this:  effective kills per second multiplied by time left (180 seconds)  This measurement, while not perfect, gives some indication of how the weapon could perform in perfect conditions. The conclusion will cover my findings in more detail. 

 

Data @ 100 Meters (TL;DR in conclusion)

Spoiler

 

// 100M Standing
wpn: arifle_AK12_F - acc: 50 percent (4/8) - time: 00:05
wpn: arifle_AK12_F - acc: 57 percent (4/7) - time: 00:05
wpn: arifle_AK12_F - acc: 40 percent (4/10) - time: 00:06
wpn: arifle_AK12_F - acc: 56 percent (9/16) - time: 00:09
wpn: arifle_AK12_F - acc: 60 percent (6/10) - time: 00:07
wpn: arifle_AK12_F - acc: 62 percent (5/8) - time: 00:04
  54% accuracy --10  shots --6  seconds

 

// 100M Crouched
wpn: arifle_AK12_F - acc: 66 percent (4/6) - time: 00:04            
wpn: arifle_AK12_F - acc: 44 percent (4/9) - time: 00:06
wpn: arifle_AK12_F - acc: 45 percent (5/11) - time: 00:06
wpn: arifle_AK12_F - acc: 80 percent (4/5) - time: 00:03
wpn: arifle_AK12_F - acc: 57 percent (4/7) - time: 00:04
wpn: arifle_AK12_F - acc: 50 percent (4/8) - time: 00:05
  57% accuracy --7,5  shots --4,6  seconds

 

// 100M Prone
wpn: arifle_AK12_F - acc: 40 percent (4/10) - time: 00:07
wpn: arifle_AK12_F - acc: 44 percent (4/9) - time: 00:05
wpn: arifle_AK12_F - acc: 50 percent (4/8) - time: 00:05
wpn: arifle_AK12_F - acc: 60 percent (6/10) - time: 00:06
wpn: arifle_AK12_F - acc: 45 percent (5/11) - time: 00:07
wpn: arifle_AK12_F - acc: 62 percent (5/8) - time: 00:06
  50% accuracy --9  shots --6  seconds

 

 

Data @ 200 Meters

Spoiler

 

// 200M Standing
wpn: arifle_AK12_F - acc: 37 percent (6/16) - time: 00:16    
wpn: arifle_AK12_F - acc: 55 percent (5/9) - time: 00:10
wpn: arifle_AK12_F - acc: 36 percent (9/25) - time: 00:27
wpn: arifle_AK12_F - acc: 29 percent (7/24) - time: 00:26
wpn: arifle_AK12_F - acc: 33 percent (7/21) - time: 00:22
wpn: arifle_AK12_F - acc: 41 percent (5/12) - time: 00:14
  39% accuracy --18  shots --19  seconds


// 200M Crouched
wpn: arifle_AK12_F - acc: 28 percent (4/14) - time: 00:15
wpn: arifle_AK12_F - acc: 35 percent (5/14) - time: 00:15
wpn: arifle_AK12_F - acc: 25 percent (5/20) - time: 00:22
wpn: arifle_AK12_F - acc: 60 percent (6/10) - time: 00:11
wpn: arifle_AK12_F - acc: 41 percent (5/12) - time: 00:14
wpn: arifle_AK12_F - acc: 40 percent (4/10) - time: 00:11
  38% accuracy --13 shots --14,6  seconds


// 200M Prone
wpn: arifle_AK12_F - acc: 38 percent (5/13) - time: 00:14
wpn: arifle_AK12_F - acc: 54 percent (6/11) - time: 00:12
wpn: arifle_AK12_F - acc: 23 percent (5/21) - time: 00:23
wpn: arifle_AK12_F - acc: 31 percent (5/16) - time: 00:19
wpn: arifle_AK12_F - acc: 21 percent (7/33) - time: 00:38
wpn: arifle_AK12_F - acc: 24 percent (8/33) - time: 00:38
  32% accuracy --21  shots --24  seconds

 

 

Data @ 300 Meters
 

Spoiler

 

// 300M Standing 
wpn: arifle_AK12_F - acc: 15 percent (6/40) - time: 02:11
wpn: arifle_AK12_F - acc: 23 percent (10/43) - time: 02:17
wpn: arifle_AK12_F - acc: 21 percent (9/42) - time: 02:16
wpn: arifle_AK12_F - acc: 15 percent (4/26) - time: 01:25
wpn: arifle_AK12_F - acc: 17 percent (6/35) - time: 01:53
wpn: arifle_AK12_F - acc: 22 percent (6/27) - time: 01:28
  19% accuracy --35.5  shots --115 seconds


// 300M Crouched
wpn: arifle_AK12_F - acc: 21 percent (7/32) - time: 01:42
wpn: arifle_AK12_F - acc: 23 percent (9/39) - time: 02:07
wpn: arifle_AK12_F - acc: 25 percent (10/40) - time: 02:10
wpn: arifle_AK12_F - acc: 5 percent (3/54) - time: 03:00  * FAILED with 2 kills *
wpn: arifle_AK12_F - acc: 13 percent (6/43) - time: 02:22
wpn: arifle_AK12_F - acc: 19 percent (7/36) - time: 01:59
  18% accuracy --41  shots --133 seconds


// 300M Prone
wpn: arifle_AK12_F - acc: 20 percent (7/34) - time: 01:52
wpn: arifle_AK12_F - acc: 24 percent (6/25) - time: 01:23
wpn: arifle_AK12_F - acc: 29 percent (5/17) - time: 00:56
wpn: arifle_AK12_F - acc: 21 percent (9/42) - time: 02:18
wpn: arifle_AK12_F - acc: 29 percent (5/17) - time: 00:58
wpn: arifle_AK12_F - acc: 18 percent (9/49) - time: 02:38
  23% accuracy --31  shots --101  seconds

 

 

Data @ 300 Meters w Optics & Attachments

 

Spoiler

 

// 300M Prone + bipod  -- #bipod was not deployed
wpn: arifle_AK12_F - acc: 18 percent (6/32) - time: 01:44
wpn: arifle_AK12_F - acc: 17 percent (6/35) - time: 01:54
wpn: arifle_AK12_F - acc: 36 percent (7/19) - time: 01:01
wpn: arifle_AK12_F - acc: 23 percent (6/26) - time: 01:24
wpn: arifle_AK12_F - acc: 31 percent (7/22) - time: 01:11
wpn: arifle_AK12_F - acc: 25 percent (8/32) - time: 01:46
  25% accuracy --28  shots --92 seconds


// 300M Prone + HAMR (opticType = 1)
wpn: arifle_AK12_F - acc: 41 percent (5/12) - time: 00:38
wpn: arifle_AK12_F - acc: 11 percent (5/43) - time: 02:19
wpn: arifle_AK12_F - acc: 19 percent (9/47) - time: 02:36
wpn: arifle_AK12_F - acc: 15 percent (7/45) - time: 02:26
wpn: arifle_AK12_F - acc: 19 percent (9/47) - time: 02:35
wpn: arifle_AK12_F - acc: 17 percent (7/41) - time: 02:14
  20% accuracy --39  shots --128 seconds


// 300M prone + optic_LRPS (opticType = 2)
wpn: arifle_AK12_F - acc: 16 percent (6/37) - time: 01:57
wpn: arifle_AK12_F - acc: 30 percent (10/33) - time: 01:45
wpn: arifle_AK12_F - acc: 15 percent (5/33) - time: 01:47
wpn: arifle_AK12_F - acc: 23 percent (5/21) - time: 01:07
wpn: arifle_AK12_F - acc: 46 percent (6/13) - time: 00:42
wpn: arifle_AK12_F - acc: 30 percent (4/13) - time: 00:43
  27% accuracy --25  shots --80 seconds

 



Findings
The effect on stance on accuracy is slight, but present. In particular around the 300 meters mark the effect is more pronounced. The nature of the experiment introduces an unintended penalty to stance modifications.  because of the way weapons are added and that the soldier begins back turned to the enemy, the initial shot (sometimes two shots) fired from the crouched or prone stance will miss due to animation changes. 


There are two categories of optics, type 1 and 2, and these have an effect on weapon performance.  The effect may not however always be beneficial, as in the case of the AK12 adding a HAMR (type 1) optic will at 300 meters negatively affect performance.  A bipod has no relevant effect on AI accuracy, and when fired from a prone position. The bipod is not engaged. 
 

At 300 meters the AK12 rate of fire is distinctly slower than at other ranges. At all tested ranges the rifle fires in a semi-auto manner. 

 

Conclusion

There is a small effect on AI weapon accuracy based on stance, but generally only between standing and prone.  At closer ranges, with small margins, changing stance or other animation interactions play a far greater effect on accuracy loss. An AI that frequently shifts between crouched, prone or standing will be considerably less deadly. 

 

I was surprised to find that the HAMR optic did not increase accuracy at 300 meters. When equipped on a 5.56 or MX rifle there was a clear improvement in accuracy at 200 meters. It is worth noting that the LRPS, a type-2 optic config wise, did see some accuracy improvements.  To answer @mmm, I suspect the reason is that optics alter the ranges where the weapon is consdiered to be  short, mid or far ranges.  Perhaps there is some intrinsic accuracy loss when the weapon is fired at further ranges.  The AK12 fares poorly, mostly for the accident of precise ranges, accuracy values and bullet damage. 

 

The biggest takeaway is of course the inadequacy of the testing suite. It was designed to compare weapons across mods, not the precise nuances between weapons and stances. A new, more easily scalable suite would need to be constructed. Preferably one which can develop numbers in a quicker and more statistically significant manner. 

 

Abstract potential

As mentioned in the introduction, to make sense of the numbers I created a sort of abstract potential.  This created a nice string of numbers,  but I must stress that the small sample size and considerable variety in test results (a single good set can skew the numbers quite considerably!), means that these should be viewed as guidelines and not final. (Higher is better)

 

Ak12 standing 100M: 95

Ak12 crouched 100M: 101

Ak12 prone 100M: 79

 

Ak12 standing 200M: 68

Ak12 crouched 200M: 64

Ak12 prone 200M: 51

 

Ak12 standing 300M: 12

Ak12 crouched 300M: 10

Ak12 prone 300M: 20

 

Ak12 prone 300M with #bipod : 24

Ak12 prone 300M with HAMR : 14

Ak12 prone 300M with LRPS : 29

 

From this it can be read that there is no real difference between standing and crouched stances.  Prone initially fares poorly, because of the aforementioned time spent changing stance, but comes out slightly ahead as distances being to increase.  The numbers are however so close that no real firm result can be read. 

 

Overall, this test is best viewed as a failure with some interesting implications and lessons learned. 

 

 

Future improvements 

- Run the test with set AI skill 1 
- Run the test in a manner which produces more numbers (and faster) 
- Create a better aggregate number which can be used to compare weapons 

 

 

  • Like 1

Share this post


Link to post
Share on other sites

I dunno... I'll call it inconclusive? This is a good deal of testing done by you already nkenny it''s simply unfair to ask you do another 100 to get closer to statistical average. But with the extra AI dispersion a good deal of randomness is involved, it's totally within what RNG could do to you. I'll stick by my assumption that optics and stances don't have a significant impact on AI accuracy, if any.

 

But arguably your test made a good case for avoiding close combat with AI at all costs... I think gameplay has already allowed us to develop such instincts but you properly quantified it. 1.5s is your life expectancy after AI starts engaging you at 100m(that explained my fear for towns and Tanoa). Given how AI handles detection at such range you're pretty much instantly engaged as soon as you get within line of sight of an AI. 300m range didn't reduce AI's hit probability to exactly 1/9 of 100m, but the time to successfully engage a target is extended by even more than that. Actually funny enough we discussed all the potential factors but none affect AI's ability to kill more than range., well at least for assault rifle. I'd say weapons and their configuration for AI use is another one, and depending on type maybe scaled differently between ranges.

Share this post


Link to post
Share on other sites

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now

×