Bug #4483

AI shooting accuracy: unit SKILL value has negligible effect on shooting accuracy

Added by fabrizioT almost 6 years ago. Updated over 3 years ago.

Status:Closed Start date:09/12/2009
Priority:High Due date:
Assignee:Dwarden % Done:

0%

Category:AI Control
Target version:1.60.87580
Affected ArmA II version:1.60 BETA First affected build:beta 83544
Reproduced by another DH user:No First affected ArmA II version:
I am using some Mods:No Single / Multi Player?:
I am using: BIForumURL:
Reproducible for you:Yes NGUrl:
Related to content of DLC: WIKIurl:

Description

In ArmA2 editor based unit skill value has negligible effect on shooting accuracy.
Lowly trained AI units equipped with assault rifles are able to kill enemies from close to medium range (300m) with the very first burst, in a split second.
AI shooting precision is so messed that in many circumstances lowly skilled units are able to shoot with same or better accuracy than highly skilled ones.

Tested within a custom target range script.
To better highlight the problem these settings where used in ArmA2 profile file:

1 2 skillFriendly=1; 3 skillEnemy=1; 4 precisionFriendly=1; 5 precisionEnemy=1;

Please take the following results as a raw example/repro of the problem:

West rifleman, firing on target from 150m. distance, CROUCHED stance:

  • 25% SKILL: 2,9 bullets needed to hit the target (average on 25 tries)
  • 100% SKILL: 4,7 bullets needed to hit the target (average on 25 tries)

West rifleman, firing on target from 150m. distance, PRONE stance:

  • 25% SKILL: 1,7 bullets needed to hit the target (average on 25 tries)
  • 100% SKILL: 2,1 bullets needed to hit the target (average on 25 tries)

West rifleman, firing on target from 300m. distance, CROUCHED stance:

  • 25% SKILL: 3,3 bullets needed to hit the target (average on 25 tries)
  • 100% SKILL: 2,6 bullets needed to hit the target (average on 25 tries)

West rifleman, firing on target from 300m. distance, PRONE stance:

  • 25% SKILL: 2,9 bullets needed to hit the target
  • 100% SKILL: 2,7 bullets needed to hit the target

The problem defeats the possibility to tactically balance numbers with better skills.
Elite units are currently worthless.

Solution:

  • drastically lower shooting accuracy for lowly trained AI units. They should not able to take down enemies with the very first burst from close to medium range.

4483.zip (26.1 kB) fabrizioT, 08/06/2011 09:39

test_AI_skill.utes.7z (688 Bytes) wiper, 08/30/2011 14:28


Related issues

related to ARMA2 Community Issue Tracker - Bug #1751: Suppression fire precision issue Closed 06/03/2009

History

Updated by Fireball almost 6 years ago

  • Due date set to 10/12/2009
  • Status changed from New to Assigned

Updated by Fireball almost 6 years ago

  • Subject changed from AI shootinmg accuracy: unit SKILL value has negligible effect on shooting accuracy to AI shooting accuracy: unit SKILL value has negligible effect on shooting accuracy

Updated by Dwarden almost 6 years ago

  • Assignee set to Dwarden

Updated by fabrizioT almost 6 years ago

Thx Fireball for title correction and to Dwarden for taking a look here ;)

Updated by Dwarden almost 6 years ago

  • Priority changed from Normal to High
  • Target version set to 1.05 BETA server
  • Reproduced by another DH user set to No
  • CPU set to Please specify!

Updated by Fireball almost 6 years ago

  • Due date changed from 10/12/2009 to 11/12/2009
  • Affected ArmA II version changed from 1.04 BETA to 1.04.59026

Updated by fabrizioT almost 6 years ago

Happy to see this is being worked on.

Updated by fabrizioT almost 6 years ago

A little "addendum": please make ALL skill-related params ("aimingShake", "aimingAccuracy", "spotDistance", "spotTime", "courage", "commanding", ...) change in a appreciable way depending on skill value. Thx.

Updated by fabrizioT almost 6 years ago

Just a question: why is target version 1.05BETA server ?
Just wondering ...

Updated by kju almost 6 years ago

Well I guess server is meaningless in this case.

1.05 BETA is saying it should be in one of the next beta patches.

Updated by kju almost 6 years ago

  • Target version changed from 1.05 BETA server to Upcoming version

Updated by kju over 5 years ago

  • Due date changed from 11/12/2009 to 01/02/2010
  • Affected ArmA II version changed from 1.04.59026 to 1.05 BETA

I guess nothing has changed here, right?

Updated by fabrizioT over 5 years ago

Nothing changed from what i can see.

Updated by kju over 5 years ago

tx

Updated by kju over 5 years ago

  • Due date changed from 01/02/2010 to 03/02/2010

Updated by Frederf about 5 years ago

Clearly skill and accuracy being independent is a good thing? That's why there are two sliders in the difficulty config, one for skill and one for accuracy. I would venture the real problem is not that skill is independent of precision, but rather that there is no direct and unit-specific way to change precision in the editor. Or perhaps there should be a 3rd combined "skill+precision" adjustment if you don't want to have to change both?

Updated by kju about 5 years ago

  • Due date changed from 03/02/2010 to 10/01/2010
  • Status changed from Assigned to Feedback

Any difference in latest OA?

Updated by kju almost 5 years ago

  • Due date deleted (10/01/2010)
  • Status changed from Feedback to Expired

No feedback, therefore expired.

Please reopen if still an issue.

Updated by kju almost 5 years ago

  • Target version deleted (Upcoming version)

Updated by fabrizioT about 4 years ago

  • Priority changed from High to Normal
  • Affected ArmA II version deleted (1.05 BETA)

I confirm that's still an issue in ArmA2 OA/CO v. 1.60 beta

Updated by Fireball about 4 years ago

  • Due date set to 10/30/2011
  • Status changed from Expired to Assigned
  • Affected ArmA II version set to 1.60 BETA
  • Reproducible for you changed from No to Yes

Updated by Suma about 4 years ago

1) The issue repro mentions a custom script. Can we have the script here, or another repro?
2) The issue repro mentions setting all skills and precisions in the user profile to 1. I think it would be preferable to test with a default game setting, and report results depending on a difficulty level.
3) I already see some strange things of how are skill values handled in the engine, but I would still prefer to have a repro to be sure I am seeing the things you are complaining about

Updated by fabrizioT about 4 years ago

Got it ;)
I'll package a repro.

Updated by Robalo about 4 years ago

As soon as you lower aiming values inside the CfgAISkill config class the skill slider in editor proves more useful.
I did not write down results from my tests, just observed AI at shooting range when trying various changes.

Updated by Suma about 4 years ago

Anyone can provide some reliable, easy to understand repro (measuring time to kill, or measuring how many units are needed to win a battle?). I think I may have a fix ready based on what I understand about the problem from the source files, but not having a repro it is impossible to verify the fix is really working, and I therefore cannot publish it.

Updated by fabrizioT about 4 years ago

@Suma,

what i can provide is a automated "target range" with an AI unit firing on target.
Distance and skill configurable.

The script would count the shots fired before hitting the target, averaging the value on 10 firing sessions.
Good way to check accuracy, but with the limitation that fatigue/morale are not in the equation.

Let me know if that would be somewhat useful for you, i can be able to post it tomorrow.

Updated by Suma about 4 years ago

Yes, that sound like a very good testing case.

Updated by fabrizioT about 4 years ago

UPDATED REPRO

See attached mission: 4483.zip

REPRO USAGE

  • set unit skill for "shooter" within editor
  • eventually tweak the settings. To do that look into "shooter" init field and you'll see:
       _nil = [this, spotter, "targetE", [50, 100, 200], ["up", "middle", "down"], 20 ] execVM  "main.sqf";
      

    You may want to alter the last 3 params, which are respectively the preset <distance>, <stance> and <number of iterations>.
    "Shooter" will fire on <number of iteration> targets foreach <stance> and each <distance>.

  • let the mission run and wait for completion, then read the stats into .rpt file. You may want to accelerate time 4x on default settings.

REPRO KNOWN ISSUES & LIMITATIONS

  • Limitation: The repro counts the number of bullets needed to AI in order to hit a target. It includes bullets already in flight at the moment the target is hit.
  • Issue: sometimes the unit just refuses to fire on target. That happens more frequently when target is far > 150m. It probably can't identify it as "foe"; i mitigated the issue by adding a "spotter" unit.

EXPERIMENTALLY GATHERED DATA

2 tests for both 1 and 0.2 skill settings. VETERAN difficulty.

Skill=0.2 (minimum in editor)

targetrange session begin
Skill: 0.2 - Distance: 50 - Stance: up - Iterations: 21 of 20 - Shots fired: 60 - On target avg: 34%
Skill: 0.2 - Distance: 50 - Stance: middle - Iterations: 21 of 20 - Shots fired: 60 - On target avg: 34%
Skill: 0.2 - Distance: 50 - Stance: down - Iterations: 21 of 20 - Shots fired: 60 - On target avg: 34%
Skill: 0.2 - Distance: 100 - Stance: up - Iterations: 21 of 20 - Shots fired: 72 - On target avg: 28%
Skill: 0.2 - Distance: 100 - Stance: middle - Iterations: 21 of 20 - Shots fired: 60 - On target avg: 34%
Skill: 0.2 - Distance: 100 - Stance: down - Iterations: 21 of 20 - Shots fired: 60 - On target avg: 34%
Skill: 0.2 - Distance: 200 - Stance: up - Iterations: 21 of 20 - Shots fired: 329 - On target avg: 7%
Skill: 0.2 - Distance: 200 - Stance: middle - Iterations: 21 of 20 - Shots fired: 141 - On target avg: 15%
Skill: 0.2 - Distance: 200 - Stance: down - Iterations: 21 of 20 - Shots fired: 64 - On target avg: 32%
Targetrange session end. Duration: 852.797 seconds" 

targetrange session begin
Skill: 0.2 - Distance: 50 - Stance: up - Iterations: 21 of 20 - Shots fired: 63 - On target avg: 32%
Skill: 0.2 - Distance: 50 - Stance: middle - Iterations: 21 of 20 - Shots fired: 60 - On target avg: 34%
Skill: 0.2 - Distance: 50 - Stance: down - Iterations: 21 of 20 - Shots fired: 60 - On target avg: 34%
Skill: 0.2 - Distance: 100 - Stance: up - Iterations: 21 of 20 - Shots fired: 100 - On target avg: 20%
Skill: 0.2 - Distance: 100 - Stance: middle - Iterations: 21 of 20 - Shots fired: 78 - On target avg: 26%
Skill: 0.2 - Distance: 100 - Stance: down - Iterations: 21 of 20 - Shots fired: 62 - On target avg: 33%
Skill: 0.2 - Distance: 200 - Stance: up - Iterations: 21 of 20 - Shots fired: 167 - On target avg: 12%
Skill: 0.2 - Distance: 200 - Stance: middle - Iterations: 21 of 20 - Shots fired: 109 - On target avg: 19%
Skill: 0.2 - Distance: 200 - Stance: down - Iterations: 21 of 20 - Shots fired: 114 - On target avg: 18%
Targetrange session end. Duration: 783.362 seconds
Skill=1 (maximum)
targetrange session begin
Skill: 1 - Distance: 50 - Stance: up - Iterations: 21 of 20 - Shots fired: 60 - On target avg: 34%
Skill: 1 - Distance: 50 - Stance: middle - Iterations: 21 of 20 - Shots fired: 60 - On target avg: 34%
Skill: 1 - Distance: 50 - Stance: down - Iterations: 21 of 20 - Shots fired: 60 - On target avg: 34%
Skill: 1 - Distance: 100 - Stance: up - Iterations: 21 of 20 - Shots fired: 60 - On target avg: 34%
Skill: 1 - Distance: 100 - Stance: middle - Iterations: 21 of 20 - Shots fired: 73 - On target avg: 28%
Skill: 1 - Distance: 100 - Stance: down - Iterations: 21 of 20 - Shots fired: 42 - On target avg: 48%
Skill: 1 - Distance: 200 - Stance: up - Iterations: 21 of 20 - Shots fired: 67 - On target avg: 30%
Skill: 1 - Distance: 200 - Stance: middle - Iterations: 21 of 20 - Shots fired: 64 - On target avg: 32%
Skill: 1 - Distance: 200 - Stance: down - Iterations: 21 of 20 - Shots fired: 46 - On target avg: 44%
Targetrange session end. Duration: 586.157 seconds

targetrange session begin
Skill: 1 - Distance: 50 - Stance: up - Iterations: 21 of 20 - Shots fired: 69 - On target avg: 29%
Skill: 1 - Distance: 50 - Stance: middle - Iterations: 21 of 20 - Shots fired: 72 - On target avg: 28%
Skill: 1 - Distance: 50 - Stance: down - Iterations: 21 of 20 - Shots fired: 63 - On target avg: 33%
Skill: 1 - Distance: 100 - Stance: up - Iterations: 21 of 20 - Shots fired: 58 - On target avg: 35%
Skill: 1 - Distance: 100 - Stance: middle - Iterations: 21 of 20 - Shots fired: 60 - On target avg: 34%
Skill: 1 - Distance: 100 - Stance: down - Iterations: 21 of 20 - Shots fired: 41 - On target avg: 49%
Skill: 1 - Distance: 200 - Stance: up - Iterations: 21 of 20 - Shots fired: 182 - On target avg: 11%
Skill: 1 - Distance: 200 - Stance: middle - Iterations: 21 of 20 - Shots fired: 478 - On target avg: 5%
Skill: 1 - Distance: 200 - Stance: down - Iterations: 21 of 20 - Shots fired: 41 - On target avg: 49%
Targetrange session end. Duration: 807.087 seconds

OBERVATIONS

  • 50m: no significative difference between min. and max. skill: 32%-34% (avg 33%) vs. 28%-34% (avg 31%)
  • 100m: inconsistent results for same skill/stance + minor average difference between min. and max. skill: 20%-34% (avg 27%) vs. 28%-49% (avg 38%)
  • 200m: inconsistent results for same skill/stance + minor average difference between min. and max. skill: 7%-32% (avg 19.5%) vs. 5%-49% (avg 27%)
  • Summed results of both runs: a skill=1 unit is 7% more accurate at 50m, 21% more at 100m, 5% more at 200m.

EXPECTED

  • drastically lower accuracy for lowly trained unit. Currently there's no big difference in firing accuracy between a skill=1 unit (elite trained troop) and a skill=0.2 unit (untrained mob).
  • On same skill and same stance results should be consistent in subsequent mission runs, while they're not (e.g. see results of 2 runs for skill=1, 200m. "middle" stance)

Updated by Fireball about 4 years ago

Since it's mentioned in the original description:

degrade spotting ability for lowly trained AI units in a appreciable manner.

This is not measured in the repro (time to spot) and would be probably be interesting to see, if adjustments should be done in this section.

From my experience (with Insurgency), the time from spotting to shooting (aiming time) is noticeable with different skill settings within regular missions. This might not be the case if the two foes are already facing eachother (as in repro scenarios).

On same skill and same stance results should be consistent in subsequent mission runs, while they're not (e.g. see results of 2 runs for skill=1, 200m. "middle" stance)

Here I disagree; it might give a better lab case, but it's what the AI makes a bit more human-like; most humans (skilled or not) are inconsistent within the percentage you mentioned, specially in a fire fight scenario. Maybe the more skilled an AI/human is, the less inconsistent results it should yield.

The bottom-line of your results is, that the AI should probably by distance be gradually be less accurate with increasing distance, in the heat of the battle (rifle accuracy should - hopefully does - make a difference too in percentage spread).

Sidenote: Outside of a battle, i.e. an ambush, with modern assault rifles, skilled soldiers should be able to hit by 97-99% on a still target, say at 300m, given sufficient time to aim.

Updated by fabrizioT about 4 years ago

This is not measured in the repro (time to spot) and would be probably be interesting to see, if adjustments should be done in this section.

It would be interesting, i agree, but i think it's not easily implemented in current repro.
I did not mention spotting ability anymore because i thought it was a bit OT, since the core issue for this ticket is firing accuracy.
However i'll investigate and possibly create a different ticket.

Here I disagree; it might give a better lab case, but it's what the AI makes a bit more human-like; most humans (skilled or not) are inconsistent within the percentage you mentioned,
specially in a fire fight scenario. Maybe the more skilled an AI/human is, the less inconsistent results it should yield.

Well i'm not saying it should be identical, just more consistent. The problem is the "accuracy spread" for the same skill is just very variable, up to 300% as you can see. This has further implications: since accuracy is quite close independently on skill, such a big spread risks to override completely the skill benefits, messing up the balance of missions.

All in all please consider that a big "accuracy spread" would be interesting and human-alike as long as we consider a real fight, while here we are seeing raw target range performance which is not already influenced by fatigue nor morale. I'd agree on your point if accuracy spread would be tied to fatigue and morale.

However i agree on your point - the higher the skill, the lower the accuracy spread should be.

Sidenote: Outside of a battle, i.e. an ambush, with modern assault rifles, skilled soldiers should be able to hit by 97-99% on a still target, say at 300m, given sufficient time to aim.

I see, but my point is mainly about poorly trained / untrained soldiers. In short i think they perform too close to (very) skilled soldiers, making hard to carefully balance missions.

EDIT: Needless to say all the stuff i'm posting here other than repro, observations and experimental data it's just an opinion and any discussion is welcome.

Updated by Fireball about 4 years ago

  • Description updated (diff)

Agree completely with all your statements, just wanted to make sure, we're doing this right and are talking about the same thing in this ticket. I'll remove the statements about time to spot from this ticket then.

Not intended as a general discussion, that's for the BIF.

Updated by Suma about 4 years ago

  • Status changed from Assigned to Resolved

The skill computations for profile values skillEnemy/skillFriendly over 0.7 were broken for very long time, since OFP:Elite Xbox (it was not noticed back then, as such skill were not used with the exception of Mercenary difficulty on Xbox). The whole formula combining skill from editor and profile was reworked now in 83390.

When briefly testing with the firing range, I have seen significant differences between 0.2 and 1.0 editor skills. Testing on more difficulty levels should be done, and it would be also good to test enemy skills (e.g. to change a side of the player to OPFOR in the firing range mission and place the player someplace where he cannot be seen by the firing unit).

Note: currently friendly units skill in the profile is always 1.0. We are checking when and why this was done and if it perhaps should be changed to scale with the difficulty level a bit as well.

Updated by Sickboy about 4 years ago

Very very nice, thank you Suma!

Updated by zGuba about 4 years ago

  • Target version set to Upcoming version

Updated by fabrizioT about 4 years ago

Sorry for my previous bogus message (now removed).
I just re-tested, VETERAN setting. I'm getting mixed results:

* at 200m. range there's a very sensible difference in accuracy, depending on skill. That's fine, but i notice that aiming shake is huge and the shooter needs quite a lot of time to take its shots.

  • at 200m. results are strange for a skill=0.2 unit:
      Skill: 0.2 - Distance: 200 - Stance: up - Iterations: 21 of 20 - Shots fired: 1381 - On target avg: 2%" 
      Skill: 0.2 - Distance: 200 - Stance: middle - Iterations: 21 of 20 - Shots fired: 277 - On target avg: 8%" 
      Skill: 0.2 - Distance: 200 - Stance: down - Iterations: 21 of 20 - Shots fired: 121 - On target avg: 17%" 
      Targetrange session end. Duration: 1089.21 seconds" 
      

    There's quite a huge gap between "up" stance and other ones. A unit in "prone" stance is 11x more accurate compared to "up" stance. Not sure wheteher this makes sense.
    After a few more testing i now think that for "down" (and possibily "middle" stance accuracy) is still a bit too much.

  • up to 100m. results are dubious and inconsistent, at least for for prone stance (0.2 skill unit having actually better accuracy than maxed skill one).
  • All in all i think that CQB range performances are still quite close. I wonder how many times a skill=1 unit will statistically prevail on a skill=0.2 unit with these settings. I'll check it out (but please solve the prone stance problem first).

See below for reference.


 targetrange session begin" 
 Skill: 1 - Distance: 50 - Stance: up - Iterations: 21 of 20 - Shots fired: 66 - On target avg: 31%
 Skill: 1 - Distance: 50 - Stance: middle - Iterations: 21 of 20 - Shots fired: 60 - On target avg: 34%
 Skill: 1 - Distance: 50 - Stance: down - Iterations: 21 of 20 - Shots fired: 93 - On target avg: 22%

 Skill: 1 - Distance: 100 - Stance: up - Iterations: 21 of 20 - Shots fired: 67 - On target avg: 30%
 Skill: 1 - Distance: 100 - Stance: middle - Iterations: 21 of 20 - Shots fired: 51 - On target avg: 40%
 Skill: 1 - Distance: 100 - Stance: down - Iterations: 21 of 20 - Shots fired: 59 - On target avg: 34%
 targetrange session begin" 
 Skill: 0.2 - Distance: 50 - Stance: up - Iterations: 21 of 20 - Shots fired: 102 - On target avg: 20%
 Skill: 0.2 - Distance: 50 - Stance: middle - Iterations: 21 of 20 - Shots fired: 72 - On target avg: 28%
 Skill: 0.2 - Distance: 50 - Stance: down - Iterations: 21 of 20 - Shots fired: 63 - On target avg: 33%

 Skill: 0.2 - Distance: 100 - Stance: up - Iterations: 21 of 20 - Shots fired: 258 - On target avg: 8%
 Skill: 0.2 - Distance: 100 - Stance: middle - Iterations: 21 of 20 - Shots fired: 83 - On target avg: 25%
 Skill: 0.2 - Distance: 100 - Stance: down - Iterations: 21 of 20 - Shots fired: 36 - On target avg: 56%

I'm not considering this fixed, awaiting for feedback. Thanks.

Updated by Suma about 4 years ago

Some more fixing and tuning in 83544.

However, even before this tuning I consider the results achieved quite good. Skill 0.2 is a bit extreme, therefore I am not surprised it has very hard time hitting at 200 m. On the other hand, hitting standing man on 50 m is very easy even for inexperienced shooter.

There could be one more thing interesting to measure: besides of how many shots are needed to bring down the target it would be also interesting to measure time needed to bring it down.

Updated by Robalo about 4 years ago

USMC sniper, beta 83500, difficulty 1 for skill and precision:

Skill: 1 - Distance: 200 - Stance: down - Iterations: 11 of 10 - Shots fired: 10 - On target avg: 100%" 
Skill: 1 - Distance: 300 - Stance: down - Iterations: 11 of 10 - Shots fired: 10 - On target avg: 100%" 
Skill: 1 - Distance: 400 - Stance: down - Iterations: 11 of 10 - Shots fired: 14 - On target avg: 72%" 
Targetrange session end. Duration: 93.936 seconds" 

Skill: 0.25 - Distance: 200 - Stance: down - Iterations: 11 of 10 - Shots fired: 35 - On target avg: 29%" 
Skill: 0.25 - Distance: 300 - Stance: down - Iterations: 11 of 10 - Shots fired: 121 - On target avg: 9%" 
Skill: 0.25 - Distance: 400 - Stance: down - Iterations: 11 of 10 - Shots fired: 75 - On target avg: 14%" 
Targetrange session end. Duration: 666.715 seconds" 

Updated by fabrizioT about 4 years ago

Some more fixing and tuning in 83544.
However, even before this tuning I consider the results achieved quite good. Skill 0.2 is a bit extreme, therefore I am not surprised it has very hard time hitting at 200 m. On the other hand,
hitting standing man on 50 m is very easy even for inexperienced shooter.

After further testing i think i agree with you. Only think that are still problematic are the inconsistencies for prone stance. Looking forward to test 83544.

There could be one more thing interesting to measure: besides of how many shots are needed to bring down the target it would be also interesting to measure time needed to bring it down.

We already have aggregate session duration within repro: you may want to restrict testing to just one distance and stance, then divide it for number of targets.

Example of init ( 200m distance, standing, 20 targets ):

_nil = [this, spotter, "targetE", [200], ["up"], 20 ] execVM  "main.sqf";

"Duration" / 20 would be average time to down a target.

Updated by fabrizioT about 4 years ago

New data from beta 1.60 build 83569 (VETERAN settings).
Restricting to the prone stance performance across skills, which is still problematic.

Skill: 0.2 - Distance: 200 - Stance: down - Iterations: 101 of 100 - Shots fired: 1066 - On target avg: 10%
Targetrange session end. Duration: 501.947 seconds" 

Skill: 0.6 - Distance: 200 - Stance: down - Iterations: 101 of 100 - Shots fired: 663 - On target avg: 16%
Targetrange session end. Duration: 383.151 seconds" 

Skill: 1 - Distance: 200 - Stance: down - Iterations: 101 of 100 - Shots fired: 362 - On target avg: 28%
Targetrange session end. Duration: 343.218 seconds" 

Here's the "shooter" init line for the repro:

_nil = [this, spotter, "targetE", [200], ["down"], 100 ] execVM  "main.sqf";

Observations

  • Speaking of average time to down a target:

    Skill 0.2: 2,12 secs. per hit
    Skill 0.6: 1,73 secs. per hit
    Skill 1.0: 1,03 secs. per hit

  • Issue #1 (time-to-hit too low for min. skill): 2 seconds to take down a target from 200m. is already a very low value for a min. skill unit (untrained mob). It's barely enough for the target (whatever the skill) to understand being under fire and possibly go prone, no margin for reaction, tactical movement nor escape.
  • Issue #2 (minor accuracy variance across half skill range): The combined effect of 10% overall accuracy and 2 seconds hit time makes 0.2 skilled units just marginally worse than 0.6 skilled (mid-range) units on same stance.
  • Issue #3 (stance playing a too big role): Consider at the following additional data:
      Skill: 0.6 - Distance: 200 - Stance: middle - Iterations: 101 of 100 - Shots fired: 1344 - On target avg: 8%" 
      Targetrange session end. Duration: 547.305 seconds" 
      

    Let's compare the stats of a prone skill=0.2 unit to a crouched skill=0.6 unit at 200m.: the former is the "winner".
    This means stance plays a crucial role across skills.
    Judging how randomly stance is currently handled by AI i find this undesirable as it's balance breaking.

    I would suggest as a rule of thumb to tweak accuracy settings in order to have roughly (at 200m.) prone accuracy for skill=0.2 < standing accuracy for skill=0.6 and prone accuracy for skill=0.6 < standing accuracy for skill=1.
    Fatigue, morale and suppressive fire effects should be enough to add randomness to firefights, we have to better differentiate raw accuracy across the skill "slider" in my opinion. Otherwise skill will still have a trascurable effect on gameplay.

Updated by wiper about 4 years ago

  • Category changed from Config to AI Control
  • Priority changed from Normal to High
  • Target version changed from Upcoming version to 1.59 BETA
  • Affected ArmA II version changed from 1.60 BETA to 1.59 BETA
  • CPU changed from Please specify! to Q9650 Core2/Quad Intel
  • System RAM size set to 4
  • GPU VRAM size set to 2
  • Audio card changed from Please specify! to Asus Xonar D2X
  • Size of OS swap file changed from Please specify! to 4GB
  • First affected build set to beta 83544

setSkill array values set in mission again ignored.
Was fixed in beta [83390] but reintroduced, most probably with beta [83544] ("Improved: AI skill settings in difficulty options easier to use.").

The last public beta that does it well is [83500].

See test mission attached:
The guy in front of you has "aimingshake" and "aimingaccuracy" both down to 0.01 ! (see units ini field)
Yet In beta [83544] or higher he almost insta kills.
Version [83500] does it well for such low aiming skills: he never hits, even if you run towards him for a while.

Updated by alef about 4 years ago

  • Target version changed from 1.59 BETA to 1.60 BETA
  • Affected ArmA II version changed from 1.59 BETA to 1.60 BETA

1.60 beta means 1.59.79600 up

Updated by kju about 4 years ago

  • Due date deleted (10/30/2011)
  • Operating system deleted (Win7 64 bit)
  • CPU deleted (Q9650 Core2/Quad Intel)
  • System RAM size deleted (4)
  • GPU VRAM size deleted (2)
  • Audio card deleted (Asus Xonar D2X)
  • Size of OS swap file deleted (4GB)

Updated by kju about 4 years ago

wiper your test mission was not attached. please try again.

Updated by wiper almost 5 years ago

test mission here, hopfully this time it sticks

Updated by Suma almost 5 years ago

Fixed in 84144 - the new calculation did not assume the skill input could be lower than 0.2, which is the minimal value you can enter in the mission editor.

Updated by Sickboy almost 4 years ago

  • Status changed from Resolved to Closed

Updated by kju over 3 years ago

  • Target version changed from 1.60 BETA to 1.60.87580

Also available in: Atom PDF