Bug #2209

AI units can't spot standing enemy at 250m. with clear line of sight

Added by fabrizioT about 6 years ago. Updated over 5 years ago.

Status:Closed Start date:06/22/2009
Priority:Normal Due date:
Assignee:- % Done:

10%

Category:AI Issues
Target version:1.05.62017
Affected ArmA II version:1.05 BETA First affected build:
Reproduced by another DH user:Yes First affected ArmA II version:
I am using some Mods: Single / Multi Player?:
I am using: BIForumURL:
Reproducible for you: NGUrl:
Related to content of DLC: WIKIurl:

Description

Tested on airstrip, easily reproduceable in editor:

- Just place 2 units (max skill)of enemy sides onto an airstrup (it assures clear LOS).
- Set their direction so they just stare at each other.
- Set their distance to 250m or more.

Preview the mission.

Result: they won't detect their enemy.

I'll add a sample mission if necessary, but i think that setup is so basic and obvious that's probably not needed.

los_bug_230m.utes.zip (846 Bytes) fabrizioT, 06/22/2009 10:36

cfg.zip (3.1 kB) fabrizioT, 06/23/2009 15:50

test_visibility_1.04beta.Chernarus.zip (1.4 kB) fabrizioT, 08/27/2009 09:14

testDetection.utes.zip - Measure detection distance on Utes airfield (1.8 kB) Suma, 10/19/2009 09:25


Related issues

related to ARMA2 Community Issue Tracker - Bug #2597: AI not shooting even when enemy is in sight and CLOSE RANGE Duplicate 07/05/2009 11/15/2009
related to ARMA2 Community Issue Tracker - Bug #2562: Moving AI see too well Closed 07/05/2009 08/24/2009
duplicated by ARMA2 Community Issue Tracker - Bug #1750: AI with clear LOS cna't see enemy over distance on a stra... Duplicate 06/03/2009

History

Updated by fabrizioT about 6 years ago

Here comes the mission sample, anyway.

Updated by kju about 6 years ago

  • Status changed from New to Feedback
  • Assignee set to fabrizioT
  • Priority changed from High to Normal

fabrizioT please attach your arma2.cfg and .arma2profile file to see your AI settings.

More info: A2 Community Issue Tracker.

Updated by fabrizioT about 6 years ago

Currently at work, however i have tested this with both precisionXXX and skillXXX = 1 in arma2Profile.
I play with visibility = 2500m, all set to normal except PP = low and shadows = high.

Updated by fabrizioT about 6 years ago

Here you go with my config files (ArmA2.cfg + .ArmA2Profile).
regards.

Updated by kju about 6 years ago

  • Status changed from Feedback to Assigned
  • Assignee deleted (fabrizioT)

thanks

Updated by acf about 6 years ago

A few thoughts:

1) Is this related to 2236 in some way?

2) Does engagement behaviour change dramatically if you use 2 opposing groups of 2?

I have not done any systematic tests yet, but a 2-unit AI group seems much more aware than a single unit.

Updated by kju about 6 years ago

1) dont think its related to 2236.

Updated by fabrizioT about 6 years ago

1) Agree with kju, Definitely not related to 2236.

Updated by acf about 6 years ago

1) I disagree with myself, it's not related to 2236.

Notes:
All tests conducted with exposed standing units.
No demo mission prepared as tests were conducted in the editor.

Theory 1: Grouped units seem more alert than single units.
Test 1: 1x group of 2x USMC riflemen ~250m from 1x group of 2x Russian riflemen.
Result 1: Neither side detects the other, they just stand there.
Conclusion 1: Grouping does not dramatically change AI intelligence.

Theory 2: What if one group is moving?
Test 2: As test 1 but give one group a move WP ~10m away.
Result 2: Stationary group detects & engages moving group.
Conclusion 2: Movement is the big giveaway.

Theory 3: Is there a threshold beyond which a stationary unit is not detected?
Test 3: 1x US and 1x Russian full-skill riflemen were placed in plain sight of each other. The distance between them was measured and displayed as a hint. The position of one of the riflemen was varied and the result observed.
Result 3: In approximate terms, at a distance <212.5m, units would detect each other and engage. At distances >212.5m, they played statues with no clear winner.
Conclusion 3: There appears to be a definite threshold beyond which a stationary unit will not be detected even if it is in the open.

Theory 4: How different is it with a large group?
Test 4: Same as Test 3 but substituting standard infantry groups in formation for the single units.
Result 4: The threshold distance increased to ~230m.

Theory 5: One might expect AI snipers to have to engage stationary exposed targets. Snipers should be better at concealment and better at detection...
Test 5: As Test 3 but using sniper v sniper then sniper v rifleman.
Result 5:
US.........Russian...Engaged at
Rfmn.......Rfmn......~212.5
Sniper.....Sniper....~141
Sniper.....Rfmn......~249 advantage to sniper
Rfmn.......Sniper....~249 advantage to sniper
Rfmn.......Sniper....~120 advantage to moving rfmn - sniper without ammo & forced to stand

Conclusion 5: Snipers have only a marginal advantage over riflemen.

Overall conclusion: This issue is due to some sort of detection threshold for stationary targets at moderate ranges. It could be considered a rare issue because most units in most missions will be moving, but when it does occur it is quite obvious. (These tests have also shown up the reluctance of units to quickly seek cover when engaged in the open, instead they usually just go prone.)

Updated by fabrizioT about 6 years ago

Nice test, acf.

Let me add that the problem is not only related to stationary vs. stationary units, that's just the culprit.

If you make one of the units move perpendicular to the direction the other is facing (eg. make RUSFOR unit cross the airfield) you will find that the standing unit will not be able to spot the moving unit unless it being less than about 300-350m far (exact threshold could be checked).

IMHO with clear LOS units should be able to spot (spot != identify targets!) enemies from at least 600-700m. for gameplay purpose.

Updated by kju about 6 years ago

Just to make sure. You are aware that AI reacts visual, audible, motion,
can share information, does guess enemy position, has several steps of
identification (not, some object, unit, infantry, hostile men).
Also binocular, various unit "skills" (values) and weapon type (scope)
come into play.

So we agree that you are saying the distances need tweaking?
An sniper needs to be detected less good?

Updated by acf about 6 years ago

Re: AI awareness - yes. That's why I've kept variables to a minimum otherwise I won't know what I'm measuring. All I can infer is that visual detection seems to have a deadzone for stationary units. There may also be deadzones for moving units, aural detection, etc. but how far do we go? I can make some reasonable assumptions about detection variables, I can 'invent' algorithms but I don't know how BIS do it.

In the tests above, I'd argue that the sniper should be more detectable - easier to see - as he was standing on a runway. It suggests that the algorithm does not consider 'silhouette' - the contrast between the target and his background, only cover that obstructs the LOS.

Direction of view is another big element - I has single units move up behind APCs without being spotted. Again, BIS no the algorithm so they should be able to predict the deadzones. That's not meant to sound like flaming criticism, I appreciate it is not a simple exercise.

Updated by kju about 6 years ago

Might be good to open a separate report for this acf.

This ticket is about AI inf not seeing far enough on open ground, right?

Updated by fabrizioT about 6 years ago

"This ticket is about AI inf not SEEING far enough on open ground, right?"

Correct. This applies for both units spotting steady AND moving targets.

I think it's better not take in account other variables here.
I know there are also issues with units behaviour, hearing capabilities, concealement, ... but to me it's better to segregate the different problems into different tickets in order to keep things as simple as possible.

Regards.

Updated by Rg about 6 years ago

I see that this is old, but I thought I should share some information that can be useful.


Be careful for what your wishing for. The AI see, hear, and engage much further then OFP and ArmA1. To push the limits even further is a bit dangerous ground. If it is increased too much it can break missions or make them extremely hard. A balance was made for gameplay purposes. You have the ability to sneak up close to AI without being detected like ArmA1, but now with longer distance spotting, hearing, and engagements. The ability to get to a relatively close ambush point undetected is a very important part of the game. You should be rewarded for playing slow and stealthy if that is you choice. If you want have a tall stance and move around quicker, the AI will obviously detect you much sooner and at farther distances.

It is impossible to make all scenario's work perfectly, since some people will be think AI are not seeing far enough and others will think AI are seeing too far. And depending on the scenario, they may both be right. Even though the AI is probably better and more flexible then most games, there is still some stuff to be desired.

It's probably best to not focus on small simple tests like this since when playing a mission, there are a dozen other factors taken into account that makes results very different from simple tests. Keep in mind, one of the biggest factors is the ability of AI to share information. Once you get spotted by AI and they start sharing info around to the others, all bets are off. Your worries about AI not spotting you far enough is probably mute at that point. That's why the element of surprise against AI is so important to have at your disposal.

I will however take a look at your test mission later, but generally speaking, everything looks about right here from the discussion so far.

Things to consider about the AI...
---------------------------------------
1) Many unit types and factions have many different values, so spotting and identifying can differ slightly or a lot from unit to unit and faction to faction.
2) Movement speed and stance play very a big role. A non moving person maybe be spotted ~200m-300m, whereas a person sprinting might be spotted at ~500m-600m with anything in between. Also, a person standing will be spotted way before a person prone and a person crouched will be spotted in between the two.
3) Even if AI spot you at 300m, it doesn't necessarily mean that they know if your friendly or foe. At a certain range (which is calculated), he will be able to tell and this distance is increased I believe with more units in a squad rather then just one.
4) AI share information among them (too much and too accurately in my opinion).
5) AI do not see contrast between units and the background, so it doesn't matter to AI how your camo actually looks ingame.
6) How many objects that are around (how "busy" the area is) is not included when calculating AI spotting ability. So an open desert and a forest are exactly the same as far as I know. Sorry for this info, but it's sadly true.

"In the tests above, I'd argue that the sniper should be more detectable - easier to see - as he was standing on a runway. It suggests that the algorithm does not consider 'silhouette' - the contrast between the target and his background, only cover that obstructs the LOS."

Sadly, you are correct. As far as I know, AI do not see contrast between units and the background.

A possible solution may be to assign a value (maybe RGB value) to a unit class and ground texture that they are standing on that can be compared and then calculated with acc,camo,sens,ect.. to give more realistic results. While I would love to see something like this, it may be too late and too time consuming to implement now.

**BTW, if your looking for much more increased AI spotting and identifying ability, ACE2 mod will surely be your answer since this was increased considerably in ACE1.

Updated by alef about 6 years ago

  • Due date set to 09/23/2009

Updated by fabrizioT about 6 years ago

Problem is related to "sensitivity" / "accuracy" default values and it's still present in v1.03.
Units with low skills can't correctly identify standing enemies 200m. far, with clear LOS.

By lowering "accuracy" or raising "sensitivity" values for various classes it's possible to force units to identify enemy soldiers at least up to 400m (and vehicles way further).
I think that default values for "sensitivity", "accuracy" (and "camouflage") have to be tweaked.

Updated by fabrizioT about 6 years ago

UPDATE.

I'm sorry to say that v. 1.03 and 1.04BETA did nothing to address this issue at all.
Two months passed by and in some cases things are actually worse (i am talking about infantry vs. infantry combat).

The problem now is HUGE for urban battles after v.1.03.
While stationary units detect moving units up to 200-240m (fine), moving units (even with maxed skill) are completely blind and unable to detect stationary enemies till incredibly low distance (30-60m!). Maybe an overdone fix for http://dev-heaven.net/issues/show/2562 was done ?

I attached a simple mission demonstrating this exact problem, please just run it, it's simple and self-explanatory.

What's worse is that skill value for both stationary and moving units seems not making any significative difference in their spotting ability.

So please:

  • Raise significatively the AI ability to spot STATIONARY units and especially STANDING units. Now they're pretty invisible.
  • Movement penalty for spotting ability, as requested in http://dev-heaven.net/issues/show/2562 is fine to me, but now it's overdone.
  • make SKILL have some appreciable weight in spotting ability!

Hope to get some feedback on this soon :(

Updated by kju about 6 years ago

  • Affected ArmA II version changed from 1.02.58064 to 1.04 BETA

Updated by kju about 6 years ago

No one knows when/if BI addresses this one.
We do not know their priority list either.

Updated by nettrucker almost 6 years ago

Hi there
things changed drastically after introducing patch 1.03. I've experienced that the AI is far more less reactive when it comes to spotting and indentifying enenmy units. In patch 1.02 the AI was far more agressive maybe a bit too much after changing the parameters in patch 1.03 the AI became somehow dump. I hope that the settings will be adjusted in a future patch and there will be found a compromise.
Kind regards
nettrucker

Updated by Dwarden almost 6 years ago

  • Assignee set to Dwarden

Updated by fabrizioT almost 6 years ago

Since this is being looked in (thx Dwarden): can you explain why spotting ranges in urban environment ar so different (lower) than in open terrain ?
Maybe a problem with building LODs ? Or is that somewhat intended and simply overdone ?
I wish i could understand that.

Updated by Fireball almost 6 years ago

  • Due date changed from 09/23/2009 to 10/23/2009
  • Status changed from Assigned to Feedback
  • Affected ArmA II version changed from 1.04 BETA to 1.04.59026
  • Reproduced by another DH user set to No

Feedback wanted. Do units see really differently in urban grounds than in the field? Does this need fixing (again)?

Updated by fabrizioT almost 6 years ago

Yes they do, you can test it yourself or run the repro case i posted 1 month ago.
Problem was not fixed at all.
I think it was not even addressed.

Updated by Fireball almost 6 years ago

  • Status changed from Feedback to Assigned
  • CPU set to Please specify!

Updated by Fireball almost 6 years ago

  • Due date changed from 10/23/2009 to 11/23/2009
  • Reproduced by another DH user changed from No to Yes

fabrizioT wrote:

While stationary units detect moving units up to 200-240m (fine), moving units (even with maxed skill) are completely blind and unable to detect stationary enemies till incredibly low distance (30-60m!). Maybe an overdone fix for http://dev-heaven.net/issues/show/2562 was done ?

I agree. I've just run your repro mission and I concur that 60-30m is too late. I've put the "bystander" which you placed to watch the scene next to the U.S. guy and ran along with him and I think I had recognized the enemy guy laying on the ground there (at this point in time) at around 120m.

Updated by Fireball almost 6 years ago

It could be that there was an algorithm implemented, just like proposed in #2562 and if I do simple math taking some averages (i.e. 100/220*55 = 25) therefore the perception while jogging would have been reduced to 25% compared to standing (100%).

Updated by fabrizioT almost 6 years ago

I hope to see this fixed soon.

It makes no sense having grunts mostly not being able to detect enemy infantry in open ground at 250-300m or more (in average they spot enemies in the 100-200m. range).
In urban areas things are even worse.

Detection distance for infantry should be bigger than average small weapons effective range, or any maneuvering will be close to worthless, since in ArmA2 cover is scarce!

Updated by Fireball almost 6 years ago

fabrizioT wrote:

Detection distance for infantry should be bigger than average small weapons effective range, or any maneuvering will be close to worthless, since in ArmA2 cover is scarce!

I also hope that you don't want this to be improved too much - running units detecting laying down units at >120 m (as I stated earlier, based on my test) doesn't make much sense - and they should only detect them at that range if they are running in that direction (FOV angle about 60°). This should be influenced (a bit) by skill adjustments and can be jagged by some randomizing, of course.

Updated by Suma almost 6 years ago

  • File testDetection.utes.zip added

I have attached my test case for this. In my testing it seems the difference between moving and stationary is not very big, but the detection distance is too low.

My testing scenario:

- run ahead (hold W)
- wait for messages confirming OPFOR / BLUFOR has been detected by the other side

Results seen by me:

BLUFOR detected, dist 160
OPFOR detected, dist 120

Updated by Suma almost 6 years ago

  • File deleted (testDetection.utes.zip)

Updated by Suma almost 6 years ago

  • File deleted (testDetection.utes.zip)

Updated by fabrizioT almost 6 years ago

@Fireball:

What i'm noticing, as i said, is that since v1.03 AI is no more able to spot moving / steady infantry targets over 250-300m.
Actually in average things are worse, spotting range is well < 200m.

It's weird when you are part of an AI group and you can easily see enemy moving say 400m-500m. on open terrain in front of you, while your AI mates can't at all.
Then you have to manually "reveal" the enemies so your mates can react.

This is the no. 1 problem with Arma2 AI to me.

I agree that AI spotting prone units from distance would be an issue too, but it's minor compared to the above problem and - more important - may be addressed differently.
IMHO What we need here is:

1) AI having MUCH better spotting capabilities, even when moving.
2) stance-related and speed-related "multipliers" better tuned to assure that prone units are really less "spottable" that crouched or standing ones.

Also, like i posted before:

- Movement penalty for spotting ability, as requested in http://dev-heaven.net/issues/show/2562 is fine to me, but now it's overdone.
- make SKILL have some appreciable weight in spotting ability!

@Suma:

i am reading you now.
Glad you are confirming detection distance is too low.
Please check out also what happens in urban environment, detection range will be even worse.

Hope it will be fixed, thx.

Updated by Fireball almost 6 years ago

I'm just weary of the bug report after 1.05 that "AI magically spots the enemy over ludricous distances in worst visiblity conditions while jumping on one leg and blindfolded"...

Updated by Dwarden almost 6 years ago

  • Status changed from Assigned to In progress
  • Target version set to 1.05 BETA
  • % Done changed from 0 to 10

Updated by kju almost 6 years ago

  • Target version changed from 1.05 BETA to Upcoming version

Updated by fabrizioT almost 6 years ago

Way better AI spotting capabilities since build 60588, due to "sensitivity" multiplication.

Was such a broad change in "sensitivity" balanced with tuning of "accuracy", "camouflage" and mayber other params or are we simply rolling back to a ArmA2 v.1.02 situation?

Without proper balancing this huge "sensitivity" change will alleviate CQB problems at the price of hindering stealthy tactics ...

Any update on the progress of this issue?

Updated by Suma almost 6 years ago

  • Status changed from In progress to Resolved

No, we are not simply rolling back to what was before. Many parameters were fixed or fine-tuned. Extensive testing is now needed to see how well it works now.

Updated by kju almost 6 years ago

  • Due date deleted (11/23/2009)
  • Assignee deleted (Dwarden)
  • Target version changed from Upcoming version to 1.05 BETA
  • Affected ArmA II version changed from 1.04.59026 to 1.05 BETA

Updated by fabrizioT almost 6 years ago

Suma wrote:

No, we are not simply rolling back to what was before. Many parameters were fixed or fine-tuned. Extensive testing is now needed to see how well it works now.

Thanks for the info, Suma.
Are you still working on this issue / plan to release more fixes soon?
I'd like to know before testing current implementation.

Updated by kju over 5 years ago

  • Status changed from Resolved to Closed
  • Target version changed from 1.05 BETA to 1.05.62017

Also available in: Atom PDF