investigating Ground units accuracy?

[Ghost_awsss]

Is anyone else having issues with ground units(not AAA) being way too accurate? It makes helos useless for ground attacks. Even Humvee with 50cal can shoot me down before I can even get in range to fire. 

[Wychmaster]

This is a long known issue. As far as I know, the AI always uses the perfect target solution with some kind of random factor depending on the chosen skill level.

[Shadow KT]

No, not having any issues. It is not the units.

[Wychmaster]

vor 19 Stunden schrieb Shadow KT:

No, not having any issues. It is not the units.

The units in DCS are way to accurate with their aim. That doesn't mean that you can not evade their fire. 

 

I read in some thread here that they also seem to consider accelerations. So some random pushing and pulling on the stick should cause them to aim way off. Have not tested it my self so far, but maybe I ll try it today.

 

If you have no issues, you have probably adapted.

[Shadow KT]

Agree to disagree. 

Don't know how many tests you've done and how much combined arms you've played, but show us a track and lets see if I can reproduce. 

 

Units have other issues and accuracy is not one of them, at least 80% of the time. There are some issues, which may lead you to believe that they are too accurate, but the core problem is not accuracy.

 

Majority of the time it is just oblivious players having no SA, threat analysis and just flying in a straight line.

Major problem for the AI is their all seeing eye, as in ... if they can see you in line of sight, they will react immediately, no matter your position in relation to them. 

Their other issues, which might lead you to believe that they are too accurate is that their accuracy is applied to a burst. A burst is usually around 10ish rounds, which is too big of a burst and the same accuracy is applied to the whole burst. Their accuracy dispersion is alright, I would say on average they are actually pretty bad, but they just need one lucky roll, to pepper that oblivious player, flying in a straight line, with 10 accurate shots. 


In any case, if you are taking small arms fire, you are too close. Unless it is radar guided SPAA, 90% of the time the AI hits is if you are flying straight at it, or literally on top of you.

Without tracks, you are just shouting pointlessly into the air. Give tracks with examples and people can look into them. Just saying... oh yeah, AI is too accurate.... yeah, no.
 

Edited July 5, 2021 by Shadow KT

[Wychmaster]

vor 14 Stunden schrieb Shadow KT:

Without tracks, you are just shouting nothing into the air. Give tracks with examples and people can look into the. Just saying... oh yeah, AI is too accurate.... yeah, no.
 

Okay, give me some time to create asome tracks.

[Pikey]

the WW2 gunners on boats, subs and bombers recently (last 3 months) were brought inline with being able to miss more credibly and it was generally noted with everyone this was improved. I didnt look at the 50 cal hummers though, albeit, to be in range of the .50 you need to be really close anyway.

[Wychmaster]

vor 27 Minuten schrieb Pikey:

the WW2 gunners on boats, subs and bombers recently (last 3 months) were brought inline with being able to miss more credibly and it was generally noted with everyone this was improved. I didnt look at the 50 cal hummers though, albeit, to be in range of the .50 you need to be really close anyway.

50 .cal hummers are the ones where it is most obvious, cause they use an analog gun with no stabilization, magnification and guidance assistence. I will test how accurate they currently are and post a track as requested by Shadow KT.

[Wychmaster]

Am 4.7.2021 um 09:36 schrieb Shadow KT:

Agree to disagree. 

Don't know how many tests you've done and how much combined arms you've played, but show us a track and lets see if I can reproduce. 

 

Units have other issues and accuracy is not one of them, at least 80% of the time. There are some issues, which may lead you to believe that they are too accurate, but the core problem is not accuracy.

 

Majority of the time it is just oblivious players having no SA, threat analysis and just flying in a straight line.

Major problem for the AI is their all seeing eye, as in ... if they can see you in line of sight, they will react immediately, no matter your position in relation to them. 

Their other issues, which might lead you to believe that they are too accurate is that their accuracy is applied to a burst. A burst is usually around 10ish rounds, which is too big of a burst and the same accuracy is applied to the whole burst. Their accuracy dispersion is alright, I would say on average they are actually pretty bad, but they just need one lucky roll, to pepper that oblivious player, flying in a straight line, with 10 accurate shots. 


In any case, if you are taking small arms fire, you are too close. Unless it is radar guided SPAA, 90% of the time the AI hits is if you are flying straight at it, or literally on top of you.

Without tracks, you are just shouting pointlessly into the air. Give tracks with examples and people can look into them. Just saying... oh yeah, AI is too accurate.... yeah, no.
 

 

Did some first tests today and saved the tracks (see below). Test was a single stationary HMMV shooting at a KA-50 that was  flying (90kts at 1000ft) a rectangle  around the HMMV close to the maximum threat range. Test ended after the ammo was depleted. I tested all AI skill levels and tried hitting the target myself. The number in the track name represents the number of hits I counted in the debriefing. As you can already see, I achieved almost as many hits as both of the highest AI levels.

 

So in this simple test the AI aims not exceptionally well compared to a human player. Obviously, I can't compare how hard it would be to hit with a real .50cal with all the recoil and the iron sights instead of stable mouse aiming. So lets keep the discussion on human player vs AI.

 

I have to admit I was just repeating stuff that I read elsewhere without performing any tests myself. So I will correct this mistake by performing some more tests. The problem I see with the given example is, that the helicopter is flying at constant speed and altitude. Once you know where to aim it gets easier. This advantage would be gone if the target flies wild evasive maneuvers. On the other hand, with some training, you might get better at it. I do not know so far, how to set up a more or less reproduceable test, but I will figure something out.

 

The next test would be to aim from a moving vehicle. My guess is, that the AI accuracy won't be affected while it will get harder for me. But let's see.

 

However, after this first test session, I think that your assesment of the situation is probably right. So thx for reminding me that repeating stuff without validation is dangerous.

average_hmmv_4_max_range.trk good_hmmv_2_max_range.trk excellent_hmmv_10_max_range.trk high_hmmv_10_max_range.trk me_hmmv_9_max_range.trk

[Ghost_awsss]

On 7/5/2021 at 1:24 PM, Wychmaster said:

 

Try to do some attack runs in Mi 8, Mi 24 or UH 1. 

Was doing some flying around in Mi 24. I was getting destroyed by BMP 3 any time I was trying to attack it, it was just sniping me off with one or 2 shots. 
I forgot to save tracks, I might do some more testing on the weekend...

[Shadow KT]

Neither is the Mi-8 or UH-1 meant to be attacking armored vehicles with heavy weapons. 

In the Mi-24 you have plenty of range to engage with Shturm.

[Ghost_awsss]

13 hours ago, Shadow KT said:

Neither is the Mi-8 or UH-1 meant to be attacking armored vehicles with heavy weapons. 

In the Mi-24 you have plenty of range to engage with Shturm.

Well they do have AP rockets, so potentially they can be used against light armor targets.   They're not meant  to be an attack helicopters because in real life there are plenty of alternatives. Early Huey were tested  against light tanks in Vietnam.

Back to ground units...

They also are not meant to be a AAA units.

I'm not even sure that the BMP can even do this..

 

[Shadow KT]

Quick google search says 60 degrees elevation for the BMP-3. Seems like 60 to me

 

 

This is for the BMP-2

Edited July 10, 2021 by Shadow KT

[NineLine]

On 7/5/2021 at 11:24 AM, Wychmaster said:

Did some first tests today and saved the tracks (see below). Test was a single stationary HMMV shooting at a KA-50 that was  flying (90kts at 1000ft) a rectangle  around the HMMV close to the maximum threat range. Test ended after the ammo was depleted. I tested all AI skill levels and tried hitting the target myself. The number in the track name represents the number of hits I counted in the debriefing. As you can already see, I achieved almost as many hits as both of the highest AI levels.

 

So in this simple test the AI aims not exceptionally well compared to a human player. Obviously, I can't compare how hard it would be to hit with a real .50cal with all the recoil and the iron sights instead of stable mouse aiming. So lets keep the discussion on human player vs AI.

 

I have to admit I was just repeating stuff that I read elsewhere without performing any tests myself. So I will correct this mistake by performing some more tests. The problem I see with the given example is, that the helicopter is flying at constant speed and altitude. Once you know where to aim it gets easier. This advantage would be gone if the target flies wild evasive maneuvers. On the other hand, with some training, you might get better at it. I do not know so far, how to set up a more or less reproduceable test, but I will figure something out.

 

The next test would be to aim from a moving vehicle. My guess is, that the AI accuracy won't be affected while it will get harder for me. But let's see.

 

However, after this first test session, I think that your assesment of the situation is probably right. So thx for reminding me that repeating stuff without validation is dangerous.

average_hmmv_4_max_range.trk 64.98 kB · 4 downloads good_hmmv_2_max_range.trk 43.25 kB · 2 downloads excellent_hmmv_10_max_range.trk 44.93 kB · 2 downloads high_hmmv_10_max_range.trk 48.67 kB · 3 downloads me_hmmv_9_max_range.trk 390.48 kB · 3 downloads

Thanks for your honesty and effort to check this out, its refreshing!

[Volator]

On 7/5/2021 at 8:24 PM, Wychmaster said:

I achieved almost as many hits as both of the highest AI levels.

Were you using the game-computed firing solution (i.e. "locking" the target and firing at the aiming circle) , or did you aim and hit completely manually without any aids? 

[Wychmaster]

vor 1 Stunde schrieb Pilot Ike:

Were you using the game-computed firing solution (i.e. "locking" the target and firing at the aiming circle) , or did you aim and hit completely manually without any aids? 

I did not use any aids. That would have been counter-productive. Don't forget, that my initial goal was to prove that the AI aims too good (which I couldn't ). You can see what I did in the track (it's about 6mins IIRC). Just make sure that you just run it and do not switch camera perspectives or it will desync. I can also upload the test mission so that you can try it yourself.

 

The thing is, that it isn't that hard to hit as you might think. Okay, I fired 500 bullets and hit only 9 times, but the AI wasn't any better. So the AI having god-like aiming seems to be a myth, at least if the first test is somewhat representative. I will do some more testing and provide the tracks, but it will take some time, since I am quite busy at the moment.

 

My current guess is, that the anaysis of Shadow TK is pretty accurate. Most people (me included), simply get too close and fly directly towards the target which makes it easy to hit you. I have a test planned for this case too, so we have some facts to discuss instead of pure speculation.

 

I also partially repeated the first tests with the BTR80, which seems to be the most feared ground vehicle in the game. IIRC the lowest AI scored just a single hit at the max engagement range. I think I scored 4 hits, but I have to double-check it.

 

As I said in my previous post, I don't know how hard it would be to hit in RL, but compared to a human CA player, the lower AI levels seem to be much worse so far. Let's see what the next tests will show us.

 

 

Edited July 11, 2021 by Wychmaster

[Blackeye]

On 7/11/2021 at 12:57 PM, Wychmaster said:

As I said in my previous post, I don't know how hard it would be to hit in RL, but compared to a human CA player, the lower AI levels seem to be much worse so far. Let's see what the next tests will show us.

 

That's my gripe with this. Comparing how well the AI does compared to a human playing DCS isn't really the point, at least from the perspective of a pilot.

 

I did a quick test with an M113 and it hit a Mi-8 consistently at 500m - while driving on rough terrain.

Now I've manned the gun in one of these way back when (albeit with a different mount) and I'm convinced that while it's good enough to intimidate infantry at 300m it can certainly not consistently hit a helicopter at 500m while driving. Same goes for infantry hip firing in auto at incredible ranges.

 

Add the almost perfect SA with no suppression and those units become a bigger threat than they should be - right now it feels like the guy standing on a pickup is more dangerous than AAA encampments.

 

For CA players ED probably would need to add extra dispersion or random sight movements for units that do not have fully automated and stabilized sights to tone them down where they should be.

 

Edited July 13, 2021 by Blackeye

[Szczerba]

There was a Mod or Script that made it way better a long time ago, since then I've lost it but it was such a small change that made so much difference, really just sick of being killed by light weapons in an Armoured fast moving heli. Radar Guided infantry with AK-74's that destroy my Hind engines is really just ruining it for me.

[Shadow KT]

I must be playing something else then.

[Dragon1-1]

On 7/11/2021 at 12:57 PM, Wychmaster said:

The thing is, that it isn't that hard to hit as you might think. Okay, I fired 500 bullets and hit only 9 times, but the AI wasn't any better. So the AI having god-like aiming seems to be a myth, at least if the first test is somewhat representative. I will do some more testing and provide the tracks, but it will take some time, since I am quite busy at the moment.

You didn't account for one thing: you're aiming with your mouse, not a real MG on a real vehicle. That's a major difference. DCS AI might not be too good compared to a human on the computer, but it's definitely too good compared to RL. Vehicle motion, weapon recoil (which, surprisingly, is a factor with mounted weapons, because the mount is usually not inline with the barrel), imperfect sight picture and a myriad other factors all conspire to degrade the accuracy of an RL machine gun. Infantry weapons are even worse, as anyone who ever shot a rifle from a standing position on a shooting range can attest to. It's doable, but not easy, and you shouldn't expect great accuracy with a handheld automatic weapon.

Also, one thing that jumped out to me, particularly with non-AAA units: line of sight rate. Most ground units use telescopic sights of varying types and magnifications. To hit something, you need to put both it and the aimpoint in the sight. The problem with BTRs shooting at aircraft isn't elevation, but their telescopic sight - for a fast moving target, the aimpoint and the target will not be in the sight picture at the same time for this reason, making it impossible to aim. A Hornet making a Mach 1 pass on deck should be safe from most ground fire even when flying straight. Slowmovers usually won't get any benefit from this, but a low pass right over the enemies' heads should be much less risky than it is now - it's simply happening too fast for them to react. Not so in DCS, where you'd get shot full of lead fire at a ridiculous angle.

[Wychmaster]

vor 1 Stunde schrieb Dragon1-1:

You didn't account for one thing: you're aiming with your mouse, not a real MG on a real vehicle.

Actually I did:

Am 11.7.2021 um 12:57 schrieb Wychmaster:

As I said in my previous post, I don't know how hard it would be to hit in RL, ... 

I have never fired a 50cal in my life, nor any other mounted machine gun,  only some assault rifles. I can imagine that it is really hard to aim, when you are driving over uneven terrain. But unless somebody who actually fired a 50cal at a chopper or something comparable makes a statement here, we can only guess.

I won't argue that the AI certainly cheats in ignoring some real life limitations of the weapon systems that you mentioned. But it is not godlike as claimed.  If they shoot a lot, you might get hit. Maybe the percentage is higher than in real life but if you give the AI time to fire 200+ bullets at you, there might also be an issue with your approach, don't you think?

The biggest problem in my opinion is that the AI sees you instantly, reacts instantly, has no fear of dying and partially ignores LOS blocking terrain. Not the overall accuracy.

 

vor 1 Stunde schrieb Dragon1-1:

A Hornet making a Mach 1 pass on deck should be safe from most ground fire even when flying straight.

Agreed, but I didn't had much problems when doing that, only when messing with AAA units. I can test it and post a track if I have the time. Just to get a solid base for a discussion.

[Dragon1-1]

If you hang around completely stationary, then it's no different than shooting at a target on the range, and it's pretty easy. However, where they are truly "godlike" is in their ability to estimate range and lead. Leading a target is hard, and even with a proper AA sight, largely dependent on experience. Try shooting skeet, or actual birds for that matter, with a shotgun, and imagine how it'd be if you had a rifle instead. Even shooting a running man, or a deer, with a rifle at a known range takes a lot of skill. It takes a very good gunner to land even one hit with something like the ZU-23 on the first burst, and that's on a predictable target that's not moving very fast. Real gunners would take some time to "find the range", and they would do that by shooting a burst, looking where it passed the target, and correct based on that. Even a hovering helo would, in a realistic situation, likely see one, maybe two bursts zip by as the gunners try to guesstimate the range. The AI makes those adjustments instantly.

Data from Vietnam shows that AAA fire was only effective in massive volumes, and I mean massive. Dick Jonas once flew a mission over Tchepone where the FAC estimated 4000 rounds of AA fire was shot at them between the two passes - and not one hit. Just about the only thing his AC did (Jonas was the GIB) was to bring them in out of the sun. It didn't always go this well, especially for A-1s strafing Vietnamese gun sites, but AA fire with no radar guidance was historically much less effective than DCS makes it. 

[NineLine]

Hi, can I have tracks of these issues please?

[Wychmaster]

Here are some tracks. I used 100 vehicles in total with 5 different types (HMMV, BTR-80, T-55, T-72, Bradley), 20 of each. I flew a Hornet with full afterburner lower than 100ft. In the first test I flew directly over their heads. I passed them twice. Once coming in at their 12 and once at their 6. In the second test I passed them at a distance of 300 (closest) to 450 (farthest) meters. I tested the lowest and the maximum AI skill level. Here is what I observed:

direct pass with average AI: 3-4 hits

direct pass with excellent AI: 4-11 hits 

300m pass with average AI: 0 hits (that's not 100% true, in one pass there were 4 hits... on some other ground units in the way )

300m pass with excellent AI: 1 hit 

 I should further mention that I did several more of the direct passes, I just realized later that I should save all the tracks. So for the excellent AI the hits on the direct pass were on average 6-7.  Also, I just used the keyboard, so the flight wasn't the smoothest (In case you wonder about my poor piloting skills).

So for starters, not flying directly over the enemy helps a lot. I can't judge how realistic the number of hits is. That's up to you to decide. I just delivered some data. Also note, that not all units fired. I did some counting for one test and it were 54 bursts. So 26 per pass.

@NineLine One interesting thing was that always during the second pass (I took a new plane for each pass) the hornet started rolling without any input shortly before I got to the targets. Wasn't easy to counter it with the keyboard. Is there a known bug report regarding uncommanded sudden rolling? If not, I can create one if you confirm that there was something strange going on. It is in all off the tracks I attached, at least I can't remember that the problem didn't occur in one of the tests.

 

EDIT:

Regarding the described rolling problem during the second pass: Its not a bug. I just realized that I most likely hit my own wake turbulence that I created with the first pass. DCS  really has an awesome level of detail.

 

 

 

AI_test_fly_over_excellent_2.trk AI_test_fly_over.trk AI_test_fly_over_excellent.trk AI_test_fly_over_excellent_3.trk AI_test_300m_pass_average_3.trk AI_test_300m_pass_excellent_1.trk AI_test_300m_pass_average.trk AI_test_300m_pass_average_2.trk AI_test_300m_pass_excellent_2.trk AI_test_300m_pass_excellent_3.trk

Edited March 1, 2022 by Wychmaster

[NineLine]

Just looking at these now, thanks.

For example, I just watched AI_test_fly_over_excellent_2.trk.

No hits on the first pass as you approached low from the rear. Only part of the group reacted or had time to react. 

The second pass, you come head on at them, more units fire so more chance of you getting hit, its just the odds right. 

So just in that example, I am not sure with a flood of rounds in the air, 3 hits wouldnt be unreasonable, but again, there is a lot of units there.