Enemy Accuracy way to good.

[max power 21]

The soldiers are larger than the closest range marker on the dragunov's sight. He's firing in a ghillie's suit, silhouetted on a rocky ridge. It's like, "Oh noez! Chewbacca is pissed off and he's shooting at us from that gravel bank!"

If he was shooting from under some grass or in a forest and from further away there would be something to complain about.

[maddogx 13]

Dude, you were shooting from like 100 meters away. That's spitting distance. I would be embarrassed if they didn't hit you first shot.

Probably more than 100m, but still there are several noticable things about that video. For one look at the ammo counter - it fades out. That means he was playing on expert difficulty. Secondly he fired at the AI from well within their field of view. The AI could easily have seen the muzzle flash and immediately known his position.

Last but not least, it looks like the AI that killed him fired a single shot. That's a strong indicator that it was a sniper, as AIs with AKs usually fire bursts.

[TangoRomeo 10]

Doesn't sound like single shot to me. With increasing distance AI bursts just get shorter. Noticeable muzzle flash against the sky during bright daylight?

That's just not right.

[icebreakr 3159]

In real conflict when you ambush/suprise someone in even a 10-20m distance with some cover he still wouldn't respond quickly like this AI :).

Sniper in the youtube video was on open, but I bet he would get wasted even when hiding behind a big bush. Lets face it, AI just knows too well where you are, even if he doesn't have a visual he knows exactly where you are and where you will appear ;)

For example to see how AI cheats in A1 (and A2), test out new Unsung mod with very dense vegetation and observe reporting of your AI teammates what can they perfectly spot in a thick jungle - you won't even see an enemy while your AI teammate will already report back with "got him" :)

Hopefully BIS will address this years old issue... one solution we've used in the past is to make bigger dispersion for AI weapons.

Edited June 7, 2009 by IceBreakr

[odjob 0]

I did a quick test by placing myself as a force recon (with nvg) about 10 meters from a russian soldier (without nvg) at moonless night, i stood infront of him and looked at him thru my nvg and he didnt see me. When i took a step he took a posture i havent seen in ofp/arma1, kinda leaned forward and looked in my general direction but still couldnt see me. It looked as he was trying to listen were i was, i took a few more steps and he shot me. I tried a few variants of crawling slow/fast crouching and it was different each time, once i crawled (slow) right up to him and got the menu for "greeting" him, when i did he responded with "yup?" lol.

All put together i think it was scarily real :)

[LeadCommando65 10]

I noticed a strange situation, when playing around against enemy static machine guns (east). Everytime they kill me with the first shot, directly to the head. No matter if i'm hiding in the woods, or running in open range. And it seems they see me even if i'm hiding or lying in the woods 300m away -> headshot... I don't believe this was intended.

... no, I didn't turn on Super-AI. I'm playing on Cadet.

[maddogx 13]

Doesn't sound like single shot to me. With increasing distance AI bursts just get shorter. Noticeable muzzle flash against the sky during bright daylight?

That's just not right.

1. Just watched the video with the sound very loud. That is a single shot. There are possibly more gunshots after the player died, but they are definitely lower than from the "killer" AI.

2. Against the sky :rolleyes:? I just took the time to locate the exact place that video was taken. Wasn't too hard: it's in the south of Chernarussia, just east of Cap Golova. And guess what: behind the player there's a large rocky cliff face.

3. I just checked the distance - it's roughly 160m.

[fabrizio_t 58]

I just made a test case to check out how unaware AI group (leader, rifleman, gunner) with maxed skill reacts against a single and well placed sniper shot taken from long distance (750m). Sniper wears ghillie suit and is concealed by only by grass.

I'll post a video about that later.

These were my findings: after a single lethal sniper shot AI group became aware of exact position of the sniper (me) and the enemy gunner returned fire from 750m, wounding him.

As you may imagine i was worried.

I retried the same scenario and this time i intentionally missed the target, hitting ground very near to the group. Nothing happened, AI was still unaware. I took another shoot, missing again. At that point the AI became aware of my presence and run towards my "general" direction, not exactly towards me. They took no shot at me. Stopped mission when AI was about 300m from me, it did not gave any sign to know my exact position.

So here is what i think: AI is really "cheating" in the case you hit or kill one of its units. In that case it instantly become aware of your exact position.

That's an issue that should be fixed.

EDIT:

i did a third test: took a single lethal sniper shot from a concealed position (behind a tree). This time AI was NOT able to pinpoint the sniper position.

See the linked video for a test summary:

Edited June 7, 2009 by fabrizioT

[dmarkwick 261]

I notice another thing about that video, that the gunfire sound from the enemy return fire didn't seem to be subject to the speed of sound. As soon as the puffs of gunfire were seen, the gunfire was heard.

[fabrizio_t 58]

Let me add here too asidenote to the test i posted before:

the gunner you see returning fire at the end of test 1/3 me got the sniper (me) wounded.

He needed a few volleys, but he did.

So he wounded a man wearing ghillie suit and prone into grass, from a distance over 750m, while standing up and firing on auto with a machinegun ...

Now you may better understand me when i say that AI firing precision both for suppressive and aimed shots is just a bit too much.

Consider that in this test "precisionEnemy" was set to 0.8, while default is 1.0 ;)

[froggyluv 2136]

Let me add here too asidenote to the test i posted before:

the gunner you see returning fire at the end of test 1/3 me got the sniper (me) wounded.

He needed a few volleys, but he did.

So he wounded a man wearing ghillie suit and prone into grass, from a distance over 750m, while standing up and firing on auto with a machinegun ...

Now you may better understand me when i say that AI firing precision both for suppressive and aimed shots is just a bit too much.

Consider that in this test "precisionEnemy" was set to 0.8, while default is 1.0 ;)

This is really a puzzle. I just did the reverse test of yours, that being enemy AI sniper out at 500-600 with a DoFire command on me and 1 AI mate. When I lowered my squaddies skill slider to very low, he generally never spotted the sniper after repeated shots at me. When I upped his skill to max, he would call out "Enemy, man, far" usually after 1-3 shots but never returned fire.

[TangoRomeo 10]

2. Against the sky ? I just took the time to locate the exact place that video was taken. Wasn't too hard: it's in the south of Chernarussia, just east of Cap Golova. And guess what: behind the player there's a large rocky cliff face.

Unless the firing position is located in an area with very low ambient light, it´s really unlikely to spot most small arms muzzle flashes during bright daylight without optical aids, especially at the distance and background settings involved in the video.

The AI reaction showcased there is not comprehensible to me. I´m all for a challenging game, but on equal odds. If i can't see through grass, bushes etc., but the AI can, that's no joy. If i don't get a benifit from wearing a ghillie suit and hugging the dirt, while having a hard time acquiring even standing AI soldiers in standard BDU's against certain backdrops - that's something that might be worth another look.

[fabrizio_t 58]

Two more vids:

single shot, lethal, taken with M107, distance 1000m.

At least in these cases AI has some more problems in spotting the sniper and / or returning fire. But hey, now i am 1Km away!

I think that more testing is needed to better understand tha AI spotting mechanics on distance.

However at the moment i think it's safe to say that under some circumstances the AI behave not realistically, being able to see / hear better than expected.

Now let's try to narrow which these "circumstances" exactly are.

BTW: by the way, as a sniper i never position on top of a hill, but just some meters lower, since i don't want to gracefully put my silhouette gracefully the sky.

I am pretty sure that AI won't care however ...

Edited June 7, 2009 by fabrizioT

[fabrizio_t 58]

This is really a puzzle. I just did the reverse test of yours, that being enemy AI sniper out at 500-600 with a DoFire command on me and 1 AI mate. When I lowered my squaddies skill slider to very low, he generally never spotted the sniper after repeated shots at me. When I upped his skill to max, he would call out "Enemy, man, far" usually after 1-3 shots but never returned fire.

try making your squaddie a gunner.

i bet he will return fire then.

[froggyluv 2136]

try making your squaddie a gunner.

i bet he will return fire then.

Actually after running it 30+ times, the only timed my AI returned fire is when I looked at the enemy through bino's whether he was previously declared or not. AI skill definitely seems to have big effect on the spotting though.

[turin turambar 0]

i did a third test: took a single lethal sniper shot from a concealed position (behind a tree). This time AI was NOT able to pinpoint the sniper position.

See the linked video for a test summary:

Something fishy it's going here. The enemy seems to have a "bonus to detect" when the shot kills someone, that machinegunner did detect you at 750 mts, it just that he couldn't hit accurately you given that distance. In the other hand, the same shot that hits the ground is mostly ignored.

Thanks for the test, it's good to have factual proof.

[walker 0]

Hi all

For an experiment to be valid it must be repeatable videos are illustrative of the results but not verifiable.

Release a copy of the experimental mission so others can test and verify.

Scientific Method

We should all be using scientific method to support our arguments

So use this as a Template

----------------------------------------------------

For the attention of:

Who needs to take notice, who is the intended audience

Name of the Experiment

Instructions to reproduce the "..."

Required Resources:

A list of what is needed to repeat the experiment.

At the very least a link to the experimental mission as a download

Summary

A description of what to expect

Method

Describe how to do the experiment

Observations

What you have observed to happen

Results

The effects of the experiment

Further:

Any additional thoughts or advice

Cause

Your hypothesis of what is causing the effect

Solutions

Suggestions on how to solve it if it is a problem

Known Workarounds

Problems

Problems with each workaround

Additional Material

Materials that provide additional support for the results

------------------------------------------------------------------------

You do not need to include each field and you can add you own; it is guidline not a bible.

Here is an Example of how to conduct an experiment from a bug in ArmA I that hope gets fixed in ArmA II

Hi all

As Requested by Armored_Sheep

Instructions to reproduce the "local bug"

Required Resources:

* ArmA1.15

* A Dedicated Server

* Mission “AI Local Bug†(Link below)

http://www.zeus-community.net/z/AI_Local_Bug.zip

* 2 ArmA client computers

* 2 human players

Contents of mission

* 2 BLUEFOR Groups (but can be any side)

* Group 1 is 1 player plus 4 AI Slots (2 playable slots one to test one to observe bug if needed)

* Group 2 is vehicle group (2 playable slots one to test one to observe bug if needed)

Method

* Start Server

* Two Players log on to the Dedicated Server

* Load attached mission “AI Local Bugâ€

* 1st Player joins group named as “1-1-A†in the “Team Leaderâ€

* 2nd Player joins group named as “1-1-B†in the “Stryker Driverâ€

* Start mission

* Vehicle is stationary

* 1st Player tells all four of his AI to disembark keys (all four 1) [`] [4] [1]

Results

AI will not disembark

AI will only disembark if the the driver of the vehicle (2nd Player in this case) gets out (in this example)

Further:

AI will not disembark if any driver crew (human or AI) is in the vehicle and not part of the 1st players group (therefore local to 1st player)

Cause

The cause of this bug is believed to be priority of locality for the vehicle being driver; who is not local to the Player commander with AI in the cargo of the vehicle.

http://community.bistudio.com/wiki/Locality_in_Multiplayer

Known Workarounds

You can have the crew eject while vehicle is in motion

Problem this leads to AI death and injury.

You can script an ejection

Problems

1) Not available in ArmA editor

2) Requires better than beginner knowledge of Real Virtuality engine scripting

3) Using Setpos can lead to AI and players setposed in to building walls etc.

4) Eject is ugly as players all land on top of each other requires a delay to clean up.

ACE has a working example of a scripted work around using the “Jump Out†action menu (all, six {number of Jump out user action }) [`] [6] [#].

Problem Ensuring Jump out is first action in user action menu

Additional Material

A video of this bug can be made available to BIS developers if required through Placebo via Skype.

Thank you to Terox for aid making example mission and testing it with me.

Kind Regards walker

Kind Regards walker

Edited June 7, 2009 by walker

[walker 0]

The enemy seems to have a "bonus to detect" when the shot kills someone.

Hi all

Some questions:

Would a body with an entry and exit wound give additional information on where you were shot from in the field?

Would the bullet still be flying at supersonic speed at 750m?

Kind regards walker

Edited June 7, 2009 by walker

[Alex72 1]

It is quite good, though I wish I could set a random accuracy, to simulate some people being a good shot, others average, others terrible. They all seem to be the same good shots.

ArmA1's AI got this later on in a patch. Some were good some bad some average (automatically that is - no slider). So it will come. ;)

It must be hell to code AI to become realistic. Pretty sure about that. And we are at patch 1.01 now and i have briefly tested ARMA2 and loved the AI during that short amount of time. I bet there are cheats from the AI and that they do some weird things - but remember the previous game - it got much much better and im sure BIS wont let us down.

Alex

[dmarkwick 261]

On a prectical level, we know that the Hit event has a few parameters passed to it, one of which is the unit that fired. I might suggest this is the simplest conclusion as to how the units get their target so quickly.

Now, what to do about it, another question :)

[froggyluv 2136]

Trying to tweak the AI response to feel human has got to be the hardest thing as there is that fine line of " They're not responding fast enough!" to "They're Uber spotters!". I think distance from fired shot should be a huge deciding factor to how fast they respond. One problem we see in CQB especially indoors is the AI's super-slow response time even when the threat is immediate.

Edit Actually CQB is much improved so I meant indoors.

Edited June 7, 2009 by froggyluv

[afp 1]

I watched the videos too, it's so obvious the AI is too accurate and reacts too fast.

Its very unlikely to belive that in combat condition, when someone sees his mate dying in front of him, he reacts in split second, sees the muzzle flash, gets proned, acquires target and delivers a first deadly shot. These all happen faster than the shooter have time to acquire him while proned and undercover, after the first shot...

I guess first reaction should be to take cover anyway.

Edited June 8, 2009 by afp

[malkuth 0]

Can't wait for some of you that don't have the game, to play and see how asome the AI is at shooting at you. LOL. 75% of the time You don't even know what hit you.

Then you can come to this thread and say how great it is... And that the inhuman reaction time is Great.... And realistic and fun... Don't forget the fun part...

LOL.

[thyco 0]

Can't wait for some of you that don't have the game, to play and see how asome the AI is at shooting at you. LOL. 75% of the time You don't even know what hit you.

Then you can come to this thread and say how great it is... And that the inhuman reaction time is Great.... And realistic and fun... Don't forget the fun part...

LOL.

and the chances are that the 505 version is release and patch 1.02 is released on the very same day that fixes these problems and people will come to the thread and say i told you so when they dont experience any of the superhaman dection AI

Edit: in fact i think patch 1.01 final has fixed some of it up for me so far.....havent had any turrets follow my every move behind concrete walls while crouched and no shadow visible for the enemy to follow.

Edited June 8, 2009 by thyco

[Majormauser 0]

Can't wait for some of you that don't have the game, to play and see how asome the AI is at shooting at you. LOL. 75% of the time You don't even know what hit you.

Then you can come to this thread and say how great it is... And that the inhuman reaction time is Great.... And realistic and fun... Don't forget the fun part...

LOL.

The reaction time and the tracking/Spotting need to be looked at. Also the Grass being invisible to the AI, but yet I can't see through it to view them is quite cheap. I'm sort of ok with the 750 yard without a scope sighting but come on 750 yards away the sniper in the video takes one shot.... the AI is the size of ants in the scope and they still spot him?!