AI Discussion (dev branch)

[Varanon 892]

some driving issues posted on STEAM discussions, unfortunately the author has no time to create repro mission :(

I can provide a repro for the first problem, since I had exactly that in a mission I'm doing. You can find it here. Altis, unfortunately, since that was where I noticed it.

To test, just order your teammate to next waypoint (f2 -> 1 -> 9). He will hit the edge of the house in the bend and take ages to get off of it.

[Tyl3r99 41]

Ai thinks the HEMTT is shorter than it already is ;)

[Varanon 892]

Ai thinks the HEMTT is shorter than it already is ;)

Yeah, they should take the curve at a bigger radius. That's one of the first things you notice as a normal car driver whn you drive a truck for the first time....

[Tonci87 163]

http://www.youtube.com/watch?v=LOcNSfT3xx8

http://www.youtube.com/watch?v=wnRW4qpGVSA

some driving issues posted on STEAM discussions, unfortunately the author has no time to create repro mission :(

@Dwarden

If you want to fix something driving related then take a look at this repro:

http://forums.bistudio.com/showthread.php?159710-Development-Branch-AI-Discussion&p=2515524&viewfull=1#post2515524

There is a whole road that AI can´t properly drive on without geting stuck or destroying the vehicle.

[bad benson 1733]

Very, very, very occasionally I'll get an enemy unit enter a building and come upstairs looking for me. So I know they can do it :) but they seem reluctant to even enter the building at all most of the time. They prefer to dance around outside :D

yup. a real shame to be honest. just some simple restrictions could easily keep house usage from being game/mission breaking (being one of the excuses).

1. only a few men of a squad are allowed to get inside a building for stationary combat.

2. if the player is inside a house AI squad leaders send units to look for you.

3. squad leaders never go inside on their own to keep squads moving and avoid getting stuckon houses.

4. when the leader moved on a certain distance units are forced outside. again to keep them getting stuck.

5. set a time limit for camping a house.

just some ideas to keep it a dynamic addition rather than a game changer. AI indoor behavior is very limited after all.

[jgaz-uk 132]

They are of course adjustable:

http://forums.bistudio.com/showthread.php?147537-Tutorial-How-to-run-ArmA3-on-a-dedicated-server

and

http://community.bistudio.com/wiki/server.armaprofile

Both on these forums...

Except at the moment in ArmAIII on a Dedicated server a number of settings dont have any effect.

[Woodpeckersam 18]

yup. a real shame to be honest. just some simple restrictions could easily keep house usage from being game/mission breaking (being one of the excuses).

1. only a few men of a squad are allowed to get inside a building for stationary combat.

2. if the player is inside a house AI squad leaders send units to look for you.

3. squad leaders never go inside on their own to keep squads moving and avoid getting stuckon houses.

4. when the leader moved on a certain distance units are forced outside. again to keep them getting stuck.

5. set a time limit for camping a house.

just some ideas to keep it a dynamic addition rather than a game changer. AI indoor behavior is very limited after all.

Well if you check my feedback ticket, you can see that my idea will eliminate most if not all problems with AI inside buildings (and out of buildings of course).

My idea is not a NEW idea, but it (i hope) is new for ArmA 3.

http://feedback.arma3.com/view.php?id=12962

[relain 1]

Hey guys,

maybe you can help me out: I just ran some test with the dev built AI. Maybe I am totally off - but is it possible that the attached scope doesn't have any influence on the accuracy of the AI? *scratches head...

[mr_centipede 31]

As far as I know scope doesn't affect accuracy. But they do affect how far can they detect threats

[relain 1]

As far as I know scope doesn't affect accuracy. But they do affect how far can they detect threats

I am a bit suprised. In my test the AI shoots the same accuracy with a Iron sight and with Scope - even on 400m. There should be a difference, or am I wrong?

[maddogx 13]

I am a bit suprised. In my test the AI shoots the same accuracy with a Iron sight and with Scope - even on 400m. There should be a difference, or am I wrong?

Agreed. AI accuracy should definitely be affected by optics.

[Tyl3r99 41]

and depending what type of optics can vary the AI accuracy! :D

i.e. - normal scope = 0.35 sniper scope = 0.75 for arguments sake

[Tonci87 163]

and depending what type of optics can vary the AI accuracy! :D

i.e. - normal scope = 0.35 sniper scope = 0.75 for arguments sake

There is currently so much wrong with the AI...

Look at this Thread:

http://forums.bistudio.com/showthread.php?166487-ARMA3-Coop-Tactics-how-to-fight-the-AI

The last two posts contain the important issues.

That video FPDR

[relain 1]

I am a noob with this AI Request Tracker thingy - there should be a AI request for this, could someone do it or find out, if there is already a AI request "Optics should influence the accuracy"?

[fabrizio_t 58]

There is currently so much wrong with the AI...

Sadly.

One thing that bogs me (since i masochistically still play as a AI group leader) is that whenever i command my whole group to move to some spot, while in combat mode, they do move there in perfect formation (wedge) ignoring cover and as walking as slow as possible (not to break that beautiful formation). They skip bounding overwatch alltogether, or at least they fully prioritize hard formation.

Obviously this kind of behaviour is useless.

[Tonci87 163]

Sadly.

One thing that bogs me (since i masochistically still play as a AI group leader) is that whenever i command my whole group to move to some spot, while in combat mode, they do move there in perfect formation (wedge) ignoring cover and as walking as slow as possible (not to break that beautiful formation). They skip bounding overwatch alltogether, or at least they fully prioritize hard formation.

Obviously this kind of behaviour is useless.

Try the same with enemys around

[Azzur33 1]

AI accuracy has been higher lately, again, I guess, some time ago they had difficulties to hit when the enemy was far. I'm not sure. Maybe.

At 400 meters a rifleman may not see you or try to shoot you back, unless you dance around a while.

But when they do detect you, it's too often one shot, one hit. That shot may not usually kill you and you have time to try to shoot back .. unless there are more than one lucky-guess-sharpshooter AI.

At 300 meters you can quite easily hit/kill an enemy if he is standing up and not moving. It's a lot more difficult to hit when he is prone facing you and shooting back.

Still ...for AI, it seems like the opposite. I dare to say you live longer if you are standing, instead of laying low.

I did test (at airstrip), 2 AI riflemen 300 meters away, facing you.

They notice you immediately and start shooting.

If you stand still, average life expectancy was 15-30 seconds. Results varied a lot, from 10 seconds to 4 minutes 20 seconds

If you went prone right away, you died in 5-10 seconds. Every time.

With one AI, the results were more even, but still you had better chances standing still. Well, atleast when prone, you are able to try to shoot and kill the minimal target ..and pray for the lucky shot. Which won't happen, because before you can "zoom" and have a good aim, you are dead.

[Smurf 12]

What are the default values for the precision parameters?

[mr_centipede 31]

What are the default values for the precision parameters?

Depends on your difficulty settings

[klamacz 448]

AI accuracy has been higher lately, again, I guess, some time ago they had difficulties to hit when the enemy was far. I'm not sure. Maybe.

At 400 meters a rifleman may not see you or try to shoot you back, unless you dance around a while.

But when they do detect you, it's too often one shot, one hit. That shot may not usually kill you and you have time to try to shoot back .. unless there are more than one lucky-guess-sharpshooter AI.

At 300 meters you can quite easily hit/kill an enemy if he is standing up and not moving. It's a lot more difficult to hit when he is prone facing you and shooting back.

Still ...for AI, it seems like the opposite. I dare to say you live longer if you are standing, instead of laying low.

I did test (at airstrip), 2 AI riflemen 300 meters away, facing you.

They notice you immediately and start shooting.

If you stand still, average life expectancy was 15-30 seconds. Results varied a lot, from 10 seconds to 4 minutes 20 seconds

If you went prone right away, you died in 5-10 seconds. Every time.

With one AI, the results were more even, but still you had better chances standing still. Well, atleast when prone, you are able to try to shoot and kill the minimal target ..and pray for the lucky shot. Which won't happen, because before you can "zoom" and have a good aim, you are dead.

Interesting. I will repeat myself, please provide repro mission and state the version of the game you are using.

[Smurf 12]

Depends on your difficulty settings

So.... what are the default values for each difficulty?

Last time I changed them on Veteran it was around .7-8 (?) which is too high no matter what and contribute to this aimbot AI feel. As always was.

[Bouben 3]

Please Azzur, create a repro for the behaviour you are observing. This looks like a serious issue.

[fabrizio_t 58]

Try the same with enemys around

That's what i did.

This is an old issue, it was already like this in ArmA2

---------- Post added at 15:13 ---------- Previous post was at 15:08 ----------

AI accuracy has been higher lately, again, I guess, some time ago they had difficulties to hit when the enemy was far. I'm not sure. Maybe.

At 400 meters a rifleman may not see you or try to shoot you back, unless you dance around a while.

But when they do detect you, it's too often one shot, one hit. That shot may not usually kill you and you have time to try to shoot back .. unless there are more than one lucky-guess-sharpshooter AI.

At 300 meters you can quite easily hit/kill an enemy if he is standing up and not moving. It's a lot more difficult to hit when he is prone facing you and shooting back.

Still ...for AI, it seems like the opposite. I dare to say you live longer if you are standing, instead of laying low.

I did test (at airstrip), 2 AI riflemen 300 meters away, facing you.

They notice you immediately and start shooting.

If you stand still, average life expectancy was 15-30 seconds. Results varied a lot, from 10 seconds to 4 minutes 20 seconds

If you went prone right away, you died in 5-10 seconds. Every time.

With one AI, the results were more even, but still you had better chances standing still. Well, atleast when prone, you are able to try to shoot and kill the minimal target ..and pray for the lucky shot. Which won't happen, because before you can "zoom" and have a good aim, you are dead.

Please BIS, double check for AI regressions between this build:

http://forums.bistudio.com/showthread.php?149636-Beta-Development-branch-changelog&p=2489229&viewfull=1#post2489229

and current beta build.

Hopefully you'll find some.

[Azzur33 1]

Interesting. I will repeat myself, please provide repro mission and state the version of the game you are using.

Dev version, I tried it few times again with todays build, didn't see much difference between prone and standing this time.

I didn't think repro helps much in this case. It was so basic 1 (or 2, grouped) AI riflemen against 1 player rifleman at measured distance (300) on the (Stratis) airstrip.

No waypoints, no obstacles. Aligned and facing eachother.

Unit skill at half way (editor edit unit), difficulty veteran, enemy AI skill 0.85.

- And the results varied a lot, maybe too much to draw any conclusions, other that laying low doesn't have any advantages, on the open airstrip.

That is not exactly a normal game situation.

http://koti.welho.com/ktiihone/dra/standprone.Stratis.7z

I would think that it is harder to hit a smaller target, but (on that hard flat surface) AI was as lethal no matter what my stance was.

Skill settings, personal difficulty settings, I don't know how much effect they have in this. Maybe I should lower them.

And normally you wouldn't lay on the airstrip receiving bullets.

Edited October 8, 2013 by Azzur33

[royaltyinexile 175]

Skill settings, personal difficulty settings, I don't know how much effect they have in this. Maybe I should lower them.

The point relates to the next chunk of work I'd like to invest in the AI. Mr Klamacz is rather tied up with various tasks now but, on the side, he's taking a closer look at the skill configuration and how it really manifests in game.

The goal will be to overhaul the way configuration options are presented to different groups - players and mission makers - giving us a more predictable test-bed and mission makers a more meaningful (and better described) set of configuration options. I'll share an extract from the initial analysis. Please share any general reactions.

Work is a bit delayed on this for a few reasons (like TF Radio Protocol, such work requires a cross-departmental task force - UI, Program, QA, Sandbox Design) but the thought is that it will be, in the long run, beneficial to finally lock down these fundamental aspects of AI so that the ongoing tweaks and refinements can be implemented and tested with more confidence. The AI (like some other aspects of the game: video options, etc) is really an incredibly powerful beast, but massively complex to fine-tune. We need to cast a splendid new light on the basic inputs to better unlock the potential of the AI in the wild. :)

Best,

RiE

Edited October 8, 2013 by RoyaltyinExile