AI Discussion (dev branch)

[teabagginpeople 398]

24 minutes ago, Tankbuster said:

Wonderful. A change that breaks things. Any chance of some changes that fix things?

 "it is a fix for broken behaviours with skill close or equal zero". 

 

1 hour ago, Alwarren said:

 

I fully agree and I am a bit worried to read things like "Some of those things (especially spotting related) might be well out of balance now". I hope that ample testing is given to these things, and that BIS doesn't only rely on community testing. For example, we don't use dev branch for our server ever since it means having daily updates and that is just too much . 

 

As others have pointed out, re-scaling skill values will destroy a delicate balance that most of the clans I know of have worked out over the years, so it better be worth it. To be perfectly honest. I would have preferred to see that kept as it is, as broken as it might appear, and concentrate on other aspects of the AI. But that is just me I guess. I do applaud any work being done on the AI, I just wish the focus was elsewhere.

Al I understand your beef with this kind of change that you have set up your AI as you and your clan want them. and this will be going into an unknown now .

 

But here is the kicker. All your tinkering with these settings put you in the front line of people most knowledgeable with this to help with feedback community side so as to make make these changes beneficial. 

 

From my view . this could be a jumping off point which if successful can possibly start some drive towards other potential Ai improvements.

[Grumpy Old Man 3545]

2 hours ago, klamacz said:

Yes, skills are still interpolated through cfgSkills, and skill sliders from settings.

Yes, the change may break or unbalance any mission which uses AI skill. 

 

Might download the devbranch to see for myself how big of an unbalance this is going to be (if at all?).

Any chance of a conversion table so I know what current value represents a value before change? Or maybe even the formula would be more sufficient so I can just hook it up to automatically recalculate my old values to the new ones?

Thanks for the answer, much appreciated!

 

1 hour ago, teabagginpeople said:

 "it is a fix for broken behaviours with skill close or equal zero".

 

0.5 isn't "close to zero" on a scale from 0 to 1.

 

1 hour ago, teabagginpeople said:

From my view . this could be a jumping off point which if successful can possibly start some drive towards other potential Ai improvements.

 

From my view they're trying to bring the AI to a state where they can say it's well enough to leave the AI in an "as is" state before sending Arma 3 into retirement and to focus on a possible successor.

 

 

Cheers

 

[Tankbuster 1746]

1 hour ago, teabagginpeople said:

 "it is a fix for broken behaviours with skill close or equal zero". 

 

All AI skills are close to zero, that's how we make them work and we spent a long time making tiny adjustments to get them just right. My skill level is another matter.

[Variable 322]

A conversion table is an absolute must given that such a fundamental change come that late into the game's life cycle.

 

I don't want to find ourselves wasting the time of 20 people over several coop nights just so we can find the new sweet spot.

 

BIS please don't mark it as an optional deliverable.

[klamacz 448]

3 minutes ago, Variable said:

A conversion table is an absolute must given that such a fundamental change come that late into the game's life cycle. BIS please don't mark it as an optional deliverable.

 

Current plan is to tweak obvious problems till the point that standard skill markers behave similarly with previous implementation. 
As standard skill markers you can consider 0.2, 0.5 and 1.0 skills (so lowest, medium, and highest). We measure reaction times, spotting times, accuracy, decision delays etc and try to get them close to what you seen before with similar values.

That means that conversion table would not be necessary.

 

18 minutes ago, Tankbuster said:

All AI skills are close to zero, that's how we make them work and we spent a long time making tiny adjustments to get them just right. My skill level is another matter.

 

The big change is the curve of skill, we predict that with current code, actual reaction time, or accuracy would be linearly proportional with skill you've set. It means that you won't have to spend time tinkering with your settings for weeks, because you will get more what you actually set.

 

19 minutes ago, Grumpy Old Man said:

 

Might download the devbranch to see for myself how big of an unbalance this is going to be (if at all?).

Any chance of a conversion table so I know what current value represents a value before change? Or maybe even the formula would be more sufficient so I can just hook it up to automatically recalculate my old values to the new ones?

Thanks for the answer, much appreciated!

 

 

0.5 isn't "close to zero" on a scale from 0 to 1.

 

 

From my view they're trying to bring the AI to a state where they can say it's well enough to leave the AI in an "as is" state before sending Arma 3 into retirement and to focus on a possible successor.

 

 

Cheers

 

 

Thank you, unbalance will be serious. Also, expect bunch of tweaking commits following that change, and be not too attached to momentary effects you see. Anything might get tweaked as we go through it. 

[Vasily.B 529]

Dear @klamacz,
 

I'm happy you have replied to most of questions here. I will not ask "when", but will ask "are You" planing to fix :
AI Infantry:
- Rooftop pathfinding
- Rearming
- Selfhealing
- Clearing buildings
AI vehicle:
- BRAKING (fill fix raming friendly infantry, fromation keeping etc.)
- Using HE rounds against infatry (tanks to be precise)
- Attention to nearby infantry
- Air freezed choppers (hovering and waiting for death to come)
- pathfinding for wheeled vehicles (also for tracked - bypassing forrest and vegetation that cant be destroed - fixing vehicles stuck on trees)

Thats all problems for AI I have noticed from this thread after long lack of presence, most frustrating, most intriguing, and forever unreplied.

If you have lack of AI programmers, maybe its good to connect with some other studio, same as You have done with some DLC? Just thinking out loud and looking for some idea...

[Grumpy Old Man 3545]

1 hour ago, klamacz said:

Thank you, unbalance will be serious.

 

Allright, back to the 155mm it is, then.

That'll show 'em.

 

Cheers

[Alwarren 2767]

3 hours ago, teabagginpeople said:

But here is the kicker. All your tinkering with these settings put you in the front line of people most knowledgeable with this to help with feedback community side so as to make make these changes beneficial. 

 

In how far? Every single one of the AI skills is just a number. As it is now, there isn't even a correlation between relative increase of that number and its effect. A change from 0.2 to 0.21 might have more effect than a change from 0.2 to 0.19. 

The values we are using have been tweaked over a long time of trial and error. If you notice that people get one-shot by a guy 400 meters away with an AK and iron sights, you know that aiming accuracy is too high. If that guy five meters in front of you don't hit, you are too low. Finding the middle way is really just trial and error, there is no knowledge involved.

 

Most of all though this takes a long long time. I don't feel like going through that experience again. 

 

I also fail to see the necessity of that change. It has been like this since an eternity, and while it certainly isn't a "good" system, it is established. Changing this now, so late in the game (literally) will just reset most of the experience people had, and without at least some sort of idea how to convert the old values to the new ones, that is nothing that I want to participate in, more so since the outcome will not improve the AI in any regard whatsoever, it will just re-scale the range of the already existing AI skills.

[iV - Ghost 50]

The Ai shouldn't see you:

 

• through bushes
• through trees (leaves)
• through gras
• through boards in front of the windows
• In the dark
• through channel pipes and other stuff from the editor

[Variable 322]

Come to think of it, the only portion of the community that will benefit from it are bran new players. How many of those exist after four years?

[en3x 209]

Variable what do you mean? There are couple of invade and annex public servers filled with 60 people daily that play against AI.

[lex__1 422]

9 minutes ago, iV - Ghost said:

The Ai shouldn't see you:

 

• through bushes
• through trees (leaves)
• through gras
• through boards in front of the windows
• In the dark
• through channel pipes and other stuff from the editor

A shot from a weapon with a silencer should complicate your detection.

[scottb613 285]

 

Probably the topic MOST players want real improvements for - .

 

Regards,
Scott

 

 

[iV - Ghost 50]

@lex__1:

We are seen through these objects before we become loud. We don't need to shoot first.

I think the Ai don't know that they exist.

 

[old_painless 182]

6 hours ago, Grumpy Old Man said:

From my view they're trying to bring the AI to a state where they can say it's well enough to leave the AI in an "as is" state before sending Arma 3 into retirement and to focus on a possible successor.

 

I am not reading anything about actually changing or improving AI behaviour. It seems this is just a fix that makes adjusting skills more easy/intuitive to work with. 

 

So when the dust settles, the AI will be no better or worse than what we have now. Taking cover, reacting to being under fire or driving will be just the same. At least, that is how I read this update (hope I'm wrong and there will also be improvements :-) )

[Variable 322]

Variable what do you mean? There are couple of invade and annex public servers filled with 60 people daily that play against AI.

And since all these servers tweaked their AI to their liking, changing the scale will set them back to the point they have to refactor the AI again.

[froggyluv 2136]

 

 Lets settle down and see where this takes us - we're still early in dev branch and dont know what their full intentions are yet. Perhaps they need a more standardized approach as to better create newer ai reactions and possibly behaviours for upcoming Tac-Ops. Progress in the mired world of AI is most definitely gonna take some (probably alot) of steps back but if the future intended result is an upgrade well than its all worth it.

[fn_Quiksilver 1636]

On 8/3/2017 at 0:45 PM, FallujahMedic -FM- said:

Moderator hat off. 

I say this personally, as a player and not a moderator, but since when did BI ever answer to you? The product that was advertised and which I purchased (way back in 2013) has been delivered and delivered in spades. Anything that came after that purchase date is a bonus. BI are unique in that they continue to improve their product. Sometimes it works out as a positive and other times a negative. 

Ever bought a car, only to have the next model year have better features?

Maybe it's my age and I'm out of touch, who knows. But, I come from a time where when you bought a game, that was its final version. Want an update? You'll have to buy the new version for that. So, when BI offer features and updates free of charge, I'm not one to look a gift horse in the mouth. 

Moderator hat back on. 

 

I stopped asking for forward AI development years ago ... I think its not too much in the RV4 engine to get a vehicle to drive around a set of waypoints for 5 minutes without breaking wheels or seizing up for no apparent reason.

 

Is that unreasonable?

[fn_Quiksilver 1636]

52 minutes ago, Variable said:

And since all these servers tweaked their AI to their liking, changing the scale will set them back to the point they have to refactor the AI again.

 

yes, refactoring for no net gain would seem to be a little odd.

 

Takes a long time to get AI accuracy which feels good to fight, with all the variables at work (CPU, AI numbers, engagement ranges, player count, etc)

[en3x 209]

I see what you mean Variable.

[Alwarren 2767]

4 hours ago, iV - Ghost said:

The Ai shouldn't see you:

 

• through bushes
• through trees (leaves)
• through gras
• through boards in front of the windows
• In the dark
• through channel pipes and other stuff from the editor

 

 

They actually don't, IF the objects are properly configured and have  the correct LOD's.

 

The problem is that the AI hearing is WAY over the top and by far too precise. They can spot you by your movement sounds unless you crawl slowly, and they immediately know you are an enemy. There are demo videos on YouTube that show this. AI hearing is, unlike vision, not impaired by things like rain and thunderstorms, wind, engine sounds, environmental sounds, nothing. 

 

[Greenfist 1863]

43 minutes ago, Alwarren said:

The problem is that the AI hearing is WAY over the top and by far too precise. They can spot you by your movement sounds unless you crawl slowly, and they immediately know you are an enemy. There are demo videos on YouTube that show this. AI hearing is, unlike vision, not impaired by things like rain and thunderstorms, wind, engine sounds, environmental sounds, nothing. 

 

Well you can just walk up to them from behind without being detected, but that might require skill setting low enough to render them blind in certain situations.

 

The hearing is impaired by one thing, by the way: Objects and terrain; the same things that block their vision. Therefore I assume it uses the same LOS checking mechanics as the visual detection.

So why can't it be dynamically impaired by the surrounding sounds? Like every environmental sound, gunfire, loud vehicle would increase the "aural fog" around them to reduce the hearing.

Perhaps that would just need too much extra cpu cycles we can't afford.

[iV - Ghost 50]

Ghillie Suit

The ghillie suit should give you much more stealth effect then other clothes.

It should be possible to get invisible if your behavior is correct.

[Alwarren 2767]

 

12 hours ago, Greenfist said:

Well you can just walk up to them from behind without being detected, but that might require skill setting low enough to render them blind in certain situations.

 

Only when the are not Aware and the skill is low enough.
 

 

 

Quote

The hearing is impaired by one thing, by the way: Objects and terrain; the same things that block their vision. Therefore I assume it uses the same LOS checking mechanics as the visual detection.
So why can't it be dynamically impaired by the surrounding sounds? Like every environmental sound, gunfire, loud vehicle would increase the "aural fog" around them to reduce the hearing.
Perhaps that would just need too much extra cpu cycles we can't afford.

It's not only dynamic stuff. Rain would be easy to add, since it is a global value. Other factors like the oil wells on Takistan would also be possible, as well as ocean. There is definitely wiggle room for a few basic tweaks which would make the playing field more level.

Also the accuracy with which they detect you by hearing is too good.

Sent from my SM-T580 using Tapatalk
 

Edited August 5, 2017 by Alwarren
Edited for typos. I was half asleep when I wrote this.

[teabagginpeople 398]

1 hour ago, Greenfist said:

 

So why can't it be dynamically impaired by the surrounding sounds? Like every environmental sound, gunfire, loud vehicle would increase the "aural fog" around them to reduce the hearing.

Perhaps that would just need too much extra cpu cycles we can't afford.

This and as pointed out. Ai  hear a sound and Just immediately  know it's enemy.  would  be nice to have some sligh change of awareness to  search /  curious / need to confirm state inbetween.