AI Improvement

[-Coulum- 35]

Yeah they do run to the hills with ASR, it´s pretty hillarious if you are in the Tank >

Ahh I see. I was thinking "duck down behind the window sill until the tank rolls by" or "lay in the bushes motionless", rather than try to run in random directions hoping eventually we get to a place that hides us before we are mowed down.

The Ai having the ability to really hide/conceal itself would not only have great effects against tanks and superior forces but also in firefights.

[Wiki 1558]

I just hope the AI pathfinding will be improved for them to drive and fly better - especially in convoy - and I hope a problem we have in ArmA 2 will be solved:

the one where drivers stop their vehicle when an animal (like a little rabbit) is on the road but don't stop when a unit is standing in front of them.

[metalcraze 290]

The bigger problem is that the pathfinding grid is too wide in ArmA2. I had a ticket somewhere on dev-heaven which showed a driver being unable to drive in between two bushes even though his car had more than enough space to do that without clipping them.

This also affects infantry - they are unable to go through any opening in the fence unless it's at least 2 men wide.

Smaller grid will solve a lot of pathfinding issues.

That and the ability for AI to freely walk on objects that aren't too high above ground. Currently AI is unable to even step onto some low pavement - while for the player it's as easy as pressing W. This and the issue above if solved will make maps like Namalsk, Podagorsk and Avghani fully playable with AI let alone will allow for much more detailed maps in the future.

Edited November 4, 2012 by metalcraze

[giorgygr 61]

The bigger problem is that the pathfinding grid is too wide in ArmA2

I have a theory..

Terrain makers (BIS included) usually making large terrains with large cell-block i.e. 10m accuracy for performance reasons.

I really think if 1m (or less) accuracy (aka microterrain) was used per cell-block,the pathfinding would be (at least) 100% more accurate with

only side-effect you would have severe performance problems.

Such example maps with small cell-block is Bushlurker's Geotypical Microterrains .

---

Of course in ArmAIII "they" say the island will be 2.4*Takistan -and for game-sales-sake they should expect better A.I. capabilities.

So either i m wrong about my theory..or they are deploying different tricks to achieve that.

Edited November 4, 2012 by GiorgyGR

[Serclaes 0]

Unfortunately if you decrease the grid size say by half, processing time should be somewhat doubled if not more. Also don't forget the update of the grid.

Path finding is really tricky and there really isn't a lot "optimization" you can do except throw more processing power at it.

Edit: Probably the terrain cell size performance hit doesn't come so much from AI as from the interpolation needed between the cells (which probably is computed by CPU). Smaller cells = More cells can be seen = more interpolation. AI still has to navigate around bushes independent of the cell size. Though I guess a big cell size with a lot of cells (= a big map) affects the AI more.

Edited November 4, 2012 by Serclaes

[-Coulum- 35]

I have a theory..

Terrain makers (BIS included) usually making large terrains with large cell-block i.e. 10m accuracy for performance reasons.

I really think if 1m (or less) accuracy (aka microterrain) was used per cell-block,the pathfinding would be (at least) 100% more accurate with

only side-effect you would have severe performance problems.

Such example maps with small cell-block is Bushlurker's Geotypical Microterrains .

I haven't done any real testing of this, but I do find that ai seem to be better at driving on maps like aliabad vs. Takistan. Or at least when the ai mess up, they don't mess up as big. Could be that since Aliabad is smaller, and more detailed (I believe) The ai are able to have more precision in their movements like you suggest. Hopefully, even though the maps are much bigger, arma 3 will have a finer terrain grid, not only for ai but just for general gameplay purposes.

I also once heard poor path-finding also had to do with animations... Is there any validity to this?

[tsmhicks 1]

I can't wait to see what the AI is upgraded to. They are already pretty decent from past mil sims but not perfect. I know in some of the mods I play in AI are pretty "dumb" . I think the AI alot of people (general public) want are AI that seem like we are playing with a buddy or in a "multiplayer" match. Basically make them seem super real that its creepy ahha

[Shagulon 1]

I have to say that tonight I tried the latest beta (no mods installed) and play good morning takistan, first fighting mission in OA campaign.

It was a HUGE improvement. I'm guessing that at the moment the devs are really getting their teeth into the AI and fixing/tuning it in preparation for Arma 3, and it seems they are making really very good progress.

Hopefully you guys can get the beta installed and give it a go, tell me if I'm imagining it :)

Better yet, try the new beta with some other AI mods and see if they work well together? They have really made some excellent changes it seems.

[Wiki 1558]

really?

I never use beta patch, so I didn't test it.

how good is the AI improvement? is it really a huge behaviour change?

[giorgygr 61]

In our team we are using Beta's and some A.I mods*

ACE+ASR_AI+TPWC_Supress+TPW_LoS

+config_AI_skillz at 0.8.

Due to fact that-lately-we are trying VERY HARD to stay alive (especially when moving carelessly) while on Air

or ground..I m sure BIS guyz are on the right track.

;)

[froggyluv 2130]

really?

I never use beta patch, so I didn't test it.

how good is the AI improvement? is it really a huge behaviour change?

They've turned a corner. Well worth checking out.

[mr_centipede 31]

They've turned a corner. Well worth checking out.

I'm sorry, english is not my first language, but what exactly do you mean? The AI make a proper turn around the corner? How if I may ask?

[cadmium77 16]

I have a theory..

Terrain makers (BIS included) usually making large terrains with large cell-block i.e. 10m accuracy for performance reasons.

I really think if 1m (or less) accuracy (aka microterrain) was used per cell-block,the pathfinding would be (at least) 100% more accurate with

only side-effect you would have severe performance problems.

Such example maps with small cell-block is Bushlurker's Geotypical Microterrains .

This is very interesting. What would make it all the more effective would be variable density of mesh. Obviously in open areas you could have 10m^2 mesh while in more dense areas like in towns or on roadways you could have 1m^2 mesh. This would be analogous to the way PTEX works in texturing or in Dynamesh in modeling in 3dCoat where extra polygons exist in greater density only where they're needed. The same could presumably be done for AI oriented meshes.

http://farsthary.files.wordpress.com/2012/10/31.png (698 kB)

Edited November 7, 2012 by Cadmium77

[maddogx 13]

I'm sorry, english is not my first language, but what exactly do you mean? The AI make a proper turn around the corner? How if I may ask?

It's just an expression. :)

turn the corner

• Fig. to pass a critical point in a process.
  
• If something or someone turns the corner, their situation starts to improve after a difficult period.
  

Source

In other words they've made great progress. And I agree, playing against the AI is now seriously challenging and fun.

[mr_centipede 31]

@maddogx

I see... thank you for clarification :)

[Tonci87 163]

Justr try the Beta Centipede. The Improvements are clearly visible. And you can deactivate/delete it anytime if you experience problems.

[giorgygr 61]

This is very interesting. What would make it all the more effective would be variable density of mesh. Obviously in open areas you could have 10m^2 mesh while in more dense areas like in towns or on roadways you could have 1m^2 mesh. This would be analogous to the way PTEX works in texturing or in Dynamesh in modeling in 3dCoat where extra polygons exist in greater density only where they're needed. The same could presumably be done for AI oriented meshes.

http://farsthary.files.wordpress.com/2012/10/31.png (698 kB)

Yes Cadmium..that you describing (and the image of course) is exactly what i was thinking :)

[metalcraze 290]

really?

I never use beta patch, so I didn't test it.

how good is the AI improvement? is it really a huge behaviour change?

- AI either attacks from their maximum possible range or sits tighter at the pre-engagement position instead of constantly closing in on the enemy. They still can pursue enemies of course, they just don't try to rush within 10m of contact anymore.

- AI gives much more preference to a nearby cover which in f.e. urban terrains results in them stacking along walls and moving in tight formations

It really makes a difference. It's certainly not a new 1.59 but it's close.

[Wiki 1558]

that sounds great!maybe the patch I was goddamn waiting for?

and in ArmA 3, the AI will act better than this?

can't wait for:

1) the final patch to be released

2) ArmA 3 release

[Psilocybe 3]

Can somebody give me direction on "config_AI_skillz at 0.8" please?

I tried looking in my config files to no avail.

[giorgygr 61]

Can somebody give me direction on "config_AI_skillz at 0.8" please?

I tried looking in my config files to no avail.

There is no a-just-one function that does it.

What i meant is:configuring your game and/or server's difficulty settings to 0.8.

For each mode Recruit/Regular/Veteran/Expert(Mercenary)

For your PC you can make adjustments either via your game menu-either from ARMA2 Other profiles(located in My Documents)->*yourProfileName->

->*yourProfileName.ArmA2Profile (open with a simple text editor and you will find the values for Each difficulty class).

Doing it for your server is somewhat the same for Your PC but you can make your life reeeeeal easy using a server launcher like

Tophe's ARMA2 Dedicated Server tool

[Wiki 1558]

here is the feedback of Shagulon:

"I think I may have been a little hasty. I think perhaps someone broke something rarther than fixed it, but the end result seems to play much better that the current released AI.

I find that soldiers seem to stay exactly where they spawn/were placed in campaign. They never move and always seem to do the same thing each time i play a mission.

The first few times I played the game seem more polish and finished, as I guess it plays out more like the mission designers wanted. However, you lose the "anything can happen" sandbox feel.

Anyone else find this? I think it needs a bit of a review? "

so, what about the last beta patch?

has any of you experienced the same?

thanks

[maturin 12]

Just make better missions with Guard, S&D waypoints or scripts, rather than plunking down directionless units and expecting something to happen.

[giorgygr 61]

here is the feedback of Shagulon:

"I think I may have been a little hasty. I think perhaps someone broke something rarther than fixed it, but the end result seems to play much better that the current released AI.

I find that soldiers seem to stay exactly where they spawn/were placed in campaign. They never move and always seem to do the same thing each time i play a mission.

The first few times I played the game seem more polish and finished, as I guess it plays out more like the mission designers wanted. However, you lose the "anything can happen" sandbox feel.

Anyone else find this? I think it needs a bit of a review? "

so, what about the last beta patch?

has any of you experienced the same?

thanks

Look..

I really haven't tested the behavior you describing but..

..IF A.I. does always "what is have been told to do"..THAT'S_DA news. :cc:

On the other hand the "anything can happen" (aka sandbox feel) ..should happen if the mission maker haven't given enough "specific" directions on "what-AI-does-next" and under "What-circumstances"

So in example..

If mission maker just puts an OPFOR group idling and during tests he constantly changing BLUFOR groups composition and/or approach-and as result the OPFOR behavior/reaction would be the same then YES..it's fail :q:

Edited November 8, 2012 by GiorgyGR

[mr_centipede 31]

I've always had a test mission of a BLUFOR squad assaulting an OPFOR squad defending the town of Kamenka at Utes (the northern town). usually I put the blufor squad attacking the town from the east along the main road into the town. Before the patch, the OPFOR almost always got killed. With recent patch, the OPFOR almost always win, because of cover from buildings in the town whereas for the blufor, they have no cover whatsoever... but the firefight does take a bit longer... with a brief period(s) of lull moments. (I kinda like that atmosphere, before if there were no more shots, that means the enemy are already gone)

But if I make an aproach from the southeast through that lonely barn and a few stonewalls, the win ratio is 50-50... They take cover effectively, advance carefully... overall smarter AIs.

I like it a lot. Cant wait for ARMA 3, with ragdolls, smoother animations etc... 2013 will be the best year ever :)