How's ARMA2 AI shaping up?

[Col. Faulkner 0]

[...]

9. AI leaders should not call out individual soldiers to flank/attack. They should be calling out at least a fireteam to flank.

10. Make suppressing fire work against and for AI.

11. AI work in buddy teams and fire teams.

The suggestions are all good, but these last three I would opine are

particularly important. The lack of these features is a huge immersion

killer in the game as it is now (and the main reason why I lost a lot of

interest in it).

[PhilippRauch 0]

cool idea there was ... introducing the new CoverLOD inside Oxygen.. well THAT would be really nice! :) ... well i can dream sometimes, cant i?

[lwlooz 0]

Hello,

This completely baffles me . For me computer controlled characters that have prescripted static moves ,

can't adapt to new environments very well sound more like automatons and nothing like AI.

I also believe its foolish to think AI could ever behave like a human.

I for one just wish the AI would act more as team/squad in the future. Stay together ,

force-allocate (instead of the frequently mentioned "Single Suicidal Engage") and relocate when possible mainly.

If the individual AI can shoot from behind cover now and even seeks it sometimes when CPU power is avabile that is spiffing ,

but unless the Squad AI is sorted that will mean absolutely nothing. AI individual rambos are just going to die in a more spectacular way I would assume.

If I had another wish I would hope that we had more waypoint-types like "Attack","Defend","Delay" so the AI has more parameters to work with.

But I agree that its probably the suppression thing that is going to make us a little less frustrated.

It also strikes me as odd that having lead AI in the "Real Virtuality" engine for the last couple of years a lot ,

that anyone would want less control of his/her AI teammates.

I get the fact that the squad command interface needs improving , I fully agree ,

but to even think of going with a "Ring-System" like I heard or have a very simplified command only would completely and indefinitely stop pretty much anyone from using AI.

Yes , the AI needs to take more care of itself , micro-managing is a horror no matter how you look at it , but you still should be able to tell them whatever you want them to do if you chose to do so.

If we were to get more autonomous AI ,

what is needed in my view is not only the option to tell them to "STOP FUCKING UP THIS FIREFIGHT! WHAT THE HELL ARE YOU DOING!?! OR DEAR GOD NO!" :) ,

but you should be able to set what they are allowed to do and whatnot before , their SOP if you will. I hardly see that happening in a simplified system.

The last proper A&D (That is 2 teams of humans in Multiplayer leading a number of AI units)

I have played was in OFP and that is very much because the formation.fsm breaks up your formation against your will and in the early stages of the game the AI just fired at whatever it fancied.

No Control , No Tactical Decisions , Boring Shooter.

In conclusion I think the AI in Arma2 isn't going to be up to our standards , one might say we don't have the technology for that yet and one is better looking for having all AI controlled by humans (squad control , high command(for platoon) , CoC CE4 for company and up :)).

And for that I can only second that the user interface needs some serious extra context-sensitivity put into it and if I would have to guess one thing that would be the most cost-effective thing is to supply the player with proper Attack/Target/Suppress options via that system.

But since none of that is going to happen , I probably won't have much fun in Arma2 either. Shame :)

Edited May 6, 2009 by lwlooz

[topeira 10]

After a bit of looking around in sandbox2 I noticed that these AI points are only on interiors (and the deck of the carrier). Otherwise if outside they are (mostly) free to roam with Forbidden area's shaped around objects and an area for them.

I'm guessing this is the movement area so that they can move around in a larger distance but after looking at this more I saw no crouching either unless they were by one of the "forbidden objects" and if I had to guess I'd say FC2 would be the same way though without the forbidden area's.

You're correct that I am going on assumption, since we can't make single player or view single player FC2 maps we won't really know other then going on the premise that it runs on the same engine (regardless of what they say we all know its true) and their AI acts similar in certain situations.

As for Stalker I've never played that nor GTAIV but if I had to guess I would say a similar aspect.

Oh and your virtual sensor idea is good but sounds very intensive especially if you have hundreds of AI at once.

weird. i am pretty sure that FC2's AI is using forbidden areas for huts, shacks, rocks and steep mountins but no cover nods since they dont really use cover.

and come one, nodunit - its obvious that FC2 and crysis ain't using the same engine. they look nothing alike (crysis looks much much better but the cry engine probably cant stream large areas like the dunia engine can).

lwlooz, i agree that the interface and command method in arma is pretty damgn bad. its very detailed but its very hard to use quickly and in too many times i had to actually SEARCH the thing i wanted the AI to do. also its very NOT contact sesitive. it could have been done SO much better. the command method in OFP2 is probably gonna be a lot easier and a lot more effective since its more contact sensitive.

u probably wont be able to tell the AI in OFP2 all the tiny things like in Arma2 but my guess is that you wouldnt need too either.

[ProfTournesol 956]

lwlooz, i agree that the interface and command method in arma is pretty damgn bad. its very detailed but its very hard to use quickly and in too many times i had to actually SEARCH the thing i wanted the AI to do. also its very NOT contact sesitive. it could have been done SO much better.

Interface command isn't relating to the AI question of this topic, unless this topic is about "how the player can control its squad".

the command method in OFP2 is probably gonna be a lot easier and a lot more effective since its more contact sensitive.

u probably wont be able to tell the AI in OFP2 all the tiny things like in Arma2 but my guess is that you wouldnt need too either.

Pure speculation IMHO.

cool idea there was ... introducing the new CoverLOD inside Oxygen.. well THAT would be really nice! ... well i can dream sometimes, cant i?

In fact, OFP AI knows how to use map objects to cover - using probably geo lod or fire geo lod -, just order them to find cover. The main trouble is that they won't find cover AND fire at the same time.

Edited May 7, 2009 by ProfTournesol

[thaFunkster 0]

...

I for one just wish the AI would act more as team/squad in the future. Stay together ,

force-allocate (instead of the frequently mentioned "Single Suicidal Engage") and relocate when possible mainly.

...

but you should be able to set what they are allowed to do and whatnot before , their SOP if you will. I hardly see that happening in a simplified system.

...

And for that I can only second that the user interface needs some serious extra context-sensitivity put into it ....

Yeah lwlooz, I totally agree with you on these points. AI working as small teams / squads / even pairs would be great.

Context sensitivity would be a huge jump forward in the menus, but you should still be able to trawl through and access any of the commands if you so wish. Ideally, I think a hotkey system should be also introduced, so that you can assign hotkey combinations to your commonly used commands, this would also help alot.

[whisper 0]

As with cover nodes, this is quite a lot of extra work for level designers.

I may be mistaken, but from what I see of the available scripting commands in ArmA (boundingBox and such), current ArmA objects have geometry LODs that are used by AI as cover, they should be able to select a nearby cover and hide behind (at least, the scripting commands available seems to indicate it's possible).

Thus all object may already be a bit like "cover nodes", isn't it?

[Scrub 0]

IIRC.. And there is no doubt I'm bad with names and dates, just ask my wife. Jan Prazack said in Deasfasts vid (I think I'm right again.. heaven help me) that all objects are considered cover, sounded to me like the AI looks for occlussion, or blockage of LOS. Very smart if so. No nodes necessary. :dance:

[Baff1 0]

I may be mistaken, but from what I see of the available scripting commands in ArmA (boundingBox and such), current ArmA objects have geometry LODs that are used by AI as cover, they should be able to select a nearby cover and hide behind (at least, the scripting commands available seems to indicate it's possible).

Thus all object may already be a bit like "cover nodes", isn't it?

Not even remotely.

They don't use cover in ArmA at all.

The best they do is lie down. (Which is a good start).

Rather than have each AI calculate line of sight between itself and everthing else it can see, and constantly recalculate this (using geometric LODs as it's referrence points), and all the other AI's all constantly rescanning the enviroment to do exactly the same....

We assign good places to stand next to wall models or barrel models or at windows etc.

Then, the AI simply knows where to stand when it is in combat mode. Instead of doing massive and constant mathmatical trigs, it simply goes somewhere that looks good from a list of pre-assigned effective possibilities nearby.

If you look at the friendly AI in the PC version of Ghost Recon Advanced Warfighter, you can see that they are constantly scanning for LOS, as a consequence they all start standing in the middle of the road.

They have broken LOS with their enemy and hence consider themselves to be in cover.

You can see them advancing up the middle of road, strafing and peeking to keep cover from an enemy far ahead of them. It's impressive to watch. They are working through all the angles, rather like a human Ravenshield player would.

But they aren't in cover, so when another enemy approaches from a different angle they are just road kill out in the open, and not behaving as a real squad member intuatively would, moving from hard cover to hard cover.

In the sequel to that game for PC, Ghost Recon Advanced Warfighter 2, you can see that they have added cover nodes.

Now when you tell an AI to move, instead of just telling it to move to a grid reference, and once it gets their lets it's cover mechanism kick in, you see place markers under your reticle that show up next to low walls and behind rocks etc.

It is now instinctive and intuative to get your AI to advance from cover to cover.

And unlike in the first version, it actually works. They use it.

Your squad AI is no longer duckwaddling down the middle of the road "keeping cover" (breaking line of sight) from one enemy AI while the other 3 all shoot it. It has taken cover.

The player is no longer fighting the AI to get them to do simple things without lemmings suicides. Another player has already gone through the map and spotted all the best places for an AI to position itself and hardcoded them into the game. By default, the AI chooses from a list of actual human decisions.

Unreal AI in Brothers in Arms takes this simple to use mechanic one step further and assigns them a suppression meter. So that if they are under fire they duck down low, and if they aren't they pop up to fire.

All that is needed now is for if they get flanked they retreat and we will get some very dynamic infantry engagements. Fluid.

Edited May 7, 2009 by Baff1

[maddogx 13]

IIRC.. And there is no doubt I'm bad with names and dates, just ask my wife. Jan Prazack said in Deasfasts vid (I think I'm right again.. heaven help me) that all objects are considered cover, sounded to me like the AI looks for occlussion, or blockage of LOS. Very smart if so. No nodes necessary. :dance:

If that is true and the AI also really shoots from cover, firefights in Arma2 could be a whole lot more exciting that in ArmA1. :D

[Scrub 0]

Well, you know, that's kinda the point. :) Why are people arguing about old tech in the new game forum?

Start at about 8:40 to get the context.. yay!

Edited May 7, 2009 by Scrub

[Yapab 10]

Scrub, that video gave me a totally different impression of ARMA2 in terms of gameplay than what I was imagining it to be (never played ARMA1).

I'll post a new thread as it would be offtopic to ask the q here :) Thanks for video, looks great!

[topeira 10]

Not even remotely.

Rather than have each AI calculate line of sight between itself and everthing else it can see, and constantly recalculate this (using geometric LODs as it's referrence points), and all the other AI's all constantly rescanning the enviroment to do exactly the same....

We assign good places to stand next to wall models or barrel models or at windows etc.

Then, the AI simply knows where to stand when it is in combat mode. Instead of doing massive and constant mathmatical trigs, it simply goes somewhere that looks good from a list of pre-assigned effective possibilities nearby.

If you look at the friendly AI in the PC version of Ghost Recon Advanced Warfighter, you can see that they are constantly scanning for LOS, as a consequence they all start standing in the middle of the road.

They have broken LOS with their enemy and hence consider themselves to be in cover.

You can see them advancing up the middle of road, strafing and peeking to keep cover from an enemy far ahead of them. It's impressive to watch. They are working through all the angles, rather like a human Ravenshield player would.

But they aren't in cover, so when another enemy approaches from a different angle they are just road kill out in the open, and not behaving as a real squad member intuatively would, moving from hard cover to hard cover.

In the sequel to that game for PC, Ghost Recon Advanced Warfighter 2, you can see that they have added cover nodes.

Now when you tell an AI to move, instead of just telling it to move to a grid reference, and once it gets their lets it's cover mechanism kick in, you see place markers under your reticle that show up next to low walls and behind rocks etc.

It is now instinctive and intuative to get your AI to advance from cover to cover.

And unlike in the first version, it actually works. They use it.

Your squad AI is no longer duckwaddling down the middle of the road keeping cover from one enemy AI while the other 3 all shoot it. It has taken cover.

The player is no longer fighting the AI to get them to do simple things without lemmings suicides. Another player has already gone through the map and spotted all the best places for an AI to position itself and hardcoded them into the game. By default, the AI chooses from a list of actual human decisions.

Unreal AI in Brothers in Arms takes this simple to use mechanic one step further and assigns them a suppression meter. So that if they are under fire they duck down low, and if they aren't they pop up to fire.

All that is needed now is for if they get flanked they retreat and we will get some very dynamic infantry engagements. Fluid.

if i read u right, than u are saying a better solution is to make the designers place cover nods (IE - almost every Unreal engine game like GOW, R6VEGAS, BIA:HH etc)...

while this might work in corridor shooters its not really working in open games, is it?

also, if u ask me, GRAW1's AI is much much better than GRAW2. im talking about enemy AI, mostly. in GRAW2 they are too dumb, dont use cover and barely move. after playing 40% of the game i became very sure if this. after completing is i had no doubt in my mind that enemy AI in GRAW2 was 4 steps back.

are u really sure it is working better????

plus, in BIA:HH the AI is so scripted that i give them no credit. same as MOH:A - they are mearly choosing a cover that was decided by the devs. the AI cant recognize cover on its own (unlike GTAIV's AI, as far as i can see) thus making them completely dependant on what the devs made ready for them. this method is completely inefficient with ARMA2, OFP2 or GTAIV or any other open game.

weird thing is that in GTAIV, that somehow has the AI very autonomous and independant and have the ability to find cover themselves, YET the dev made vey linear and scripted mission. R* could have made non linear open missions and the AI is already ready for that, but they didnt. weird.

[whisper 0]

What is the cover node of a rock? North of it? South of it?

Same question with a tree, ofc ;)

[Baff1 0]

What is the cover node of a rock? North of it? South of it?

Same question with a tree, ofc ;)

All sides of it are likely to have cover nodes.

[whisper 0]

Right, which side an AI should use, then, and based on which criteria?

What difference would it make with LoS calculation based on objects LOD?

[Baff1 0]

if i read u right, than u are saying a better solution is to make the designers place cover nods (IE - almost every Unreal engine game like GOW, R6VEGAS, BIA:HH etc)...

while this might work in corridor shooters its not really working in open games, is it?

also, if u ask me, GRAW1's AI is much much better than GRAW2. im talking about enemy AI, mostly. in GRAW2 they are too dumb, dont use cover and barely move. after playing 40% of the game i became very sure if this. after completing is i had no doubt in my mind that enemy AI in GRAW2 was 4 steps back.

are u really sure it is working better????

plus, in BIA:HH the AI is so scripted that i give them no credit. .

The open warfare aspect of OpF doesn't need an AI improvement.

It works fine in empty fields with nothing to hide behind.

Urban warfare in ArmA, however is crap. Utter crap. Beyond useless.

The AI difference between GRAW 1 and GRAW 2 is only found in the friendly squad AI as far as I've seen. The enemy AI doesn't use cover nodes, it's the same old enemy AI as in the previous title.

And yes the friendly AI works a lot better and it's not a corridor shooter either. It is an excellent example of AI evolution from an LOS system to a cover node one.

BIA:HH suppression AI is unscripted. In the moments when they are allowed to run about freely, they use cover nodes. Hide behind crates and tables in warehouses etc.

The one thing you will never see in BIA is an AI standing about in the open.

Right, which side an AI should use, then, and based on which criteria?

What difference would it make with LoS calculation based on objects LOD?

With squad based AI, you could personally choose which side yourself just as you do in GRAW2 or R6 vegas.

With enemy AI or uncontrolled friendlies, it could use a trig calculation to choose which side. The other side to what it's enemies are on.

(And hopefully, if it can't make this calculation, because it is flanked, it will fail morale and retreat until it can).

The difference between pure LOS and a cover node, is that if the AI breaks LOS with the enemy 15 feet away from the rock. It will be happy to stay 15 feet out in the open. (cf GRAW).

While if it uses a cover node system, it will attempt to hide behind the rock not just break LOS. (cf BIA).

It will head straight for the rock and take up position there.

It will go to the most defensive position in the locality not simply sidestep 5 feet to the left in the middle of the open to break LOS.

There is a difference between breaking LOS and seeking cover.

A human soldier does not stop out in the open because he can't see the person who was shooting him anymore, he hides behind the rock.

Edited May 7, 2009 by Placebo
Merge

[whisper 0]

My point was that there are plenty of script function already available in ArmA that will give you the bounding boxes of objects and corresponding hide points of these bounding boxes in reference to an ennemy position.

It's not a system where you stop as soon as you break LoS. It's a system giving you a hiding point near an object that will hide you from a dynamic ennemy position.

I think that is what BI tried to achieve already in ArmA. From the few tests I did, it's badly functional, but I don't find the idea stupid.

Seeing the scale of ArmA, putting static cover nodes on every object of a 200 sq km map (even automated) is just.... meh...

Once again, the unique scale of ArmA is something that'll turn effective solutions seen elsewhere into a mess. You either put your cover nodes all manually. In which case, you'll finish your map in something like 10 years.

Or each object have predefined sets of cover nodes. Meaning you're limited in said object placement, for avoiding things like placing a cover node inside another nearby object, having conflicting cover nodes, things like that.

1 good use of cover node indeed would be house windows. But I don't think it should be generalized for every cover on a ArmA map.

[Baff1 0]

What ever system ArmA uses, it is terrible. I'm, not looking for any more of that. That is an example of a system that doesn't work.

I agree that it is a lot of work to add cover nodes and also AI tracks to all the maps.

However it's what's needed. Making 3D models is also a lot of work. At least they get paid, I suppose.

Although it's a lot of work, lots of the GFX in ArmA is generic. From buildings to rocks to tree's and sandbags. Building the nodes into objects rather than manually placing them all afterwards could save some time.

And yes, it will give the AI only a limited set of standing positions at each object.

However each of those limited choices will be a humanly decided point.

And however few defensive positions it offers them, each one will be one more than they have currently. An improvement.

Neither will it be the result of an almost random trigonometric calculation, each position will be the result of human intellect. they will be placed where it looks natural not just where the AI happens to end up.

And while I agree that cover nodes do not provide the absolute perfect solution, they are the most advanced available.

It works, and while limited, it is functional and provides an enjoyable game experience in the games that use it, unlike ArmA's which does not.

Windows are good, as are low walls, or roof tops and sandbags etc.

There are a lot of places they are worth having. I don't really forsee them being placed in open terrain very much at all.

Although in urban area's their concetration must needs be very dense.

Edited May 7, 2009 by Baff1

[froggyluv 2136]

I've been experimenting alot with the AI in the FFN mod and have to say they do a much better job of using rocks, trees etc for cover, but the problem is Arma's pathfinding is based in meters so often, units are 'just off' to the left or right of desired cover. I'm hoping the new micro-ai, that is, the unit's ability to gauge more precise measurings of where to hide as well as lean out, will add significant improvement.

[Placebo 29]

Baff1 edit your posts rather than double posting please.

Seeing as this is far into the realms of AI speculation and wishlist type things (in comparing it to other games), it can be moved to Suggestion. General is more for what we know and what we've found out type topics.

[craig3000 10]

Well I'm really looking forward to this game

Never played the first one but the second looks amazing and I hope I'm not wrong!

[Scrub 0]

And while I agree that cover nodes do not provide the absolute perfect solution, they are the most advanced available.

Wrong. Seeing cover for cover is more advanced. This is what Arma II is doing. It has no quirky nodes that won't allow the friendly user-made tank that just drove up into the battle to be used as cover. If my guess is correct about how it's implemented, ArmA II will. Either way, the AI will look for cover and creep to get LOS on it's target.

[stanko 10]

"On vacation May 13th-20th, so if you have any problems with the forums contact someone else, I'll be in Croatia fishing!"

Hey Placebo, I know that this is not the topic, but greetings from Croatia.

[whisper 0]

What ever system ArmA uses, it is terrible. I'm, not looking for any more of that. That is an example of a system that doesn't work.

I agree that it is a lot of work to add cover nodes and also AI tracks to all the maps.

However it's what's needed. Making 3D models is also a lot of work. At least they get paid, I suppose.

Although it's a lot of work, lots of the GFX in ArmA is generic. From buildings to rocks to tree's and sandbags. Building the nodes into objects rather than manually placing them all afterwards could save some time.

And yes, it will give the AI only a limited set of standing positions at each object.

However each of those limited choices will be a humanly decided point.

And however few defensive positions it offers them, each one will be one more than they have currently. An improvement.

Neither will it be the result of an almost random trigonometric calculation, each position will be the result of human intellect. they will be placed where it looks natural not just where the AI happens to end up.

And while I agree that cover nodes do not provide the absolute perfect solution, they are the most advanced available.

It works, and while limited, it is functional and provides an enjoyable game experience in the games that use it, unlike ArmA's which does not.

Windows are good, as are low walls, or roof tops and sandbags etc.

There are a lot of places they are worth having. I don't really forsee them being placed in open terrain very much at all.

Although in urban area's their concetration must needs be very dense.

So it's your way or no way?

right, enjoy the lack of implementation of it then :P