COD like combat in Arma 2

[froggyluv 2136]

I agree there should have been a functional suppressive system hardcoded in Arma. There are good scripts out there but i see little reason BIS couldn't have implemented one to make it standard especially for MP COOP play.

Realistic cover is another matter entirely. Even the common laymen should be able to recognize the Herculean task it would take to get potentially hundreds of units to properly use cover in a huge open 3d world such as Arma. First off, pathfinding would have to be tighter than a clam with lockjaw especially in the intricate urban landscape. Then AI would have to be able to recognize the geometry of ALL shapes and adjust their body to be protected as well as their weapon to the proper height to fire over/around. Now add in finding cover from aerial attacks and youve got enough programming for a seperate game.

Afaik, there has never been a lifelike cover system used successfully in a "Full World" milsim which encompasses as much as Ofp/Arma does. I've never played Red Orchestra, so maybe I'm wrong, but I think most companies opt for 'just MP play' so they don't have to deal with the issue. I do own Swat 4, Graw, R6 Vegas, Brother 'n Arms, and I have to say those games do possess cover systems, are extremely linear and thereby a completely different animal unsuitable for comparison.

I'm not blaming anyone for wanting this, just wasn't realistic expectation for OFP 1.5.

[Beagle 684]

In Vietnam the bullet kill ratio were 16.000 to 1 and now you're telling us it's fifteen times less accurate. Maybe it really is overestimated.

As you can see, infantry combat is still evolving. In every new war they need more ammo to kil you!

Must be the concept of supressive fire, leaves you with no ammo when the real engament beginns.

[walker 0]

Hi MBot

No one is defending it.

We are pointing out that a real AI as is used in ArmA is more complex than a corridor shooter based script.

For a cover solution in the form of an AI the systems analysis is more complex because of the range of possibilities in ArmA. In all the games liljb15 mentions an AI is not involved.

The way a real AI works is that it senses the world. It then applies its Neural Net and comes up with a solution to the problem.

In the games liljb15 mentions there is no true AI; so saying they have better AI is like saying a brick has more brains than a slug. The entity does not sense the world. The AI is told where to go and what to do in a script which may be complex but is not an AI and cannot solve new problems it can only do what is in the script.

Complex worlds such as in ArmA require a true AI. Those games that liljb15 mentions do not. They are constricted worlds where one can know all points and interactions required.

If an FSM (it means finite state machine remember that limited, finite, a known number) is used it is possible by the use of fuzzy logic to create more complex script. A true AI is better in that it reacts via a neural net. It learns about its environment or is taught about it, or it is presented with information and experiences that force it to understand the world it inhabits. It is not told about the wall it just knows what a wall is and how to use it. Make a new island and AI does not need new scripts. In COD they have to reprogram the whole thing over again every time they make a mission. I can make missions in ArmA on any island people make that is why ArmA has an editor and tools to create new object and environments. There is no script for each of these, the AI just generalises them via their class and its experience.

Most of those making the pronouncements that COD AI is better than ArmA AI are not aware of or understand the complexity of an open environment.

Consider how a corridor shooter COD uses a wall for cover:

1) It is told there is a wall by the script.

2) It is told by the script to move to the wall

3) It is told by the script to kneel behind it

4) It is told by by the script to fire at any targets from this position

5) If the wall is flanked. Most often it will just continue to kneel and rotate to face any targets. In order to prevent this the programmers will create corridors that prevent flanking so it hides the fact there is no AI involved.

6) If there is a bit of fuzzy logic in the script and maybe an FSM is used to create the script; it may react to being flanked. It may stand up or lay down but any such reaction is scripted. Anything outside the script and in the worst case the game may crash or most likely the AI is an easy kill like shooting rats in a barrel, as much of COD is and why it is sooo boring.

7) For each wall there is a separate script, some walls and their scripts may be repeated but that makes for boring game play. it is why corridor shooters are boring. It also why such games have such a short lifespan and why their environments are so small.

In ArmA the environments are vast so instead of the few tens of scripts you require in COD you would require millions.

In ArmA you can always flank a wall. The AI is at least to some degree aware there is an environment it will move down hedge rows use dead ground and it will flank and most importantly it is aware it can be flanked and if it is flanked will react to it without resorting to a script.

It does this because it has either learned, been taught by a subject matter expert or has been placed in scenarios that force it to conclusions about its world. How is this done? Welcome to the world of Neural Nets. It is a long and complex process. Most good AIs are a mix of Neural Nets and Expert FSM systems with some Fuzzy logic thrown in for spice. The ideal AI is almost pure NN but even humans do not have that though we do have an instinct(our FSM logic) override from experience and training.

Can ArmA be made more properly aware of walls and an urban environment yes, it is complex problem though what is an urban environment? Write down mathematically what an urban environment is and you will find out how complex it is. As humans we just see it and understand it, but few of us can describe it mathematically.

Consider the humble corner. You know how to use one as a human player, now try to describe it in terms of the mathematics.

We understand corners from our brains Neural Net. We learned about corners with those coloured bricks as kids, the ones we put in our mouths, as we began to sense the world, they were the ouchy bits, as distinct from the flat flubery bits, we then learned about them from tottering around our nursery they hurt more than bumping into the flat wall, though flubery bits hurt as well when going fast, more ouchy, note to self, take care when going near ouchy bits, we began to understand that really ouchy bits were where two flubery bits joined and if we really bright we understood they came in outy and iny versions but were both related, then we learned language, ouchy bits had a very important name "CORNERS" put those top of the list. Then we went to school and learned the mathematics of corner but most of us thought that was boring and we would rather play footie, or hand-clap games.

What is a corner? How many sides has it got? how many positions has it got? Is its aspect internal or external to your entity? For your info a two dimensional corner (top down view) has three positions. One internal and two external. What is the the location of the threat? How do you navigate to place the corner between you and the threat? What about multiple threats?

Of course island makers and in particular object makers can help us by adding each corner and its cover direction to them its a lot of work but required for ArmA. For a simple box building with no doors windows or internals there are 8 cover positions four corners times two sides for each corner. Now add in some doorways and windows, corridors archways a second and even a third floor stairs internal corners both 90 degree internal and 90 degree external and we are only talking 90 degree corners here no goddamn slopey walls, what about furniture in houses EEK! Or street furniture or rubbish in streets omnidirectional cover if you lay down but only if the cover is between you and the threat but how many threats are there? How far do you look? Is the movement to get behind the threat worth the loss of concealment from this and other threats. As human players we do it in ArmA without any seeming thought but like our ability to see it was programmed into us with thousands of years of evolution and all those games of hide and seek and war we played as kids. And every one of us has a computer in our heads that dwarf all the computers in the world put together.

DYNAMIC ENVIRONMENTS! What happens when the wall falls down? Is the wall cover to all threats? It might stop a pistol bullet but what about an M2 or an RPG or a shell, how may hits will it stop?

That is what many of us are working on. There are some scripts and even FSMs out already. The first stage is generalised scripts to deal with recognising an urban environment, then the simple easy to understand bits of the problem but as froggyluv said the task is herculean. We are little guys chipping away at mountains with a hammer and chisel, but I know mountains in Wales where a few hundred guys like that removed whole mountains in less than 100 years.

We are monkeys and BIS have given us the typewriters, some of us are certain to write sonnets. Want to join the million monkeys?

Kind Regards walker

[Second 0]

Walker told it well, i decided to drop my AI-boosting hobby because of those reasons.

I didn't use mathematic (i bad in it), but much more rough way and still it was pain in the ass and CPU to get AI to understand even most simple thing: This is building corner or knee high object (like wall).

But nice to see that other much more skilled and intelligent persons are working on it.

[jtec 0]

Im pretty sure to make things simple the odjects could be defined so the ai has random ideas on where to sit, hide or run to. Take a wall for instance you could have a small hole in it with a corner piece one end. These areas could have some defined areas that can be randomly picked by the ai. So it could move into a set anim for that area it is in. Would be so cool seeing a ai running for cover behind a wall then crawling to its chosen point where it will either, cover from the hole or hide in the corner, or do a action that can make use of said wall. Pretty possible now im guessing with a decent scripter and a few models to go with it?

[Ezekiel 0]

Bravo Walker, very nicely put.

I have a question tho... could the AI be taught to recognise a wall/tree/bush/building using someone sort of 'sonar' echo laser rangefinder type thing? That way, if they enter a 'take cover' state in their FSM and this cover is close enough, they can be scripted to move towards it. If they can triangulate where the cover and where the enemy is (only x and y axis are needed, height isn't relevant since MOST things are cover if you can go prone behind them), they can move in such a way as to keep the cover directly between them and the enemy (or as many enemies as possible if there are many). Then they can be scripted to move towards the cover until they are against it, all the while assessing whether the cover is still between them and the enemy.

Then i gets more complicated, cos they need to be able to move (either by changing stance or walking/crawling) in such a way as to get an enemy in their LOS, otherwise they'd just play hide and seek with you. Surely they can just use trial and error? Go from prone to crouch, look in direction of where enemy was last seen. If enemy is visible, shoot, if not move. If they're taught to move along the edge of a wall, surely the can be taught to keep going until an enemy unit is spotted?

Just a shot in the dark, I'd like to learn more about how AI works.

[AWDrift 0]

IMO, in a combat sim AI being able to drive across the map with a single waypoint doesn't matter much compared to AI being competent and realistic in battle. I'd glady exchange huge maps for one's the size of a city, forest, or large field if it meant AI being realistic in combat. After all this isn't test drive unlimited or an island tourism sim, it's a combat sim and basic things like surpression and cover should be priorities over having a huge map.

[ebud 18]

Not that it will help much, but I made an opfor addon that used the civilian FSMs and it worked pretty well in combat, well enough as a starting point for trial and error editing. It worked so that once contact is initiated, the opfor will fire on the enemy as they are falling back to find cover. They don't do the "drop on the ground" routine usually either and just run to the closest bushes/rocks/walls and then slowly look to engage the enemy. I only did some quick testing by placing 4 groups of 8 units, 2 groups using the civ fsms and 2 normal groups. The normal east units were picked apart as they never looked for cover. The addon units which use the civ fsms fared much better as they stood the whole time and retreated for cover objects at the first sign of the enemy.

Not a perfect solution, but it is something someone could look into if they are looking to edit FSMs.

[ck-claw 1]

As vague as ever Walker!

[benreeper 0]

In my post I was not speaking to anyone in particular, I was railing against unfair comparison's. I play this game (and OFP before it) everyday to the exclusion of all else: If I'm to be called a fanboy then so be it.

@PASTOR

you created a mission in ArmA in which the AI performed no flanking maneuvers correct? I was just wondering what results you got when you created the same tests in COD?

--Ben

[BraTTy 0]

Someone said that they made missions and no flanking happened.

I swear I have seen some flanking, whats probably happening is that if the AI has more than one objective they are more likely to flank.If you set a waypoint to stay and defend and nothing happens after that, that may be all they do.

And like some points I tried to make earlier that a mission maker could script some alternates and make us a sweet mission.(maybe something I may do soon)

I am apoligizing to liljb15 from everyone that we do agree that AI does need some work, we have all said that.You just set off some reflex actions from us all. Surely the dev's has it noted and on the todo list to make best it can be.

Asking for COD like behavior is a downgrade.

We have all asked for better CQC abilities but its hard with the size of area.

I particularly don't like their chopper flying skills (lately they just keep tipping) And that is something only lately in latest patch(s)

AI is not perfect , we will all agree, but so far the best I have seen in games. (especially cause of the freedom involved)

VBS was originally used to train Marines for situational awareness I believe. And alot of law enforcement and other branches of military use it too.

America's Army I haven't played in a while (there isn't no AI in it , right?) Was made like someone said for propoganda to get us to join.It wasn't used to train Army

We are all waiting for the next best realistic simulator whether its Arma2 or not

But honestly you can have COD like missions already, some have stated that anims are too slow, maybe so in Arma but in OFP take a look at the towable artillery addon I had to script some anims to hook and unhook the PAK36, they were like 80% or more reliable

Someone just needs to design some good missions (I can complain that BIS didn't include enough good missions in Arma)

More reason for us users to make good missions to go with the great game

[snakefang 0]

Quote[/b] ]Please show me a solider that cantt o adjust his sights on his weapon.

Show me a game where that's actually a feature.

Quote[/b] ]Please show me a solider that takes 4 seconds to throw a freaking grenade.

*Points to every soldier in the world*

Seriously. You try pulling it out of whatever pocket it's in, pulling it's pin, throwing your arm all the way back and hurling it across a field in under four seconds.

JOTR, and ALL NOVALOGIC GAMES feature realistic bullet drop, WINDAGE, and ELEVATION FOR ALL WEAPONS, sniper rifle, M4, MP5, M203, everything.

As for grenades, its not in your pocket, or at least i hope it isnt, they should be located on your belt, and if i had 2 brain cells, id look down, grab the thing, and throw it.

Even then, put a base ball in your pants cargo pocket, it sure as h@#% doesnt take me 4 seconds to pull it out and throw, and im not trained solider.

Also, i dont think you need 4 arms to carry a freakin MP5, and an M4, if you would need 4 arms to carry those, youd need more to carry a M24, and an AT-4, since they weigh a crap lot more.

Also, i dont think ArmA soliders carry 80 pounds of equipment, heck unless there carring the M107 or Jav, they probly arnt carring more then 50 pounds. My dad was 82nd AirBorn, he carried an M60 machine gun, with 1,000 rounds, and the base plate of a Mortar, and he could run 400m, drop to the ground, set up with M60 in about a second, and give ACCURITE BASE OF FIRE OUT TO ABOUT 500 meters! More so if need be.

[Ezekiel 0]

I think one of the features mentioned for Game 2 (Arma 2 now) was encumbrance, i.e. stamina based on weight of equipment. Perhaps some of the restrictions on carrying equipment will be relaxed. After all, the limit of one assault rifle + optional AT4 was put in place to encourage realism. Often in ArmA your actions are restricted not out of lack of decent programming or coding, but to prevent unrealistic behaviour.

If it were possible to carry two rifles, then everyone would be doing it (I can just see the Evolution games now, with everyone running around with an M24 and M4/MP5 combo). If there were an obvious disadvantage to carry a second or even third rifle on your shoulder (such as faster tiring or decreased accuracy) then it'd be something only done when necessary.

EDIT: ^ He has a point, when someone said 'show me a game with adjustable scopes' I instantly thought back to the original Delta Force (and its many offspring). Hell, in a way that game was very similar in concept to OFP.

[4 IN 1 0]

Quote[/b] ]Please show me a solider that cantt o adjust his sights on his weapon.

Show me a game where that's actually a feature.

Quote[/b] ]Please show me a solider that takes 4 seconds to throw a freaking grenade.

*Points to every soldier in the world*

Seriously. You try pulling it out of whatever pocket it's in, pulling it's pin, throwing your arm all the way back and hurling it across a field in under four seconds.

JOTR, and ALL NOVALOGIC GAMES feature realistic bullet drop, WINDAGE, and ELEVATION FOR ALL WEAPONS, sniper rifle, M4, MP5, M203, everything.

As for grenades, its not in your pocket, or at least i hope it isnt, they should be located on your belt, and if i had 2 brain cells, id look down, grab the thing, and throw it.

Even then, put a base ball in your pants cargo pocket, it sure as h@#% doesnt take me 4 seconds to pull it out and throw, and im not trained solider.

Also, i dont think you need 4 arms to carry a freakin MP5, and an M4, if you would need 4 arms to carry those, youd need more to carry a M24, and an AT-4, since they weigh a crap lot more.

Also, i dont think ArmA soliders carry 80 pounds of equipment, heck unless there carring the M107 or Jav, they probly arnt carring more then 50 pounds.  My dad was 82nd AirBorn, he carried an M60 machine gun, with 1,000 rounds, and the base plate of a Mortar, and he could run 400m, drop to the ground, set up with M60 in about a second, and give ACCURITE BASE OF FIRE OUT TO ABOUT 500 meters!  More so if need be.

sorry m8, but did you really tried to throw a grenade b4?

4 secound is about the time you take to grab a nade from your vest, pull the pin, aim and throw it high and far

sorry for not able to roll it, but we dont have to either

about ballistic, yes stock ArmA might not get them 100% correct, but we can still tweak it(and those guns in the game you suggested doesnt seems to have "adjustable" sight too)

p.s. since this is topic about AI, plz try to stick on topic

[lecholas 2]

Someone just needs to design some good missions (I can complain that BIS didn't include enough good missions in Arma)

More reason for us users to make good missions to go with the great game

You mean... script the AI? Like some Liberation mod from OFP missions which had soldiers taking cover and poping up from behind sandbags? Or using Blanco's basic script for trenchfights? I really don't know if you think that Blanco's work makes any sense... It's scripting. Just like COD scripting... only (due to limitations of ARMA engine) it's worse than COD scripting... Sorry but I couldn't help myself making the comment

And one other thing. Im pretty sure that ARMA (as well as OFP) has (unscripted) flanking. But what are the purposes of flanking in RL? Supprising the enemy from the direction he do not expect you from and... cathing him with his pants down - attacking him from the direction which he has no cover from. What's the point of flanking if the enemy don't have cover from any direction? And if it is almost impossible that he will survive untill the flanking has been completed (firefights in ARMA last for seconds...). In such a case just firing at the enemy without wasting your time for useless maneuvers seems much resonable. Flanking is only one element of more complicated military tactics. If there are no other elements of the tactics (especially suppressing fire and taking cover) the one element itself is almost useless. "Fire without maneuver is indecisive. Maneuver without fire is fatal.".

And about Walker's elaborate post Noone here says making a good AI is easy. You're probably right in most of your points (apart from some unimportant ones - e.g. games like COD are quite easy to beat not because the AI is so bad but because you have 'health' higher than the enemy and because they are so designed not to frustrate action-oriented players). But while I agree with you that the task of giving the AI awarness of cover in the environment wouldn't be easy I saw some quite encouraging empirical evidence that AI in OFP/ARMA could be improved in this matter. Did you tried solus' SLX where AI uses house positions for cover (BTW OFP/ARMA uses some mechanisms simmilar to other so called 'corridor' games like CODs when it comes to buildings - pathways for moveing and nodes for places where cover can be taken). What if every such a building (or other piece of cover) had some predefined 'house positions' (even on corners) with some more info in each ([animation covered, animation for fireing, direction]; direction would be relative to model and would require to be translated to the direction to world). As ARMA islands use not so many types of buildings (which already have some building positions implemented) this does not look like an impossible thing. There were some more efforts made by the community (but BIS do apparently do not consider this topic as an important one) - second's suppression script, suppressive fire implemented in group link scripts, script for taking cover by General Barron, script for urban movement done by someone whose name I can't remember (it was on old OFPEC forums). The last one was also made by the idea of making 'nodes' telling the AI what to do at the point.

[MBot 0]

And one other thing. Im pretty sure that ARMA (as well as OFP) has (unscripted) flanking. But what are the purposes of flanking in RL? Supprising the enemy from the direction he do not expect you from and... cathing him with his pants down - attacking him from the direction which he has no cover from. What's the point of flanking if the enemy don't have cover from any direction? And if it is almost impossible that he will survive untill the flanking has been completed (firefights in ARMA last for seconds...). In such a case just firing at the enemy without wasting your time for useless maneuvers seems much resonable. Flanking is only one element of more complicated military tactics. If there are no other elements of the tactics (especially suppressing fire and taking cover) the one element itself is almost useless. "Fire without maneuver is indecisive. Maneuver without fire is fatal.".

A very well thought out post, I very much agree. Aswell as with your previous well made post earlier in that topic, lecholas.

I don't think it would be a bad idea to design every single object in the world with predefined cover points attributed with informations about directions and strenght etc. After all this is not so unrealistic at all, it kind of mimics the human reaction. As a first reaction to danger we don't make LOS calculations where the best possible cover is. We just see some sandbags then dive for it, simply because we have the predefined knowledge "sandbags=good cover". Only once we are there we start to finetune our position according to the specific situation, finding the best possible cover our obstacle can provide. Wouldn't be such a approach be feasible to our game world? The workload to define such cover points in the creation of each object would be managable.

This would at least cover objects placed on the map, terrain cover itselfe would be a different matter.

[lecholas 2]

A very well thought out post, I very much agree. Aswell as with your previous well made post earlier in that topic, lecholas.

Thanks. Conceptual work is what I'm getting payed for .

I also think that the thing that makes an AI more human like is unpredictabillity. This can be achieved by randomnes. So AI should sometimes make mistakes, choose not the best cover etc. Predefined points could then have own 'placement radius' (like in waypoints) so the AI would not look like glued to specific objects. It could sometimes stick out off the cover. There's a room for a lot of ideas here.

This would at least cover objects placed on the map, terrain cover itselfe would be a different matter.

Very true. I'm not sure if OFP/ARMA's AI takes any of the terrain factors into consideration while taking cover (SLX has taking advantage of height as one of its features). I've been playing OFP/ARMA from the day the OFP demo was released and I didn't notice it, but I may be wrong (I think the AI in stealth mode will try to stick to valleys but I'm not sure if it's not caused by the fact that they stick to trees and bushes and they usually grow in the valleys). In my countless experiments with the OFP/ARMA editor I tried to identify possible terrain cover myself and to mark it with game logics (which I added to cover list in General Barron's script)... it had some very nice effects (e.g. on FDF's desert island which had 3d roads sticking out of the ground) but it was... ... sorry it doesn't want to come out of my mouth... scripting!...

[4 IN 1 0]

i think by doing that, the game will be 100% unplayable due to the lag these created

[lecholas 2]

i think by doing that, the game will be 100% unplayable due to the lag these created

I wouldn't be so sure. First, ARMA original AI search for cover (in combat mode) on their own and it doesn't create so much lag. This searching can be disabled leaving some free processing power. Second, adduced SLX mod proves, at least to me, that some scripts, that in my opinion should burn our processors, actually do work. I'm really looking forward to ARMA version of SLX to see what will solus come up with. Third, we are talking here about engine-side mechanisms, not about scripting workarounds. These should work much faster.

[froggyluv 2136]

I'm really looking forward to ARMA version of SLX to see what will solus come up with.

I've been fortunate to be a SLX Beta tester and I think your really going to like the dynamic element of the AI

[Commander-598 0]

And one other thing. Im pretty sure that ARMA (as well as OFP) has (unscripted) flanking. But what are the purposes of flanking in RL? Supprising the enemy from the direction he do not expect you from and... cathing him with his pants down - attacking him from the direction which he has no cover from. What's the point of flanking if the enemy don't have cover from any direction? And if it is almost impossible that he will survive untill the flanking has been completed (firefights in ARMA last for seconds...). In such a case just firing at the enemy without wasting your time for useless maneuvers seems much resonable. Flanking is only one element of more complicated military tactics. If there are no other elements of the tactics (especially suppressing fire and taking cover) the one element itself is almost useless. "Fire without maneuver is indecisive. Maneuver without fire is fatal.".

I have been flanked and killed by the AI in OFP before so it definitely works...I only recently got a hold of ArmA so I have yet to really see what the AI can do in it...but I doubt it's much better...

[Heatseeker 0]

I have been flanked and killed by the AI in OFP before so it definitely works...I only recently got a hold of ArmA so I have yet to really see what the AI can do in it...but I doubt it's much better...

The a.i. flank even better in Arma, the second you shoot at their position they will knowabout your exact location for good, even when you move out of their visual range they will find you, this is why the a.i. is more active compared to OPF, their little radars track you for longer.

One problem i have is the new zoom on sights, to be acurate over distance (wich is needed since the a.i. snipe better than ever) you have to zoom in on the sights and this increases tunnel vision, the result is a big loss of peripheral vision wich makes it easier for one of them sneaky bastards to close in on you from any direction. Zoom on sights was an OPF:E thing that shouldnt have found its way into Arma.

The best tactic is the same old tactic though, kill them all real quick, if you fail to do so run for the hills!

[froggyluv 2136]

Spoock found these nice videos of ARMA 2 in which AI is leaning and using cover

video

[Heatseeker 0]

Spoock found these nice videos of ARMA 2 in which AI is leaning and using cover

video

A small step for a bot...

Its nice to see them using cover and leaning, but it would be even better to see them stay behind cover when shot at (supressive fire).

[froggyluv 2136]

Spoock found these nice videos of ARMA 2 in which AI is leaning and using cover

video

A small step for a bot...

Its nice to see them using cover and leaning, but it would be even better to see them stay behind cover when shot at (supressive fire).

Yeah I thought that too buts it's definitely a step in the right direction and they have over a year to compliment it with a nice suppression system.