Scrap the AI System

[captmoose344 29]

Something that has not really changed much from the beginning has been the AI in the game. They still have the same frustrations i noticed from Arma II. Very clunky, frustrating and non useful units for players to use. From the other end of the coin they still dont have sliders for accuracy or an ability to adjust aggressiveness of the AI via the Editor. Everything is still development of scripts to try and fix the AI or make them better which is fine. I would just like to see the devs come out with some real enhancements to the system since the game has been out now for a few years and still people complain of AI being too accurate, knowing where they are and having laser accuracy. 

 

I just recently saw a video from the old Operation Flashpoint where the AI control system looked way better than what we have today with scroll menus and all. Why was that system never recalled to the building board and implemented as the AI seem to respond when ordered. Not sit lying down on the ground till the gun fire stops. Or spend 15 min while the AI attempts to path into the helicopter or truck you just want them to jump into. Heck even AI medics that are designed around supporting the player would be an incredible implementation. 

 

 

If its not the AI or there actions thats the issue then the pathing system should be scraped and rebuilt to a more simplistic system. Perhaps the way the AI have been designed they have too many variables figured into what should be a simple calculation of moving between two points on the map. 

 

Any other ideas people have on how we could simplify and improve the AI system of Arma 3 Please comment in the area below. 

[fn_Quiksilver 1636]

The AI driving at the moment (1.66) is in a Broken state, once a few bugs have been addressed with it I don't think its that bad.

 

Unfortunately we're in a situation where work was started, to overhaul the AI driving, several bugs were introduced, but then for some reason that work has been abandoned and we are left with the broken driving.

 

As for making the AI better, many game studios have figured out that committing company resources (skilled manpower) to developing game AI doesn't generate a return on investment in many games. Better profit from developing PvP and 'emergent' gameplay, which is IMO a reason why there has been virtually no progress in the core ArmA AI in many years (several BI devs have done great work extending the core AI with some features though).

 

BI need only look to the server browser (and their analytics) to see what sells. People are playing PvP/Survival/RPG, not much AI presence in any of those.

 

just my thoughts from where I sit (not in the BI office so don't know if the above is accurate)/

[froggyluv 2136]

Yes but how many people bought the game 2.5 million or more? How many people on the servers few thousand? There are many many SP gamers out there and yes I hear about AI all the time. The problem is most casuals just instinctively sense whether AI "feels right" or not but cant or dont bother to really articulate the issues that bother them. This doesnt mean AI dont matter on the contrary Red Dead Redemption sold 13 million and apparently had very good AI which people talk about. Ghost Recon Wildland's AI is being panned wildly ahead of time and people and sales will notice.

 

In short: Artificial Lives Matter

[maffa 29]

 

8 hours ago, fn_Quiksilver said:

The AI driving at the moment (1.66) is in a Broken state, once a few bugs have been addressed with it I don't think its that bad.

 

Unfortunately we're in a situation where work was started, to overhaul the AI driving, several bugs were introduced, but then for some reason that work has been abandoned and we are left with the broken driving.

 

As for making the AI better, many game studios have figured out that committing company resources (skilled manpower) to developing game AI doesn't generate a return on investment in many games. Better profit from developing PvP and 'emergent' gameplay, which is IMO a reason why there has been virtually no progress in the core ArmA AI in many years (several BI devs have done great work extending the core AI with some features though).

 

BI need only look to the server browser (and their analytics) to see what sells. People are playing PvP/Survival/RPG, not much AI presence in any of those.

 

just my thoughts from where I sit (not in the BI office so don't know if the above is accurate)/

 

Arma is not a twitchy competitive FPS. It has nothing to do with COD or BF, or Overwatch for whats worth.

 

Ai is paramount important to Arma.

 

Half the matches in clan nights are COOP. Other complex missions like Antistasi totally rely on AIs. High Command module is a great promise never fullfilled: you cant even order a unit to get in a vehicle for god's sake, not to mention all the times they get stuck and wont move. Not every clan can field enough players to have TvTs all the time, and the very low bottom line is 20 players, 10 per side, just to have an Unreal Tournament like small scale mission. In a coop it's not unusual have x1.5 up to x3.0 AI for every player, and the scope of coop missions gets grander and grander. And yet we still have AI inside buildings laying low below windows, never getting up and sometimes firing by clipping their guns from walls.

Edited February 11, 2017 by maffa
the day i can leave a post without editing typos will be a very fine day indeed

[deady 0]

Squad AI isn't that bad, but you have to manage them well.  Driving has always been bad in every OpFlash + Arma :)  They over react to objects and people and will try to get around them but just encounter more obstacles, eventually getting stuck.

[kerodan 25]

15 hours ago, captmoose344 said:

AI being too accurate, knowing where they are and having laser accuracy. 

Yes, people complain a lot about this issues but never mention the situations where they occur. I have been playing ARMA coop for over 4 years now (so A2 OA and A3) and never perceived a game-breaking flaw concerning these systems. In my experience, most of these complains could be easily resolved by setting up the AI configuration properly.

 

So I would agree with you, that BI could invest more resources in making the AI configuration easily accessible and document the subskills in more detail (for example, I am still looking for a definitive answer to the question if the view distance setting influences the AI's visual spotting range). However, I would not agree on scraping the whole AI system as I am repeatedly astonished by what the AI can actually do given the unpredictable situations thrown at it.

 

On a different note: Why is everybody calling the current iteration of AI driving 'broken'? They still drive far better than before work started on improving the driving system and I haven't noticed any glaring issues with the current version. But I still hope for further improvements, of course. ;)

 

[HeroesandvillainsOS 1504]

14 hours ago, fn_Quiksilver said:

As for making the AI better, many game studios have figured out that committing company resources (skilled manpower) to developing game AI doesn't generate a return on investment in many games. Better profit from developing PvP and 'emergent' gameplay, which is IMO a reason why there has been virtually no progress in the core ArmA AI in many years (several BI devs have done great work extending the core AI with some features though).

 

BI need only look to the server browser (and their analytics) to see what sells. People are playing PvP/Survival/RPG, not much AI presence in any of those.

 

just my thoughts from where I sit (not in the BI office so don't know if the above is accurate)/

I think more people would play with AI if they didn't suck so much and maybe valued their life even one bit. They are always perfectly happy to walk directly into their own death. It's glorified babysitting, where fire fights are over almost as soon as they start.

 

You can see this in ALiVE. Let the units spawn from both sides and teleport to where two AI sides are fighting. 10 individual AI on each side and almost all of them are dead within minutes. They don't try to take smart cover, or even move into buildings that have AI pathfinding so they can bunker down and save themselves.

 

ALiVE is great for this because it doesn't touch low level AI at all.

 

All this, let alone other basic things such as medics that automatically heal others, or hell, even units themselves that have FAK's are perfectly willing not to use them on themselves because they'd rather lay down and die.

 

If indeed the statistics show that most people don't play with AI it's because the AI is awful, not because PVP billion times better. People want challenges and smart things to strategically shoot at, and the AI are not even close to providing that kind of experience.

[krycek 349]

Infantry AI had the most improvements in current title,last year they also started to work on the much needed fixes for AI driving and hope they continue.I haven't fired up A3 since july so I'm not up-to-date if they broke it again but if they did,this should be a priority after 4 friggin' years after release.

Sadly the driving and flying part of their AI system didn't got much attention throughout the series.It's understandable this is a complex system with so many units in a sandbox environment but their work on it throughout the years it's simply not excusable considering the timeline from 2001 to present day.I'm way past the "...but give them time to work on it".How much time is needed?Another 10 years,20,30?

 

I think one of the important reasons for lack of progress or big changes in this area is also the sad state of gaming industry and that there hasn't been even now a worthwhile competitor to Arma series that could manage big scale of combined arms ops.The last guys that tried to challenge BI were those devs that created Soldner many years ago and those of us that remember that game we also remember the sad joke it ended up to be.

 

Also going the easy way and scrap the whole thing wouldn't do them much good,like others said many play with AI,be it single or coop.Transforming it into some BF isn't the way.

[Vigil Vindex 64]

I was really really disappointed when A3 came out that the AI never got the kind of revision that I was hoping for, particularly in terms of performance. When you load up the game in the editor you can see the game runs fine. And even if you load in some AI with no enemy AI or activity the game runs fine. But when you run AI that are active either against other AI or against players you can see a dramatic decrease in performance. Granted these calculations for path finding and engagement behavior will have a performance cost associated with it I still think this is an area that should have been better targeted in A3. Simply by the fact that the maps are so large they really need to be populated with more AI to make them feel alive. The work being done on the dynamic simulation just now could help achieve the perception of more AI on the map in certain circumstances where fast movement across the map in jets for example won't cause issues with caching of AI. And I am glad that some work is still happening on these sort of systems. And after that "no magic bullet" statement I have came to accept that we will never see a revolution in AI performance across the board in A3. So I am watching the DayZ team with much hope for their efforts in making new AI in the Enforce scripting engine to see what is capable of happening in the future with BI products.

[fn_Quiksilver 1636]

16 hours ago, froggyluv said:

Yes but how many people bought the game 2.5 million or more? How many people on the servers few thousand? There are many many SP gamers out there and yes I hear about AI all the time. The problem is most casuals just instinctively sense whether AI "feels right" or not but cant or dont bother to really articulate the issues that bother them. This doesnt mean AI dont matter on the contrary Red Dead Redemption sold 13 million and apparently had very good AI which people talk about. Ghost Recon Wildland's AI is being panned wildly ahead of time and people and sales will notice.

 

In short: Artificial Lives Matter

 

the server browser also shows milsim activities (when big ops are organized), occasionally a few times a week there is a milsim server which breaks into top 25 by population, but only for a couple hours.

 

there are many thousands on the server browser playing PvP/Survival/RPG at all times of day, compared to occasionally a few hundred doing milsim/coop operations.

 

Maybe once GR Wildlands is released the game industry can assess whether hiring skilled programmers (who went to school to study  Game AI, machine learning, etc) is worth it for future titles. 

 

All we know is for the duration of A3 development BI did not hire one Dev who specializes in AI ;)

 

There is also this perpetuated myth of the 'unseen majority' playing Singleplayer. We don't have visible data on SP numbers so I can't comment, but I find it unlikely that there are near as many SP players online at any one time compared to MP players.

 

And the driving is broken, too many reasons to list but tell a few vehicles to patrol back and forth along the same road and you will see.

[Vigil Vindex 64]

 

4 hours ago, fn_Quiksilver said:

All we know is for the duration of A3 development BI did not hire one Dev who specializes in AI ;)

 

I thought they did hire someone. I commented back in 2013 about it and got a response which you can read here: 

 

[road runner 4344]

4 hours ago, fn_Quiksilver said:

There is also this perpetuated myth of the 'unseen majority' playing Singleplayer. We don't have visible data on SP numbers so I can't comment, but I find it unlikely that there are near as many SP players online at any one time compared to MP players.

Why would SP need to be online? I'm a single player, I don't need to join a server, or be online to use the game. I also spend 50% of the time in and out of the game testing addons I'm making etc,  It's simple really, look at how many copies have been sold globally, then look at the amount of people on servers, so yes, I'd say that the vast majority of end users of Arma3 are indeed single players

[warlord554 2065]

^^^^^ sums it up for me also

[Vigil Vindex 64]

I don't think he stated that SP needs to be online. He stated the obvious fact that it is hard to prove that there is a majority of people playing single player offline. Even if you take away the number of people online from the number of global sales you can't "assume" that all the rest must be playing offline. It is possibly more reasonable to "assume" that those players are simply no longer playing the game. Perhaps subtracting online steam player numbers from the players in servers would give an idea of how many people just play single player but it would not account for those who play offline. Regardless the point I think is that AI improvements would benefit both offline and online play. And that we are seeing the majority of online play focused in areas of PVP type scenarios, possibly due to the issues around MP AI scenarios.

[road runner 4344]

18 minutes ago, Vigil Vindex said:

I don't think he stated that SP needs to be online. He stated the obvious fact that it is hard to prove that there is a majority of people playing single player offline. Even if you take away the number of people online from the number of global sales you can't "assume" that all the rest must be playing offline. It is possibly more reasonable to "assume" that those players are simply no longer playing the game. Perhaps subtracting online steam player numbers from the players in servers would give an idea of how many people just play single player but it would not account for those who play offline. Regardless the point I think is that AI improvements would benefit both offline and online play. And that we are seeing the majority of online play focused in areas of PVP type scenarios, possibly due to the issues around MP AI scenarios.

To a degree a lot of that's true, but again, there's absolute no need to go online as a SP, just because you don't see it, doesn't mean its not happening, numbers wise.

You simply cannot make a statement to contradict that the majority of players are single players because there's no way to prove it, there is, but it's more or less being dismissed.If the bulk of customers were online players, especially through steam, the information on numbers online should be easy enough to get.

How many people play PVP out of this Multiplayer, or simple KOTH only? The whole big picture of the "Online" community starts to shrink. PVP/KOTH don't need fancy AI enhancements, CO-OP and SP do, so yes there's scope for a better AI, but it's never in a month of Sundays going to be as to what the majority want. As in a proper AI, that will see them think for themselves like the human player, for starters, the cost to develop that would be phenomenal, also the customer database would shrink, to probably only the Military/DOD.

Even in VBS2, the AI wasn't that much better, sure they would have better tactical skills, but, nothing overly fantastic. You can't even get AI to fly a steerable parachute, or climb up/down a ladder in/out of the water, on onto/off a roof, when it gets to that sort of level, I'd be over the moon, but it's not going to happen.  

The AI are for all their motradness, still better than the majority of other games, if anything they're still too "smart".

In real life, not every soldier is Daniel Boone, but the OPFOR AI still has this uncanny ability to drop you at 400M + with an AK, in real life, with iron sights, I can assure you, unless you're in the Higher Trained element, hitting targets at 400M+ with an AK is luck more than skill. You either need to dumb basic units down, to the point that their shots fired, to kill ratio is like 1000/1 you may laugh, but during the Falklands war the shots fired to kill ratio was like 20000/1 .

The AI will always be either too fucking stupid, but outstanding marksmen, or full on Tier 1 scoot and shoot, and they kill you easily, and you'll complain ? it's a catch 22 really.

Scrapping/discontinue trying to make them do the impossible it would in a way be a logical step, but we'd have to accept they're stupidity.

One thing I have noticed is that a lot of these 3rd party AI improvements are really good, but end up conflicting with each other.

[tpw 2315]

As an exclusively SP player I have to agree that AI could do with some work. Everyone, including the devs, knows this. Infantry AI are generally reasonably competent, with occasional flashes of either superhuman ability or subhuman stupidity, but they can at least be scripted into something that suits your tastes to a degree. On the other hand vehicular AI pathfinding and collision avoidance is atrocious and very difficult to compensate for by scripting. And to top it all off, game performance plummets once AI go into combat mode and start shooting at each other. To address the thread title though, I don't think the AI should be scrapped so that BI can concentrate exclusively on the milsim experience for a particular subset of their MP userbase.

 

There's a similar vibe in some segments of the sim racing community, where AI is absolutely critical to the SP experience. For every discussion on AI improvements there'll be "why bother" comments from MP players who can't understand that some people simply can't / don't want to play online with others for numerous perfectly legitimate reasons.

[froggyluv 2136]

2 hours ago, Vigil Vindex said:

 

 

I thought they did hire someone. I commented back in 2013 about it and got a response which you can read here: 

 

Thats a nice find and pretty revealing thanks. Whatever problems the devs will admit to having with the AI is really probably only 10% of the reality of the problem. So those statements are pretty telling (along with current MP trajectory) that they just aint digging in even if some of their coders would like to -they aint getting the green light of resources. Sounds like they dont want to start pulling cards from a tremendously huge house of cards because the risk to the stability of the game is too great. Thats why I recommend AI behaviour modules -leave the present state for Vanilla game but start gradually adding some Attack/Defend Urban/Forest/Mountains, Sniper, Spec Ops, Civilian and we'd all be happy meethinks.

 

There are also some industry standard things that need to be added such as AI recognizing and RUNNING from grenades. Better orienting their rifles on known enemy area is needed as well because sometimes they just seem to erhm, space out and walk perpendicular (to known threat) with their rifles up..

 

 

[fn_Quiksilver 1636]

On 2/12/2017 at 10:08 AM, road runner said:

Why would SP need to be online? I'm a single player, I don't need to join a server, or be online to use the game. I also spend 50% of the time in and out of the game testing addons I'm making etc,  It's simple really, look at how many copies have been sold globally, then look at the amount of people on servers, so yes, I'd say that the vast majority of end users of Arma3 are indeed single players

 

respectfully,

 

 

 

:)

[road runner 4344]

Can you disprove the comments though quicksilver with 100% verifiable facts? I keep hearing that there's more Online Multiplayers than single players, but little to back this up, I merely offered one way that it could possibly be dis-proven?

I'm easy either way, it makes no odds who has the greater numbers, I'd just like to see stats that lean one way or the other.

[bmcook 14]

35 minutes ago, road runner said:

Can you disprove the comments though quicksilver with 100% verifiable facts? I keep hearing that there's more Online Multiplayers than single players, but little to back this up, I merely offered one way that it could possibly be dis-proven?

I'm easy either way, it makes no odds who has the greater numbers, I'd just like to see stats that lean one way or the other.

According to https://arma3.swec.se/server/list, ~5300 players online and according to http://steamcommunity.com/app/107410, ~13000 in game.

[Rydygier 1317]

From the perspective of a person, that have no access nor knowledge about actual AI code in Arma, yet, as another exclusively SP player I would like to say as follows:

 

1. For me - AI is one of the most if not THE most important aspect of Arma... Oh, OK, it is the most important aspect of Arma to me. 

 

2. I don't know any other game with better/as good AI. It is in many aspects awful, but it is awful in terms of its own league, that this AI defines by itself, a league, where Arma's AI is the only member. In other words: it's that good, it annoys at things, other AIs don't even try to reach. And since appetite here grows as we eat... 

 

3. My guess is, some AI problems, we observe in Arma 3 indeed may be caused by the fact, the system is so complex right now (comparing to OFP). Perhaps too complex to handle, and from such complexicity we got both performance and behavioral issues. 

 

4. From the other hand however to me, if we move bugs and issues aside, the primary problem is - AI too often is still too inhuman (as for driving, reaction to danger, a way, it's perception works...). And that means, AI should be even more complex. 

 

The question is, if it is even possible simultanously to improve AI performance, to make it less CPU-hungry, more reliable/less clunky/easier to control and yet to make it more complex/natural as for behavior? If there's no smart way to do so, perhaps more complex levels of behavior should be dynamically enabled/disabled depending on current CPU load. 

 

The second question - considering current status quo, could such goal be achieved by tweaking and enhancing, what we have now, or indeed it's beyond human possibilities and in order to do so developers would need to write an AI completely anew, this time having in mind more sophisticated/efficient outcome, than when they designed current system?

 

From my scripting experience - if I have to turn some really complex script into something else and/or better, easier and faster often is rather to write, what I need, from a scratch, than trying to adapt existing code without getting lost in its logic. Especially, if changes, I want to make, require reworking fundamental code design (flaws and limitations caused by design itself). I mean - to write a new script I can construct it piece by piece, considering one at a time, but to adapt existing, ready, complex construction of many interrelated pieces, I need to grasp it as a whole first to make sure, I know, what I'm doing with it, and what exactly consequences each change will have. Much harder. 

[semiconductor 309]

I can't express how much I'm tired of trying to command a bunch of dumb retards who can't get simplest task done and constantly force me to listen to their incomprehensible and meaningless gibberish.

 

In a stark contrast with these idiots, the AIs are at least trying.

 

[ineptaphid 6413]

7 minutes ago, semiconductor said:

I can't express how much I'm tired of trying to command a bunch of dumb retards who can't get simplest task done and constantly force me to listen to their incomprehensible and meaningless gibberish.

 

In a stark contrast with these idiots, the AIs are at least trying.

 

HAHAHAHA! I was about to remove this comment for adding nothing to the discussion...until I took time to re-read it :)  It still doesn't add much-but it's funny :D

 

As one of these so called- "unseen majority myth" single players..I have over 3100hrs in Arma 3-totally in SP. While the ai could do with some work-nobody disputes that, I think it is actually pretty fantastic.

No other game expects as much from it's ai- every unit needs to be able to fly, swim, run,drive,shoot, navigate massive and rugged terrains, navigate buildings, stay in formations, utilise 9 different stances and present a challenge to the player. And for the most part -it does all that quite well.

 

Of course there are always instances of squad ai taking ages to get past a small fence or whatever, or enemy ai bullseyeing you from 400 mtrs on a dark night-but this does not happen too often-and certainly not as much as it used to.

 

A couple of ai mods will greatly enhance your experience-although the only one I use at the moment is Halek's "Discipline" mod. It reduces the amount of ai radio chatter, but still lets them say the important things, and it removes the perpetual grunting and panting when you have been slightly injured .

[Neviothr 102]

27 minutes ago, semiconductor said:

I can't express how much I'm tired of trying to command a bunch of dumb retards who can't get simplest task done and constantly force me to listen to their incomprehensible and meaningless gibberish.

 

In a stark contrast with these idiots, the AIs are at least trying.

 

 

 

That's why I play by myself, without AI friendlies. All it requires is balancing the mission, and being your own Zeus once in a while - something I've created an addon for.

[ProfTournesol 956]

1 minute ago, Neviothr said:

 

That's why I play by myself, without AI friendlies. All it requires is balancing the mission, and being your own Zeus once in a while - something I've created an addon for.

 

Well i'm afraid you missed the irony in his post