Bohemia, have you bothered at all with the AI in the last 3 months?

[bad benson 1733]

i really think they should be as aimbot-like in very close range as they can be now in long range. it should be very dangerous to get close to them. i would be fine with super human aim when you are very close. of course cuppled with human spotting ability.

i think in general the problem with the turning speed is that it is the same for long and close range. so if they have to adjust to someone running left to right 400 m away they only have to move their aim slightly. if the same happens close they have to almost turn 180°.

here's a crappy picture to illustrate what i mean in case my englisch fails ;). the angle is much bigger when the target is close.

the problem is the overall way their aim is made human-like. what ever the method is, it just dowsn't work right. it's just set up/tweaked for lange range and simply fails for close range

it has to be replaced by super reactive turn reflexes and long range has to be made harder for them by using AI weapon sway and dispersion. it seems like different distances almost aren't a factor at all. that's why the identical turn speed for long and close range leads to better aim at long range which makes zero sense.

[Vigil Vindex 64]

The fact that it is only recently that a new programmer has been brought on-board to tackle the AI is a bit strange to me. I was hoping that BIS would have a programmer putting a fair share of their time into the AI on a regular basis, not having to hire new staff to fill that need. And I have to wonder, as a programmer myself, what kind of state your documentation is in for the AI code. I've worked at places where there is good docs and its easy to drop in and start work on other peoples code, but I've also worked at places where code docs are non-existent and new devs are left having to sift through spaghetti with no good overview of how the code interacts with the rest of application, which takes serious time and effort, and can even stop progress all together if its a critical system, such as the AI. I remembering seeing some AI FSM docs on the wiki but it looked like it wasn't being actively worked on some time ago.

[sbua16 19]

This thread title is -all- wrong. It should be: "Bohemia, have you really put in enough effort into the AI in 12 years?"

IMO that answer is a resounding 'nope'.

How about we get something innovative in ArmA 4 (no chance for 3) or at least something close to what we have to a heavily modded up ArmA 2 at the moment.

AI using buildings

AI communicating with each other

Coordinating attacks

AI actually using cover very affectively in the majority of cases

AI utilising and calling in Artillery

AI with realistic aiming and ranging abilities

Random movement that doesn't make any sense whatsoever kept to a minimum

AI utilising smoke as cover

Actual Medic System

AI that are reliably able to get from point A to point B on foot, in the air, and on wheels

Just some of the things that we have in ArmA 2 currently w/ mods that we don't have in ArmA 3.

In all honesty, I really don't think i'll be playing much of ArmA3, it just doesn't have the right feel in it's current iteration (and I don't think it will for a couple of years to come). It's aesthetically pleasing, yes, but that's all I think it's get going for it at the moment. I just don't feel nearly as immersed as I do in my current rendition of ArmA 2

[royaltyinexile 175]

The fact that it is only recently that a new programmer has been brought on-board to tackle the AI is a bit strange to me.

It's not just strange to you :)

It's true that - at the beginning of the project - certain things were not exactly prioritised as perhaps they 'should be' - mistakes have certainly been made

I want to be open about the approach to AI, if only to 'manage' expectations. It's clear that the AI deserves a lot more attention, but it's one thing saying it and another thing creating a workable design to actually improve it. There are lots of little identifiable issues from the community - and these really help to make the bit-by-bit improvements; but a consistent, focused overhaul requires careful inter-departmental thinking and the time and space to break the game. (time and space often being in rather short supply).

It was like with Radio Protocol. Such a huge, monstrous legacy feature that clearly needed to be improved for years. Originally there was no work planned on it at all, but we saw an opportunity to make a clear, fairly modest plan, felt the benefits would be obvious and the task manageable. That kind of work you can execute with (relative) confidence. It's still not perfect, but it's a worthy iteration.

AI is a different kettle of fish, as it touches many many aspects of the game. It's far more important - in the sense that it has a major and direct effect on gameplay - but it's also really dangerous to tweak, because it can break everything at once. If more time was allotted earlier on in development, more fundamental changes can be made. That time wasn't budgeted, and now we must talk honestly about refinement and evolutions.

That's not to underplay the effect that these can have - I've said in the past - many quirky AI behaviours are rooted in mis/under-configured assets, which is where the balance designers come into play. It's also not to kick the AI too hard. Given the nature of Arma, it does a job, but at critical times it can be doing too much or not enough, which can in that moment tear the whole thing apart.

Overhauling the more fundamental behavioural systems, however, requires focus, care and attention. It'd be great to get my teeth into it together with a small Task Force (as with radio protocol: the - part-time - work of 3-4 guys across various departments), but it's simply not feasible at this time.

I've also worked at places where code docs are non-existent and new devs are left having to sift through spaghetti with no good overview of how the code interacts with the rest of application, which takes serious time and effort, and can even stop progress all together if its a critical system, such as the AI.

I can't speak for the documentation (but it's in Czech), and yes, the AI is well known to be a deep-rooted beast. I don't envy our new programmer's task, but I do fully support the endeavour.

Best,

RiE

Edited July 16, 2013 by RoyaltyinExile

[old_painless 183]

I would pay for a DLC that had improved AI as its sole feature. If we add convoy behaviour to that list (two posts up) it would get my money. How about "ArmA 3: Going Prone" or "ArmA 3: Staying Frosty" or "ArmA 3: Special Forces" (double meaning to that one)

Edited July 16, 2013 by Old_Painless
more DLC marketing material

[Azzur33 1]

"Bohemia, have you really put in enough effort into the AI in 12 years?" 'nope'

AI using buildings

AI communicating with each other

Coordinating attacks

AI actually using cover very affectively in the majority of cases

AI utilising and calling in Artillery

AI with realistic aiming and ranging abilities

Random movement that doesn't make any sense whatsoever kept to a minimum

AI utilising smoke as cover

Actual Medic System

AI that are reliably able to get from point A to point B on foot, in the air, and on wheels

Large OFP/Arma-community (with Lots of really talented people) has never achieved "Fantastic"-all-in-one -level AI-mods either.

Asking developers for some sort of miracle and being overly disappointed for not having it, is not very constructive.

How would a game company achieve an almost perfect AI then?

Starting from zero, focus on AI only. Spending years to make AI adapt to every possible (or impossible) situation, fine tuning reactions, fine tuning even the human-like errors they could make in the field. Adding behaviours to special situations, ironing out all the weirdness and bugs, and so on.

And then start making the Game. Huge graphical effort of 3D-models, textures, interesting story, fluent user interface and so on. Over the polished, fantastic, human-like AI.

Yeah, that could work. So, where is the game company that could do that? Why isn't there any?

AI in Arma-games is the best I've seen, considering the scale and diversity of the game worlds.

And sure there are flaws and some missing features, and some stupidity sometimes. Totally understandable. AI really is a complex thing.

It can and will be fine tuned, but it never will be perfect. The main thing is, it still usually WORKS.

Edited July 16, 2013 by Azzur33

[Sneakson 1]

The fact that it is only recently that a new programmer has been brought on-board to tackle the AI is a bit strange to me. I was hoping that BIS would have a programmer putting a fair share of their time into the AI on a regular basis, not having to hire new staff to fill that need. And I have to wonder, as a programmer myself, what kind of state your documentation is in for the AI code. I've worked at places where there is good docs and its easy to drop in and start work on other peoples code, but I've also worked at places where code docs are non-existent and new devs are left having to sift through spaghetti with no good overview of how the code interacts with the rest of application, which takes serious time and effort, and can even stop progress all together if its a critical system, such as the AI. I remembering seeing some AI FSM docs on the wiki but it looked like it wasn't being actively worked on some time ago.

I think their AI guy broke his collarbone. Not much to do lol.

[froggyluv 2136]

How would a game company achieve an almost perfect AI then?

Starting from zero, focus on AI only. Spending years to make AI adapt to every possible (or impossible) situation, fine tuning reactions, fine tuning even the human-like errors they could make in the field. Adding behaviours to special situations, ironing out all the weirdness and bugs, and so on.

And then start making the Game. Huge graphical effort of 3D-models, textures, interesting story, fluent user interface and so on. Over the polished, fantastic, human-like AI.Yeah, that could work. So, where is the game company that could do that? Why isn't there any?

Yeah, that's what I was always hoping BI was doing - santa AI elves locked away for 12 years in the workshop and emerging with something yuletide glorious! And your right -why hasn't ANYONE done this? Ai has gotten worse over the years in PC gaming -not better. Where's the evolution of R6 or Swat 4 AI? hell even BF2 bots were fun to play with -what happened to their AI? IMO the emergence of PvP gaming (no flame war here) has basically unburdened developers with the hardest part of pardon the redundancy, developing. As for SP roleplaying type games, where's the evolution? Shouldn't all important bots be able to actually open/close doors (not teleport), pop and duck from guard towers, sneak around rooms and ambush player by now? They're still the god awful world chasing mouth breathers with no movement nuance that they've always been?!

Any fan of SP should be outraged by the lack of development here as really no one besides BI is pushing the envelope at all. That's why despite disappointing progress here, I still support.

[Polygon 11]

A honest question to BIS: what refrains you from opening up the possibilities a bit for community to improve code on both shaders and AI? Is it the fact that you possibly sell the commercial use license of RV engine for others to utilize?

I'm fairly sure there's enough knowledgeable enthusiasts around here to get their hands on this stuff without breaking everything in one *magical* touch.

I see some prioritization issues in the development of A3. For instance, was PiP feature really worth the gain on the gameplay / visuals instead of improving AI or terrain tech? Underwater areas are understandably included - to give the player a WOW factor and attract some new crowd. Not really sure how that did worked out. It was Wasteland mostly attractive to new folks?

[bigshot 64]

we normally would avoid threads with aggressive titles

Hi RiE,

While I can certainly understand that, you should understand (and im sure you do) that after many years of leaving "kind" & "polite" messages for you on these forums and watching them go ignored/unanswered by any BI Rep. (as well as many hours spent with the alpha/beta tracker to try and help) tends to leave a customer no choice but to get a "tad" aggressive (we've both heard and read much harsher words, afterall) in order to get some very simple answers regarding the AI.

That said, I thank you for finally chiming in but only wish you would have addressed the 2 points I brought up regarding the AI to begin with...it would also be nice if you could create an AI related sticky for us where you (BI) would post about the AI progress every so often so we can all see it and understand what can be fixed/changed and what cannot be. An easy place to find new AI related info from BI where there can be NO RESPONSES (reading thru 50 pages of responses is ridiculous) and it can be read in plain layman's language for those of us who do not understand programming/code "talk".

Now, I know you said you've hired a programmer that's currently trying to wrap his head around that code in order to work on certain aspects of the AI targeting....that's good to hear. Can you please elaborate some more into my questions though?...specifically 2 things:

1. Are you aware of the path finding issues that cause foot units to get permanently stuck inside of buildings they were never supposed to enter...and are you aware of the AI driving issues as well?.... can these things be fixed or no? are you intending to try and fix it or no?...if you are, then what sort of time frame are we looking at? (weeks or months or years)

2. Will there be a new difficulty setting added in the UI for the AI's "precision/accuracy"?...if not, will you at least replace our ability to set both "skill" and "precision" in the profile config file (the way it used to be about 18 months ago before you automated it in A2)?

Thanks again, RiE for responding...and thanks to everyone else in the thread for HELPING to attract the response as well. I hope BI will talk more openly to us about the AI in an easy to understand way (even when it's not good news).

[tremanarch 6]

guys programming such an AI is incredibly tough! i am sure it would be worth 4 PHD titles.. its extremely crazy - it tangents scientific areas! dont expect too much.

think of how happy guys in informatics are if they have crappy robots playing soccer..

the ai as of now - im sure - is big project that could be advertised at universities.

the cleverst thing would be to implement some kind of software interface (a comfortable and mighty one) that the community can access to rewrite some og the ais behaviour themselves.

[smacker909 3]

Hi RiE,

2. Will there be a new difficulty setting added in the UI for the AI's "precision/accuracy"?...if not, will you at least replace our ability to set both "skill" and "precision" in the profile config file (the way it used to be about 18 months ago before you automated it in A2)?

.

This would be awesome. At least to start with :)

[Dwarden 1125]

and if you want speed things up, know or get some new good AI bug tickets on tracker,

then you may always hit me (pm,skype,irc,email w/e you find) ...

i will look on that, assign to QA ... link together more AI tickets if needed ... discuss with rest of team

it's not like we blind, deaf and ignore You :p

[sbua16 19]

Large OFP/Arma-community (with Lots of really talented people) has never achieved "Fantastic"-all-in-one -level AI-mods either.

Asking developers for some sort of miracle and being overly disappointed for not having it, is not very constructive.

How would a game company achieve an almost perfect AI then?

Starting from zero, focus on AI only. Spending years to make AI adapt to every possible (or impossible) situation, fine tuning reactions, fine tuning even the human-like errors they could make in the field. Adding behaviours to special situations, ironing out all the weirdness and bugs, and so on.

And then start making the Game. Huge graphical effort of 3D-models, textures, interesting story, fluent user interface and so on. Over the polished, fantastic, human-like AI.

Yeah, that could work. So, where is the game company that could do that? Why isn't there any?

AI in Arma-games is the best I've seen, considering the scale and diversity of the game worlds.

And sure there are flaws and some missing features, and some stupidity sometimes. Totally understandable. AI really is a complex thing.

It can and will be fine tuned, but it never will be perfect. The main thing is, it still usually WORKS.

Yeah, people who put their free time into putting content into the game for us to enjoy. We don't pay these people, they don't get anything out of it (usually), they're releasing these mods to us to enjoy, to make the game better.

All those features are present in modded ArmA 2, features that should be and could be in ArmA 3, Heck it's practically there for the picking. I hate to sound like I know what I'm talking about, but surely it's possible to see these features that people introduced and say "Yeah.. we see how that's done, we'll introduce something similar that's polished". Instead of "Yeah guys, now the AI spin slightly faster so we won't have as many of those YouTube videos of people running around them hilariously", I don't give a shit about that. Give us fire and manoeuvre, give us AI that don't randomly run off for no reason, give us transport helicopters that don't do almost exactly what they did 12 years ago under fire.

Here's a test, put 2 teams out in a field, one Opfor and one Blufor. Play both these scenarios in OFP and ArmA 3. The result is borderline indistinguishable. Both will use bushes and covered terrain for cover, both will be frustratingly slow on their approach, both will crawl more then is needed, they both will look cool ONLY in random sporadic times, like once every blue moon. You can't even suppress AI for gods sake, I did that in Brothers in Arms circa 1963 (sic). All we're asking for is some sort of leap forward, something other than "yeah.. we er.. made it so that they can sometimes get into cover and lean properly". A good number of times I can't even get from one end of Agina Marina to another without AI getting stuck in some movement loop that either forces me to kill them, or just end the mission in frustration.

Edited July 16, 2013 by Sbua16

[bigshot 64]

and if you want speed things up, know or get some new good AI bug tickets on tracker,

then you may always hit me (pm,skype,irc,email w/e you find) ...

i will look on that, assign to QA ... link together more AI tickets if needed ... discuss with rest of team

it's not like we blind, deaf and ignore You :p

Actually...my only 2 questions asked in this thread are STILL being ignored :-) (see my post #1 and #60)

I've done the tracker thing for a few months...no longer have the time to donate, however...and I'm only interested right now in getting the 2 questions answered at least.

Thanks once more.

[dr. hladik 231]

The fact that it is only recently that a new programmer has been brought on-board to tackle the AI is a bit strange to me. I was hoping that BIS would have a programmer putting a fair share of their time into the AI on a regular basis, not having to hire new staff to fill that need.

What is strange? We are always happy to hire new capable programmers. And yes, as there are many things that should be improved, he is currently helping me with some AI bugs.

I think their AI guy broke his collarbone. Not much to do lol.

Not AI, but sound guy and he is brave enough to keep working even when he can hardly type.

Don't know where you took this information, but please next time be more careful with this kind of gossip.

Also please note, that not all engine changes are listed in change log.

After engine change is made it usually takes additional work (new data, map changes, balancing, testing) that is not present exe beta patch. In that case it cannot be simply announced with source code commit, as you would not notice anything.

Also, as we really like you, some changes are made even for older version of VRE (OA, TKOH) and after that they are merged to A3. Those are unfortunately also not visible in change log.

Edited July 16, 2013 by Dr. Hladik

[Tyl3r99 41]

i think we should all be co-operative :) lets get all the valid (HIGH VOTE) AI tickets and can we email these to the new dev? or even to you dwarden?

thank you for giving some feedback its always a great feeling when a dev replies lol

[old_painless 183]

and if you want speed things up, know or get some new good AI bug tickets on tracker,

then you may always hit me (pm,skype,irc,email w/e you find) ...

i will look on that, assign to QA ... link together more AI tickets if needed ... discuss with rest of team

it's not like we blind, deaf and ignore You :p

Dwarden, I have not entered any items on AI myself, but if you go to http://feedback.arma3.com/view_all_bug_page.php and filter by Category="AI Issues", then a couple of hundred items show up. As for pathfinding, one example is http://feedback.arma3.com/view.php?id=11197

[bad benson 1733]

omg. it's Dr. Hladik!

i have so many good memories with this dude from chernarus apocalypse :D

edit:

think of how happy guys in informatics are if they have crappy robots playing soccer.

bad example. robots have to function in the real world. game AI has to function in a world creted by who makes the AI.

i do agree that AI is a tough thing to do. but we should differentiate between people asking for a perfect AI and people who just want old bugs to be fixed. i really like the potential the AI shows and always showed. but it's just time to polish what is there. and that goes for buildings too. enterable buildings have building positions in them that are for the AI.

how hard can it be to add an option to a garrison type waypoint to make AI man some positions? or make a squad leader give one or two squad members an order to cycle through all building positons of a house the player is shooting from? these things are 100% possible (has been done) and can always be made optional, if there is a fear of breaking missions.

it's just the fact that AI won't use any of what is already there at all.

Edited July 16, 2013 by Bad Benson

[bigshot 64]

i think we should all be co-operative :)

thank you for giving some feedback its always a great feeling when a dev replies lol

Oh yes, they did a wonderful job of double talking to calm the storm didn't they?...yet they really never answered a single question asked....and some of you actually fall for this old trick?

Getting close to 8 pages now...still waiting for 2 answers to 2 very simple and reasonable questions for which I started this thread with. 12 years is a long time to wait I admit, apparently I am quite the optimist!

[dr. hladik 231]

i think we should all be co-operative :) lets get all the valid (HIGH VOTE) AI tickets and can we email these to the new dev? or even to you dwarden?

thank you for giving some feedback its always a great feeling when a dev replies lol

Send them to me.

In particular I would appreciate repro mission for AI stuck in a building.

[-Coulum- 35]

Thanks to devs for all your feed back in this thread.

I wasn't expecting much in terms of ai "smarts" for arma 3 release - the devs made it pretty clear ai features and behavioral changes weren't happening before release, and, although I am not necessarily thrilled by that, I understand it and am eagerly looking forward to post release patches.

But one thing I am confused about is "Task force balance". I thought that one of that team's many objectives was to round out the ai a bit to make their skills more human and less robotic - but I personally don't really see any changes, in terms of spotting or accuracy - the ai are still more like robots than humans (esp in the spotting.).

What exactly is the situation with this ai balancing? I was hoping to see more "test, gauge, readjust and repeat" concerning this but it doesn't seem like anything is happening... Are the changes just too "internal" at this point for a player like me to notice? Is there plans for more "aggressive" tweaking?

Edited July 16, 2013 by -Coulum-

[dr. hladik 231]

Oh yes, they did a wonderful job of double talking to calm the storm didn't they?...yet they really never answered a single question asked....and some of you actually fall for this old trick?

Getting close to 8 pages now...still waiting for 2 answers to 2 very simple and reasonable questions for which I started this thread with. 12 years is a long time to wait I admit, apparently I am quite the optimist!

1. AI path finding - Could you please send me a repro mission? As you can imagine, there are many things that affect pathfinding (engine AI, vehicle configuration, obstacle configuration, map) and this is still not a final product. So it is really hard to just tell how/when this can be improved. (I'm sorry if I missed it somewhere).

2. AI precision settings - just in this thread some are asking for higher precision when attacking moving target (answer is in #41), some are asking for smaller precision. This is more (not all) about design decision than about engine changes and you cannot expect that someone will change it in just one day.

[froggyluv 2136]

2. AI precision settings - just in this thread some are asking for higher precision when attacking moving target (answer is in #41), some are asking for smaller precision. This is more (not all) about design decision than about engine changes and you cannot expect that someone will change it in just one day.

I think this is exactly the sort of thing we were (are) hoping that your Team Balance would calibrate. We sympathize with the full spectrum demands that the community may think is the best way to go but are hoping that a well versed team can better decide. Maybe even pluck a few of the AI guys from around here to work with your team :confused:

[smacker909 3]

How about precision slider in settings.. so we can adjust ourselves? I was under the impression (read in a thread) it was removed from A2 because developers wanted to simplify difficulty settings.