AI Discussion (dev branch)

[old_painless 182]

What better way to test than to use it full scale. Wiki will find things that will be hard to come by otherwise

[Bouben 3]

You'd be surprised by how many people think that the dev branch is a stable and tested version.

That is their problem then. Not my problem, not a problem of BIS.

[oukej 2911]

btrw, the issue with drivers should now be gone ;)

[Fushko 59]

That is their problem then. Not my problem, not a problem of BIS.

Yeah, that was quite obvious.

btrw, the issue with drivers should now be gone ;)

Oukej you're on an AI fixing spree lately :cool:

[Wiki 1558]

What issue are you talking about?

[oukej 2911]

http://feedback.arma3.com/view.php?id=20638

Oukej you're on an AI fixing spree lately :cool:

Not me, it's The Four Musketeers from the other room. They don't always flirt with the AI, but when they do... ;)

[NeMeSiS 11]

http://feedback.arma3.com/view.php?id=20638

Not me, it's The Four Musketeers from the other room. They don't always flirt with the AI, but when they do... ;)

They should take the AI out to a proper date. :)

[killzone_kid 1331]

AIs are hilarious! No seriously! I was testing something, so I spawned a buddy rifleman and an enemy Ifrit, unarmed though. Ordered my buddy to attack Ifrit, nothing. 2, Attack! Nothing, 2, Fire! Nothing. 2, Move there! 3 sec later "Enemy MRAP!". Really, I just ordered you to attack it you dumbass. So he refused to shoot. Ok fine no AT weapons, but what about tyres? Chuck a grenade maybe? Anyway, he goes to the Ifrit and starts circling it, while still refusing to engage. The other guy in Ifrit notices him standing in front of the vehicle, so he starts reversing, then turns around and starts to flee. It is like watching dumb and dumber, really.

[kremator 1065]

Truely sad state of affairs :(

[potemps 1]

Hello everyone.

There is a lot of talking about what can be done with A3 AI, and what cannot be done - but for first you should chek this link, and look at videos inside.

https://sites.google.com/site/wargamingpluspc/

Arma, and OFP is best games what I ever played, but the most problem in all of these games is AI :( I really hope, that some day BIS will finally change the whole AI mechanics to better one.

Regards

[Variable 322]

I'm replaying Operation Flashpoint: Dragon Rising again, and seeing how well the AI behaves there truly frustrates me in regards to the shameful state of Arma 3 AI.

They take cover, the suppress, they are getting suppressed, they are responding to orders under fire, their accuracy is human like, they report when they find bodies, they breach and clear buildings(!!!), they move fast when ordered to. And the list goes on and on.

I think it's about time BIS assign a programmer to open up the AI code and start treating issues that have been there since Arma 2 and before.

Sorry for the harsh words, but issuing more DLCs with the current state of the AI feels like a spit in the face.

[old_painless 182]

So true. About time that AI is regarded as game content in and of itself.

[miketim 20]

Talk about a spit in the face, your speaking about OF: DR, wow :)

Then again, I bet there are probably reasons for that,

Lets say you have wall model #1

Skira isnt that big, so they only have, lets say 4 wall models.

They can add the postions for AI to take cover and all those things, whereas arma 3 probably has 30 types of walls so, its not like they can implement that now.

Thats a bad example though, because some AI mods allow AI to peek around walls.

But I am just trying to say obviously certain things cant be done, specifically because of how much stuff is already present and if it would have to be changed to make it "compatible" (ie: building models) even though we dont usually think of AI related stuff as anything but code

[bad benson 1733]

Talk about a spit in the face, your speaking about OF: DR, wow :)

Then again, I bet there are probably reasons for that,

Lets say you have wall model #1

Skira isnt that big, so they only have, lets say 4 wall models.

They can add the postions for AI to take cover and all those things, whereas arma 3 probably has 30 types of walls so, its not like they can implement that now.

Thats a bad example though, because some AI mods allow AI to peek around walls.

But I am just trying to say obviously certain things cant be done, specifically because of how much stuff is already present and if it would have to be changed to make it "compatible" (ie: building models) even though we dont usually think of AI related stuff as anything but code

pretty sure that arma walls already have cover positions and that that is what makes AI use them, sometimes. what makes you think that preparing assets for the AI to work better is a bad/cheap/"out of scope for arma" method?

it's exactly what BI needs to do. hell afaik the so called navigation mesh they indroduced to dayZ (can you believe it? soon dayZ's zombies will be smarter at pathfinding than arma's soldiers FPDR) is exactly that. a precalculated method to give AI more meaningful info to get around the world without having to calculate it on the fly. teh info is just there from the start. the whole "indoor AI" is based around fixed positions inside the models since at least arma 1.

i remember a discussion in here about cryengine and how "they are just placing nodes to make AI work in houses better" in it, eventhough it's exactly arma's approach with the slight difference that it's severly limited and utterly broken. and before anyone tries to pull out the scale card. these kind of predefined things are exactly what large scale games need since it avoids complex calculations and analysis of the environment for each AI constantly on the fly.

it's a pretty sad attitude to belittle other games where stuff actually works better (as intended) just to spare BI the well deserved criticism for their AI that has almost not evolved beyond its biggest flaws.

now just to make this clear. this is not against you personally or any of the brave guys at BI working on fixing this mess of AI they haven't created themselves. it's just frustrating to only see bandaids applied to a system that has big problems at its core, year after year.

Edited September 13, 2014 by Bad Benson

[-Coulum- 35]

I'm replaying Operation Flashpoint: Dragon Rising again, and seeing how well the AI behaves there truly frustrates me in regards to the shameful state of Arma 3 AI.

Ouch!

Of course I would love to see Ai improvements as well.

And just because there isn't any hugely noticeable improvements as of late doesn't mean nobody is working on it I am sure (well hope.)

[Variable 322]

Talk about a spit in the face, your speaking about OF: DR, wow :)

Then again, I bet there are probably reasons for that,

Lets say you have wall model #1

Skira isnt that big, so they only have, lets say 4 wall models.

They can add the postions for AI to take cover and all those things, whereas arma 3 probably has 30 types of walls so, its not like they can implement that now.

Thats a bad example though, because some AI mods allow AI to peek around walls.

But I am just trying to say obviously certain things cant be done, specifically because of how much stuff is already present and if it would have to be changed to make it "compatible" (ie: building models) even though we dont usually think of AI related stuff as anything but code

Is this an architecture game or a damn combat simulator? For heavens sake, if the size of the island is indeed what prevents BIS to create meaningful AI behavior (and it doesn't) I would trade 2/3 of the glorious Altis just to see AI acting reasonably.

Do you know what I just had in OFP: Dragon Rising? I had an AI manning a static heavy MG. That particular MG can be oriented in an arc of 90%. We got flanked and an enemy guy snuck up behind that fellow. Do you know what he did? He left the damn static gun, drew his rifle and shot that enemy. Right after that he got back to his position in the stationary MG. Now you want to tell me that that's not possible due to Altis size as well? And we are talking about a game that competed with Arma 2 for f*** sake! In 2009! This just makes me sad.

Don't tell yourself stories, BIS simply don't allocate sufficient and adequate resources to open up the AI code and fix. The reasons are are just beyond me.

Ouch!

Of course I would love to see Ai improvements as well.

And just because there isn't any hugely noticeable improvements as of late doesn't mean nobody is working on it I am sure (well hope.)

I'm sure nobody is working on it. Maybe just some patching on the config level.

Edited September 13, 2014 by Variable

[oukej 2911]

I'm sure nobody is working on it. Maybe just some patching stuff on the config level.

Not true.

Apart from teaching the AI all the new functionality (slingloading, firing from vehicles... ref. roadmap) the focus so far has been mainly on fixing issues. Most recently it's been low level AI optimizations and the pathfinding in buildings (which had been broken for a long time in buildings without the floor level). There are more improvements coming and some "new" features are in the process or under investigation. (I know, I know - and you've hinted it - some of them have been an industry standard and "should have been there from the start").

But to put into manpower perspective - it's been only relatively recently that we were able to expand the team and even now our programming dept. is about 3 times smaller than the one from OFP-DR.

Btw, the cover system in Arma is generic. The AI will happily walk your map and take cover behind objects you created. Not that the compatibility (also backwards, data and scripting compatibility - not to break much of the already created content) and universality didn't come at costs :/

As for the DayZ zombies - running straight up to you is all they need ;)

Edited September 14, 2014 by oukej

[bad benson 1733]

As for the DayZ zombies - running straight up to you is all they need ;)

that's why i said "smarter at pathfinding" because i knew someone would say what you said :p . i think in general saying that pathfinding is bad because they also have to shoot and do other things is not accurate. i think the problems of pathfinding are elsewhere and i think that all the improvements a nav mesh brings can easily be applied to any type of AI.

pathfinding itself is a basic AI feature and it makes no sense to sacrifice proper pathfinding for complex stuff that isn't even working since it's blocked by stupid anims or delayed by who knows what. i hate to be the car analogy guy, but what good is an expensive high powered car if it's missing something basic like tires to even drive?

Btw, the cover system in Arma is generic. The AI will happily walk your map and take cover behind objects you created. Not that the compatibility (also backwards, data and scripting compatibility - not to break much of the already created content) and universality didn't come at costs :/

yea the problem here is the difference between actual code and what the player experiences as actually something he can see them do effectively. i think people are mostly talking about the leaning feature when they mention cover since you barely (if at all) see AI "conciously" take cover quickly behind stuff when it's needed.

so it comes down to the animation being an indicator of them trying. i mean i realise that there might be something going on under the hood concerning cover but you have to forgive people, if it's not visible to them because the timing is so off that you will never notice an AI thinking "oh i'll do my cover routine now" because all you see is them leaving cover under fire all teh time, which leads to thinking that there is no system in place at all because the outcome is the same.

in arma AI feels retarded mostly because of stuff not functioning not because it's overly simple. so the question is what of its complexity on the code level actually reaches the player.

again. i realise that you guys didn't write this from the ground up for a3 and that you are trying your best to improve it anyways, and you are succeeding in many isolated cases. and please don't let our frustration keep you from doing it. it's just an indicator that there is a lot of passion for AI in this community. so your work is more than welcome.

now back to being constructive: would it be possible to fix AI getting stuck when inside houses? i really love what you guys did to CQC reflexes with the AI. it's for me one of the best features of arma 3. problem is that even if you just want stationary dudes in houses in your mission so people can clear the houses, you get a lot of AI that are stuck unable to rotate. that makes them basically sit there accepting any punishment. HUGE immersion killer.

i learned to live with next to no dynamic indoor AI but that bug has to go. they have to react fast like outside. that would improve the existing limited indoor combat capability by a huge amount.

[Sterlingarcherz101 15]

Not true.

Apart from teaching the AI all the new functionality (slingloading, firing from vehicles... ref. roadmap) the focus so far has been mainly on fixing issues. Most recently it's been low level AI optimizations and the pathfinding in buildings (which had been broken for a long time in buildings without the floor level). There are more improvements coming and some "new" features are in the process or under investigation. (I know, I know - and you've hinted it - some of them have been an industry standard and "should have been there from the start").

But to put into manpower perspective - it's been only relatively recently that we were able to expand the team and even now our programming dept. is about 3 times smaller than the one from OFP-DR.

Btw, the cover system in Arma is generic. The AI will happily walk your map and take cover behind objects you created. Not that the compatibility (also backwards, data and scripting compatibility - not to break much of the already created content) and universality didn't come at costs :/

As for the DayZ zombies - running straight up to you is all they need ;)

I hope none of. The devs take any of the fair criticisms laid before them personally. And maybe understand that it is because people like the game they are so passionate about the game. they want for it to succeed therefore get frustrated when iparts are broke for so long.

Glad to know yous are at least getting Down to some much needed improvements in the ai .

There is a big list from ai , game performance,sounds, that have been really lacking . So it can be annoying when dlc gets announced and you look at the current state of play thinking throwing more cogs into the machine that does not work properly.

[Sniperwolf572 758]

that's why i said "smarter at pathfinding" because i knew someone would say what you said :p . i think in general saying that pathfinding is bad because they also have to shoot and do other things is not accurate. i think the problems of pathfinding are elsewhere and i think that all the improvements a nav mesh brings can easily be applied to any type of AI.

pathfinding itself is a basic AI feature and it makes no sense to sacrifice proper pathfinding for complex stuff that isn't even working since it's blocked by stupid anims or delayed by who knows what. i hate to be the car analogy guy, but what good is an expensive high powered car if it's missing something basic like tires to even drive?

I agree, this is the main thing that makes the AI appear "dumb" to me. I know under the hood that they are taking a million things into consideration to plan their actions, but when they execute movement, they always seem to be undecisive and they needlessly change direction as if going:

1. Ok I'm going to go there!
   
2. Wait! This might be a better path. TUUUUUURN
   
3. NO NO THIS THIRD ONE IS. TUUUUUURN
   
4. oh wait, the original one was actually a better path. TUUUUUURN
   
5. Go to 2.
   

They eventually get there, but it's like watching a driver deal with over and understeer and constantly fail at it by overcompensating.

I'd say strafing at times would also be a much better option for the AI rather than going to a spot then turning or moving and turning in a radius. A separation of AI aiming vector and AI movement vector could also make them appear so much less dumb and prepared, but I believe this is too much to consider before the issue above is somehow dealt with.

Edited September 14, 2014 by Sniperwolf572

[roshnak 41]

Sorry for the harsh words, but issuing more DLCs with the current state of the AI feels like a spit in the face.

It's worth noting that the issuing of more DLCs is probably what is financing the further development of the game.

yea the problem here is the difference between actual code and what the player experiences as actually something he can see them do effectively. i think people are mostly talking about the leaning feature when they mention cover since you barely (if at all) see AI "conciously" take cover quickly behind stuff when it's needed.

I would say that I frequently see AI taking cover, but far too often I see them taking cover on the wrong side of an object.

I also think that even forgoing special animations, just having the AI utilize the stance adjust system would go a long way towards making it more obvious when they are "using cover."

Edited September 14, 2014 by roshnak

[froggyluv 2136]

pretty sure that arma walls already have cover positions and that that is what makes AI use them, sometimes. what makes you think that preparing assets for the AI to work better is a bad/cheap/"out of scope for arma" method?

it's exactly what BI needs to do. hell afaik the so called navigation mesh they indroduced to dayZ (can you believe it? soon dayZ's zombies will be smarter at pathfinding than arma's soldiers FPDR) is exactly that. a precalculated method to give AI more meaningful info to get around the world without having to calculate it on the fly. teh info is just there from the start. the whole "indoor AI" is based around fixed positions inside the models since at least arma 1.

Agreed, Very interested and hoping for cross development.

i remember a discussion in here about cryengine and how "they are just placing nodes to make AI work in houses better" in it, eventhough it's exactly arma's approach with the slight difference that it's severly limited and utterly broken. and before anyone tries to pull out the scale card. these kind of predefined things are exactly what large scale games need since it avoids complex calculations and analysis of the environment for each AI constantly on the fly.

Exactly! And who says Cry can't handle scale with ease:

Dragon Rising did indeed offer a few neat AI trcks and playbooks but came at quite a cost I seriously doubt us BI fans would find satisfactory. Cryengine has decent AI for that type of game: small scale shooter with player vs a few gunners at a time in select areas. Hard to compare games with such distinct foundational plans for their respective AI. Cryengine ain't the end all be all -Star Citizen, which is being built as a CE flagship has basically had to join hands with CE to create a new version just to get basic features working and keep pace with the much smaller, in-house built little space sim Elite Dangerous.

[killzone_kid 1331]

As for the DayZ zombies - running straight up to you is all they need ;)

Not sure this is a very well thought statement even if meant in jest. I will not going to start a debate about what zombies should be like, lets just all agree that "smart" zombies present a challenge and make for better experience

[fn_Quiksilver 1636]

Not sure this is a very well thought statement even if meant in jest. I will not going to start a debate about what zombies should be like, lets just all agree that "smart" zombies present a challenge and make for better experience

What like stacking on corners/doors and charging in when player is looking the other way? :)

[killzone_kid 1331]

What like stacking on corners/doors and charging in when player is looking the other way? :)

I thought more like flanking, but I like your idea, though it might actually be a bit too smart for them.