AI Discussion (dev branch)

[Alwarren 2767]

But metalcraze, did you not see them currently working on AI? There are dedicated feedback threads here in this forum. Supression for instance. Or they said they pretty much rewrote big parts of driving code.

 

He was specifically talking about AI pathfinding in terrain, which prevents AI to walk through a large enough gap between to rocks because of pathfinding issues. From what I see nothing is being done on that. Not saying a driving AI wasn't long overdue, or suppression, but what Ondrej essentially said there is that some places will not be accessible to AI period.

 

With the game being infantry-focused, I share metalkraze's concerns.

[froggyluv 2135]

 Dont think its fair to say "NO pathfinding improvements since Arma 2" - still remember guys fumbling over 6 inch ledges and 1 foot drops and getting stuck in Arma 2 Urban was far more common to get stuck. Arma 3 urban my team rarely, if ever, gets stuck in even the most busy of areas. Hell I just seen my squaddies cross over some pretty high roots in the jungle.

[madrussian 347]

But isn't it logical to move AI navigation forward in parallel to other advances?

My sentiments exactly.  With all the advances we are seeing lately (which are awesome and greatly appreciated!), AI path-finding on foot seems to be getting left in the dust.  Agree with froggy, there have been a number of small steps forward since A2.  However, very little progress regarding path-finding in and out of open buildings (getting stuck inside), and now (apparently) we have large swaths of the map inaccessible to the AI via foot.

 

It occurs to me this must be a particularly challenging area to advance.  Regarding such AI navigation improvements (particularly on foot), have we hit some sort of technical barrier?  Or are they simply not a priority (for now)?

 

Altis already allowed a player to exploit AI by just entering a house and whacking AIs from there and the chance that AI will even try attempting to enter the same house was always negligible. But now we are talking whole areas available to exploiting. It's not even some mountain path filled with non-terrain objects - the path is clear, just has rocks less than 4m apart. You can just sit there and pop AI heads with some marksman rifle without any punishment.

Seems like this will be big problem for zombie missions on Tanoa.

[fn_Quiksilver 1636]

 

Altis already allowed a player to exploit AI by just entering a house and whacking AIs from there and the chance that AI will even try attempting to enter the same house was always negligible. But now we are talking whole areas available to exploiting. It's not even some mountain path filled with non-terrain objects - the path is clear, just has rocks 

 

designed for Life/Wasteland/Exile/BattleRoyale/KOTH :)  Mil-sim/CTI/COOP will just have to use other parts of the terrain.

[Alwarren 2767]

designed for Life/Wasteland/Exile/BattleRoyale/KOTH :)  Mil-sim/CTI/COOP will just have to use other parts of the terrain.

 

I fail to see how this is funny

[old_painless 182]

He was specifically talking about AI pathfinding in terrain, which prevents AI to walk through a large enough gap between to rocks because of pathfinding issues. From what I see nothing is being done on that. Not saying a driving AI wasn't long overdue, or suppression, but what Ondrej essentially said there is that some places will not be accessible to AI period.

 

With the game being infantry-focused, I share metalkraze's concerns.

Slightly disturbing sentiment BI reveals here. Any community developer who is made aware that some parts of his island is inaccessible for the AI would promptly change things around (remove rocks, add ladders etc). But BI sees it as a perfectly fine tradeoff to make areas inaccessible for the AI in exchange for pleasing screenshot gamers.

 

Hm ...

[lexx 1363]

Sometimes you people are some real negative nancys.

[chompster 29]

I fail to see how this is funny

Because it's not. It's sad. because it's true.

[pansyfaust 69]

Sometimes you people are some real negative nancys.

Well what do you know, underpromising and "overdelivering" hasn't turned out so well, has it?

[old_painless 182]

Sometimes you people are some real negative nancys.

Heh, you may be right. We are getting a lot of goodness these days, but this thing rubbed me the wrong way

[metalcraze 290]

Clearly there are 3rd party pathfinding solutions available and seeing how VBS3 makes use of one - it should be possible to find a way out if dedicating more programmers to a new AI project is not an option.

Or spend those money buying another tank with a cemented barrel, what do I know

[R3vo 2654]

Or spend those money buying another tank with a cemented barrel, what do I know

 

I am pretty sure the T-72 was cheap compared to the developement costs of a new pathfinding solution.

[froggyluv 2135]

Clearly there are 3rd party pathfinding solutions available and seeing how VBS3 makes use of one - it should be possible to find a way out if dedicating more programmers to a new AI project is not an option.

Or spend those money buying another tank with a cemented barrel, what do I know

 

 Well I uninstalled and lost my VBS key over a year ago -byebye $350 game :/    but last I played, the pathfinding was atrocious despite their 3rd party addition which was utterly underwhelming. I had soldiers in a desert area refuse to rejoin formation with close to no objects in their way -"Can't get there" -the DESERT. Forget the name of that 3rd part Dev, I remember contacting them and they were pretty rude so I give them the finger. Their highlight was ordering a soldier to a node at the top of a tower...something accomplished easily in Arma 2.  Again I no longer own but if i had to gauge Arma 3 infantry pathfinding > VBS Infantry.

[fn_Quiksilver 1636]

162 RC:

 

I get this in my RPT when I spawn the new CSAT pacific units:

 

 8:07:39 Sound: Error: File: A3\Dubbing_Radio_F_EXP\data\CHI\Male01FRE\RadioProtocolFRE\Normal\130_Com_Reply\Confirmation2_2.ogg not found !!!
 8:07:39 Sound: Error: A3\Dubbing_Radio_F_EXP\data\CHI\Male01FRE\RadioProtocolFRE\Normal\130_Com_Reply\Confirmation2_2.ogg: sSize 0
 8:07:39 Sound: Error: File: A3\Dubbing_Radio_F_EXP\data\CHI\Male01FRE\RadioProtocolFRE\Normal\130_Com_Reply\Confirmation2_2.ogg not found !!!
 8:07:40 Sound: Error: File: A3\Dubbing_Radio_F_EXP\data\CHI\Male01FRE\RadioProtocolFRE\Normal\130_Com_Reply\Confirmation2_3.ogg not found !!!
 8:07:40 Sound: Error: A3\Dubbing_Radio_F_EXP\data\CHI\Male01FRE\RadioProtocolFRE\Normal\130_Com_Reply\Confirmation2_3.ogg: sSize 0
 8:07:40 Sound: Error: File: A3\Dubbing_Radio_F_EXP\data\CHI\Male01FRE\RadioProtocolFRE\Normal\130_Com_Reply\Confirmation2_3.ogg not found !!!

[Alwarren 2767]

 Again I no longer own but if i had to gauge Arma 3 infantry pathfinding > VBS Infantry.

 

Well the argument "It's worse in X" doesn't really hold, does it? The point is that the pathfinding in Arma 3 leaves a lot to be desired, and metalkraze is right on the money saying that this should be improved. This game boasts with its infantry combat, but still AI infantry does not manage to get through certain areas, and more than once if you tell them to go some place you get a "negative" after half an hour.

 

Infantry not being able to walk along a mountain path on Tanoa just looks atrocious in 2016, and especially in single player it is one of the fundamental sources for frustration. At least with Varanon's Autocombat addon, you can now prevent them from doing their stupid dance when you want them to get into a vehicle fast or get out of mortar fire, but you can still not get them to run between to rocks, and that is an issue.

[froggyluv 2135]

 The VBS and 3rd party option was not presented by me, I merely responded to it stating that it's not necessarily a better route to go. And of course Infantry pathfinding can always improve _ Ive been calling for more attention to building movement since, well, forever. Saying that Arma is below the game niche standard is false imo tho, as there just arent many games in which the AI do things like navigate complicated rock formations well and the like. Sure in tightly scripted shooters you may see some cool shit, but this is the hardest endevour in game development and we've seen almost all the greats give in to either neglecting or straight out removing AI in favor of pure MP in which they can design anything they like -in short, easier.

 

 As much as we all want improvement here, if the AI tech cant keep up with the level design tech -do we prefer they dumb down the terrain, keeping it imo pretty dull and lifeless but functional, or do we compromise with 85% moveability but beautiful design and more memorable sites and trekking locations. Personally as long as they add more nodes to building tops - prefer the current option.

[das attorney 858]

Note to AI users.

 

It seems that if an AI infantry group are on an isle/island (ie: a body of land surrounded by water) and a waypoint dictates that they need to travel over water to get to a new destination (on another isle/island) , then they will move to the shore closest to the other island and wait there in formation,

 

If you change their uniform to a diver uniform, only then will they swim across the body of water to reach their waypoint.

{_x forceAddUniform "U_O_Wetsuit"} forEach (player nearEntities ["CAManBase", 500])

If you do the same thing with a group mounted in a vehicle, then they will not move in the first place and stay put. I didn't try changing their uniforms but one would hope they didn't try to drive across the sea just because they were wearing wetsuits...

 

That's from my tests though so ymmv.

[Vasily.B 529]

Note to AI users.

 

It seems that if an AI infantry group are on an isle/island (ie: a body of land surrounded by water) and a waypoint dictates that they need to travel over water to get to a new destination (on another isle/island) , then they will move to the shore closest to the other island and wait there in formation,

 

If you change their uniform to a diver uniform, only then will they swim across the body of water to reach their waypoint.

{_x forceAddUniform "U_O_Wetsuit"} forEach (player nearEntities ["CAManBase", 500])

If you do the same thing with a group mounted in a vehicle, then they will not move in the first place and stay put. I didn't try changing their uniforms but one would hope they didn't try to drive across the sea just because they were wearing wetsuits...

 

That's from my tests though so ymmv.

Nice catch!

[Alwarren 2767]

The VBS and 3rd party option was not presented by me, I merely responded to it stating that it's not necessarily a better route to go. And of course Infantry pathfinding can always improve _ Ive been calling for more attention to building movement since, well, forever. Saying that Arma is below the game niche standard is false imo tho, as there just arent many games in which the AI do things like navigate complicated rock formations well and the like. Sure in tightly scripted shooters you may see some cool shit, but this is the hardest endevour in game development and we've seen almost all the greats give in to either neglecting or straight out removing AI in favor of pure MP in which they can design anything they like -in short, easier.

I know, but that doesn't mean that third party AI modules cannot be a good solution.I am a programmer myself, I have even worked for a while in the game industry and I am well aware of the challenges associated with it. But this isn't a question of whether call of duty does a good job or not. The problem itself is not unsolvable, and calling it "the hardest endeavor in game development" simply isn't true. We're not talking about making up large-scale tactics or strategies here. We are talking about an AI finding a way from A to B. I grant you it isn't easy, but it is far from being unsolvable.

 

There are plenty of games that have excellent pathfinding, that solve their issues one way or the other. One game I worked on solved it by meta-information in the map itself. basically adding AI hints (for taking cover in that case).

 

Yes, a game like CoD has it easier. So? Does that mean to settle for less just because it is harder?

 

As much as we all want improvement here, if the AI tech cant keep up with the level design tech -do we prefer they dumb down the terrain, keeping it imo pretty dull and lifeless but functional, or do we compromise with 85% moveability but beautiful design and more memorable sites and trekking locations. Personally as long as they add more nodes to building tops - prefer the current option.

You make this sound like a binary choice, either AI-friendly and ugly or AI-unfriendly and beautiful. That isn't the case. It's not like these things are mutually exclusive. Maybe you would need to widen the gap between rocks in some places, maybe you would need to remove a rock or move it to a different place, but there is no way to sell this "either-or" logic since it isn't true.

 

Back in the Arma 2 days AI was taught to go prone to crawl under fences. Prior to that they would just pass through the fence as if it wasn't there. There was talk about teaching them to step over fences as well, but it never go implemented. It would have required e.g. the addition of memory points in the model so that the AI knows which direction the fence is and can step over it. It was never implemented obviously.

 

But the current maps contain next to nothing in terms of "AI hints". A simple first step would be to introduce AI guide objects that say "if you want to go there, move along this path". It's not difficult to implement - unless the implementation is really convoluted, which cannot be excluded in code that has been growing for fifteen years. Not saying that this is a flaw, I have written code myself that has been around for years and has little things retrofitted here and there making the whole thing pretty much a "mess that works".

 

Bottom line, you are making it appear as if AI movement and visual aesthetics are mutually exclusive to justify/excuse the poor performance of AI path finding in these areas.Obviously, it greatly depends on what kind of player you are, if you play PvP only, then you couldn't care less, but I am a coop player and I DO care about that, as do all single players. I am not even going to come up with the argument that "I paid 100% and not 85%", since I am a supporter, but the whole "visuals over AI playability" argument simply doesn't hold.It's an argument I would respect for a community map like the Taunus map that said that AI is secondary for them and they prefer visuals (a trade-off they wouldn't have to do if the AI pathfinding was better), but it certainly is not an argument for a commercial expansion to a game - you cannot simply say "X % of the map is inaccessible to AI because we want it to look prettier" instead of fixing the AI issues with navigation. What you are doing is justifying the argument for a community map; there, I would agree with you. For a Bohemia map, I do not.

 

Don't get me wrong. I do appreciate all the work that goes into the game, especially three years after release, and I do appreciate the fact that they are trying hard to make it possible for AI to drive through a town with all roadsigns and fences still intact, and that 85% figure is not scientific anyway, it is probably more than that. It doesn't matter, there is an obvious problem that prevents AI from maneuvering in certain areas of the map, and I don't think it is asking too much to look into this, and I don't think you have to turn Tanoa into an utterly ugly map just to make AI work.

 

EDIT: Let me add something before the usual "haters gonna hate" posts start. I am not saying this is a major issue, or that BIS is trying to starve Coop players. They clearly care about AI or they would not have taken on the driving. I am not suggesting a sinister scheme here. All I am saying is:

- Fixing the pathfinding is something that they should look into.

- There is no major trade-off between look and functionality

- The 85% are very likely exaggerated. No idea where that figure came from.

 

As to how to solve this issue, I think (personally) that a preprocessing of the map before putting it into the game would be a good start. An algorithm can identify the problematic places and insert AI hints. Even if that takes a few hours, it is a one-time process at the end of the development of a map and WELL worth the effort. I know a number of games that did exactly that.

[froggyluv 2135]

 The 85% is not literal as I did not literally walk around and survey 100% of the land :rolleyes:  (Actually if you include water and total landmass -I would guestmate AI can go 98% of everywhere)

 

 And yes, like you I only play SinglePlayer, not as choice, but as necessity of not falling to Beta status in my household where gaming mics are frowned upon. I agree on the Ai crawling under the fence being the really last true, Infantry behavioural advancement and also wish they had gone further in this direction. Almost get the feeling like the 'higher (or highest) up's have put the lock down in this area especially after seeing they can make substantial profit in a game without the headaches of AI - yes, the hardest part of game development.

 

 The fact that you worked in the business doesn't mean you are an expert on Ai development in a sandbox game environment. I have nor ever will be due to bad thinking genes but as a life long gamer at the ripe age of 45, I can tell you that pathfinding with countless AI on a giant map rarely if ever includes all areas -unless of course, the map is just a giant sliding board of rolling hills. So talking about this game being behind industry standard is just disingenuous -if im wrong, kindly point me in the direction of a game in which all house, rock formations and seas are not only accessible, but allow the AI to thrive in a meaningful way. Im not saying this to be an argumentative dick -but as a lifelong fan, follower and hopeful wisher of good AI games.

 

Edit: On further thought, the challenge was to be 5% dick

[haleks 8212]

  kindly point me in the direction of a game in which all house, rock formations and seas are not only accessible, but allow the AI to thrive in a meaningful way.

 

GTA. :ph34r:

But, to be fair, it is the exception rather than a standard in the game industry.

 

EDIT : Crysis deserves an honorable mention too, but I'd say it's about it.

[froggyluv 2135]

 Right but now imagine having up to 100+ AI under your command in GTA, can they still go anywhere? All buildings in a meaningful way? Havent played since IV but do remember the cops having a hard time around certain landscapes -but they did about as good as can be expected. They had their shortcuts as well -could I go and hideout in any building or just the certain ones they left open for me?

 

  Dont remember towns or cities in Crysis but Crysis 3 thought it was just a painted backdrop -meaning all those buildings were not enterable. Personally I just think its alot to ask that all buildings be enterable AND the AI be able to navigate it all -  but hey, if they're taking orders...sure I'd like that as well!!

[haleks 8212]

 Personally I just think its alot to ask that all buildings be enterable AND the AI be able to navigate it all

 

I totally agree on this; the titles I mentionned are exceptions like I said, and they sure have their limits (smaller maps, simpler AI routines, less enterable buildings like you pointed out).

No game has the perfect recipe, but Arma does perform well given the sheer scale of it.

[Alwarren 2767]

And yes, like you I only play SinglePlayer,

I actually hardly play any singleplayer, I mostly play coop.

 

Almost get the feeling like the 'higher (or highest) up's have put the lock down in this area especially after seeing they can make substantial profit in a game without the headaches of AI - yes, the hardest part of game development.

I don't get that impression, if that were the case they could have skipped trying to improve the driving AI. It is more likely that they don't do it because the issue isn't that glaring that it needs immediate attention.

 

And without wanting to offend you, why do you claim that AI is the hardest part of game development, if you have no experience in said field? I am not saying it is easy, but then, we are not even TALKING about AI here - we're merely talking about path finding. The AI part has already decided "I am at point A and I need to be at point B", and now it checks for a way to get there. There is no artificial intelligence involved here. All it has to do is find a path. I am sorry, but while this is not a trivial task, it is a well understood and much researched topic due to the fact that it happens in a million different applications from simulation, navigational software, to strategy games, roleplaying games, and everything inbetween.

 

The fact that you worked in the business doesn't mean you are an expert on Ai development in a sandbox game environment.

LOL, no it doesn't. However, your argument is "just because you worked as a doctor doesn't mean you are an expert in trauma treatment". No, I am not, but I am much more qualified than most that have no experience in programming, no experience in game development, and no computer science background.

 

as a life long gamer at the ripe age of 45,

I am a lifelong gamer at the ripe age of 47. Point being?

 

I can tell you that pathfinding with countless AI on a giant map rarely if ever includes all areas -unless of course, the map is just a giant sliding board of rolling hills. So talking about this game being behind industry standard is just disingenuous -if im wrong, kindly point me in the direction of a game in which all house, rock formations and seas are not only accessible, but allow the AI to thrive in a meaningful way. Im not saying this to be an argumentative dick -but as a lifelong fan, follower and hopeful wisher of good AI games.

This statement alone shows you don't know what I am talking about, and whether there is a precedent case or not is completely irrelevant for the discussion. Any number of games has pathfinding algorithms. Neverwinter Nights 2 for example used a map that was similar to Arma (height field with placeables) and used a seperately generated walk mesh that was created in a post-processing step. In the end it boils down to a simple graph traversal (even a LINEAR one since there is no direction involved, unless you want to make them jump off low cliffs which we don't even need here).

 

So yes, while my expertise is not in AI (which, as I pointed out already, is not even what we are talking about), I am proficient enough to make a statement because I *know* how this stuff is implemented, or could be implemented in Arma. I could implement this in a game engine,

 

The discussion, in the end, is a futile one. It's not going to happen unless Bohemia Interactive decides it is required, and apparently they have not. Development work is never free, and so they might think that the investment in a better path finding is not worth the gain, or they already worked on it for DayZ (I know that there had been work done in that direction in DayZ) and intend to use their Enfusion engine for further development anyway (AFAIK that is their plan).

But please, if I make such a statement, even though I am no longer working in game development (I wish I was, it's more fun than kernel-level OS work), I do know what I am talking about. I base my arguments on facts and experience in programming. You don't.

[froggyluv 2135]

I am not saying it is easy, but then, we are not even TALKING about AI here - we're merely talking about path finding.

 

 

 Ai PAthfinding to be exact. And yes, minus my degree in Computer Science I can safely say this is a subcategory of a successful AI. At the end of the day, a successful perception of potent AI includes many things besides FSM's and that includes AI friendly terrain and other smoke and mirrors ie FEAR. Sometimes common sense takes one further than degrees..

 

Neverwinter Nights 2

 

 

 Oh lord here we go...Sure top down scrollers, 2d platform jumpy games and sprite dominant strategy games do it better...yup, your right. Again, give me a fully flushed out 3d world that is open ended with an intricate landscape in which AI can go anywhere Player can? Yes, precedent and game industry standard DO matter when criticising as such.

 

, I do know what I am talking about. I base my arguments on facts and experience in programming. You don't.

 

 Great! And I base my experience on gaming and being a cunty consumer with common sense - until the expert programmer can show me precedence, well than all those accolades dont account for much. What 3rd party solution has shown to be effective in a world open fully 3d FPS with unlimited AI?

 

I am a lifelong gamer at the ripe age of 47. Point being?

 

 ..that you REALLY need to get a life!

 

:P