AI Driving - Feedback topic

[pierremgi 4887]

Lol, sometimes things appear worse than they are!

 

I replayed the same convoy conditions as the video above:

Convoy leader has speed limited to 30km/h in order to allow the trucks to catch up. // vehicle leader only

Convoy separation is set to 10. // all vehicles but leader

Group is in safe mode the entire time.  // Personally, I often add a driver disableAI "autoCombat", so the gunners can fall into combat mode and the drivers stay on path.

 

It's a mess?

Add just a formation column. Well... Seems far better, no?

 

 

 

[Tankbuster 1746]

That's actually pretty smart, Pierre.

Could I get you and the others with useful info to contribute this to this thread?

https://forums.bohemia.net/forums/topic/216226-ai-convoy-best-practices/

[SuicideKing 233]

iirc the official word is to use a column formation. can't find the post though.

[oukej 2911]

7 hours ago, .kju said:

A3 leaderships essential non investment in AI area suggests they dont intend to use it for Enfusion.

Primary reason for lately low "investments" in AI in A3 has been a way too low cost-benefit ratio (spend days, weeks fixing seemingly simple issue) and relatively high risk connected to code changes (fixes causing issues on remote places).
AI devs have been the first ones to put their hands on Enfusion, building up the AI from the very foundations.

 

[froggyluv 2136]

43 minutes ago, oukej said:

Primary reason for lately low "investments" in AI in A3 has been a way too low cost-benefit ratio (spend days, weeks fixing seemingly simple issue) and relatively high risk connected to code changes (fixes causing issues on remote places).
AI devs have been the first ones to put their hands on Enfusion, building up the AI from the very foundations.

 

 Understandable and literally the best AI news ive heard since microterrain pathfinding

[fn_Quiksilver 1636]

11 hours ago, oukej said:

Primary reason for lately low "investments" in AI in A3 has been a way too low cost-benefit ratio (spend days, weeks fixing seemingly simple issue) and relatively high risk connected to code changes (fixes causing issues on remote places).
AI devs have been the first ones to put their hands on Enfusion, building up the AI from the very foundations.

 

 

nothing unexpected :) i wish the driving bugs were resolved but it makes sense. i never pushed for broad AI improvements/fixes due to the above.

 

hopefully enfusion AI isn't just zombie AI, given its zombie dayz legacy (although i would take zombies if they could drive reliably)

 

i appreciate the candid points regarding cost-benefit, helps put some questions to rest :)

[kremator 1065]

Would be nice to know if Enfusion will be superior regarding AI abilities. 

[xon2 102]

I appreicate the notion that ai devs ''build stuff up from the ground'' for the enfusion engine. Good to hear that this is something that is not abandoned in the future! I generally accept the notion that you have to call it a day at some point and make that call of whether to leave it as it is and invest resources into the future or try to fix something in a soon to be deprecated engine version.

But i think it depends on the degree or severity. I'd be ok with the general driving issues that we see with all vehicles happening now and then. It is what it is, but for most parts it does not totally break the game. But convoys are totally broken and you cannot use them in any fashion that is not totally small distance and dependent on luck. Convoys getting broken by random vehicles simply stopping moving breaking the entire convoy....there must be a way to fix this. This is a very specific issue i reckon that would not necessarily touch the general ai driving routines! If the devs got that fixed i'd rest my case and patiently wait for the next arma installment!

[dragon01 902]

Just a few posts above. Use of the "column" formation for convoys is now mandatory. They aren't completely broken, they just require more setup than they used to. I agree that AI driving isn't in a very good state, but that particular issue has been addressed. In fact, as shown in the video above, the vehicles do exactly what they're told to - if you have them in the default "wedge" formation, they'll drive off the road and try to form a wedge. :) I've had occasional problems with convoys even with proper setup, far cry from nonfunctional, though.

[Tankbuster 1746]

11 hours ago, oukej said:

Primary reason for lately low "investments" in AI in A3 has been a way too low cost-benefit ratio (spend days, weeks fixing seemingly simple issue) and relatively high risk connected to code changes (fixes causing issues on remote places).
AI devs have been the first ones to put their hands on Enfusion, building up the AI from the very foundations.

 

So we are to wait, "potential next Arma game is still years away" playing the game that had an integral part working servicably well a year ago and you (by which I mean BI) guys broke. Note that I accept you probably didn't do this intentionally, it was done by, as you say above, 'high risk connected to code changes ~causes issues on emote place'.

 

Surely you can see that this is unacceptable? You can see how having to see news about 'old man' and related stuff reads to us? Sure, we can bash out single player stuff that will be played by a few dozen people. Can we look after the Co-op community? Nope.

 

I'm not stupid. I know that the people who make the stuff on the 2018 roadmap are not the same people who can fix that driving AI.

 

What I am saying is you should fix it. Do what it takes. If that means delaying Enfusion or Arma 4, so be it. Hire some guys. You're making a noise about a hiring a community manager? Get the people in who can fix the driving AI.

[krycek 349]

@dragon01 Actually they are, they still randomly stop breaking the whole convoy even with safe\column. So it's pretty broken. I had a test on Tanoa from Nicolet to La Rochelle with 3 suvs and the last one stopped in the last town. It's somewhat improved but it still happens, column ain't a guarantee for randomly stopping vehicles.

This is the big problem here, even stuff that worked it's now broken.

 

Oukej, what prevents A4 not to have the same fate? Yeah you have devs concentrating on the AI part now,  maybe those devs will move on and new ones will be hired. New ones will say again  "yeah we don't know what those guys before us did,can't fix it otherwise we might break the whole AI".

[fn_Quiksilver 1636]

49 minutes ago, Tankbuster said:

So we are to wait, "potential next Arma game is still years away" playing the game that had an integral part working servicably well a year ago and you (by which I mean BI) guys broke.

 

its been 2 years this month since the ai driving was broke :)

[xon2 102]

''...relatively high risk connected to code changes (fixes causing issues on remote places).'' But isn't that exactly what dev. branch is for? If you guys could at least investigate why convoy vehicles randomly not start driving at all or stop somewhere along the route, come up with the best fix and put that up on dev branch and leave it there for 2 - 3 months to see whether this breaks other remote stuff, that would be a start. We are not asking to have this fixed like in a week, but some honest effort that after enough testing time might lead to something. I feel thats not too much to ask for.

[dragon01 902]

11 hours ago, krycek said:

Oukej, what prevents A4 not to have the same fate? Yeah you have devs concentrating on the AI part now,  maybe those devs will move on and new ones will be hired. New ones will say again  "yeah we don't know what those guys before us did,can't fix it otherwise we might break the whole AI".

I can tell you that 1). Enfusion is highly modular, so changes in other places shouldn't affect AI (not saying it won't spaghettify by ArmA7, but the architecture is inherently modular). 2). It's being done with a relatively good idea of what vehicles are going to be like (note that OFP didn't have too many vehicle problems, which is when the AI was made). 3). As a science, AI development has considerably advanced since OFP days, so more refined methods can be used.

 

Do remember that there was no revolution in ArmA AI at any point between OFP and now. There was some evolution, but it's running an outdated core made under outdated paradigms. Do note, for example, that there is one vehicle type (VTOL) that wasn't in OFP and was implemented later (in ArmA1, to be precise, with the Harrier). Also note that this vehicle type is one that the AI pretty much can't cope with without very heavy scripting, and even then it's basically a variation on the airplane. Granted, OFP did cover almost all the bases (which is why we're only getting a replacement now), but it's becoming increasingly clear that it just isn't enough today.

[fn_Quiksilver 1636]

4 hours ago, dragon01 said:

I can tell you that 1). Enfusion is highly modular, so changes in other places shouldn't affect AI (not saying it won't spaghettify by ArmA7, but the architecture is inherently modular). 2). It's being done with a relatively good idea of what vehicles are going to be like (note that OFP didn't have too many vehicle problems, which is when the AI was made). 3). As a science, AI development has considerably advanced since OFP days, so more refined methods can be used.

 

Do remember that there was no revolution in ArmA AI at any point between OFP and now. There was some evolution, but it's running an outdated core made under outdated paradigms. Do note, for example, that there is one vehicle type (VTOL) that wasn't in OFP and was implemented later (in ArmA1, to be precise, with the Harrier). Also note that this vehicle type is one that the AI pretty much can't cope with without very heavy scripting, and even then it's basically a variation on the airplane. Granted, OFP did cover almost all the bases (which is why we're only getting a replacement now), but it's becoming increasingly clear that it just isn't enough today.

 

a couple of the things which bothered some of us, was that its been clear there hasn't even been an attempt to address some rather simple AI bugs. a core part of the game (everything non-pvp) neglected almost completely. 

 

for instance, simple QA and testing (or even just devs playing their own title) could have discovered and fixed this bug:

 

 

^ thats not a hard fix, simply adjusting the slope and using the target as Z reference instead of ASL=0

 

but zero fixes can be made if BIS has committed zero resources for such a core part of the game.

 

 

on top of that, its also the least open and least moddable parts of the game. since it is truly abandoned, why was there not some effort to expose the systems for modders to fix? is there valuable intellectual property in the spaghetti? embarrassment? all unknowns and question marks

 

it appears zero effort in that regard as well.

 

note this is just retrospective, not suggesting BIS raise the effort level from zero in these regards

 

[dragon01 902]

5 hours ago, fn_Quiksilver said:

^ thats not a hard fix, simply adjusting the slope and using the target as Z reference instead of ASL=0

Wrong. If this was not a hard fix, it would've been fixed. You're assuming the AI code has been written 100% sensibly and logically, when it's not the case, considering its history. For example, the Z reference is very likely "quietly assumed" by the whole bloody code to be unchanging and permanently set to 0 (just a guess, it's the first such shortcut that came to my mind). Whatever's the reason, it probably made sense in OFP, given the limits processing power and memory back then. Maybe there's a way to fix it, but it is likely hard, will break other stuff and the people most qualified for fixing it have been shunted away to Enfusion a while ago.

 

It's not the first game with legacy AI that I worked with, and hardcoded assumptions were the big limitation there, too. It was difficult to depart too far from the originally intended gameplay without running into them. And that was a relatively simple space shooter.

[fn_Quiksilver 1636]

2 hours ago, dragon01 said:

Wrong. If this was not a hard fix, it would've been fixed.

 

by whom?

 

Quote

 

You're assuming the AI code has been written 100% sensibly and logically, when it's not the case, considering its history. For example, the Z reference is very likely "quietly assumed" by the whole bloody code to be unchanging and permanently set to 0 (just a guess, it's the first such shortcut that came to my mind). Whatever's the reason, it probably made sense in OFP, given the limits processing power and memory back then. Maybe there's a way to fix it, but it is likely hard, will break other stuff and the people most qualified for fixing it have been shunted away to Enfusion a while ago.

 

just fanboi fluff, nothing here to respond to

 

what we do know from the outside looking in, is that there has been very little--close to nil--effort put into either resolving or improving things, and similarly low effort put into exposing these systems for modders to fix/replace themselves. and the timeframe i use is 2012-present (6 years)

[Sparker 185]

You don't just say the magic word 'modular' and expect something to be magically better based on that. For instance, the issue of airplane issuing a turn-away maneuver 300 meters from target instead of 500, is it because it's not modular?

Also I think there is no way to let modders access the AI code since it's written in C, not SQF. Unless they specially implement the AI code as a library, give us its source code and let us connect a different library to the game.

"a way too low cost-benefit ratio (spend days, weeks fixing seemingly simple issue) " - the AI topic clearly affects mission creators like Barbolani. Upset mission creators -> less missions -> less players -> less money, I thought it should look like this from a company's point of view.

 

I think this whole conversation is pretty pointless. Clearly, nobody is going to do anything with AI any more, ever. I have made a negative review on Arma to raise the awareness of other people about this problem. It's not something I'd like to do to my favourite game, but what else can we do?

[chops 111]

The thing I find most infuriating in all this is that AI driving once worked OK. It was stuffed up in a patch and the devs seemed to have shrugged and moved on.

 

It really hinders my enjoyment of the larger-scale command engine/combined arms type of battles that Arma excells at, and is on offer nowhere else.

 

I guess the only point of this thread is to vent our collective spleen at the glaring and noe permanent faults.

Edited June 24, 2018 by chops
Spelling

[dragon01 902]

6 hours ago, Sparker said:

You don't just say the magic word 'modular' and expect something to be magically better based on that. For instance, the issue of airplane issuing a turn-away maneuver 300 meters from target instead of 500, is it because it's not modular?

No, but if the code is modular, there's a chance it is as easy as changing a single value in one place in the code (like the uninitiated often expect it to be). If it isn't, then you're as likely as not to break, say, the landing subroutine. Or, the way things tend to go in these cases, helicopter landing damage. It's not a buzzword, but a paradigm that does tend to make the code easier to expand. Typically, there was little thought given to expandability of a given game engine in the old days - it just had to work right as shipped, and afterwards it was set in stone. For example, in OFP you had the A-10 - and planes that fly more or less like the A-10 don't have this problem. A modular design would have enabled a total rewrite of, plane AI, without worrying it'll break all the other systems in the game (of course, it would've helped if the value wasn't hardcoded in first place, but it wasn't a concern when they were making OFP...). 

 

8 hours ago, fn_Quiksilver said:

just fanboi fluff, nothing here to respond to

 

what we do know from the outside looking in, is that there has been very little--close to nil--effort put into either resolving or improving things, and similarly low effort put into exposing these systems for modders to fix/replace themselves. and the timeframe i use is 2012-present (6 years)

What are your programming credentials again? I might not be a programmer by trade (though what I do does involve programming), but after a few university courses, looking into codebases of two games and extensively modding of them for years (along with a bunch of lesser forays), I did start to see certain patterns and know enough to make an educated guess. I couldn't speak definitely because I'm not a BI employee, but I would be really surprised if they were somehow thinking differently from all other programmers on the planet. Make/mod (as in, total conversion, not add one new gun) a few games yourself, and then you can tell people what is easy and what isn't. People like you keep railing about "easy fixes" and "low effort" while having no idea of how hard these things can actually be. This isn't helping anyone, whether you're talking about a game, a mail program or anything else.

[pierremgi 4887]

 

[xon2 102]

So as the OP suggest, the big ai driving refactoring was implemented via the 'AICarSteeringComponent' system. My question; wasn't there a way to disable this new system and have a car use legacy code for driving?

[oukej 2911]

12 hours ago, xon2 said:

So as the OP suggest, the big ai driving refactoring was implemented via the 'AICarSteeringComponent' system. My question; wasn't there a way to disable this new system and have a car use legacy code for driving?

 

There's also an initiative for you to try configuring the vehicle PIDs. We'll be happy to hear your thoughts or even go ahead implementing your PID settings for indiv. vehicles which would have better behavior than the current vanilla default.
You'd have our big thanks - and from players too!
Moreover you'll be allowed to claim #CommunityFixedItForYaBIS ;)

 

[Grumpy Old Man 3546]

14 minutes ago, oukej said:

 

There's also an initiative for you to try configuring the vehicle PIDs. We'll be happy to hear your thoughts or even go ahead implementing your PID settings for indiv. vehicles which would have better behavior than the current vanilla default.
You'd have our big thanks - and from players too!
Moreover you'll be allowed to claim #CommunityFixedItForYaBIS ;)

 

Thing is, it was working fine before the AISteeringComponent has been introduced.

It was good enough somewhere in the middle of 2017 but got worse since then, without any indication of fixing it/reverting to the previous system.

Same goes for the out of the blue AI skill "refactoring" but let's leave it at that.

 

A2 also didn't have this feature, convoys are working fine in A2, same for vehicles that get a waypoint behind their direction, requiring a 180° turn.

In A3 AI will struggle doing a 3 point turn and simply stop moving.

 

Disabling it using useAISteeringComponent doesn't seem to do anything.

I believe the error to be somewhere inside the AI brains or whatever you seem to call it,

something that has been requested to make available for modifications through sqf scripting but for some reason hasn't happened yet.

 

At this point doing anything with multiple vehicles in a single group is a pain in the rear if it involves that group moving from A to B.

 

It's good that we have the possibility to tune the AI steering component,

but why bother if the devs themselves, who introduced and developed this system can't seem to find a working solution?

 

Cheers

[barbolani 198]

On ‎22‎/‎06‎/‎2018 at 7:03 PM, pierremgi said:

Lol, sometimes things appear worse than they are!

 

I replayed the same convoy conditions as the video above:

Convoy leader has speed limited to 30km/h in order to allow the trucks to catch up. // vehicle leader only

Convoy separation is set to 10. // all vehicles but leader

Group is in safe mode the entire time.  // Personally, I often add a driver disableAI "autoCombat", so the gunners can fall into combat mode and the drivers stay on path.

 

It's a mess?

Add just a formation column. Well... Seems far better, no?

 

 

 

 

I swer I made the same thing you did with the same results tan in the first case....