AI Driving - Feedback topic

[fn_Quiksilver 1636]

On 2/23/2018 at 2:33 AM, krycek said:

  In that case if this ancient engine (which I've agree, there are some legacy bugs going to OFP like the AI switching between binocs/nvgs) that you can't drop a fart around it and AI will collapse, then work on Enfusion should have started much sooner, perhaps before A3 release.

You're forgetting a tiny detail, I've been here since OFP, I'm not mentioning this for some bragging crap but for me the "next one will be better" excuse grows extremely tiring after OFP-A1-A2(OA)-A3. A lack of competition probably caused this engine to live so long.

 

At one point there is a line where you stop, how much understanding for the devs are you capable after you invest lots of your time (and money) in their products. That investment also translates into "ok I trust the devs that the next project will indeed be better". So far all I'm readying from you are excuses and more excuses and how Enfusion is second coming. If that is enough for you, maybe you don't have any expectations about their games and you're happy no matter how good/broken a BI product is.

 

past behavior is best predictor of future.

 

at this point there's no reason to expect AI to be a feature in ArmA 4 at all. THere's also no evidence of BI hiring an AI programmer since 2011, so there's that ...

[Sparker 185]

 

On 21.02.2018 at 10:04 PM, dragon01 said:

You still don't get the point. If the AI is good at the start, it will stay that way. Enfusion is modular, which is the main point here. Do you think they deliberately broke the AI (it's ridiculous if you do, BTW)? ArmA3 has expanded greatly upon OFP, but other than large naval vessels, there really aren't any major things that could be added at this point. Enfusion AI simply has to be designed to correctly handle everything that is in ArmA3, and it'll be mostly good for the rest of the game's lifetime unless they come up with something bizarre. Current ArmA AI, on the other hand, was iterated on with every single RV version since OFP, not to mention as a hardcoded part of the game, there's a chance that even changes in code not directly related to AI can break it. Enfusion mostly avoids that by being modular. That's not to say AI subroutines won't require fiddling with at some point, but it should be a lot more robust system than RV.

How exactly does the magic 'modular' design going to revive the AI? After all, the AI 'module' will have to do exactly the same things. Get data from same kinds of entities, analize what's going on around, and perform the same actions as now. Maybe the new engine's AI will be build not with FSMs but with something else?

Of course if they redo the AI from 0 it might improve things, because, as you said, the core of AI has been there from OFP times.

As I see it, arma's AI is pretty good, apart from some cases when it severely fails.

[oukej 2910]

30 minutes ago, fn_Quiksilver said:

There's also no evidence of BI hiring an AI programmer since 2011, so there's that ...

 

I hope you have checked every single developer that was hired or whole development teams acquired by Bohemia Interactive in the last 7 years to make such a statement.

[.kju 3244]

Not point in arguing semantics - what is true that since OFP to the player there has been no notable improvement to AI (outside micro AI, but even that still has quite some issues/limitations).

To be fair Ondrej (Suma) has done a lot of tweaking and fixing in A2 and OA days for specific AI issues, and A3 has also seen a fair share of fixes and tweaking, along with overall basic to reasonable support of new features by AI.

 

However for long time players the experience has been very disappointing and frustrating to say the least. The core is that they game became more and more complex, especially the landscape the AI has to cope with,

yet the AI was not made able to handle it properly or at least somewhat well (path finding or movement for infantry, cars, even tanks get stuck all the time or fall over from physx interaction, convoy capability has become significantly worse and worse, AI engagement and combat behavior (reaction time, engagement ranges, different type of weapon use, tactics, proper vehicle use of different types) to name some of the prime areas),

yet also most areas have almost no improvement or none at all (AI formations, AI commanding, high level AI behavior (AI info share, combined arms, etc), acquire equipment, rearm, resupply, repair,heal and revive behavior, etc).

 

In the end the fact of the matter is that many areas to the player are the same or worse than OFP - also after significant performance improvements at the end of A2 and OA patches each for AI (and player) numbers, A3 has been again a considerable step backwards sadly.

 

OFP/early RV engine was extremely remarkable for all its capabilities at the time - unfortunately the code quality and complexity surely have taken their toll to get this all done with the very limited manpower and time.

I would assume also one of the main reasons why it has not seen much significant improvement since.

 

Anyhow Carrier Command seemed to have a decent AI on different levels, cant make any judgement about Take on Mars in this regards - Enforce, and thus Enfusion, in general seems to have the capability to bring progress again.

Of course only the future will tell.

 

We all hope and pray BI leadership sees the priority, allocates the manpower, acquires the talent and skill this area deserves.

 

PS: The community has done a lot very impressive work at least like DAC, GL4, ASR, bCombat, VCOM, AISS, C2, CE, ALiVE, SLX, TPW mods, HETMAN, RUG DSAI to name a few, but much more :

 

 

[fn_Quiksilver 1636]

there is still 2 AI driving bugs outstanding for a long time now. Bridge crossings + vehicle stops and wont continue along its route, for no apparent reason

 

no amount of words or friendly banter/argument changes that ^

 

 

also are we deleting my posts again? thats not nice :(

[.kju 3244]

Yeah ignoring fundamental AI pathfinding issues, by deleting posts from "critics", wont help the overall impression people get from the game.

[Tankbuster 1746]

6 hours ago, fn_Quiksilver said:

there is still 2 AI driving bugs outstanding for a long time now. Bridge crossings + vehicle stops and wont continue along its route, for no apparent reason

 

no amount of words or friendly banter/argument changes that ^

 

 

also are we deleting my posts again? thats not nice :(

I laughed at your division by zero post because I read it in the topic notification email. Sadly, that post was deleted too.

[mad_cheese 593]

To add to @.kju's report - it is reproducable, but I for one am not clear where right or wrong is here. 'Wedge' is the standard formation in A3. Giving the group or the waypoint a 'Column' formation fixes this issue and the vehicles drive on the road nicely. I am not sure at all but i guess that allows for flexibility and is actually not all bad? Should the group auto-detect that it's only built from vehicles that are arranged a certain way, label that as a convoy and then automatically set it as a column? Really wondering why stable produces the 'correct' result in the first place. Not trying to talk smack here Kju, really trying to wrap my head around this.

 

Anyways, I for one am pretty damn impressed with the improvement of TRACKED vehicle pathfinding. 

 

 

Sorry for this test not being vanilla but since I am reporting a state of zen I figure that's acceptable. The Linebacker usually creates mayhem and does not manage to follow the convoy. Not anymore, I love it.             

 

It may not be perfect, both of my unsophisticated tests ended with some sort of issues eventually. But MAN, I can see the work that was put in and would like to encourage whoever is working on that to keep up the great work. There, I said it.

 

 

 

[.kju 3244]

@mad_cheese thanks for another confirmation. Alwarren also found out that formation type matters

However in this stage of the lifetime of a game this is nothing you change as default behavior as you break way too many missions.

 

Also its probably not a reasonable default for convoy behavior. Even less for non tanks.

 

If this should be made available, there has to be a new scripting command to expose functionality like this

(like "_group respectFormationInConvoy true;" or "_group enableConvoyBehavior false;").

[fn_Quiksilver 1636]

2 hours ago, .kju said:

@mad_cheese thanks for another confirmation. Alwarren also found out that formation type matters

However in this stage of the lifetime of a game this is nothing you change as default behavior as you break way too many missions.

 

Also its probably not a reasonable default for convoy behavior. Even less for non tanks.

 

If this should be made available, there has to be a new scripting command to expose functionality like this

(like "_group respectFormationInConvoy true;" or "_group enableConvoyBehavior false;").

 

wait, are you saying "column" formation fixes some AI driving issues?

 

my expectations are fairly modest, I only want 2 bugs fixed (given that dev time since 2016 is extremely limited), these two:

 

 

 

 

I should probably just go away anyways, considering my criticism isnt wanted, but if those two bugs were resolved, I wouldnt have any real constructive criticism left for A3 (given limited dev resources).

 

 

[.kju 3244]

no clue about "some" - it "helps" with the newly introduced issue Ive posted above

[Alwarren 2767]

3 hours ago, fn_Quiksilver said:

wait, are you saying "column" formation fixes some AI driving issues?

 

The mission that .kju posted had the formation set to wedge. That didn't pose a problem before, but with the dev branch the convoy actually respects the settings and the vehicles go offroad.

 

As a general rule of thumb, I would always set convoys to "Safe" and "Column".

[.kju 3244]

Note: I didnt change anything in the mission (on purpose). So its default behavior of units.

 

My own missions have also safe/careless.

[oukej 2910]

1 hour ago, Alwarren said:

5 hours ago, fn_Quiksilver said:

wait, are you saying "column" formation fixes some AI driving issues?

 

 

The mission that .kju posted had the formation set to wedge. That didn't pose a problem before, but with the dev branch the convoy actually respects the settings and the vehicles go offroad.

 

As a general rule of thumb, I would always set convoys to "Safe" and "Column".

+1

If you use "Column" (or other formation without left/right offsets) then a convoy PID will be governing the group movements (It was introduced along with the main path-following and speed PIDs - first for wheeled, more recently for tracked). So exactly what Alwarren said I'd put as an "official" recommended way how to do convoys.

Then you have setConvoySeparation for more control over the convoy looks without having to touch formation configs.

[oukej 2910]

5 hours ago, fn_Quiksilver said:

my expectations are fairly modest, I only want 2 bugs fixed (given that dev time since 2016 is extremely limited), these twoo

2

Both issues seem like a data problem - first with a map (road paths, bridge paths, connections...). We've recently changed the offsets in roadlib, that should help. Second one looks like a flaw in the vehicle's PID (AIBrain, AIDriving in the diag .exe can reveal more why it's moving so funky)
 

[mad_cheese 593]

1 hour ago, oukej said:

+1

If you use "Column" (or other formation without left/right offsets) then a convoy PID will be governing the group movements 

Well there we go! That works for me (personally)

 

55 minutes ago, oukej said:

Both issues seem like a data problem - first with a map (road paths, bridge paths, connections...). We've recently changed the offsets in roadlib, that should help. Second one looks like a flaw in the vehicle's PID (AIBrain, AIDriving in the diag .exe can reveal more why it's moving so funky)
 

Anything particular we can gather to help? Especially in regards to bridges :D

[.kju 3244]

Quote

However in this stage of the lifetime of a game this is nothing you change as default behavior as you break way too many missions.

 

[fn_Quiksilver 1636]

4 hours ago, oukej said:

Both issues seem like a data problem - first with a map (road paths, bridge paths, connections...). We've recently changed the offsets in roadlib, that should help. Second one looks like a flaw in the vehicle's PID (AIBrain, AIDriving in the diag .exe can reveal more why it's moving so funky)
 

 

thanks for the intel

 

i was on diag yesterday having a look. what is the Blue dot in the AIDriving diag?

[pr9inichek 133]

@oukej Hi, big thanks for you work!!!
Can you please check this post - 

We VERY need this commands for diag.

 

disableRendering 
diag_drawmode "NORMAL", "GEOMETRY", "VIEWGEOMETRY", "FIREGEOMETRY", "PATHS", "ROADWAY"  

 

And "AIWaypoints" or something for check brain of pilots, as example Hellcat pilot

 

Some errors:

Spoiler

 

 

 

 

 

 

 

 

 

 

 

[kremator 1065]

^^Bridge vs AI 10:1 is so soul crushingly sad !

[Imperator[TFD] 444]

On 2/26/2018 at 4:24 AM, Alwarren said:

As a general rule of thumb, I would always set convoys to "Safe" and "Column".

 

I seriously thought that this was common knowledge and practice?

[.kju 3244]

@Imperator[TFD]

the default behavior changed a few dev branch updates ago. see report and videos a few posts back

[The Man Without Qualities 110]

Well, setting safe and column is a WORKAROUND used for a long time.

Nevertheless I expect that a developer - pretty much aware of the obstacles his beloved AI facing - is providing routines/methods alowing the AI to master such obstacles IN ANY CASE.

When AI groups passing narrow places they leave the formation and regroup when possible.

If a bridge cannot be passed with 10 vehicles side-by-side I expect a routine that detects a bridge in the calculated path and is adapting formatin settings etcpp.

BIS had 10+ years to think about a solution. It is time to deliver.

[kremator 1065]

^^ come on now.  Everyone knows that tanks mortal enemy is the 'lean-to' or the awning!  Normandy landings wouldn't have quite so well if the beaches were scattered with lean-to or awnings!  They are well known to throw 20 ton vehicles right up in to the air - well known fact that! 

 

DEVS - So which part of the PhysX simulation is causing this ?  Is this something that we can help with - as Tanks DLC is coming up and although fancy interiors are lovely, if a tank cannot get from A to B using AI, or is catapulted into the air by a 'insert small object here' then it doesn't bode well for the final DLC.

[mickeymen 324]

On 05.03.2018 at 2:41 PM, kremator said:

DEVS - So which part of the PhysX simulation is causing this ?  Is this something that we can help with - as Tanks DLC is coming up and although fancy interiors are lovely, if a tank cannot get from A to B using AI, or is catapulted into the air by a 'insert small object here' then it doesn't bode well for the final DLC.

 

kremator, I agree with you.

 

Also as seems to me, In order to significantly improve the problem of AI driving and destroy AI-drivers stuck or jumping tanks into the sky,  BIS do not need to create tons of scripts,  only need to make a one thing - to create better destructibility of most objects in the map! 

 

Such fragile objects as tents, old barns, other dilapidated buildings, small stone fences, etc. must be destroyed in front of armored vehicle. 

Even a civilian car must destroy some dilapidated buildings, tents.

In the Arma3, player can see that large trees are being destroyed under the tank.

Then a reasonable question arises - why the old barn or awning, can not be destroyed?

I'm sure, the better destructibility of Arma3 allows to mitigate the problem of AI driving by about 50%

BIS, please consider this. This can smooth the problems of the AI drivers