AI Driving - Feedback topic

[Grumpy Old Man 3545]

7 hours ago, barbolani said:

 

@Grumpy Old Man, you havent done @pierre MGI tricks about grouping the vehicles in that vid. Pierre's trick worked for me in both editor and script spawned, I suggest you to try it.

 

I'm aware of it.

 

The whole point is, you shouldn't need to jump through hoops just to get a convoy moving,

since your average mission maker that just wants to throw a quick mission together most likely doesn't want to have to learn advanced scripting for a simple convoy,

which has been working fine at least early 2017.

 

In A2 it's as simple as placing the vehicles in a group and set them a waypoint.

It just works.

 

When only one vehicle moves and the rest gets stuck, it's a sign that there's clearly something faulty going on in the AI.

 

Cheers

 

 

[dragon01 902]

Well, now it's not so simple. That said, the steps involved should be pretty obvious (though I suppose BI could explain it in a tutorial, I suggested adding a few more built-in ones). In effect, you need to actually tell the AI to form a convoy and make sure it's in correct order. Don't expect a "thrown together" mission to be of high quality, to get good results you need to put in actual effort. In particular, anything involving the AI requires testing, and it always did. If it wasn't finicky before they wouldn't have had overhauled it in first place.

 

Also, in ArmA2 convoys worked poorly at best. Maybe they did work in a more straightforward manner, but I occasionally had problems with them (and with single vehicles, as well) even in vanilla missions. They got stuck, lost wheels on obstacles and had all manners of problems as well.

[B_Fox 132]

The only time i noticed convoys really working was in OFP/Arma Cold War Assault, but that being said why not just make a customization convoy way point? The user tells the way point how many vehicle are in the convoy, how fast the convoy should move if the convoy should stop if under fire and if so which vehicle should be excluded from the stop command.

[pierremgi 4850]

6 hours ago, B_Fox said:

The only time i noticed convoys really working was in OFP/Arma Cold War Assault, but that being said why not just make a customization convoy way point? The user tells the way point how many vehicle are in the convoy, how fast the convoy should move if the convoy should stop if under fire and if so which vehicle should be excluded from the stop command.

How long does it take to:

- group the vehicles in good order (one by one)

- set the group to safe (not mandatory) and column, low speed (from group icon)

- add a limitSpeed in vehicle leader

- add a setConvoySeparation in other init fields of vehicles?

Less than two minutes.

That said, you're right, why not an advanced waypoint (like fire mission or land)? Imho, the real added value could be the behavior of a convoy under fire. Actually, the loss of the leader is not  too bad if you add a code refreshing the separation for remaining vehicles able to move. So far, there is nothing managing the disabled but alive vehicle.

[dragon01 902]

A new waypoint type would be useful, but might be though to code. I'd say we should focus on how to improve what we have now instead of adding new functionality.

[madrussian 347]

From my perspective, two simple and well-defined issues account for easily 90% of the problem.

 

The worst offender right now is vehicles given a WP directly behind pretty much always get stuck.  That one should be easy to fix, and would have an enormous impact.

 

Fix that and the convoy thing and you'll have a lot of happy customers.

 

[kremator 1065]

But will they listen? 

[Grumpy Old Man 3545]

13 hours ago, kremator said:

But will they listen? 

The correct question is: What will they do about it?

Seeing how this is known for at least one year with little to no effort to work these things out, I doubt A3 will ever get out of its current state.

 

Cheers

[The Man Without Qualities 110]

On 15.7.2018 at 7:08 AM, pierremgi said:

How long does it take to:

- group the vehicles in good order (one by one)

- set the group to safe (not mandatory) and column, low speed (from group icon)

- add a limitSpeed in vehicle leader

- add a setConvoySeparation in other init fields of vehicles?

Less than two minutes.

That said, you're right, why not an advanced waypoint (like fire mission or land)? Imho, the real added value could be the behavior of a convoy under fire. Actually, the loss of the leader is not  too bad if you add a code refreshing the separation for remaining vehicles able to move. So far, there is nothing managing the disabled but alive vehicle.

 

 

You ever thought about how that should work in a "user friendly" MP Mission? You do that eighter with one config set valid for all type of convoys or you create a telling GUI that enables the random user to select the convoy (or even to rearrange / regroup) and to set space/speed on the fly.

 

Don't make the mistake assuming that the random gamer is a game developer or even just someone with some basic coding knowledge.

 

They wanna play - that's it.

[dragon01 902]

1 hour ago, The Man Without Qualities said:

They wanna play - that's it.

So they should stick to playing. Mission making is a significantly more advanced activity than just playing the game. Sure, some old missions got broken, but there are others that work fine.

[.kju 3244]

Again its bad design to have bad default parameters/setup.

At least BI should provide a custom waypoint as suggested to contains all the relevant parameters to customize (and with decent descriptions to the parameters).

[The Man Without Qualities 110]

1 hour ago, dragon01 said:

So they should stick to playing. Mission making is a significantly more advanced activity than just playing the game. Sure, some old missions got broken, but there are others that work fine.

 

Maybe you missed the point: As cure for BIS failure to provide a working AI in terms of vehicle driving, some of the forum members suggested to do some manual workaround by fiddling around with unit behavior parameter as well as with convoy setup parameter.

My point is that those WAs are not accessible for the casual gamer.

Your point that some missions are broken but others work so...fine? ...is void IMHO since this discussion IS ABOUT the fact that many beloved missions ARE broken as soon as working convoys are necessary.

Your point that non-devs should stick to playing is void...they can't without problems if flawless working convoys are necessary.

 

May you rephrase what you wanted to say? Maybe I did not understood the intention of your comments.

[Grumpy Old Man 3545]

2 hours ago, dragon01 said:

So they should stick to playing. Mission making is a significantly more advanced activity than just playing the game. Sure, some old missions got broken, but there are others that work fine.

 

Mission making in arma 3 is, for the most part, as basic as placing a few units, modules and triggers.

Something most folks who grew up around PCs can pick up over 1-2 weekends.

Adding voice acting, cutscenes and some animations isn't any more complicated either.

 

If you talk about developing new game modes with custom score system, logistics or resource handling and dynamic missions for a persistent server, then yes, that's more advanced, but nothing the average joe is going to take on.

 

Placing a convoy that simply works should be a baseline feature of a game that is mainly a military sandbox with AI.

AI that somehow fails to do so for the first time in this series and for some reason is being left as is by the devs.

 

It's a pity, considering how stunning the game can look.

Looking at screenshots made by certain folks I still need to look twice to find out it's actually a game and not some RL military footage.

 

You keep mentioning missions that are still working fine in the context of convoys.

Care to name a few? I'd like to take a look at those.

 

Cheers

[pierremgi 4850]

Calm chaps! My intent wasn't to argue with the possibility or not to rewrite the "beloved" old missions including a convoy.

I thought I was on a "AI driving feedback topic" and it was decent there to add my contrib for some little "workaround" to make a convoy, let's say "not too bad". I thought it was at least as useful as criticism.

Now I admit Bi can do something better and compatible with old scripts. (I doubt an advanced waypoint will satisfy all people).

-  For all players, It's a common sense to make it compatible. But compatible with what? Was there one day with something working ? At what time was the most decent behavior for convoy?

-  For a mission maker able to place a convoy and some triggers, it's not so difficult to correct his own mission, or take these parameters (as shown) into account for a future one,... waiting for BI improvements.

 

I think the setConvoySeperation (and older limitSpeed) was a great improvement. But incomplete due to the raising difficulty to take away the disabled alive vehicles.

 

If I'm right, the main problem seems the difficulty to order the convoy to move in column. That's easy to think about it while in editor now, but this point seems to break some old missions as far as the column was automatic? with safe mode (if I'm right). So this "new" behavior (the formation stays wedge by default) breaks the old convoy.

Making some column automatic in safe mode, just for convoy is a risky choice. If done, you could have some mission makers with tanks on large places as desert or plains who could complaint for stupid column instead of formation line. So, imho, the formation behavior must be independent from the combat mode. Just an opinion.

 

ARMA is so open to scripting and mission making and mods, that every choice supposed to be an improvement for most of the cases, can become a nightmare for plenty of other cases. How many players? free non-vanilla missions? How many official patches/DLC? (see the growing buildings, weapons, units... classes, with weird inheritances sometimes), how many commands/functions (with or without decent documentation but tested out by thousands people)?

The solution will come, perhaps, with ARMA IV reset. ARMA III is near to its own engine limits and complication, right?

[xon2 102]

@pierremgi I have done it your way yesterday....limitspeed, convoyseperation, safe, column, very few waypoints across half the map. It looked sure nice in the beginning and you think, they are going so slow and orderly...this time they gonna make it! But no, same issue as always, at some point some random vehicle just stops and breaks the convoy. And if those standard script commands would actually deliver, then BI would have fixed this already by changing the default convoy paramters to that working set of values. BI did not do so because that does not definitively solve the problem of broken convoys. And over all the months i have been following this issue in this thread, not once did a dev directly respond to the 'random convoy break-up issue'. Oukej wen't on about the PID values which are in all likelyhood not the cause of this convoy trouble. I have seen the same behaviour with aisteeringcomponet off. And that no dev is even aknowledging the convoy issue means in all likelyhood that it has been decided internally to leave it be. Personally, i am not interested in that third party content that is supposedly coming our way. I paid for a alot of assets spread over many dlcs and want to play the game that arma 3 promised to be...but i can't because all those driving issues make it very difficult to actually employ all the assets we already got. Why have more vehicles if i can't use them in a combined arms context cause they drive around bonkus all the time!

[pierremgi 4850]

Yep. You're right. Ai driving is not at its best and that breaks the convoy behavior, especially if you add Tank in tiny urban curves (and many other circumstances like rocks in Malden, Kunduz, forest in Tanoa...) Anyway, I'm not sure "you paid for a lot of assets" and some Arma3 Graal promised to you. I'm as bother like you with Arma limitations and weird commands unable to do the job. As I wrote last year, a command/function should "command", I meant "override the Arma's engine", for many scripts like AI choosing a launcher AND keep it in order to fire with, switch vehicles light on, weapon laser on... and no matter what the fsm/engine could be with the combat mode or the behavior. Not mentioning  some other time killing commands, never doing what you expect (or just during the job for one frame!).

On the other hand, the AIs vehicles traffic is not totally perfect in many other games. You can find a very good traffic in Ghost Recon but if the first vehicle is stopped, a long stupid queue is guaranteed.

I'm like you and many others: waiting for a better AI driving. Imho, It's less a steering problem (even if wrong parameters don't help at all) than an obstacle detection with an avoidance algorithm. And that... is not for tomorrow.

[x3kj 1247]

I agree with pierremgi, blaming assets is not fair (because with current DLC model, which imo is the most consumer friendly in the industry, assets are the only reason to buy a DLC - which also support development)

 

Quote

As I wrote last year, a command/function should "command", I meant "override the Arma's engine", for many scripts like AI choosing a launcher AND keep it in order to fire with, switch vehicles light on, weapon laser on... and no matter what the fsm/engine could be with the combat mode or the behavior. Not mentioning  some other time killing commands, never doing what you expect (or just during the job for one frame!).

Thats one of the most core issues with this hole thing imo. Nobody other than devs can do something about the problem, because it's all inaccessible and there is no way to override driving behaviours efficiently with custom logic. The script commands (like doMove etc) are insufficient for dealing with this. I'm expecting/hoping for improvements in future games in regards for AI driving. But i'm also hoping of improvements in accessability of that logic from the outside. Eventhandlers for stuff related to driving routines for example - detected obstacle, or detected crash, change of move order, etc .

And also for 2 distinct ways of exerting controll over AI externally in scripts - "by command" (like it is now  with commands "fire at thing X", "move to pos Y") and "direct" (which forcibly interrupts the logic chain in order to be able to do a specific thing exactly and precisely by simulating player inputs. Like pressing the key to move forward for an exact amount of time. ).

 

Right now we have no direct control over AI driving actions (increase throttle, stop, steer left/right etc). It can be done somewhat on command level - which breaks in certain circumstances. E.g. you can order tanks via https://community.bistudio.com/wiki/sendSimpleCommand but such commands are only accepted in special conditions - namely when AI subordinate follows the owning player. Not when it is told to stop or is told to move somewhere by the player.  I have been running into this problem when trying to get AI to turn into a specific direction in my attempt at a custom artillery script (https://www.youtube.com/watch?v=uKGzWkKDwdY).

 

[mickeymen 324]

It seems that the football gate is trapping the tank forever and this can break the idea of any singleplayer mission.

 

 

Possible types of solutions:

 

1) when the AI-driver encountering an obstacle, teach the game to create a trigger for such moments and force AI-drivers do retreat maneuvers (for example  back to 10 meters and rotate 90 degrees in either direction and again move 10 meters)

2) make the football gate to be destroyed (falling to the ground like stone fences)

 

Maybe there are other solutions, but to my great regret the attempts to change this situation are not interesting to the developers

but maybe some mod makers can create this...

 

In order to reduce the problems of AI driving in ArmA3 it is enough to make the environment is more destructable

More than half of the problems for AI drivers in ArmA3 can be neutralized with the help of a destructible environment,  the tank driver must break 80-90% of his obstacles and then the chance of a stuck will be negligible

[The Man Without Qualities 110]

1 hour ago, mickeymen said:

It seems that the football gate is trapping the tank forever and this can break the idea of any singleplayer mission.

 

Possible types of solutions:

 

1) when the AI-driver encountering an obstacle, teach the game to create a trigger for such moments and force AI-drivers do retreat maneuvers (for example  back to 10 meters and rotate 90 degrees in either direction and again move 10 meters)

2) make the football gate to be destroyed (falling to the ground like stone fences)

 

Maybe there are other solutions, but to my great regret the attempts to change this situation are not interesting to the developers

but maybe some mod makers can create this...

 

In order to reduce the problems of AI driving in ArmA3 it is enough to make the environment is more destructable

More than half of the problems for AI drivers in ArmA3 can be neutralized with the help of a destructible environment,  the tank driver must break 80-90% of his obstacles and then the chance of a stuck will be negligible

 

For 1.): A kind of watchdog is necessary for every moving unit: "if predicted moving distance per time unit was much bigger then real distance or real distance was zero THEN use reverse gear for xxx meter same path you came and try an alternative path"

 

For 2.) When I browse through the entire object library of A3 it is just 2-3 weeks of focussed work to go through and review all parameters and check collision behavior with 2-5 types of mobile units.

 

Example:

 

a.) object "bush"; human body should clip through or trigger "grass flat" like animation, car and tank should do same

b.) object "traffic light pole"; human body should rebounce, maybe trigger voice sample "ouch!", car should receive damage, but pole too, depending on mass vs. pole resistance relation, tank should simply make it flat.

 

 

[.kju 3244]

its a design problem - by default any object should be possible to drive over (with tanks at least)

however since A1 I think each model needs to be set up properly to be able to do so - too many are not

[pierremgi 4850]

4 hours ago, .kju said:

its a design problem - by default any object should be possible to drive over (with tanks at least)

however since A1 I think each model needs to be set up properly to be able to do so - too many are not

 

Sure! one more good reason for expecting ARMA IV...

At least, it could be fine to "delete" any contact/physx when an object is hidden or disabled sim...

I tried to do something with the  @mickeymen , while tank tusk in goal, I failed to delete these contacts, even hiding the goal:

 

this addEventHandler ["epeContactStart", {
  _tank = _this select 0;
   _obj = _this select 1;
hint str _this;
       if (isNull _obj) then {
      _obj = (nearestTerrainObjects [getpos (_this select 0) ,[],7,true,true]) select 0;
_tank setpos ((_this select 0) getpos [-0.2,getdir (_this select 0)]); // for test (that's OK to make the tank unstuck)

  _obj enableSimulationGlobal false; // works but the contacts remains
  _obj setDamage 1; // works (damage is set to 1) but totally useless, no effect as there is no different model

  deleteVehicle _obj;  // doesn't work
          };
        }];  

[Varis 34]

On 7/18/2018 at 12:50 PM, Grumpy Old Man said:

The correct question is: What will they do about it?

Seeing how this is known for at least one year with little to no effort to work these things out, I doubt A3 will ever get out of its current state.

 

Cheers

 

I would not read too much into it. For starters it might not make sense to spend just a little effort if you cannot fix the issue. A dev team could address such issues when it fits their schedule and there are no higher priority items in the backlog. With decreasing effort on ARMA3 such fixes will become less likely - don't think they'd be any beef for 3rd party teams.

 

On 8/8/2018 at 3:59 PM, mickeymen said:

 

1) when the AI-driver encountering an obstacle, teach the game to create a trigger for such moments and force AI-drivers do retreat maneuvers (for example  back to 10 meters and rotate 90 degrees in either direction and again move 10 meters)

2) make the football gate to be destroyed (falling to the ground like stone fences)

 

 

1) Best algorithms would probably be more complicated, at least say iterate in .5m increments or something. But if there was a trigger created and the mod script could catch that, it could be a way to fix issues in specific circumstances.

2) Destroyed or able to be driven through/over. We have very similar issues in infantry movement - any KotH infantry player will bump into it when moving between ground/building floor, through doorways, etc. There is probably a number of reasons why destructible environments are not ubiquitous in today's games (we started to have them already some 10+ years back), just updating all models could be tons of work.

 

AI driving problems I encountered just recently while playing BECTI. It's amazing how the AI cannot even follow a vehicle on a seemingly easy, almost straight road. There are at least two sides to the problem - first is the AI developing a problem in the first place and the second is how to reliably recover from it (I guess time used is less of a concern). When the AI drives its nose into an obstacle (often small rocks etc - should they be destructible?) it seems to get almost totally paralyzed. Simple backing up theoretically might make it recover. Sometimes it just wants to drive over impossible rocks (and that's already possible in ARMA3...).

 

On 7/19/2018 at 5:36 AM, pierremgi said:

As I wrote last year, a command/function should "command", I meant "override the Arma's engine", for many scripts like AI choosing a launcher AND keep it in order to fire with, switch vehicles light on, weapon laser on... and no matter what the fsm/engine could be with the combat mode or the behavior. Not mentioning  some other time killing commands, never doing what you expect (or just during the job for one frame!).

On the other hand, the AIs vehicles traffic is not totally perfect in many other games. You can find a very good traffic in Ghost Recon but if the first vehicle is stopped, a long stupid queue is guaranteed.

 

Game engines seem quite hairy and fixing all that would probably be a humongous amount of work. It's a small miracle ARMA can be modded to the current extent without things breaking left and right (and speaks volumes about good engine design).

[Varis 34]

On 7/18/2018 at 2:08 PM, The Man Without Qualities said:

You ever thought about how that should work in a "user friendly" MP Mission? You do that eighter with one config set valid for all type of convoys or you create a telling GUI that enables the random user to select the convoy (or even to rearrange / regroup) and to set space/speed on the fly.

 

Don't make the mistake assuming that the random gamer is a game developer or even just someone with some basic coding knowledge.

 

They wanna play - that's it.

 

Yes I guess in say the setting of BECTI the mod developers would have to first figure out the workaround interface and script convoys back to shape, THEN expose those functions in the custom player menu so the players can set valid parameters that work just for their type of convoy. Mind you this is not something you would expect from your average ARMA player but players of BECTI already spend time building and arranging their tiny private task force (and driving AI is kinda central to how you can utilize the available assets). In a milsim context I'd imagine that to be the group leader or convoy master, and then zeus... All in all, possible but quite a bit of work from a number of people involved. BI could even provide several ways of implementing a workaround and the relevant parts of the player community can then figure out what works the best just for their circumstances. We'll have to just go with what we get but of course it would be best to fix as deep as possible, while learning valuable lessons for ARMA4 engine design. (Exposing more engine internals kinda has a tasty ring to it, BI could review what can be achieved with such an approach. Performance might be a consideration for any large scale adoption of deep APIs though.) 

 

On 8/8/2018 at 5:30 PM, The Man Without Qualities said:

 

For 1.): A kind of watchdog is necessary for every moving unit: "if predicted moving distance per time unit was much bigger then real distance or real distance was zero THEN use reverse gear for xxx meter same path you came and try an alternative path"

 

 

I would be very surprised if there wasn't already such a watchdog or a more complex mechanism to address the same problems! Almost 30 years back we had this little game (named OMEGA or something like that) where the gameplay basically consisted of creating algorithms (code) for your robot tank to avoid obstacles and to engage enemies. Access to an OBSTRUCTED flag was one of the key elements.

[Tankbuster 1744]

On 08/08/2018 at 8:21 PM, pierremgi said:

 

Sure! one more good reason for expecting ARMA IV...

At least, it could be fine to "delete" any contact/physx when an object is hidden or disabled sim...

I tried to do something with the  @mickeymen , while tank tusk in goal, I failed to delete these contacts, even hiding the goal:

 

this addEventHandler ["epeContactStart", {
  _tank = _this select 0;
   _obj = _this select 1;
hint str _this;
       if (isNull _obj) then {
      _obj = (nearestTerrainObjects [getpos (_this select 0) ,[],7,true,true]) select 0;
_tank setpos ((_this select 0) getpos [-0.2,getdir (_this select 0)]); // for test (that's OK to make the tank unstuck)

  _obj enableSimulationGlobal false; // works but the contacts remains
  _obj setDamage 1; // works (damage is set to 1) but totally useless, no effect as there is no different model

  deleteVehicle _obj;  // doesn't work
          };
        }];  

From a script I posted a few days ago; I don't like, trust or use the Physxeventhandlers.

 

{
	[_x] spawn
		{// spawned code to try to unstick stuck vehicles
		params ["_q"];
		while {(alive (driver _q)) and missionactive} do
			{
			private ["_pushdir"];
			private _vq = vehicle _q;
			private _pvq = getpos _vq;
			private _psq = getPosASL _vq;
			sleep 5;
			if ((abs(speed _vq) < 2) and { (alive _q) and (canMove _vq) and ((fuel _vq) > 0) and ((_vq distance2D _pvq) < 8) }) then
				{// vehicle is stuck
				if ((lineintersectssurfaces [_vq modeltoworldworld [0,0,0.2], _vq modeltoworldworld [0,8,0.2], _vq]) isEqualTo []) then
					{//push it forwards a little
					_pushdir = 10;
					}
					else
					{// if there's something in front, push backwards, not forwards
					_pushdir = -10;
					};
				_vq setVelocityModelSpace [0,_pushdir,0];
				diag_log format ["*** pushing %1 a little", name driver _q];
				};
			};
		};

} foreach _sorteddrivers;

This has, in more than half of the observed cases, been enough to 'rescue' a vehicle that has got stuck on an object, stopped for no apparent reason or stopped on a bridge where missing roadsconnectedto has broken the pathing. I'm sure there some more to be done, but I'm happy with results so far. Note that the above code is taken from my mission and would need adapting to work elsewhere.

[fn_Quiksilver 1636]

19 minutes ago, Tankbuster said:

From a script I posted a few days ago; I don't like, trust or use the Physxeventhandlers.

 

{
	[_x] spawn
		{// spawned code to try to unstick stuck vehicles
		params ["_q"];
		while {(alive (driver _q)) and missionactive} do
			{
			private ["_pushdir"];
			private _vq = vehicle _q;
			private _pvq = getpos _vq;
			private _psq = getPosASL _vq;
			sleep 5;
			if ((abs(speed _vq) < 2) and { (alive _q) and (canMove _vq) and ((fuel _vq) > 0) and ((_vq distance2D _pvq) < 8) }) then
				{// vehicle is stuck
				if ((lineintersectssurfaces [_vq modeltoworldworld [0,0,0.2], _vq modeltoworldworld [0,8,0.2], _vq]) isEqualTo []) then
					{//push it forwards a little
					_pushdir = 10;
					}
					else
					{// if there's something in front, push backwards, not forwards
					_pushdir = -10;
					};
				_vq setVelocityModelSpace [0,_pushdir,0];
				diag_log format ["*** pushing %1 a little", name driver _q];
				};
			};
		};

} foreach _sorteddrivers;

This has, in more than half of the observed cases, been enough to 'rescue' a vehicle that has got stuck on an object, stopped for no apparent reason or stopped on a bridge where missing roadsconnectedto has broken the pathing. I'm sure there some more to be done, but I'm happy with results so far. Note that the above code is taken from my mission and would need adapting to work elsewhere.

 

this is what i use for vehicle "rescue"

 

https://github.com/auQuiksilver/Apex-Framework/blob/master/Apex_framework.terrain/code/functions/fn_AIXVehicleUnstuck.sqf#L50-L79

 

just counting the time that a vehicle has been stationary for. if the time goes above a threshold, the vehicle is put to a nearby clear position. doesnt look pretty, but we only do it when players arent nearby.