AI Driving - Feedback topic

[xon2 102]

The actual driving of individual wheeled vehicles on Stratis and Atlis isnt't that bad when you simply send them via one waypoint from one side of the island to the other. But once they hit something they just end up in their endless ''being stuck motion'' until you either pull the driver out and delete the waypoint and put him back inside or reset his attitude (safe, aware....) at least twice. And the other biggy, convoys just are unusable! No matter how often i try it with no mods and game restarts, one unit always simply stops holding the rest of the convoy back. And its just no fun resetting this vehicle again and again to get the convoy moving again. And couple of corners later its another vehicle from the same convoy that does that. And if you let the ''convoy'' drive as individual vehicles with there own respective waypoints, you run into all the overtaking issues wich lead to collisions that result in the ''being stuck motion'' as described at the beginning of this paragraph. I am at my wits end here...

 

Edit; ...and yesterday, i had a single ai (in the civilian version of the idap transport van) drive from Altis airport to pick me up at the ruins on Altis eastcoast. He did not manage to make it over one bridge without incident. I had to ''teleport'' him as zeus over every bridge. Only once did he almost make it, the front wheels reaching the other side of the bridge,...but then he swerved to the right. After a bit of struggling he made it finally over the bridge. But that was an outlier, not  the norm.

[Grumpy Old Man 3545]

On 5/4/2018 at 9:10 PM, xon2 said:

The actual driving of individual wheeled vehicles on Stratis and Atlis isnt't that bad when you simply send them via one waypoint from one side of the island to the other.

 

But it is.

I can barely remember the day when it was possible to send an AI driven vehicle from A to B without breaking a sweat.

 

 

Bugs like that don't really need any elaborate debugging procedures/more than one single QA guy worth his salt.

 

Cheers

[.kju 3244]

its AI tweaked for life mods - drunken driver and so 

[FirstPanic 39]

14 hours ago, Grumpy Old Man said:

 

But it is.

I can barely remember the day when it was possible to send an AI driven vehicle from A to B without breaking a sweat.

 

 

Bugs like that don't really need any elaborate debugging procedures/more than one single QA guy worth his salt.

 

Cheers

 

Yup... that's exactly how it is today. 

Take a vanilla Arma 3, take a standard map, place a standard car with driver and let this car drive for 500m on a street without any curves... straight ahead....

And the AI will fail.

 

Dear Developers: Why can't we step back to the AI driver system which we had in 2016... it was close to perfect.

 

FirstPanic

[Devastator_cm 434]

I think BI need a fresh graduate from Electronics department who can write the control function with feedback loop. Looks like the logic is not considering the speed of the vehicle and when the vehicle is fast the control function is not able to handle the correction of the direction properly.

I guess if you set the speed limit on that vehicle to 30-40 it will stay on its lane (although with some unnecessary right and left corrections during whole travel)

[oukej 2910]

The AI is using PID controllers. I believe there are many improvements still possible just by setting up the data, the PIDs. At this point most of the vehicles inherit a setup from the car base class. (Btw, Hatchback Sport is quite an unorthodox asset. More of a just for fun car with an over-the-top configuration. Hunter MRAP would be a better benchmark for AI driving issues.)

This also means that when looking at the AI path-following we have to consider the particular vehicle. Bugs in this field are well recognizable in the diagnostics - from various oscillations, deliberate under/oversteering, constantly braking and accelerating, generally driving that is not smooth. This is something we can still improve. Given the time.... It is also where we more than welcome any suggestions for an actual setup of individual PID controller values.

If AI plans its path over some obstacles, steep slopes or areas where it will obviously get stuck, if it can't reach a destination or if it doesn't even try driving anywhere, it usually indicates a problem in the path-finding. Those are predominantly caused by bad map data. It can be something introduced directly by the map or object designer (areas restricted to the AI because of objects placed under the terrain, bad path LODs in buildings or their connection to the terrain,... ), but it can also be an outcome of binarization and processing problems (incorrectly connected roads, bridges; incorrect order in the cost map...). The vanilla maps can have those issues too. With probably the most painful ones stemming from issues within the binarization. We've tried and we've been half-way to solving that. To be honest, we've probably already run out of time at this point.

We still plan to release more diagnostics. To help the community authors better solve these problems on their creations.
And again, mainly when it comes to path-following, we've still got options for improvements via PID config tweaks ;)

[xon2 102]

@oukej Do i summarize your thoughts on ai driving correctly by stating: AI driving suffers from two main issues. Not perfet PID values for a given vehicles length (turning radius), size and average travel speeds AND ''bad map data'' on which an ai relys when planning its route.

 

3 hours ago, oukej said:

The vanilla maps can have those issues too. With probably the most painful ones stemming from issues within the binarization. We've tried and we've been half-way to solving that. To be honest, we've probably already run out of time at this point.

Could you elaborate on that a bit more? Do you mean you have gone over all vanilla maps and identified all places that were not ''ai driving friendly'' and have already fixed up half of them?

And is there a way to see thoses bad map data in the debugging mode so we can identify where something is wrong?

 

Additionally; whats your best guess for the broken convoys where one random vehicle stops and holds everything behind it back as well?

 

BTW, thx very much for checking in every couple of days even though our frustrated comments must not be the most fun to read

[xon2 102]

@Grumpy Old Man From todays maitainace patch:

 

 

CHANGELOG

Tweaked: Hatchback vehicle dynamics

Tweaked: APC wheels are now more durable

Tweaked: Tank tracks are now more durable

Tweaked: AI helicopter gunners are now more prone to use DAGRs and Scalpels (https://feedback.bistudio.com/T128103)

Tweaked: Vehicle fuel loss across multiple scenarios is now better handled

Fixed: Commander of the Rhino MGS could clip through the vehicle in some cases (https://feedback.bistudio.com/T128035)

Fixed: AI will no longer execute incapacitated players in the Revive system

Fixed: AA missiles were not effective enough against air targets

Fixed: Some of the Tanks DLC assets were missing their Simple Object configuration

 

Try your Altis Airport test-run again!

[Grumpy Old Man 3545]

54 minutes ago, xon2 said:

Tweaked: Hatchback vehicle dynamics

 

Looks like it's way easier to turn now, when driving it myself. Has a tendency to easily drift, very enjoyable.

The whole oscillation issue hasn't changed though, thanks for letting me know!
 

Cheers

[xon2 102]

But thats happening for the sports variant of the hatchback. The normal hatchback is not doing that oscillation stuff. The sports variant can be tamed with the slowest speed. Normal or fast do indeed show these oscillations. Anyway, it really seems to be a lot about vehicle specific values. The new Idap van and its civilian variants simply can't cross any bridges, where as other vehicles cross them just fine. And tracked vehicles on the other hand (especially the big tanks) do not stay properly on the road but change side quite often and go off the road regulary which gets them into trouble with objects and raises hell when encountering other vehicles.

 

And thats the area that needs most improving atm i think. There seems to be an overtaking mechanic in place, but vehicles too often come up to fast behind a slower vehicle and crash into its back which sets off the ''being stuck'' motion, or pushes a vehicle off the road or makes them turn and go backwards. The code for vehicle encounters,...this is where the work needs to begin. Vehicles need to slow down way earlier when approaching vehicles before them or heading towards them. There needs to be a piece of code that clearly negotiates between vehicles which overtakes which and stay on wich side of the road....and all of that at slow speeds.

 

And at the danger of sounding like a broken record.....conboys need a whole lotta fixing.

 

Btw, to look at the diagnostic data, i need to be on dev brack to have the diag64.exe right?

 

EDIT: totally forgot to type up what i meant to write in the first place. So i spent another two hours sending vehicles across Altis as zeus via one waypoint on the other side of the island. And except for the specific vehicles i mentioned above, all where doing really well. All AAF vehicles...no issues when driving by itselves, most civilian vehicles the same. Tracked vehicles have that aforementioned going off raod problem which leads to a whole lotta crashes with wheeled vehicles. It didn't really feel like there where any map data problems at all. The main issues arise when vehicles meet, overtake or convoy. And even a small crash throws the pathfinding off.

[f2k sel 164]

Tested using hatchback.

car forceFollowRoad true;  is still broken and makes things worse.

Vehicle won't go over a bridge when using it  and selecting the slower speeds in the waypoint has no effect it still runs at full speed.

I'm also seeing a lot of spins now when AI are trying to brake, they're braking too late and too hard.

 

 

 

 

[xon2 102]

did you use hatchback or hatchback (sport)?

[f2k sel 164]

9 hours ago, xon2 said:

did you use hatchback or hatchback (sport)?

Yes sport  spinning when braking heavily usually at last WP but can happen when travelling at high speeds.

Normal Hatchback seems OK

 

Both have issues with bridges and  ignoring the WP speed setting  when using  forcefollowroad.

I wish there was a WP that didn't force the vehicle to an almost stop when you just want to change some thing like behaviour or add a script.

Now all you can do is have endless triggers if you want smooth control over a vehicle.

[Grumpy Old Man 3545]

17 hours ago, xon2 said:

But thats happening for the sports variant of the hatchback.

 

The oscillation happens for every vehicle, the hatchback sports is the only one fast enough to make the oscillation more prominent.

Vehicles shouldn't oscillate left and right in the first place, there was zero oscillation before every vehicle received its own AI driving config.

Vehicles were also keeping to their side of the road before that change happened, now every vehicle drives in the centermost line if possible.

 

Place a regular (non sport) hatchback named "test" on the VR map, use this in its init field:

driver test domove (test getrelpos [5000,direction test]);onEachFrame {hintsilent str (velocityModelSpace test # 0)}

This will make the car go 5km forward, in the same direction the car is facing. The hint displays the x velocity of the vehicle (left/right).

Watch as the oscillation gets bigger and bigger, reaching up to 0.12m/s in the example mentioned above, which is plenty of swerving for a vehicle going in a straight line.

Driving like that on a public road will most likely get you pulled over by cops.

 

 

Even on a perfectly flat map with a waypoint straight ahead that requires zero direction change, the AI will always oscillate, no matter the vehicle.

The faster a vehicle goes, the bigger the oscillation gets, that's why it's most prominent with the hatchback sports.

If the car hits an oscillation peak in front of a corner, and that peak is in opposite direction of the corner you can imagine what happens.

 

Oscillation should never happen, especially on a straight road, even more so on an empty map where the movement destination lies straight ahead.

 

Cheers

[FirstPanic 39]

On 11.5.2018 at 2:52 PM, Grumpy Old Man said:

 

 

 

Oscillation should never happen, especially on a straight road, even more so on an empty map where the movement destination lies straight ahead.

 

Cheers

 

Thanks... that's exactly the evidence that the current AI technology is the wrong way.

It broke the game and I absolutely don't see any signal of understanding from the devs.

 

Dear Dev guys... you do a great job... but please... the AI is back to Arma 2 level.

So, I would like to repeat my question: Why can't we step back to the AI driver system which we had in 2016... it was close to perfect and the current Ai is so obviously wrong.

 

FirstPanic

 

[joostsidy 685]

36 minutes ago, FirstPanic said:

 

Thanks... that's exactly the evidence that the current AI technology is the wrong way.

It broke the game and I absolutely don't see any signal of understanding from the devs.

 

Dear Dev guys... you do a great job... but please... the AI is back to Arma 2 level.

So, I would like to repeat my question: Why can't we step back to the AI driver system which we had in 2016... it was close to perfect and the current Ai is so obviously wrong.

 

 

I think the devs made their understanding very clear; they are aware of problems, but they never had a good amount of programming resources to fix this once and for all. Besides that, at this stage, programming resources are limited even further as development of Arma 3 is closing.

 

Quote

Why can't we step back

BI never steps back, just like Stalin ;-)   Besides that, BI does not know how reversing works ;-)   My guess is that stepping back is not as simple as:

newDriving = false;
oldDriving = true;

Reverting driving also requires programming resources.

 

I share the feeling that driving could have been a lot better at this stage. IMO BI nailed a lot of aspects of the game in the 80% awesome range, whereas driving is stuck at a much lower level. It becomes a blemish on an otherwise excellent game. Because they didn't say super-duper-hard that they are not doing anything in this area anymore, I have a tiny bit of hope that resources can be invested to polish the blemish away a bit. Hopefully some internal negotiations / awaiting certain developments is going on.

[oukej 2910]

2 hours ago, joostsidy said:

BI never steps back, just like Stalin ;-)   Besides that, BI does not know how reversing works ;-)   My guess is that stepping back is not as simple as:

newDriving = false;
oldDriving = true;

 

useAISteeringComponent

[oukej 2910]

2 hours ago, FirstPanic said:

On 5/11/2018 at 2:52 PM, Grumpy Old Man said:

 

 

 

Oscillation should never happen, especially on a straight road, even more so on an empty map where the movement destination lies straight ahead.

 

Cheers

Thanks... that's exactly the evidence that the current AI technology is the wrong way.

It's definitely an evidence of a vehicle where the PID hasn't been tuned well enough together with the updated PhysX configuration #blameoukej.

[Grumpy Old Man 3545]

6 hours ago, oukej said:

#blameoukej.

 

May I direct you towards the following post:

 

 

Cheers

 

[Alwarren 2767]

13 hours ago, oukej said:

It's definitely an evidence of a vehicle where the PID hasn't been tuned well enough together with the updated PhysX configuration #blameoukej.

 

Pardon my ignorance but what exactly is the PID?

[Tankbuster 1744]

https://en.m.wikipedia.org/wiki/PID_controller

It's the software that provides the constantly updating driving decisions.

NOT Practiced In Destruction. Hehe.

[Grumpy Old Man 3545]

17 hours ago, oukej said:

useAISteeringComponent

 

I should have done this video without giving it away, but as of now, useAISteeringComponent only delays the oscillation for a bit:

 

 

So there's no way to override the new driving stuff?

Thing is the AI controlled hatchback would blast down this road without problems prior to the AI getting individual steering stuff for each vehicle. It might have been way too fast for the final turn at the end of the straight, but on the straight it stayed straight, heh.

 

Cheers

 

[ProfTournesol 956]

28 minutes ago, The Man Without Qualities said:

Clear evidence that something like Q&A does not exist in BIS.

 

THANK YOU BIS!

 

 

Come on mate, BIS has been supporting this game for years, contrary to most games that are abandonware a few months after the release. Stop embarrassing yourself.

[FirstPanic 39]

20 hours ago, oukej said:

It's definitely an evidence of a vehicle where the PID hasn't been tuned well enough together with the updated PhysX configuration #blameoukej.

Thanks for your answer and for the useAISteeringComponent command... wasn't award of that... will check, but following GoM's video it doesn't seem to be what it sounds like.

 

So, just to understand your statement: The new AI driving system now forces ALL vehicle and map developer to adjust AND ALIGN every content so that it can work together... right?

 

So hundreds (if not thousands) of objects need to be reconfigured, tested with maps, etc.... which means basically: current content is more or less broken and not usabele with AI drivers.

Am I the only one who thinks that this was a wrong decision? And unfortunately it's not the first time that this kind of change happened.

 

Really sad... and actually it's the confirmation for me to totally stop with A3 mission development now. Since  1.5 years I spend most of my time to find "workarounds" for core changes in A3 which made it impossible to design a good mission... and even when you find a solution... with the next update it's screwed and basic mission elements are no longer working (e.g. a jeep can not driver over bridge).

 

Well... I think it's time to stop this tragedy and focus on something which is more robust, reliable and stable.

BTW... noone is blaming you... actually I think that everyone is happy that we finally get answers to understand what happened. However... noone should expect a "Hooray" when we learn that AI driving is and will be broken for a very long time because every vehicle and map needs to be adapted due to the new AI driving system.

 

FirstPanic 

[Grumpy Old Man 3545]

44 minutes ago, The Man Without Qualities said:

Let's summarize: BIS managed to cripple AI in general, AI drining in special to an extend that I personally consider A3 as broken.

All we got is some jokes but not even a hint that BIS would even think about fixing the mess they created.

To be clear: to me it looks not like some tuning would help, to me AI driving feels broken from A3 release with the worst state right now.

The tests that GoM did I expect to be done my BIS after every tiny change they did, with premade testmissions.

Don't tell me that those bugs are hard to detect!

 

Clear evidence that something like Q&A does not exist in BIS.

 

THANK YOU BIS!

 

 

BI made the arma series to what it is today.

They sure went through one hell of a time during the development phase of A3, not sure if any other similar sized company could have pulled through that.

On top of that the franchise has been in development for more than 15 years, by multiple developers trying to follow various visions of what they think this game should be.

If you're interested in some nightmare inducing reads, just google "legacy code horror".

 

Despite all its flaws this game alongside its devs is still a gem, especially in a time of day one DLCs and multi GB sized day one patches being the norm and various early access survival games that will never be finished flooding the market.

Just take into account all the free stuff we got since A3 came out, firing from vehicles, weapon resting, advanced heli flight model, slingloading, vehicle in vehicle transport, fatigue system overhaul, the most complex stance system any game has to offer and more.

 

Bugs like seen in the videos I posted are most likely not hard to detect, but for reasons mentioned before, fixing them might be out of the scope for the current team still working on A3.

All there's left to do is to wait for a possible roadmap past tank DLC and see what it has to offer.

 

Keeping a friendly tone in the forums usually pays off,

don't forget,  BI owns a tank.

 

Cheers