gameplay-ai

[wary ivy]

it uses the navmesh projection function

it doesn't do a line trace

but whatever recast does

[north cape]

Hi! Is there a way for AI to multi-task? In my example, I'd like the attack logic and movement logic to be performing simultaneously (so the AI can shoot while strafing for example)

My understanding is that I should wrap up all of 'shooting' or 'moving' into as single task and chuck it into a simple parallel?

[charred lava]

You can also chuck one side into another behavior tree and use "Run Behavior tree" on the left side of your Simple Parallel node.

[brittle lynx]

EQS does time-slice per item, not per test. FEQSItemIterator::operator bool() returns false if you go over the limit, even if the iterator is pointing at a valid EQS item.

This way the for loop over the items inside a test is automatically stopped if the test is taking too long. On the next frame the test starts the for loop again, but the iterator starts after the items that were already processed.

[analog root]

AI can not sense the player correctly when I use spawn actor, but it works perfectly when I put them on the map, what is the problem?

normaly , AI can detect and kill the player when I put them on the level , but when I want them to be spawn and using SPAWN ACTOR function , AI does not work good

[crimson token]

Never heard of that, I'll check it out

Anyway, how would the cpp/bp implementation work? It would just create a class like any other, that I can use normally in my queries?

[crimson token]

Any chance you're getting and using a list of all currently existing actors on beginplay?

[analog root]

what is your mean ? can you explain more?

when I use spawn actor,AI can not sense the player unless I shoot him, but when I put the AI on the level, it can sense the player as soon as seeing him

[crimson token]

Hm, I'm not sure then

Can you show me your sight logic?

[brittle lynx]

Look into the AutoPossessAi setting on the enemy character. It should be PlacedInWorldOrSpawned (instead of PlacedInWorld). This setting determines when (or if) to spawn the AIController that will control the character.

[analog root]

it is PlacedInWorldOrSpawned butstill same issue exist

[analog root]

ok I fixed that, I was working on a kit from the marketplace, a TPS kit, creator of that used a strange thing for the "on seen pawn Event" ,that his code did not work for spawning actors AI but I customize it and now it work, thanks guys

[ocean wren]

There's a checkbox that tells the controller to spawn attached to the enemy.. can't remember what its called off the top of my head though

Just set it to dynamic and it'll recompute when you change the level

[celest python]

Any ideas about what would be the best way to have some sort of "layered" target goal (an actor) that is integrated in BBs, and works with other external systems together with BTs?
Basically, I have enemies that selects a target player in the world and run behaviors to attack and follow it. But some external factors can distract the enemy, like L4D2's pipe bomb etc
I thought of three possibilities, all of them would work but I'm trying to select the best one would scale well on the long run:

1- Have "main target" and "external target" keys, update "external target"in AIController (or with a component for modularity) with some sort of utility systems and let "distractors" pass some params to do utility math, if it's score higher than current distractor, start following it. Meanwhile keep an array of external targets in AIController to switch between external targets.
2- Ignore the "external key" in BB, have "main target" key and store an array of external targets in AIController along with actual "non distractor" target and if there is no external target left, switch to main target
3- Create CDO-only "target" objects, store them in "target component"s and let each target define their own behaviors with high level commands (probably tags) like "follow" "attack" "avoid", and let AI update the closest target component's logic in AIController

Third one gives a nice bottom up approach but with some undesired side effects like possible performance issues

[ocean wren]

I've traditionally kept utility outside of the BT, so for stuff like targets, I'd build a utility decision system that takes the perception system output as input and itself outputs the current target into the BB, that way you can allow people to author different utility decisions. Usually by modifying curves (I had curves that modified perception values for stuff like being visible, in cover, if the agent was aware already or not)

You could probably do a rough pass of it using the EQS system now I think about it 🙂

[celest python]

this doesnt interop with perception system, this is for things that directly affect the AIs without any other middle-man

i.e. if a bomb exploded in a radius, AI's will always register it to their external target array

but just not sure which one would scale better

[ocean wren]

Well, its just data isn't it? add a component that deals with the data and decision making

[celest python]

So you say I should go with 2nd option then

[ocean wren]

I guess my point is that you can use another component to do some of the decision logic to then feed data into the BB

Yeah, second option

The BT just deals with the results, the other system decides what the target is

I have a squad target system that allocates enemies to the squad do this kind of thing too

because it makes sense to coordinate fire

[celest python]

Roger, thanks

[ocean wren]

Just make a nice design surface for the utility scoring so that its mostly data/designer led

I've used curves and such, but those aren't always the easiest

Its where UE shines though, because you can write custom editors for it.. and custom asset types and whatnot

[celest python]

Talking about utility systems, do you think FMath::GetMappedRangeValueClamped is enough to normalize values into 0-1 range?

EQS has an example but my system does not have negative scores

so their math doesnt match with mine

[ocean wren]

easiest way to find out is to test it

[celest python]

So far it works, quite clumsy though

there are accuracy errors

[ocean wren]

You should have some unit tests for this kind of thing anyway

[celest python]

true

[ocean wren]

I usually try and identify some common use-cases and write up tests for those, in case you get regressions later

But math for utility scoring is notriously tricky

Dave Mark's book on behavioural mathematics is a worthwhile read there, but its pretty expensive now last I looked

[celest python]

I saw some examples from Dave Mark but I guess he is not a fan of pseudo math codes, he gives the math with notations directly 😄

I'll try to check CryEngine source

maybe I can find some stuff there

surprisingly they have some solid AI module btw

even better than UE

they even have Valve's AI driven dialogue thing out of the box

[celest python]

I'm also about to finish this by the way, currently writing a parser to support defining criterions with a text format. I'll open source it later on

Went through a few crazy things from implementing 3rd party tuple matching thing for C++ to random forest algos and such 😂

[ocean wren]

Yeah, a bunch of that was done by Mikko Mononen and Matthew Jack (who now runs Kythera AI) and a few others. Definitely compare the EQS system with the Cover point system in CryEngine, basically Matthew's design and Mieszko saw it and added it to UE

If you can't get along with Dave Mark's stuff, try looking for Kevin Dill's papers on Dual Utility reasoning, its basically utility systems, but a pretty simple implementation.

You really should learn what the math notations mean though, they're not scary, just not obvious until you see them

[celest python]

I skipped a lot of math lessons at high school, so I gotta learn basics of algebra, trigonometry etc first.. probably will happen before I participate in uni exams next year 😅

I expect it will empower my development skills a lot though

[ocean wren]

Its kind of intimidating, but honestly its not that hard. The problem is that the math notation changes based on the problem domain, plus there's a lot of unspoken rules and stuff 🙂

I couldn't believe it until someone went over some of the symbols with me 🙂 "this just means the sum of all the terms" and whatnot..

[fringe patrol]

@uneven cloud hey, i found out it's something wrong with the project, i tried it in a new project and it worked, i don't know why it happens, this is why it's better to give ue5 more time to adapt, anyway thank you so much for helping me figure it out!

[granite vault]

Welcome. Good to know that it works better for you.

when are you launching your game?

[celest python]

Next year hopefully

[granite vault]

game name is?

[keen sonnet]

I tried asking this before but ill ask again. Will any tutorial that allows ai to follow "sound" work for any game that does the same thing. I want to make a horror game that tracks microphone and in game sound that is created upon like stepping on glass or something. I think I can easily do in game sounds but how would I make sound tracking for when the player speaks into their mic. And will most tutorials show me how to do it in a basic sense. For whatever reason I'm too nervous to try any tutorial.

[misty wharf]

You'll just have to try them and see

It might not give you the exact steps you need to use for your game, but it should help you understand how it works, and adapt it to your own needs

You can't really "go wrong", if one tutorial doesn't help you, just watch another

[quaint radish]

I cannot seem to get the hearing sense to work properly.

I have sight working and setup with AIPerception component within the AI controller and the AIPerception_StimuliSource component within my player.

Sight is working properly and I set up/added hearing the same way to both components.

In the AIController, I have 2 methods confirming which sense was triggered, and neither result in Noise/hearing

Both output "Sight" perfectly fine

Followed this tutorial where adding sound/hearing is discussed/shown at 17:19 https://youtu.be/iY1jnFvHgbE?t=1039

When I jump, I have it set to "Report Noise Event" within my player character.

here is the Report Noise Event node

Not using the tag at all

even as a dominant sense, Hearing doesnt trigger the event

here is the entire bp logic from event to print string

figured it out. The tutorial didnt say that this was necessary but solved it

Needed an instigator for unknown reason. Anyone understand why?

[uneven cloud]

All senses need an instigator to check if it should be ignored based on teams.

[quaint radish]

oh wow. It would not trigger at all without the instigator though. Some reason worked for the person who made the tutorial.

Its not a surprise that this got me. Not every Instigator input pin is required.

[stiff gale]

Hi - What's the concept behind underwater AI using NavMesh? Since navmesh works horizontally but not vertically...

[celest python]

probably nothing different than flying AI

[north cape]

Thank you for the suggestion! I've been trying this, and unfortunately when in a simple parallel, 'run behaviour' doesn't seem to function (it constantly gets called repeatedly (flashes yellow) but the logic within never turns yellow). Do you have any documentation or resources I could take a look at for reference?

(subtree)

[celest python]

Didn't name it yet, but so far going on with "Zombie Game" thing

[charred lava]

Maybe it doesn't work on the left side? 🤔 I had a similiar scenario like you and I think I had my Sub behavior tree running on the right side (no subtree needed here actually) and on my left side the move action.

The doc has the section "Concurrent Behaviors" https://docs.unrealengine.com/4.27/en-US/InteractiveExperiences/ArtificialIntelligence/BehaviorTrees/BehaviorTreesOverview/ which describes which option the Behavior tree has available for Parallel actions.

maybe try something like this? @north cape

[fast bridge]

Hey, can anyone help me? I made an enemy NPC. It seems to be working perfectly in the editor when I press PLAY.

However, when I package the thing, the NPC doesn't move in the packaged game.

I'd really appreciate some help.

I'm using Unreal Engine 5.0.3

[green fiber]

Hey guys. I've got a Find Player -> Catch Player AI. So Pretty simple, and everything works fine. Is anyone here who knows ai pretty good, and maybe could help me/give some tips so i can improve the ai and make it even better. Cause sometimes it looks a bit weird, and i want it to be perfect 🙂

[bleak lance]

how do i fix this

[misty wharf]

"this"?

[brittle lynx]

If you mean navmesh edges not matching the geometry, try reducing “Max Simplification Error” in the Project Settings under Navigation Mesh. If that doesn’t help, also try to reduce “Cell Size” there.

[storm gulch]

Hi! Does dynamic navmesh work with instanced meshes?

I have a procedural level generation done with those and the navmesh only appears when I tweak something in the actor that contains the instanced meshes
edit: manually tweak at runtime

[uneven cloud]

It should do if they are set to affect navigation. It's likely that the order of loading things isn't correct or that your procedural generation takes too long. In C++ you can delay when generation on load happens until after your world gen. In BP, you might just have to trigger another regen.

[lusty nebula]

How are you doing the procedural world generation? Do you have any resources you used to learn that? I can’t find much that’s good?

[storm sail]

https://gyazo.com/a129393e58eae1c50375ec848727709d
Anyone know how to get AI to run towards a player by curving? Red is what the AI currently does (runs directly toward the target) Green is what I would like the AI to do (curve toward the target and not run directly at him)

[celest python]

either create points with EQS and follow them with order by checking movement state on path following component or dive into depths of navigation system and alter how path finding works

[misty gale]

Or spawn a helper object that the AI can follow

[storm gulch]

how do I trigger navmesh regen from bp? Can't find that

[storm gulch]

basically had to figure it out myself and not that it's particularly great or anything 😔

[brittle lynx]

You could look into Steering Behaviors. What you described sounds like Seek. Here’s a tutorial: https://gamedevelopment.tutsplus.com/tutorials/understanding-steering-behaviors-seek--gamedev-849

[celest python]

not sure but iirc it's a C++ only thing

unless there is a way to "indirectly" trigger navmesh regen in BPs

I vaguely remember some people were trying to regen their navmesh so hard they ended up doing it in C++

[uneven cloud]

Not entirely certain, because I don't use BP for that kind of thing. You can try making sure regen on level load is turned on in the settings. You can also try on bounds updated and see if that works.

[uneven cloud]

Have you tried slowing down the rotation rate? It won't give you exactly the drawing, but it will be closer.

[celest python]

or setting up ground friction in CMC

if Rotation Rate is set below to 270 and ground friction is around 1-2 is should be slightly curvy

---

anyway quick question:
does integer based utility systems has any downsides compared to EQS-like floating point utility systems? just generically speaking.. I want to build a generic utility system that is going to work everywhere (basically a static function) but I'm not sure how would it scale on the long run

    template<typename T>
    static T IntegerUtilitySelect(const TMap<T, FIntegerUtilityBooleanArrayWrapper>& Map, int32& OutHighestCount)
    {
        T HighestValue{};
        for (const auto& [Type, BooleanArray] : Map)
        {
            const int32 Count = BooleanArray.Evaluate(); // get count of "true" booleans
            if (Count > OutHighestCount)
            {
                HighestValue = Type;
                OutHighestCount = Count;
            }
        }
        
        return HighestValue;
    }

The reason I want to prefer integer based ones, it's not easy to maintain and design a system that takes too many parameters like this

[storm gulch]

okay then, how to do it in C++?

[uneven cloud]

Floating point utility gives a large range of possibilities in a clamped range. I always use 0.0 to 1.0

[celest python]

So you don't normalize to 0-1 but provide a value between ranges of 0-1 from the beginning?

[celest python]

probably UNavigationSystemV1::Build()

[uneven cloud]

It depends on what I'm scoring. It gives you the freedom to normalize to 0-1 or provide a value between 0-1. If the utility function only returns 0-1 it really doesn't matter how it's getting that score.

[storm gulch]

there was a RebuildNavigation console command in earlier versions, but its not there now (I guess)

[celest python]

it's there

I just saw it while trying to find UNavigationSystemV1::Build() 😄

[storm gulch]

in UE5?

[celest python]

yes

I lost it already though

[uneven cloud]

I don't remember the specifics as I haven't done it in ~6 years, but the general approach was block navigation from building, start world gen, trigger a nav build when world gen finished, started the game when nav gen was done.

[celest python]

Alright Luthage,

[celest python]

@storm gulch
else if (FParse::Command(&Cmd, TEXT("RebuildNavigation")))

RebuildNavigation command

[celest python]

0.00s is odd

either you're using a NASA computer or you dont have a navmesh

wait does it say command not recognized after executing it

[storm gulch]

...yes

[celest python]

it still executes though

maybe some legacy code interfering

logs are defined in function body in source code

[uneven cloud]

The function is RebuildAll.

[storm gulch]

when I replace the instanced meshes with ordinary static ones it kinda works, but I guess that'll be optimisation going straight out of the window

[celest python]

UNavigationSystemV1::Build() calls that

[celest python]

at this point I'd just debug

it does bunch of things but seems straightforward to understand if you have C++ knowledge

function names are clear at least

[storm gulch]

it all works for non-instanced meshes

and only generates the navmesh for instanced meshes when I manually do something to it in the editor

is there any way to emulate that "move it in the editor" part at least?

[storm gulch]

30h without sleep and all I see is the giant middle finger from UE

Found this method but it's not working either
https://forums.unrealengine.com/t/navmesh-ignoring-instanced-static-mesh-actors-generated-through-a-blueprint/338862/7

[uneven cloud]

That "move it in editor" is just changing the location which triggers a rebuild.

[storm gulch]

this doesn't help either

[storm gulch]

why though?

adding a delay before generating also doesn't seem to do anything

[solemn wasp]

Hey everyone !

Is it possible to access a specific object from a task blueprint to use in a Behavior Tree ?
In my case, I want a task "LookAtTarget" for a turret, and I need to access different parts of it for Pitch and Yaw.

I started doing it in in the event graph of that turret and it works, but I want to use the Behavior Tree workflow

[uneven cloud]

Yes. You need to either save that object into the blackboard and send it to the task or make a way to get it.

[storm gulch]

iirc you need to just cast from the controlled pawn if that's your turret

and the target should be stored in the blackboard

[solemn wasp]

aaah right ! Totally forgot about casting, didn't know it was available for tasks

alright thanks guys

[celest python]

I have an idea

Create an editor subsystem
find a relevant delegate in FEditorDelegates that triggers when any actor's location is changed
Check if actor has instanced mesh in scene components
if true, rebuild navigation manually

this is a workaround to trigger nav rebuild

[peak oxide]

do anyone have a tutorial or documents for how to make an ai for cars with Collison avoidance

[solemn wasp]

Hey guys ! Quick question: is it always best practice to use the AI Behavior Tree even for simple behaviors ?

Because right now, I have turrets and flying drones, but their behavior is really simple (find target, shoot, and for the drones just follow a spline path). I got them working in their Event Graph, they do what I want, but I find that the Behavior Tree makes it a bit more difficult to access certain values.

Is the tree better performance wise ? Or is it the same as tick ? Knowing that the drones could be a swarm of like 20+ drones (but they all access the same variables from one place, they don't have their own arrays or anything like that).

[celest python]

Tree itself rarely affect performance

but the content of your nodes can affect

BTs provide tools to debug easier, design easier and iterate mechanics easier

If you don't need those or your behavior is simple as that, you dont need to use it

but to unify systems, especially if you have a "top down AI manager" works with BTs, it might be better to just do it in BT

so its a huge "depends™️"

[solemn wasp]

alright, thank you !

[peak oxide]

help

[celest python]

It's not something that can fit into "tutorial category"

Even if you would find one, you should avoid it imo

That's something GDCs can explain or papers

or sample projects

BlueMan has a plugin for exactly what you're looking for btw, but its paid on marketplace

[peak oxide]

shit

i just wanted to know the basics of it

[celest python]

collision avoidance is RVO

implemented in the engine

but not for vehicles

[peak oxide]

ik

[peak oxide]

i wanted to make simple traffic

[celest python]

so why not #mass ?

did u gave up on it

[peak oxide]

idk what they do there

i thought it is ai thing

[celest python]

its what citysample is lol

built on top of #mass

its exactly made for a traffic simulation

it can be easy or complex, up to you too

https://github.com/Megafunk/MassSample

[peak oxide]

my manager would not like the idea i think

he wants his own system

[celest python]

well #mass already has some people made traffic sim, and they are pretty helpful and sharing about what they have done. If you have no way to convince your manager, then probably just make cars follow a spline and do traces to stop when it's about to hit something 😄 I swear what CP77 is doing the same

[peak oxide]

i am already doing that 😆

but it would not be smart enough

[celest python]

further implementations are advanced i guess

I dont know a straightforward way

[peak oxide]

me too

[celest python]

try to search through GDC vault

for traffic sim implementations

or get inspired by mass

or copy mass and say i made it

[peak oxide]

nah i will tell him the both solutions

i am leaving at the end of the month anyway

[celest python]

wise move tbh

imagine a manager doesnt want to use built in solutions

[peak oxide]

i had to do it 3 months ago

but i thought we could do something

[peak oxide]

he just do not want to be forced by unreal stuff

and i understand it

what if unreal removed a system or stopped supporting it

[celest python]

like almost everything about AI

but thats how UE works

what matters is shipping a product

[peak oxide]

yep

[celest python]

one way another without wasting thousands of dollars you wont be able to make an equivalent of mass anyway

at least maybe you can hire people on the long run to support on your own etc

thats what some studios does for BTs

I saw studios rewrote some parts of BT

though recently they just use 3rd part ones

[peak oxide]

i argued with him once about that

i would not argue again

i might not tell him the real reason why i am leaving

[wary ivy]

why does UEnvQueryGenerator_PathingGrid run the projection step twice?

it calls Super::ProjectAndFilterNavPoints which projects the points to navigation and then it does its own projection later on

seems... weird

[ocean wren]

Doesn't make a lot of sense.. don't want to use an UE system because they might not support it? I mean they literally don't support it anyway, so use it and modify it and take ownership of it

unless the plan is to sell it to someone else, which is perhaps the only decent reason to build stuff yourself other than for performance/usability reasons

[brittle lynx]

It doesn't do that for every point. I only does that if it couldn't find a path from the context to the point. It's a fallback for the case where there are multiple floors to the navmesh and Super::ProjectAndFilterNavPoints projected the point to the wrong one. This is why the fallback uses NavMeshData->ProjectPointMulti to project the point onto all floors and check if there is a path to any of the projected values.

[wary ivy]

Yea that just occurred to me too

[wary ivy]

there's one other weird thing about it, it's that the projection extent doesn't seem to work properly. It's set at zero but still the projected points are shifter on XY axis

for example see the point on the lower left side of the testing actor

whaat, the extent X only affects the x asis of the trace

well I suppose it makes sense now, it just didn't work the way I assumed 😄

I don't understand why the property rendering has been made so that it's not visible there

this is a bit weird

FPImplRecastNavMesh::ProjectPointMulti uses all X, Y and Z of the projection extent vector but then FEQSHelpers::RunNavProjection uses only the X to construct the projection extents

[twin plinth]

Hello i want to ask how performance demanding is build-in unreal pathfinding system for real time strategy (i want use about 200-250 units in total) or it is better to try to create own A* in the form of flowmaps? Or is any good solution for RTS pathfinding on the market?

[ocean wren]

Default UE pathing probably isn't great for 200 ish units, unless you do a bit of work to batch queries and whatnot. But it'd probably be sensible to come up with a better system, HPA* and the like

[wary ivy]

you'll most likely run into problems with the actors themselves (if you use them) and how slow it is to move them in a scene

[fathom sun]

Is it correct making a platform a pawn if I need to add some logic?
For example to allow a platform to use a behavior tree to move from one location to some other. As I understand actors don't have a controller, so they can't have a controller, so they can't have some AI, so that's why I'm asking 😄

[brittle lynx]

If you have to use a Behavior Tree to move it around, then yes. Or you could just make an actor blueprint for the platform with some code in there that makes it move, without using a Behavior Tree. If all you need is for it to move back and forth between some locations that's what I'd recommend.

[fathom sun]

Yeah, I already made such thing, I mean some patrol system, but then I remembered that I can do it with AI stuff

[lethal helm]

is there a way to set the tick rate for a service from a blackboard key?

[misty wharf]

Might be possible in C++

[lethal helm]

huh yeah

getting intimate with the source code seems the best way to get stuff to work in BTs

[misty wharf]

At least for less common things like this :)

There is a function you can call in C++ to set a component's tick interval at least, so I would imagine there could be a similar function available for BT services which you could call after reading the BB value

[lethal helm]

voila
void UService_DynamicTickBase::ScheduleNextTick(UBehaviorTreeComponent& OwnerComp, uint8* NodeMemory)
{
float DynamicInterval = GetInterval();
const float NextTickTime = FMath::FRandRange(FMath::Max(0.0f, DynamicInterval - RandomDeviation), (DynamicInterval + RandomDeviation));
SetNextTickTime(NodeMemory, NextTickTime);
}

public:
UFUNCTION(BlueprintImplementableEvent, BlueprintCallable)
float GetInterval();

don't even have to reimplement my original service just need to reparent it to this thing

[misty wharf]

Oh nice

I would not have expected them to think ahead and make it BlueprintImplementable lol

[lethal helm]

they didn't

I used to the source code as a reference to make a child of service_blueprintbase that uses this blueprint implementable func

though I would expect people wanting to have dynamic tick rates to be more common tbh

[misty wharf]

Who knows, the AI systems aren't known for their forethought lol

[tawdry crag]

What's the standard way of setting up navLinks so that my AI can identify how to interact with them? i.e. this part of the path is a jump, ladder, etc.? The Things I can see in FNavigationPath is the PathPoints that have

• CustomLinkId, but I'm not sure how to set that to anything but 0, nor how it interpret it
• Flags, which I can't event see where they are set, let alone what they might mean

[misty wharf]

I've used smart links for that type of stuff

Eg. AI enters smart link for a door which triggers it to do a door open animation and such

[tawdry crag]

As in, the level tells the AI what to do?

[misty wharf]

the smart link itself can do that

and you can have different ones

[tawdry crag]

How would you query in advance what that smart link is going to do?

[misty wharf]

I don't recall how exactly it was set up but at least the ones I've used it for like doors and jumping you didn't really need to know it in advance

[tawdry crag]

hmmm, so it's like "Using a Nav Link Proxy to Allow Your Agent to Jump to Its Goal" about 2/3 of the way down in here?
https://docs.unrealengine.com/5.0/en-US/overview-of-how-to-modify-the-navigation-mesh-in-unreal-engine/

For at least the trigger condition, right?

[misty wharf]

Ah yep that's pretty much how you can do it

You can do more elaborate stuff with it in C++ but it gets messier :)

[tawdry crag]

Alright, I'll dig around in the smart link and see if it will suit my needs. Thanks

[fathom sun]

What are the methods that Move To node use? I'm trying to implement an own movement component, but the only thing it found out is that it edits the velocity vector inside base movement component. The thing is that in my implementation I also change the velocity (in the tick function), but how can I easily distinguish between my velocity and the one set by that node?

[low iris]

iirc it calls some combination of RequestDirectMove and RequestPathMove from UNavMovementComponent, but I don't recall exactly how it works

[fathom sun]

alr, I'll check it out later and tell what I got

[low iris]

yeah it looks like RequestDirectMove is where it's setting the velocity, so you can override that to do whatever you want

[runic patio]

Any information someone can recommend on AI path finding when there is lots of actors in the way? So they end up like the first image instead of the second one.

[storm sail]

decrease the collision boxes of the actors that block AI navigation, RVO avoidance can also help actors get where they need to be without getting stuck and timing out

if you're talking about like AI formation, you're gonna have to program that yourself

[runic patio]

@storm sail The collision boxes for blocking AI navigation is a nice one, didn't know about it thanks.

[ocean wren]

DetourCrowd is better than RVO

[hearty niche]

I am using set focus to make the AI aim at the player. But it seems to aim at the capsule center instead. How can I fix this?

[hearty niche]

I think eqs would be a good idea. I have heard a lot of people say that it is better to have ai positioned well in the first place. You could generate an eqs around the player and do a trace on each point to check if it is blocked by something or not.

[charred lava]

Aim at a component of the player where the position is more suited. Like Character Mesh Spine or Head.

[hearty niche]

How do I do that? As of now I am using a set focus service which handles this stuff

[winter moon]

does aidamage only work with line traces?

[charred lava]

Overwrite AAIController::GetFocalPointOnActor

[hearty niche]

does this have to be done on cpp?

[charred lava]

I think so yeah

[hearty niche]

oof

Alright on it

Any chance I can override this via an actor component? I don't want to increase dependencies

[charred lava]

dunno, I don't think so

[fathom sun]

You're right, RequestDirectMove is the function I'm looking for, it gets called every time by a behaviour tree, while apparently RequestPathMove doesn't, I have overriden both functions to just print a log message, and only the first one did.
btw do you have any ideas what given velocity to the function is based on? I'm asking because it always give me vectors with 5 digit length or something like that, and it differs from frame to frame, and it makes no sense for me, the direction is correct though

[green goblet]

anyone can please guide, how did he make the enemy light of sight, which is occluding from object, I want to make as it is, https://www.youtube.com/watch?v=iY1jnFvHgbE&ab_channel=UnrealEngine

[hearty niche]

how do I set the abort condition in a custom decorator? I am checking if the stored target location is too far from the player

[quick steppe]

Hi

I tried to follow the tutorial for mass ai in 15min

but when i press play crownd dont show up

https://www.youtube.com/watch?v=2LvUB3_PAhI&t=602s

it's this one

i litteraly copy cat everything but it seem to not be working

[hearty niche]

How can I prevent ai from moving towards the same eqs point

[brittle lynx]

What I usually do is add a "distance to reserved locations >= X" EQS Test where "reserved locations" is a custom EQS Context that returns locations from a shared tracker. When a character decides to go somewhere, they add the destination location to that tracker, when they're done moving to it (succeeded or aborted/failed) they remove it. This way locations picked by one character will be avoided by others.

However, there is a corner case with this approach, when multiple characters run the EQS at the same time, such that they end up selecting the same location before one of them reserves it.

[hearty niche]

So I could register them centrally somewhere right?

[brittle lynx]

Yes.

[hearty niche]

Subsystem time XD

[brittle lynx]

This can be caught by having the reserving of locations be done by a decorator, such that the decorator condition fails if it failed to reserve the destination (because it's reserved by someone else).

[brittle lynx]

WorldSubsystems are 🔥. I don't think you can make them in BP though.

[hearty niche]

I know cpp not to worry. And there is an AI Subsystem class to derive from.

[brittle lynx]

UAISubsystem is not an actual subsystem though. It appears to be something to latch onto the AISystem.

[quick steppe]

Anyone has an idea on why i got no mass ai ?

[brittle lynx]

@hearty niche I think subclassing UWorldSubsystem or UTickableWorldSubsystem (they actually inherit from USubsystem) is more convenient because then the subclass is automatically instantiated by the engine for each UWorld (and ticked in the latter case).

[hearty niche]

I think it's a child of World Subsystem only or what is the scope of this class? https://docs.unrealengine.com/5.0/en-US/API/Runtime/AIModule/UAISubsystem/

[brittle lynx]

It's just a UObject

https://github.com/EpicGames/UnrealEngine/blob/46544fa5e0aa9e6740c19b44b0628b72e7bbd5ce/Engine/Source/Runtime/AIModule/Classes/AISubsystem.h

[uneven cloud]

The scope is the world. It was made long before the world subsystems, so it doesn't use them. The world does own and initialize it though.

[hearty niche]

Good to know thanks 🙂

[low iris]

if I remember correctly it's just the distance from the current position to where the AI is trying to move to

[fathom sun]

Are you sure? The length between my pawn and the point it has to move to isn't even 4 digit, but the velocity length is more than that

[tawdry yew]

I need a quick refresher on basic ai, please correct me if I'm wrong:
to get the most barebones AI

• you optionally make a APawn class e.g. MyAIPawn, might aswell use 1 class for players and ai's
• create AIController class, with AI perception component.
• set AIController class in APawn class
  right?

[storm sail]

pretty much that'll do it

but there's already a default Aicontroller

[uncut python]

hi everyone, im having an issue where my AI run inside of eachother. They get stuck and they start walking really weird. I already turned on detour crowd controller, and it still doesnt help. Is there any way i can get them to avoid each other?

[rancid wing]

anyone know how to get a bb value or a boolean from the tasks of ai to another bp? pls @ me

[uneven cloud]

You get the AI controller, then get the blackboard component and use get value as bool node.

[uneven cloud]

Are they set to collide with each other?

[rancid wing]

thats exactly what im doing lol, but what do I put in as the key name ?

[uneven cloud]

Whatever the name for the blackboard key is.

[hearty niche]

how can I prevent an ai from shooting his own team mate in front of him?

A misfire should be fine but it should not just mow down everything in its way.

[celest python]

Trace and check hit?

[hearty niche]

so trace before shooring?

[celest python]

yes

[celest python]

Anyone knows should I just tweak SeparationWeight in CrowdFollowingComponent or should I change RVO settings in CMC?

[shell horizon]

Hey, while I was working on an AI system, I found something that I'm not sure if it's a bug in the engine or it's something I messed up somehow.
I've already set every changed value back in the navmesh settings, so I'm not sure what can casue this.

So not always, but sometimes the FindPathSync puts an extra point to a nav mesh poly edge, not sure if to the closest always. It does the same with all the MoveTo functions.

Does anyone know what can cause this?

Two images that shows the issue:

[uncut python]

Which component would I be checking?

Pawn collision or collision sphere

[rose terrace]

Hey guys, just want a short yes or no for now,
Is it possible to blueprint AI to try at all cost avoid friendly fire? Aka shooting an ally?

Hope its possible

Like am i expecting too much from a machine? 😅😅

[brittle lynx]

yes

[rose terrace]

Woop woop 😎

[uneven cloud]

You should use either RVO or detour crowd. Not both.

[celest python]

I thought Crowd extends the RVO framework?

[uneven cloud]

It does not. Both being on will give unexpected results.

[celest python]

Ah, alright. Thanks for the headsup

[uneven cloud]

For the crowd settings, everything tunable without C++ is in the project settings. The rest can be done via the crowd path following component.

[celest python]

Do you know if I need to make every other pawn try to avoid a single pawn, should I just tweak agent's SeparationWeight in CrowdFollowingComp or should I assing a new agent group to it? So far I wasnt able to observe which one was better, probably because RVO and Crowd was active together

[uneven cloud]

Ideally you want the capsule root component to be the only thing that blocks other pawns.

[uneven cloud]

Not sure. It's been a few years since I used either. I do remember crowd to be better, but that wasn't a very high bar.

[celest python]

Alright,

[shell horizon]

I have been looking at the source code of the CrowdManager, CrowdFollowingComponent in the past few days and pretty much most of the code that is related to the crowd avoidance.
I've tested most things, but I couldn't find what I was looking for.

So is there a way (without overriding any of it) to not use the MoveTo functions in the engine, but keep Crowd Avoidance working? So e.g. there is a spline, I want my AI to move on it, but keep the crowd avoidance ON? So it can pause the spline movement and just follow what the crowd avoidance tells it to do. And when the avoidance is "finished", then give the control back to the spline movement.

So here I'm looking for a public bool that tells me if crowd avoidance should happen, or like a direction FVector that controlled by the crowd avoidance, etc. What I found variables in c++, only works with the MoveTo functions.
I've tried to use CrowdManager->RegisterAgent, but it did not do the avoidance, only registered it. Am I missing something with this registration process?

If there's no correct way to make this without overriding any of it, is there a protected variable/function that could help with this?

(Not sure if this should be in #engine-source or here tho, sorry if this is the wrong text channel for this)

[uneven cloud]

Not without a significant amount of rework. Crowd is tied with path following.

[cinder spear]

Hey there! Is there any way to degrade the AI sight stimulus perception strength over distance? I want my enemy to detect my character faster the nearer it is to it, and its driving me insane

[shell horizon]

I see, thanks.

[uneven cloud]

You would need to make your own sight sense, which is only available in C++

[celest python]

Why do you need to avoid actual path following?

Can't you just alter the path finding and follow a spline-ish path?

#gameplay-ai message

https://youtu.be/yqZE5O8VPAU

or if you prefer zoombapup's way you can dive into string pulling algorithms etc (not a spline movement but more smoothed curves) #gameplay-ai message

[shell horizon]

Well, yeah this would be good to achieve. So by "alter the path finding" you mean change the actual path finding to work with curves? And that would work with crowd avoidance? By any chance could you give me some clue how would I start making this?

[celest python]

I dont know how path finding exatly works yet; but one way or another this will require you to dive into depths of AI framework, and you won't be able to find a good support around the community since almost no one worked on such things, at least in this discord.
From what I can see you can do some things in PostProcessPath function, which I assume will require you to override some classes and set some function pointers since most of the things are not virtual in the engine when it comes to path finding for optimization purposes

though I should say engine is ambigious about if this will work with crowd or not

you might need to disable some flags

Now I'm even questioning if string pulling is even active on most of the cases its enabled by default but everything is mutating the variable

[shell horizon]

Hmm, I wanted to leave diving really deep for the last-case scenario.
I was just wondering on this plugin: https://www.unrealengine.com/marketplace/en-US/product/ai-spline-based-movement-system

In this video(https://youtu.be/9wwzi85BviY?t=147) it can be seen that a spline based movement works with crowd avoidance. And at around 2:10 it can be seen that it's with the default DetourCrowdController. So this is why I thought it's possible to make it without diving that deep as you said.
Maybe I'll buy the plugin to see how it works tho, I'm just too curious.

I mean it obviously doesn't work like Death Stranding's system, but this isn't bad either.

[celest python]

You can always ask before buying though

When it comes to marketplace, by Epic standards people can bug the seller as much as they want 🥲

[shell horizon]

I guess yeah, but I thought it's not a question anyone would answer lol. I mean "hey how ur plugin works?" Obviously in a more formal way. But I'll give it a try.

Anyway, thanks for your help and the links! It really did help a lot and gave new options.

[uneven cloud]

You can use splines and path following together. Where the next move to location is the next spline point. It's really easy to set up.

[shell horizon]

I already have the spline generation, I only need to make moving on it work with crowd avoidance.
Or I'm not sure what do you mean by "splines and path following together", could you explain that?

[celest python]

They probably mean getting world location of "spline point" and moving towards to by incrementing an integer to follow up to last spline point in the spline component
But last time I tried that, with multiple agents moving close to each other, it didn't work as I expected

But I was using RVO

[shell horizon]

ah, yeah I've tried that, but doesn't really work as I want.

[shell horizon]

If I don't use the MoveTo function then crowd doesn't calculate. If I use MoveTo function then forcing it to move on the spline, it would be just broken.

[hazy swan]

Hey guys, is there any good tutorial you can recommend for studying the behavior tree?
I want to make simple A.I for enemy detection around my turret

[keen crow]

is it ok to put multiple navmeshes on a level with navigable areas on very distant heights? or should I put a single navmesh and extend it all the way from top to bottom? which is better performance-wise?

[wary ivy]

you mean multiple bounds?

[keen crow]

yeah

[wary ivy]

there's no difference performance wise

only difference that there could be is generation time in editor

[winter moon]

how do i make my ai stop chasing after a certain time period?

im doing ai sense on damage

[misty gale]

Do a timer?

And check vs some initial location if distance matters

[winter moon]

I’m using behavior trees

Would make this way easier if I wasn’t tbh I’d just put a timer on ai move to

[misty gale]

Cant you have incremental time tasks in bt aswell?

Or have it start in the ai bp whenever the following state is set ?

[ocean wren]

The detourcrowd ends up as a vector "where I want you to go" that gets integrated every update, you can integrate a spline following into that calculation, so its a "move towards this point on the spline" as part of the summed forces in the detour calculation. I used a seperation force so that I can push AI out of the way and a few forces to allow players to push the AI more than the other way around

[shell horizon]

So if I understand it correctly I should just override the class this gets calculated and basically modify the calculations to move along the spline instead of towards the next path point? So this way detour crowd could do it's own thing?
Please correct me if I'm wrong.

Could you tell me if you know in which class or where is this calculation that you refer to? Or the place of that vector? Just so I know which calculation you mean.

[ocean wren]

Its more about incorporating the spline following into the vector calculation, so you don't override it, but include it in the movement vectors that get summed. The basic idea is that you have a bunch of competing vectors and the logic runs to sum them together and then get a final desired velocity which is then solved with the constraints of collision avoidance etc.

Its a few years since I did it, but basically follow the DetourCrowdFollowing? component and you should eventually get a function that has like a "desired velocity" value... which it then integrates for movement..

[shell horizon]

Yeah, I know a lot of things gets calculated and a result comes based on pathfinding and detour crowd, but I'm not sure how would you modify that without overriding it.

Is there something simple I missed? Something that I can use to include anything in those movement calculations before it gets locked and decided finally for that tick?
I mean after we start the MoveTo function, we can't do anything useful with those vectors, we can "fight" against it, but that's not a solution.

The only thing I can think of at this point is to derive a class from CrowdFollowingComponent and override what's necessary to include spline to the calculations.

[signal olive]

How do you implement custom nav links that the AI can use or ignore based on specific abilities they may or may not have?

[uneven cloud]

You can either use point links (C++) or smart links (C++ or BP). Both are available in the nav link proxy. Point links you would use nav areas and filters and smart links have a function you can override to determine if an agent can use it.

[uneven cloud]

There is a tutorial on Epic's learning site called AI with Blueprints. It goes over all the AI systems, except for the new ones added in UE5.

[uneven cloud]

Crowd works with the current path, so it's not going to work without a path unless you do some rework. If you really want to not have a path, then it might be easier to do your own obstacle avoidance.

What is not clear is what about using the spline as navigation points is not giving you the results you want.

[hazy swan]

ill go through it thank you man

[shell horizon]

What is not clear is what about using the spline as navigation points is not giving you the results you want.

Not exactly sure what you mean by this. Could you rephrase it?

[spiral urchin]

I have a sequence, and I only want this sequence to execute if my bool blackboard key has a value of true.
What kind of decorator do I need for this?
Edit: Nvm, there is a perform condition check function in the base decorator. You just can't search for it, but rather have to check manually.

[shell horizon]

The result I want is basically a smoother turning/rotation (with avoidance) for mainly animal-like AIs, so for horizontally longer creatures.
Something what I could achieve with root-motion. But I don't want to use root-motion for this.

[uneven cloud]

Have you tuned the rotation rate in the movement component? You need to uncheck use control rotation yaw in the character class and check use control rotation in the movement component and then adjust the rotation rate.

[shell horizon]

Yeah, but that doesn't give too nice results I think.

[celest python]

Try setting Ground Friction to 1

or 1.75

[uneven cloud]

There is a built in decorator for checking a blackboard condition.

[celest python]

It will manipulate the velocity direction and interop with crowd easily

[shell horizon]

Changing ground friction to 1 or 1.75 in the character movement comp doesn't really do anything visible to me.

[celest python]

then crowd overriding everything about accel and velocity

which sucks

you're up to what death stranding does then

[shell horizon]

I guess yeah.

[celest python]

I'd still brute force every other option

until I see it's my only chance

no one would want to dive into string pulling and depths of AIModule

[shell horizon]

Yeah, I'll first just try with the overriding process, that isn't that hard and then the plugin, but as I saw some reviews it doesn't work too good with crowd avoidance too in some cases, so maybe I really have to work out my own solution.

[celest python]

https://twitter.com/games_inu/status/1247969381965017091

Inu Games (@games_inu)

Detour Crowd obstacle avoidance
#UE4 #UE4Study #gamedev #indiedev

▶ Play video

Twitter · Apr 8, 2020, 7:27 PM

Are you looking for something like this

[shell horizon]

Yeah, I saw this twitter post, this would be nice but without root motion. Not sure if this was done with that tho

[celest python]

no idea about that, though it's worth to annoy them about it 😄

[shell horizon]

that's true

[celest python]

learner (another regular in this discord) told they shared the configs somewhere

but I couldnt find

[shell horizon]

really? At least I have my plans for the whole weekend lol.

I guess this comment under the post makes it clear that they probably used rootmotion

[celest python]

that doesnt mean same behavior can be achieved without root motion though

[mortal umbra]

how can i make it so that if the AI hears a bullet impact he will go investigate around the area of the impact?

i know i have to use a report noise event

why is the AI debug thing sometimes empty?

it seems bugged

[hearty niche]

Is there any common function called in BTTask on finish execute?

[misty wharf]

Yep - You can use report noise and give it some tag, for example "BulletImpact". You can then test for this in your perception handling code

[hearty niche]

How do I run a separate behavior tree in a behavior tree?

[charred lava]

search for "behavior" when right clicking on the behavior tree graph

[mortal umbra]

@misty wharf i have that working but it was a bit buggy

it still might be but i think i fixed it a bit

what causes the perception debugger to do this?

it should show this instead

it just randomly doesn't show stuff sometimes until i restart the PIE

it is annoying

[empty fossil]

hello. According to this part of the doc about StateTrees, I can reuse data within the same state beteween Tasks. Could someone tell me how? Also, is there anything else on the internet about how to use StateTrees apart from that epic doc? It's very basic and shallow 😦

I have a state wander, and I want the task in GoTo to use the location found by the parent state

[shell horizon]

Is this a normal behavior of the crowd ai controller in UE4.27.2? Does anyone with 4.27.2 experience this?

[celest python]

Doesn't happen on me

But I'm on 5.0

Though dtour crowd didnt get updates

since... ages

Maybe cube is colliding with the wall

if it has collision enabled

[shell horizon]

ahh... Seems like I won't even be able to use crowd avoidance then if this is a version bug.

I've tried to make it much smaller, but doesn't work

[celest python]

Doubt 4.27 has a such bug like that

[shell horizon]

I've already set everything back to default that has to do anything with pathfinding, but I'm gonna just download 4.27.2 from epic quickly and see if this happens there too. If not, then maybe my source build has some problem.

[uneven cloud]

You need to center your camera on the AI before hitting ' to bring it up.

Crowd has always been a bit weird on edges of the nav mesh.

[shell horizon]

ohh, so this weird behavior is an actual existing thing? Probably this is a reason why they made that path point offset value in the project settings, but not like it works too good...

isn't there a way to minimize this? I can't believe this is just how it works in 4.27.2

[uneven cloud]

I've seen it a lot before and there are settings to minimize it, but it's been years since I've used recast.

Epic hasn't really made any changes to navigation since UE4 released. So I'm not terribly shocked it still exists.

[shell horizon]

I see, this is kinda sad.

[uneven cloud]

It is incredibly frustrating. RVO is even worse than crowd.

You might want to see if they made any changes with 5. The only ones I saw were with level streaming with world partition, but I wasn't looking too hard since I use Havok now. They may have done more smaller fixes with the Matrix demo.

[celest python]

Even though RVO sucks, somehow it's cheaper than I expected btw

but probably still expensive compared to other implementations

[shell horizon]

Now I'm not even sure if I want to use crowd avoidance, maybe I'm better to work out my own avoidance and make AI behavior not to be around each other that much.

[uneven cloud]

RVO is super cheap, but doesn't really do what it's supposed to. I've written my own and it's not terribly difficult to write an efficient one. Lots of information out there on obstacle avoidance.

[celest python]

Real problem about AI algorithms is they require you to understand some concepts of CS and if you're a self-taught learned things with game engine context rather than raw programming it's getting difficult 😄

I assume AI is second most difficult area after graphics programming

Also most of the resources just display an abstract example of how things can be achieved, when you try to implement them in UE you get lost in AIModule's depths 😦

Since zoombapup mentioned it first I'm looking for string pulling algorithm, which is amazing covered by many resources but UE's implementation is so weird that you can not override even most simple behaviors without tons of hassle

[celest python]

But I'm pretty confident about "mastering" AI is more difficult than mastering graphics programming because more people interested in graphics these days. Without having a senior to watch your back it's probably very difficult to get used to low-level things in short time

[uneven cloud]

There's a lot out there about steering behaviors. It's also included in any book about game AI.

UE's implementation was to take the open source Recast and put a wrapper on it.

I think graphics programming is considerably more difficult. It's such a black box of unknowns.

AI can be very difficult. But I think a lot of people over think things when they are starting out, which makes it harder.

[mortal umbra]

@uneven cloud seems like a bug to me

thanks

it makes sense when there are many AI

so it might not actually be a bug

[uneven cloud]

It can only show information for one at a time. Not a bug, but also not terribly intuitive.

[mortal umbra]

@uneven cloud do you know how i can make AI turn in place using ALS v4?

i currently have it set up to patrol to points but when it has to turn around it just slides and doesn't turn in place like it does when using set focus

[uneven cloud]

I've never used ALS, so I don't know how it works.

[mortal umbra]

@uneven cloud how would you do it without ALS?

[uneven cloud]

I use turn in place animations + a rotate to before moving, if the angle is larger than a given threshold. During movement a blendspace for turning left/right. I also adjust the rotation rate in the movement component and have that do control rotation instead of the pawn.

[mortal umbra]

@uneven cloud can you make a short video on how you have it set up? it might really help me out

[uneven cloud]

Can't, sorry. I work at a studio and can't show in progress work.

[viscid gulch]

Anyone here have any experience with the World Director Pro plugin?

[ocean wren]

For what its worth, the detour crowd stuff is actually pretty reasonable, but Epic's implementation isn't ideal.. look at the raw Recast+Detour non UE example codebase for what it can really do

That said, you got the code, write your own and make it better 🙂

[shell horizon]

just to make sure, by raw recast and detour you mean these stuff, right? https://github.com/recastnavigation/recastnavigation
Or you meant something else?

[ocean wren]

Thats the one, build the demo and take a look

[celest python]

why detour crowd always tries to follow wall lines?

even in the open source version

[ocean wren]

because it uses string pulling, its the "shortest path"

using the funnel algorithm

[celest python]

wasn't string pulling is about smoothing the path

rahter than making it shorter

[ocean wren]

no, its about making the shortest line that forms a path to the goal

[celest python]

gamedev.net failed me

so is there any algorithm or a setting to make it more smooth?

[ocean wren]

its called string pulling because it works like pulling a piece of string

it depends on what you mean by smooth

smooth in what way?

[celest python]

I guess it also makes it more costly right?

because when I looked at the code, I saw string pulling makes path finding repeat itself

until a valid "string" created

[ocean wren]

Well, if it didn't do string pulling, it probably just passes through nav poly edges.. so it'd look a bit zigzaggy

[celest python]

smooth like this ^

In theory it should be easy as just tweaking the string pulling but in practice it doesnt work at all

[ocean wren]

ah right, so you can just take the path points and fit a spline to them, just look at different types of splines and continuity

hermite, bezier, etc.. different types of splines have different properties in terms of "smoothness"

[celest python]

so basically I just need to curvature the given path point array

directly in string pulling function

[ocean wren]

Well, look at that video, they're obviously doing string pulling, but they're also calculating a spline with specific continuity (how the in and out of the spline points smooth out)

[celest python]

I guess they integrate the "spline" directly into pathfinding though

[ocean wren]

I wouldn't think so

its probably a post-process thing after the path is calculated

[celest python]

so this is literally just projecting path points to navmesh by spline points?

[ocean wren]

I mean you can do a string pulled pathfind.. then take that and spline it.. makes more sense to me

[celest python]

that's what I thought too

but I'm confused about "spline it" part

[celest python]

does it suppose to be a math operation happens with path points or just like how I described?

[ocean wren]

Well, simplest first step, is to take the string pulled path and fit a bezier curve to it

see what that looks like

[celest python]

ah, alright. I'll see by iterating

[ocean wren]

I mean, no idea if that's actually it.. but that's where I'd start looking at it

or just contact the guys who did the presentation and ask 🙂

[celest python]

Never did such a thing like that before tbh 😄

But if I end up being so desperate I'll definitely try that too 😄

[ocean wren]

Well, why the hell not? 🙂

I mean if they did a GDC demo, they're open to talking about it

[celest python]

fair point

I'll work on this next week, and if it works out, I'll open source it

[ocean wren]

most AI guys love talking about the crap they've worked on 🙂

[celest python]

#cpp recommended this by the way: https://martinfleischmann.net/line-simplification-algorithms/

probably you're familiar with it

since it's used on openCV

I guess taking the "spline interpolation" math from engine would work

if I can get the constraints work

🥳

[ocean wren]

the important part is the continuity.. the bit that determines the incoming and outgoing smootness, different spline types have different guarantees and whatnot

I'm sure wolfram has info on different splines

[celest python]

When you mean continuity, do you mean the math term?

[ocean wren]

its the term for how continuous the path is through the point

[celest python]

I should have attend to those math classes on at high school 🥲

but at least I have a good chance to practice

[ocean wren]

the idea is that you want a spline that passes through the points, but you also want to have a smooth transition as you pass through the points (so you can maintain speeds etc)

[celest python]

now I see why it needs to hug the wall

otherwise it looks so rigid

it also makes sense death stranding constrained the spline to hugged wall vertices

[ocean wren]

yeah, you want your spline to pass through the points where they hit the corner points right?

so you do a string pulled path, then spline that so it hits the control points and passes through them smoothly

[celest python]

right, it makes so much sense now

thanks zoombapup

[ocean wren]

Look forward to seeing it in action 🙂

Hmm, might be catmull-rom spline I was thinking off

couldn't remember which ones pass through the control points or not

might even be a chordal catmull-rom spline

[ocean wren]

https://lucidar.me/en/mathematics/catmull-rom-splines/

[lethal helm]

what could cause a MoveTo to just randomly pause for a couple second like this

[celest python]

assuming from the images, thats the one

[celest python]

check visual logger

[lethal helm]

meh that doesn't really tell me anything

going to debug the moveto cpp to see why it's having mini-aneurysms

[winter moon]

my AI event on perception updated never gets called despite the actor receiving damage

enemy controller:

i also have the simuli source on the actor doing the damage

[midnight scroll]

Curious if someone with more experience in Detour AI might know more correct settings for it? Am having trouble where an AI will not detour around stuff if it's move starts up against other agents and it needs to go around them. If it starts further back it seems to detour around them fine.

[lethal helm]

looks like it was some sort of redundant collision check confusing the moveto

not sure why it was even bothering to check collision when thats like the whole point of spending all the processing power on pathfinding

[mortal umbra]

@uneven cloud oke thanks anyway

i will see if i can do it with the information given

[uneven cloud]

Are you sending a damage event when taking damage?

[winter moon]

it was several issues. component had to be on the controller not actor, and i had to report the damage on applydamage, not anydamage

[shell horizon]

Okay I can't believe crowd works like this in 4.27.2. What even is this pathfinding. But it goes that way first with the basic pathfinding too.

The other thing I don't understand is this value for crowd manager:

It sometimes offsets the path point, but sometimes it doesn't and sometimes it does, but to the wrong direction.
This makes the path sometimes go out from the navmesh, but crowd avoidance makes sure it doesn't go out of the navmesh and it just starts the shaky movement.
This is really how it is in this version?

[mortal umbra]

how do i get the player speed from within an AI controller without casting?

is that possible?

[harsh storm]

It's possible, sure, but also convoluted. Why don't you want to cast?

[celest python]

Casting is very cheap by the way

if someone else told the otherwise

[midnight scroll]

Still wondering about detour AI, or if I should look into something different. Detour only seems to work for flowing AI. But in the case where I have grouped up agents, if I tell one to move to the other side of the group, they just run against the group and get stuck. If I move them away from the group a little and then have them move to the other side of the group, they correctly detour around the group.

I have two cases. Bottom and top. These are on default settings. Have done nothing but override the AIController with the detour controller. Have tried messing with several settings in the detour project settings but nothing seems to affect this behavior.

Top: Will try to detour, but doesn't seem to detour enough. Gets stuck on non moving agents.
Bottom: Will detour around the group, but still brushes up against the other agent, slowing it down oddly, but reaches it's point.

[midnight scroll]

<@&213101288538374145>

[fair zealot]

:no_entry_sign: eikonoklastes#9758 was banned.

[hearty niche]

are old blackboard values retained if I run a different behavior tree or purged?

[winter moon]

I’m trying to create an aggro system similar to WoW. Is there a way to store all actors that hit a mob in an array, and keep track of their damage on some sort of timer? Probably as a BTService

[ocean wren]

Turn off the colliders for the agents so they don't use collisions against each other. Have a "push" force on the moving agent so they can push other agents out of the way with more force. The default detour setup in UE isn't really correct.

[midnight scroll]

Hmm. Not really what I'm looking for. I may not go the detour route. Mostly just looking at having the one agent able to path around any non moving units, but I don't want them to be shoved by movement.

[uneven cloud]

If you have a reference to the player, you can cast to the character and get the movement component. Casting to a code class doesn't create a hard reference like it does with BP.

[keen crow]

i haven't played WoW but regarding

Is there a way to store all actors that hit a mob in an array, and keep track of their damage on some sort of timer?
have you considered using UAIPerceptionComponent method to get all valid damage events? The data there has the amount of damage and expiration info

[uneven cloud]

^ that. You don't need to store an array of the actors that generated threat as the perception will handle it for you. Just need to send damage events and query the information.

[hearty niche]

guys?

[uneven cloud]

I believe so.

[hearty niche]

Retained? Purged?

[uneven cloud]

I believe are retained if the different behavior trees use the same blackboard. But that is something you can easily test with the visual logger.

[hearty niche]

I am trying to keep AI states in different behavior trees that's why I asked.

[uneven cloud]

I do the same thing. If you use run behavior dynamic, they are retained. But the trees have to share the same blackboard asset.

[hearty niche]

What would be a good way to send the AI into search mode after they lose sight of the player. I was thinking of storing the player location centrally and using it as a last seen location. This would update only if at least one baddie can see the player.

[hearty niche]

Same BB so I guess I'm fine.

[harsh storm]

That's pretty much all I do. Get to last known location - if they don't see the player, find an area or two in X radius to check. If still no find, go back to wandering or w/e

[hearty niche]

I mean making the AI go into that state.

[harsh storm]

I do it if my enemy loses sight of the player

[hearty niche]

I was thinking of separating AI into zones and making a zone manager that stores this stuff.

[hearty niche]

individually? It would look weird if one guy is searching while the other shoots.

[harsh storm]

Meh - I have a subsystem that I created that helps with a bunch of AI stuff.

[hearty niche]

I do it for reserving locations for pathfinding

[uneven cloud]

Why not use the perception component? You can query it and get the last perceived location.

[winter moon]

So every damage event can be queried? Does it have refs to the actor that damaged them?

[hearty niche]

problem with subsystem for this is that it eould shake up all enemies in the map while I would want enemies across the map to remain unaware

[winter moon]

I ask because I’m also trying to divide exp gained by the amount of players that hit the mob as well

I probably need to look into ai perception more as I’m assuming it can handle all this data

[uneven cloud]

You can query all the perceived actors and check if they are perceived by the damage sense. You can also get the perception into for a given actor.

[winter moon]

Does damage sense not reset to a new actor on every damage hit?

[hearty niche]

I am trying to keep the state of all AI controlled by a director class. If all AI lost sight of the player send them all to search.

[winter moon]

It’s not like sight where it can perceive multiple actors

[uneven cloud]

It stores every target with every sense that it's perceived with.

[winter moon]

Nice

That makes my life easier

[uneven cloud]

It also will clear it after it expires. You can set how long it takes to expire on the perception component.

[hearty niche]

The problem with this approach is one AI could end up in search while others are still shooting the player.

[uneven cloud]

Does it or are you solving problems that don't exist? I find that setting a max age of a reasonable time actually keeps them all engaged.

Your approach will actually cause edge case bugs where AI that should lose sight from the player perspective won't.

[hearty niche]

I want to start search a few seconds after all AI cannot see the player. Even if one AI can see the player no AI would go into search. I might be wrong in this approach but what is wrong with doing it this way?

I am talking in theory considering there will also be in context voice lines like ai saying find him and stuff. Imagine if one guy says find him and the rest can see the player? @uneven cloud

[uneven cloud]

It's an over complication which generally cause bugs. It's also likely unnecessary. The trick with AI is to try the simple version first and only do complicated behavior when needed.

With your solution, you will have AIs that should start searching from the player perspective, but don't. This makes the player feel like the game is cheating.

[hearty niche]

So I was thinking if the AI loses sight, I could use a wait and get the last known location and make the AI walk there wait again and shift to search?

[uneven cloud]

With your example you can have the others say: they are over here. Or check if others see the player before saying the line. It's a simple fix.

[harsh storm]

If this were a problem for me, I'd just use my AI subsystem to query it and ask if any AI in my area can see the player, if so, get the location, if not, then that means no one can see them or are actively in the hunt state or w/e.

[hearty niche]

hmmm so start dialogues for find when the AI enters search?

[hearty niche]

that is my first idea which Luthage says might not be a good one so I'm curious

[harsh storm]

Go ahead and do it. See if it works for you. Tweak from there.

[hearty niche]

I think I'll make the singleton approach plan B. The perception one might be easier to prototype

[uneven cloud]

Start simple, fail fast.

[hearty niche]

Also, is there any way to check if the bttask variable blackboard key is of a specific data type. Say, I want to do location logic but my param could be either vector or actor?

I'm tired of making actor and vector counterparts.

[harsh storm]

I don't really see where luth is saying it wouldn't be a good one. The difference is I talk about doing things in the area, where your first idea sounds globally.

Which if it's global, yeah, could be lame.

[uneven cloud]

There is in C++. I don't remember what it is, but if you have source one of the built-in tasks does it.

Simple way is to get it as an actor, if not valid get it as a location.

[hearty niche]

hmmm wont that be expensive per tick? I am using it for decorators in some cases.

[harsh storm]

Don't guestimate. Evaluate.

Do the simplest implementation first

[hearty niche]

☹️

okay

[uneven cloud]

Premature optimization never works. There are cases where it would be expensive, but generally no.

[hearty niche]

alright

[celest python]

fwiw one day I was very bored and decided to calculate how much time does it take to get a value from BB and value was around nanoseconds level

[harsh storm]

So slow

We need picosecond speeds

[uneven cloud]

If you are getting it by name and you have hundreds of blackboard keys, it can get costly because it needs to iterate over all of them. But yeah mostly cheap.

[harsh storm]

FName's are cheap to compare

[celest python]

TMap lookup adds up a little bit though

[uneven cloud]

The TMap only works if you are looking it up by key ID, not name.

[distant snow]

I'm pretty new to Behavior Trees and all but... should this node be reappearing after I delete it and hit "Save"?
I restarted the editor already.

[celest python]

I got this issue on 4.27

I guess this is a serialization error

try plugging it to something else

and save

then remove it again

[hearty niche]

what about key selectors?

[distant snow]

Ah, so it seems to be a feature of Simple Parallel. Even with a brand new Behavior Tree, that "Wait" will appear when I hit Save.
But yeah, I've had plenty of fun in ABP with serialization errors, so that's what I figured too 😄

[celest python]

Honestly UE has some badass reflection system but serialization errors are everywhere 😄

Corrupted assets, corrupted properties, broken BTs etc.

30% of my development time goes to fixing serialization errors

[uneven cloud]

Key selectors look up by name in BP, but can use key IDs in C++.

[mortal umbra]

@uneven cloud i already figured it out

i could have done either get player character and then iscrouched or what i did now is just get the stimulus and then stimulus strength

[hearty niche]

any good suggestions for how to select cover points? I'm confused between doing cover detection manually vs procedurally.

Also, how do I make the AI not go into search if it is in cover?

[uneven cloud]

I would use the EQS.

How to make them not search is entirely dependent on how you set things up. There isn't one way to do anything.

[dense condor]

Can you use Launch character node for AI?

[mortal umbra]

if i use EQS do i need a testing pawn for it to work?

[brittle lynx]

No

[mortal umbra]

@brittle lynx oke, thanks

[hearty niche]

how do I do this?

[tawny forge]

has anyone done obstacle avoidance without using the NavMesh/Pathfinding? Basically I have a very simple AI that moves to a target location in a straight line, but I want it to be able to detect static obstacles/walls and go around them if possible.

[brittle lynx]

If it’s simple enough, you could get away with something like the Obstacle Avoidance steering behavior: https://gamedevelopment.tutsplus.com/series/understanding-steering-behaviors--gamedev-12732

Why not just use the navmesh though?

[tawny forge]

Thanks I'll have a look at those. The reason I'm not using the NavMesh is because the pawn is actually possessed by a PlayerController (so it's more of like a player assisted AI than pure AI). I also need to be able to tell it to move around in places with no nav mesh, such as on the back of a giant flying monster etc.

[hearty niche]

Is it possible to run a BT sequence by force on some event ignoring the last sequence on a specific event? Something like hearing gunfire?

[winter moon]

How do I get all the actors that are sensed by the AI?

[mortal umbra]

@uneven cloud do you know how i can use these functions?

it is from the lyra content sample

but i can't find where it gets referenced other than that

or this

it doesn't reference the weapon reload thing anywhere i can find

it seems like a better way to do it than this

which is what i currently have

the documentation doesn't say anything about it either

[celest python]

those are not even related with AI module

Gameplay Events exists to communicate with external systems outside of the GAS, and GameplayEventDatas are replicated between client and server

Gameplay Abilities listen for Gameplay Events with WaitForGameplayEvent ability task

inside of their graph

[mortal umbra]

then why is the lyra example using them to communicate within the behaviour tree?

[celest python]

When you send a gameplay event, any listener will receive the GameplayEventData

[uneven cloud]

Because it's communicating to the gameplay ability system via the BT

[mortal umbra]

so what should i do?

i am new to AI

[celest python]

Quick correction to myself: It's also commonly used to communicate between different abilities too, not just external systems

It's nothing but a fancy delegate at most simple level

[uneven cloud]

It's also used to trigger abilities

[mortal umbra]

abilities being for example reloading or shooting?

[uneven cloud]

Yes

[mortal umbra]

oke

is my way of doing it also correct?

with casting?

[uneven cloud]

You don't need to cast

[mortal umbra]

i don't?

how would i do this then?

https://i.imgur.com/rGlzLyh.png

if i don't cast?

i need the information stored in the character bp

[uneven cloud]

Oh. Yeah if you are calling a function on the character you need to cast. Don't need to cast if you are using the GAS event. It's early here.

[mortal umbra]

so i was wondering how i would use a GAS event instead?

do i even need it?

[uneven cloud]

Are you using GAS?

[mortal umbra]

no

i guess that answers it

[hearty niche]

umm @uneven cloud ?

[uneven cloud]

If you Google "EQS cover" there are a lot of results. You might want to start with the documentation: https://docs.unrealengine.com/4.27/en-US/InteractiveExperiences/ArtificialIntelligence/EQS/

[hearty niche]

last perceived location not cover. This is for lost sight.

[uneven cloud]

You can query the perception component for lots of information. Such as if the AI is currently perceiving a target, if it is currently perceiving it with a specific sense and the last perceived location. If you are using only BP, I'm not sure how much is exposed.

[hearty niche]

its fine if I have to use cpp but how do I query stimulus locatiom

[uneven cloud]

I typically use the function GetActorInfo which returns FActorPerceptionInfo

The FActorPerceptionInfo has a lot of functions. You can get the last known location or the last known location for a specific sense, such as sight.

[hearty niche]

I found the actor perception info. How do I know what stimulus it is?

[uneven cloud]

What do you mean? You can ask for information based on a given sense, but you need to get the sense ID. I believe you can get the ID for AISense_Sight from the perception component.

[hearty niche]

you mean the index?

[uneven cloud]

The FAISenseID

[hearty niche]

then how would I query via sense ID?

[uneven cloud]

There are a lot of functions on the FActorPerceptionInfo that take in a sense ID. Such as IsSenseActive is used to check if the target is currently seen.

[hearty niche]

Get Actor Info is cpp only?

[uneven cloud]

I do not know how much is exposed to BP.

Looks like for BP it's Get Actors Perception

[hearty niche]

How do I check if it does have a stimulus info or not?

[mortal umbra]

@uneven cloud why does EQS never work for me?

this is annoying

i have copied it exactly from the Lyra thing

it doesn't work

[uneven cloud]

I can't possibly know that. The visual logger will give you some hints.

[mortal umbra]

the testing pawn?

that doesn't work either

the plugin is enabled

can i share my screen for a bit? maybe you see what i am doing wrong

[uneven cloud]

In C++ it returns a pointer. If there is no perception info, that pointer would be invalid. I don't use BP enough to walk you through it.

[uneven cloud]

Can't. I have meetings all day. The EQS is being logged out to the vis log. All you have to do is run it and walk through what is actually happening.

[mortal umbra]

@uneven cloud how can i enable the visual logger?

i can't find it in the window menu

To activate the Visual Logger, use the menu in Windows > Developer Tools > Visual Logger. In Unreal Engine versions prior to 4.7, enter the console command "VisLog". If you are in the Editor, the Visual Logger viewport will also open.

its not there in 5.0.3

[uneven cloud]

Tools -> Debugging

[mortal umbra]

@uneven cloud i have this

the result should be the blackboard key TargetActor

it however says it is none

[quaint radish]

How would I have 2 conditionals within the same composite that act as an OR statement?

Like I want a Selector to run IF the player is in Attack or Patrol state.

I currently have a conditional (decorator) checking for idle, another for not idle.

[quaint radish]

currently decorators act as an And

requires both to be true

how do i make it so it will run if one of the decorators are true?

[opal crest]

Use a composite decorator. Then you can configure how the logic of your decorators combine inside the mini node graph inside the composite.

[fleet cradle]

Hey everyone, I wanted to make 2 characters one is playable and the other copy the same movements, should I use AI for that?

[fleet cradle]

got it working, if someone is interested to know

good call

[hearty niche]

is it possible to forget a stimulus by sense? I could only find forget all but that could cause problems in case I also have stuff with multiple stimulus

[rich veldt]

Hello, I'm lower-mid-level understanding with AI and the navigation system, and I'm looking into researching standard methods for having an AI understand that a closed door may be blocking the line of their path (and perhaps also allowing Find Path To Location Sync to "pass through" closed doors in the process), and subsequently move to open it when at a certain distance on the path to it, but I'm not having an easy time finding definite examples. I've found Nav Link Proxy but unsure if it can be used for this just yet. Any suggestions for things to look into/learn about some more?

[celest python]

Back three years ago I did spawn a NavMeshModifier thing that makes AIs impossible to move to a specific location to placed when doors are closed

then called find path to door's location

then also to be sure checked door's state if doorClosed

if two of them matches I told AI that the door is closed

there might be more modern and better solutions

but for a quick prototype this one worked out without any problems

[ocean wren]

NavModifierVolume

set a new navarea class for the volume.. change the value for the path through depending on the door state

Rama did a video on how it all works

[uneven cloud]

There are 2 routes you can go that are pretty equal. Using nav modifiers and area classes like others have suggested is a good option, but it takes a bit more setup for opening the door.

Using a nav link proxy is another option and makes it a bit easier to trigger the door to open. Point links use nav areas (C++ only). Smart links (BP or C++) don't, but are more expensive.

[rich veldt]

Awesome thanks all, I'll try both methods for learning.

[mortal umbra]

what is the best way to get a variable from the player bp to an AI controller bp?

i know i can cast, but i rather not

[hearty niche]

Any idea why I am getting this error on build? It seems to be something AI related

UATHelper: Packaging (Windows): LogInit: Display: LogOutputDevice: Error: begin: stack for UAT
UATHelper: Packaging (Windows): LogInit: Display: LogOutputDevice: Error: === Handled ensure: ===
UATHelper: Packaging (Windows): LogInit: Display: LogOutputDevice: Error:
UATHelper: Packaging (Windows): LogInit: Display: LogOutputDevice: Error: Ensure condition failed: ObservedKeyNames.Num() > 0 [File:D:\build\++UE5\Sync\Engine\Source\Runtime\AIModule\Private\BehaviorTree\Decorators\BTDecorator_BlueprintBase.cpp] [Line: 68]

[uneven cloud]

You have an error with one of your BP decorators.

[uneven cloud]

You can cast to the player or cast to an interface.

[hearty niche]

seems like it needed a Blackboard Key selector set to self actor as a workaround. I was using possessed pawn directly from the function override. Engine Bug?

[random jay]

Anyone have any good talks on AI?

I'm curious about melee AIs in games like dark-souls/elden-ring/zelda etc

[celest python]

No idea about Zelda but souls games are notorious with terrible hit boxes

[celest python]

Check pinned messages

[celest python]

Something like this would work: https://twitter.com/Wabbaboy/status/1548299662754402305?s=20&t=eAyApM3szUIz8nBUql12sA

I was saved it a while ago to recreate but never had enough time

[mortal umbra]

how can i have a CCTV camera detect a player and communicate it to an enemy AI?

should i just cast or interface it?

[harsh storm]

Do you just not use anim notifies or something? 😅

Anim notifies are one of the reasons I like Unreal, lol

[celest python]

I use it but it's also tracking the bone rotation properly and somehow able to draw debugs in anim editor which I couldnt find how back a few months ago

There is a relatively small boilerplate process to do it like how it's shown in the tweeet

Oh btw I forgot to read your ping in lounge, I was outside

Will read it in a few mins

[signal olive]

So it turns out nav links can be pretty damn buggy

Especially if you're using blueprint

Which is frustrating for someone like me who basically needs to make use of nav links, especially smart links

[dim lynx]

could someone advise me? there will be 100 AI in my game, each of them I would like it to be so that if a player approaches the AI ​​close enough, the AI ​​will gradually approach it every 20-50m >>> wait 5 sec >>> go 20-50m >> > wait 5sec and until it comes close to the player at 10m

i do it with Overlap Sphere and task but i think will be better do it with EQS

but I don't know much about EQS

here is 50 AI

[novel flower]

@dim lynx What do you use the overlap sphere for? To detect if the player 'is close' to the ai? Maybe you can use the perception system in stead?

[lament maple]

How do you see the bounding box for a nav mesh volume? I see the green area for debug with "P", but not the extents to tell how far I should scale out the volume

[harsh storm]

It shouldn't scale outside of the volume

As for actually seeing the lines, it might be that it is scaled to the point where it is off the map. Try and find it in the world outlier and select it. See if you can actually see it when it is highlighted. Perhaps the lines are too thin for you.

[lament maple]

I'm a dummy, it was just so big I didn't realize the bounds were outside my viewport

[lament maple]

Yes exactly thank you 😄

[rotund lynx]

NavMesh is driving me crazy! I try to fine tune values but nearly all the time when I change value and mesh shows agent error, I change value back and error won't go. I have to delete the navmesh and start from scratch. Anyone else experience this?

[brittle lynx]

Why don't you just check distance between the player location and character location on tick? Even without any optimizations it will be massively faster than overlap checks with huge spheres.

[ocean wren]

Hmm, it might not actually.. the broad phase of the physics engine might well prune out a lot of potential overlap checks as it does a sort and sweep

[brittle lynx]

Given that UpdateOverlaps takes 22ms per frame in the provided screenshot, I doubt that.

[ocean wren]

it'd depend on your scene and how many things are moving

and the size of the spheres

and if you sorted them into a seperate physics channel

I guess you could do the sweep and prune yourself.. its not hard

[noble totem]

Hi. I have an AI Pawn (a flying object) moving inside the navmesh volume at a 1000 units above the navmesh plane. Is there a way that I can tell it to detect the navmesh below it? It works fine when the plane is at 100units below. Antything further than that and the AI stops working.

[harsh storm]

I mean, when you get the new nav location, just take the X and Y parts of it. Then write some custom stuff to monitor the Z.

[noble totem]

Ok there are a lot of nodes that can detect the navmesh but none that actually says use it. Using the FloatingPawnMovement component btw. Also I don't need my AI Pawn to change height. Just need it to move on a plane that is 1000units above the navmesh plane.

[quick oak]

Look for ProjectPoint in the ANavigationData actor. You can specify extents that will stretch those 1000 units to find the navigation "below". YMMV if you have multiple "layers" of navigation- it should find the closest to the origin you provide though.

[noble totem]

Will try that, thanks.

[rotund lynx]

@noble totem depending on the task, you can have the transparent non-collidable plane 100 units below for nav

that's what I've done for pathing in empty space

[noble totem]

Actually it was way simpler than that. If you go to the RecastNavMesh actor in your level, go to the the Query section and increase the Vertical Deviation from Ground Compensation to the height your actor is, it works like a charm.

Thanks for all your input that led to me figuring it out. Hope this helps someone.

[alpine path]

I need rotate my NPC to target actor and keep looking at it. I use RotateToFaceBBEntry but it resets focus when finished. Do I need to implement custom node (so it never finishes) or there is better way?

[dim lynx]

how to do so that the AI approaches a given object every e.g. 20m

I mean just approaching the object

I have Distance to this object

and i have location of this object

and i have AI location

Can someone help with the calculations?

all is valid

could someone help with the rest? 😅

I have no idea for this

[alpine path]

You can use node moveTo

[dim lynx]

i used it

but I want him to go this way every 20 m

target is 100m

he should go 20m >> wait >> go 20m >> wait >> go 20m >> wait .....

and finish

if 100m complete

sorry

finish if distance is < 10

[alpine path]

What is problem? Switch to wait every 20m from Move state to Wait state

You just need local variable to mark location when NPC started movement then every tick compare to 20m

[dim lynx]

but in behavior tree

in task

I would prefer to calculate the distance to be traveled and set in a variable the location to which it is to go

[dim lynx]

the goal is only to set the first location where it will wait... everything is done in the behavior tree, when the location is set goes to this place and waits, the problem I only have with the location.

these green points may be random, but all the time towards the player

someone can help? i want do it with only calculation

I have no idea for this

[uneven cloud]

You can also set focus. There is a built in decorator that sets default focus.

[uneven cloud]

I would use the EQS to generate the points. The problem with doing it only with a calculation is dealing with situations where it can't go in a straight line.

[uneven cloud]

If you really want just the calculation: you get the direction vector by subtraction of locations. Target location - AI location. Then normalize that vector. Multiply that by how far you want to go in unreal units (centimeters). That will give you a point on the red line.

[young thorn]

I'm so confused. I'm looking into the AIPerception system and I can't find any way to consume its events in C++. Everything is about doing it in blueprint. Anyone familiar with this?

[uneven cloud]

The same events that are in BP are in C++. What are you trying to do?

[young thorn]

Ah, I see. For some reason I assumed I shouldn't touch events that are exposed to the blueprint interface. 😝 I got it working. Tyty

[fallow gust]

Hi. I have been developing a plugin to extend the AIModule in UE5. It takes a hybrid approach to Utility AI and GOAP with a focus on enabling AI behavior design through a data driven editor. I feel as if it has been maturing quite well from the AI implementation side of things as I have been integrating it into my upcoming project. The way it works is that it uses Utility AI concepts to determine a goal from a generated state provided by a new type of Brain Component. Once a Goal is determined - using either dual utility or absolute utility - it creates a plan of actions to accomplish the determined on goal via GOAP. Right now the only planner is AStar, but I am thinking of adding a Fringe search to reduce memory load when there are more agents (but it does cost on CPU) and perhaps other types of searches for creating plans. I would love to see if there is any feedback as there is a short tutorial to help you get started. https://github.com/hollsteinm/ReasonablePlanningAI Currently, for the most part, it has a lot of feature parity with Behavior Trees as far as making your AI controlled characters doing stuff in game. It also has integration with Environment Query System natively, and a small tutorial on how to (in C++) integrate with Smart Objects Module or any other AITask based plugin/module. I also have a ton of automated tests on the basic stuff so I feel it is stable enough, but it is rapidly going through a lot of changes.
It is free for anyone to use while it is in this development stage (just clone it into your Plugins folder) and I would love to hear some feedback after you have tried it out. Figured it would be a fun way to add a reactive and thoughtful AI using something other than state machines, events, and/or behavior trees.
It is also integrated with Visual Logger for debugging and plans the AI is executing are shown under the debug overlay (using the ' key in game/PIE)

Also, this project had me deep in the weeds in the AIModule... so open to help out with any questions on that as well, especially the C++ side of things.

[dim lynx]

I don't know too much how to do it in EQS, I don't understand this system

[fallow gust]

The quick start guide shows you how to generate a point and travel to any areas X distance from player and has the AI wait. Sounds like your use case. https://docs.unrealengine.com/4.27/en-US/InteractiveExperiences/ArtificialIntelligence/EQS/EQSQuickStart/

[dim lynx]

thanks

[celest python]

I wanted to take a quick look and provide feedback but it looks huge lol. At least I can say idea behind the implementation looks really good. If I can find time from daily job I'll try to read source too. Thanks for sharing with community

[fallow gust]

Thanks! It was a ton of work over the past 9 months and still growing especially after being yanked out from my upcoming game and refactored. For a slim reading I would ignore Composer directory as most of that is just objects for data driven configuration, with the exception of Composer/ActionTasks which are similar to BTTasks.

[dim lynx]

ok, I used the EQS more or less, I did the same, just a little different, but it works the same, and how to make it go towards the player every 20m?

[dim lynx]

someone can help?

[uneven cloud]

You want to generate points around your agent instead of your target. Then test the points on direction to target, distance to target and has a path to both target and agent.

[signal olive]

Any suggestions on how to make an AI react to player actions? Like, say, if there is an incoming projectile or the player has triggered a melee attack animation

[fallow gust]

There is a Behavior Tree decorator called Cooldown, use that on the node or the parent selector. Assuming 20m = 20 minutes: Engine/Source/Runtime/AIModule/Classes/BehaviorTree/Decorators/BTDecorator_Cooldown.h

[fallow gust]

I think the perception system would be a good place to do that. https://docs.unrealengine.com/4.27/en-US/InteractiveExperiences/ArtificialIntelligence/AIPerception/

[signal olive]

Well, incoming projectiles are easy enough to identify, but not enemy animation types

Also how the hell do you set up teams/factions, the senses imply such functionality but I've never seen it

[uneven cloud]

You need to tell the AI about it in some way. I use the perception system for it, but I don't believe that you can add new senses if you are only in BP.

[uneven cloud]

The Generic Team Interface

[signal olive]

That gives nothing to work with

[dim lynx]

I don't understand a bit

in QC_WechPlayer... i get all actors of class BP_Character

Could it work that it checks how many points are up to the player?

and e.g. if it is less than 5, then it moves every 1 point and 5 seconds wait for each point

[uneven cloud]

It gives you a term to search for. The perception system uses an interface that is on all AI controllers called the generic team interface.

[dim lynx]

if 6, 7 ... etc, then not move

[signal olive]

Right, it's just that there's no obvious way to set teams

[uneven cloud]

In BP there is none. In C++ it's SetTeamID.

[signal olive]

I'm really starting to consider using C++ at this rate.

[uneven cloud]

There are a lot of things, especially in AI, that are just not exposed.

You can just not use the built in filtering and use your own team functionality after perception gets everything.

[mortal umbra]

what is a good EQS tutorial?

i can't get it to work

it says that the vector of movetolocation is invalid in the behaviour tree when debugging

@uneven cloud

when i follow this exactly

https://www.youtube.com/watch?v=3-jDztCAkk4&list=PL4G2bSPE_8ums1OIn0tdj1xbpw_8W-GIP&index=3

it does not work

the AI does not move to the player at all

it just stands still

don't mind the shooting bit, that isn't important for this

[uneven cloud]

You could do, but that's not a simple way. You can just use tests to find the next point as they move.

[uneven cloud]

Is the EQS failing to run? Is the EQS not generating anything? Is it failing all the tests? The visual logger should give you some insights as to where it's failing.

[mortal umbra]

the EQS runs

it is just that the blackboard vector is invalid

try and follow the tutorial and see it will probably also happen to you

[dim lynx]

I just want something temporary until I learn EQS

[uneven cloud]

Does the visual logger say that it found an item?

[mortal umbra]

i think i git it to work for now

i am not sure though

[dim lynx]

this doesn't work

[dim lynx]

and i don't understand 100%

[mortal umbra]

@uneven cloud now the EQS thing doesn't give a score

it just says that the score is 0

[uneven cloud]

It's easier to debug if you add some debug drawing and debug print strings.

[dim lynx]

could you show it on blueprint?

[mortal umbra]

@uneven cloud what am i doing wrong here?

https://blueprintue.com/blueprint/dbzmxknt/

it doesn't generate EQS points

but if i change it to generate actors it does generate that

[mortal umbra]

nvm

i figured it out, i missed something in the class defaults

[uneven cloud]

No. Debugging is incredibly easy with all the UE tools. You need to verify all your actors are valid and draw the location on the screen. Draw debug sphere is really good for drawing locations.

You are also multiplying by 20 which is a very small number. That's 20cm not 20 meters.

[fathom sun]

I'm trying to add perception component to my ai controller component through C++, but for some reason I can't change class of a sense in their config (in BP). Whenever I add a new element to senses config it's None by default, and if I try to change to anything else it just doesn't, any ideas what may be wrong? The only thing that I did in C++ is adding a declaration in header file like that

UPROPERTY(VisibleAnywhere, BlueprintReadOnly, Category="Components")
TObjectPtr<UAIPerceptionComponent> PerceptionComponent;

and creating the component in the source file

PerceptionComponent = CreateDefaultSubobject<UAIPerceptionComponent>(TEXT("AI Perception"));

[uneven cloud]

You need to add an edit specifier to your UPROPERTY. EditAnywhere or EditDefaultsOnly

[west lark]

I am trying to make an IA vehicle that runs off spline, but instead of using the time it takes to complete, i want it to have a set speed, so i can have multiple vehicles going the same speed right noext to each other. How do i do this?Please help

[leaden mural]

I have never used a perception component in my life lol

did you think I knew about this because of #mass ?

sorry buddy 😔

what do you mean by resetting its peception anyways?

[harsh storm]

ForgetAll()

[languid mirage]

hi all, I'm using pawnsensing, I'm trying to make AI character 1 to follow and shoot only AI character 2. I could not figure it out. Can anyone gimme a hand?

[hasty badge]

How do I make a behavior tree execute smth only when a blackboard value is changed?

[keen crow]

Try implementing IGenericTeamAgentInterface:: GetGenericTeamId on both ai controller and character and return different value for different NPC types

[keen crow]

Have you tried putting this something under a selector node with a blackboard decorator that aborts lower priority?

[uneven cloud]

Blackboard decorators have observer aborts that can be set to result change (the result of the decorator) or value change.

There is a function you need to call when changing teams. RequestStimuliListenerUpdate on the perception component. Using ForgetAll will only make the AI forget the targets, not reconfigure for a different team.

[hasty badge]

I'm trying to make a character go to where I right click, follow the cursor when I hold RMB, and not call the move function when the target destination doesn't change (it makes using nav links impossible)

idk maybe I don't need to use the behavior tree for that at all, I'm very new to AI

[hasty badge]

How do I edit navmesh so characters can walk on narrow surfaces like this?

[celest python]

you probably still need RequestStimuliListenerUpdate though

[rose bramble]

hello, when i launch a game, behavior tree is extrimlee flickering

i mean market parts is just bright green and grey in next second

i mean i watch a tutorial, everybody have stable bridght green

[cursive wave]

First of all, it is unnecessary to use two sequencers, according to the picture you posted. You can delete one. If you're having a flick in the behavior tree, your task is probably failing. Can you show the contents of the task blueprint?

[rose bramble]

npc is space ship

gravity is 0

so i move it by velocity

also NPC is Character blueprint

[cursive wave]

Don’t you get blueprint runtime error?

Because character movement ref is null I mean it is not assigned to anything. You should call get character movement component from As Character pin and assign return value to character movement ref

Also you can use breakpoints to debugging nodes. Right click to node > add breakpoint or select the node and press F9

[rose bramble]

i dont think it failed, cause nps is flying to me

could it fail and fly to me anyway ?

[cursive wave]

Okay so, I think it’s flicking because there is nothing else to do. You can add wait task to left your task’s to test

[rose bramble]

yes, it not flickering, when it do wait task

but it making Fly_To_charecter task is less than a second. is it ok ?

[rose bramble]

oh wait, you actully was correct. how to solve it ?

[cursive wave]

Yeah true. For example Move To task takes time to happen but your task finishing immediately

[cursive wave]

Drag from As Character pin and call get character movement and assign the character movement ref to this component

[storm zephyr]

hello i'm trying to use a box component behind an enemy as a dynamic obstacle to serve as a way to do avoidance, and it works good enough for what i need (and better then the solutions for avoidance like RVO and the CrowdFollowingComponent), the only problem is that it hurts performance quite a bit, is there a way to reduce the impact on performance? maybe by making smaller or bigger cells on the nav mesh?

[cursive wave]

Can you show how did you use to box component. For example overlap event

[rose bramble]

it is still flickering, but no error message already

[rose bramble]

so it is not flickering when it make Wait, and make my custom task like less in a second. and that should be ok ? am i correct ?

[cursive wave]

After you set the velocity, you can look at the distance between the player in the tick event and run the finish execute accordingly. this way your task is not finished right away.

[rose bramble]

but there is nothing bad that it execute less in a second since it fly to me ? or i miss something ?

or it eat a lot of resoursce that way ? like because it execute every second ?

[cursive wave]

Yeah it’s casting almost every frame