gameplay-ai

[naive sorrel]

The same attributes and not respawning the whole of AI's is the problem i'm having

Like 1 AI out of the bunch is killed, I want it to respawn 60s after destroy with the same attributes he had set in the DataTable

[misty wharf]

You would need to store the information about what needs to be respawned somewhere. For example, when the actor gets destroyed, you set a timer to spawn a new one

You can store the spawn-related settings in the actor and get the values when it gets destroyed and store them somewhere else like in a struct for spawning it back

[naive sorrel]

Can't I use the already existing Data Table ?

[misty wharf]

You can, if you can identify which row you used the last time

[naive sorrel]

hmm

Can I add an ID sort of in the Struct ?

[misty wharf]

Maybe you could store the row name it was spawned from on the actor or something and use that when respawning

[naive sorrel]

So like cast and set a variable inside the AI that was spawned

[misty wharf]

For example yes

If your actors have a shared base class, you can also put it there, and set it to expose on spawn which would make it available on the spawn actor node directly

[gusty trellis]

hello, i use an anim montage with motion warping when i interract with the npc for a dialogue and it block anything about the npc movement after that despite the task trying to run any idea to solve this problem ?

I found the problem apparently its not a good idea to desactivate colision on the fly on a npc trying to do task lol

[naive sorrel]

hmm

very nice

[naive sorrel]

like this ?

[misty wharf]

Yeah something like that would work

[naive sorrel]

Okey Thanks I'll flesh it out

[naive sorrel]

Seems to work :=) Ty

[raw tundra]

I'm having the weirdest issue.
My shield alien enemy(the one with the bubble) isn't moving after spawning, unless I add a test spawner that spawns a bunch more of them(Shield aliens).
I tried upping the spawn rate of Shield Aliens on the original spawner, just to see if it's the number of active shield aliens, and it's not that. They all end up standing still.
The test spawner doesn't really do anything that should affect AI like this. It spawns enemies in the exact same manner as the main spawner.

Spawn logic for test spawner:

Spawn logic for main spawner:

Any suggestions? :/

[misty wharf]

You need to debug what's happening. Start by verifying it's being possessed by the controller correctly, and after that you need to check whether the logic it's supposed to run is running, and so on

[raw tundra]

I've been debugging what I can. But it shares its logic with all the other AI, which is working as intended, so there's a lot of entities crowding up the debugger.

The only difference I can think of between the spawners is that one of them was dragged into the scene, while the other is spawned from code. I know spawning from code doesn't add things like default controllers(I do that manually), so maybe there's something else it's not doing as well.

[misty wharf]

If having too much stuff is making debugging difficult, then you can make a new level just for testing this without any of the extra stuff

[raw tundra]

Like this. There's no reason for my spawner to have a controller though. But maybe it needs something else?

[misty wharf]

There shouldn't be any differences that would affect it spawned vs placed in level besides the controller possession thing you mentioned

[raw tundra]

Good idea. I should probably do that anyway.

[misty wharf]

If the spawner is not a pawn, then it can't really have a controller anyway... at least I probably wouldn't make spawners pawns :)

[raw tundra]

It's an actor. 🙂
Not meant to move around on its own. 😛

[misty wharf]

Yeah then it shouldn't really have anything that would affect it in that sense

However there can be some differences in initialization order and some things like that for things that exist in level vs spawned at runtime

Whether that would affect it it's hard to say as that would be entirely dependent on how your game works

[raw tundra]

I made an interesting discovery.
So I moved the destination for all the AI (see picture), out in the open. The center of it was inside the barn, which does affect NavMesh. After doing this, the Shield Alien started moving without the test spawner.
I think what was happening was that the Shield Alien, being really big and having a big AgentRadius, did not fit inside the barn. But maybe the little ones did and so remained unaffected.
The "Patrol Destination" is Instance Synced on the blackboard, so when I added a second spawner which set PatrolDestination on the blackboard, it may have also affected the original Shield Alien.
So, basically, what I think happened is that the big and bulky AI couldn't reach its destination, and so it didn't even try.

That probably means it's ignoring this part of the MoveTo though, which I'm not too stoked about.

[misty wharf]

Check in Visual Logger, it should include info on any pathfinding issues

Also you should be able to see it if you look at the behavior tree assuming you're using one - any BT MoveTo nodes would most likely immediately fail if the movement wasn't returning a valid path

[raw tundra]

You're right, it started flashing instead of glowing constantly when it couldn't reach the target.
The pathfinding in the viewport is being less than helpful though. I'm not sure what decides which AI is selected. It feels like it's whatever you're aiming at when activating the debugger?

[gusty trellis]

u can select the ia actor instance that u debug as u can with any other bp near the pause or resume, on ur video its ai_BasicEnnemie_C_51

[gusty trellis]

Hey guys, in this screenshot i fail all the task when a boolean is true with the service on the sequence but when the boolean go back to false it resume from the root my question is can i stop Pause a task and resume it and not go back to the root but still apply this same condition to all those task ?

Context its basicaly to stop the current task when i interact with an npc as a player

[dense owl]

You want to stop or pause the tree altogether?

[gusty trellis]

i want to pause the tree while i interact with the npc and then continue when im done with the interaction, but i want this to be the case with every task of the npc thats what im trying to say , i want to be able to pause the current task whatever the task is

[dense owl]

There’s stop and start logic but if you plan on doing this with every task, maybe you shouldn’t be using BTs then

[gusty trellis]

what is the alternativ ? i like the bt 😦 what would u do to do this ? is that not a common practice or normal thing in a video game (to be able to have dialogue with npc that have behaviour) ?

[dense owl]

I mean being able to stop the logic at any time is very different than saying you want to pause every single task

On interact stop logic, on end start logic

[gusty trellis]

i might dont know how to express corectly what i want

[dense owl]

Beyond that, you might need to look at State Trees if you want to go from one state to another

[gusty trellis]

i dont know that

[dense owl]

Then start with what I said

[gusty trellis]

that look awesome didn't know it was a thing

[delicate linden]

Hello, Im looking for Help with my behavior tree. I made a short vide to describe my issue. Any insight will be a huge help!

[delicate linden]

My fish cannot detect the newly spawned bobber actor for some reason

[dense owl]

Have you watched the video in the 5th pin in this channel yet?

[delicate linden]

I have not watched that one no, I was using a youtubers explainations and using the documentation (or trying to at least lol)

[dense owl]

I find that vid to be prly the best video source of knowledge on BTs

[delicate linden]

All the info carries over to UE5 1 to 1?

[dense owl]

Yeah

Not much has changed since

[delicate linden]

OK cool, I will watch it over lunch and see what I can learn

[gusty trellis]

btw i dont want to pause every single task i want to be able to pause like other but not at the same time, if for example my npc is doing a move to task and i interact iwth him i just want him to pause this task and resume it when im donne with it thats it like it is in any game, but im looking at what u advised and its look cool for what i want to do anyway thx for that im just saying in case i didnt expressed myself correctlyt

[dense owl]

What I advised was to use stop and start logic nodes

[gusty trellis]

oh i was talking about the state tree, im gonna try that too sry

[gusty trellis]

where are located stop and start logic nodes* in bp ?

[dense owl]

In the AIController get the brain component and call from there

[gusty trellis]

oh i see thx !

[rose drift]

Good day guys! Question about AI Hearing Perception

Does anybody know how Loudness in Report Noise Event works? Also, how to debug and draw Report Noise Event radius?

Additionally, in "break ai stimulus" we have "Strength" instead of "Loudness".

In documentation we see that Loudness "multiplies" max range, but I tested and 500 max range + 10 Loudness is not 5000 radius))0 But maybe I'm wrong

[uneven cloud]

Loudness affects if a sound can be heard. Something that has a high loudness can be heard if it's outside the max range. I don't remember the exact math to it, but it's not just a multiplier.

Strength is a general term that all senses have, which is why it's not "loudness".

You can see the perception debugging by using the gameplay debugger. You access that by using the apostrophe (') and turning on the perception category using the numpad.

[rose drift]

Thnaks a lot! Btw, where I can read more about it? Because in official documentation I see only specs

[uneven cloud]

I'm not sure of anywhere to read about it. I've only just looked at the code.

[rose drift]

But did somebody use loudness in development?)

I want to write some "Sound priority" system for AI noises (for example, rifle is more loud than a rock, so AI will trigger on rifle) and I wonder to use Loudness. But now, when I know that Loudness changes max range, maybe using tags is better Idea? Because when I will use loudness, I need to re-calculate max range when change smth (Loundess or max range)

[uneven cloud]

I use loudness in development. Have for several games now. I also prioritize perception based on the strength, so it works well.

[rose drift]

So I need to calculate "actual" max range based on loudness?

That makes sense somehow, but just little unusual for me

[uneven cloud]

I think you are overthinking it. I recommend you doing a bunch of tests to see how it works.

[languid meadow]

How do you guys do move states and moving while doing other things e.g. shooting?

Should I end my latent move tasks immediately - I have been ending tasks when the move was done but I realize that it blocks doing other stuff while moving.

[mild bobcat]

Not sure if it's best practice, but I have my shooting function with a decorator on it that aborts the lower priority movement when it becomes available. So it always interrupts the lower priority tasks.

[bitter helm]

@rose drift If it works like in most other engines then the sound level falls off based on the square root of the distance, not a linear distance.

[modest nest]

Hi
I need some help to understnad how navmeshes work

When I open my level in editor, the navmesh builds and everything works
However, if I open my level from another level during the game, the navmesh is not there and my AIs don't move
Is there a way I can force navmesh building when opening a level?

[lyric flint]

For me it returns true, even if the location is completely out of nav mesh bounds, or inside a wall, or even both at the same time. It never returns false.

[inner zephyr]

Did you figure something out about the entrance?
I am trying to use them in a StateTree and after I looked at it for a little, I couldn't find anything that would give access to it in BP. The subsystem has some functions to find the entrance but those are c++ only.
So the only way to make use of the entrance is to use wrappers (for example in a blueprint-function-library)?
EDIT: well well well, had to dig a little but there is also the GameplayInteractions plugin that adds a lot of functionality to the StateTree. One of those is Find Slot Entrance Location

[polar furnace]

what is the new version for this ?

UNavigationSystemV1::GetRandomReachablePointInRadius

    UE_DEPRECATED(4.16, "This version of GetRandomReachablePointInRadius is deprecated. Please use the new version")
    UFUNCTION(BlueprintPure, Category = "AI|Navigation", meta = (WorldContext = "WorldContextObject", DisplayName = "GetRandomReachablePointInRadius_DEPRECATED", ScriptNoExport, DeprecatedFunction, DeprecationMessage = "This version of GetRandomReachablePointInRadius is deprecated. Please use the new version"))
    static FVector GetRandomReachablePointInRadius(UObject* WorldContextObject, const FVector& Origin, float Radius, ANavigationData* NavData = NULL, TSubclassOf<UNavigationQueryFilter> FilterClass = NULL);

    /** Finds random, reachable point in navigable space restricted to Radius around Origin
     *    @param ResultLocation Found point is put here
     *    @param NavData If NavData == NULL then MainNavData is used.
     *    @return true if any location found, false otherwise */
    bool GetRandomReachablePointInRadius(const FVector& Origin, float Radius, FNavLocation& ResultLocation, ANavigationData* NavData = NULL, FSharedConstNavQueryFilter QueryFilter = NULL) const;

i found it

[gusty trellis]

hey guys is it possible to do a custom decorator based blackboard condition in bp?

[uneven cloud]

Is Valid AI Location, checks if the location is not Invalid Location. A lot of things return Invalid Location if they fail.

To check if a location is on the nav mesh, you project it to the nav mesh. There's a function in the navigation system to do that.

My guess would be that you are taking a location and adding to it in the Z axis and it can't project to the nav mesh.

What is it you are trying to do? From what I can tell the EQS is a much better tool.

[uneven cloud]

Yes.

[frozen saddle]

Do Someone made an shooter ai enemy??
I need help to make a shooter ai for game.

1. I want to make cover system and when player close to enemy, enemy change cover
2. When player in cover for long time enemy throw grenade
3. When one enemy see player he tell everyone about my location and then all enemy find cover and shoot
4. Enemy focus to last location of player he saw
5. enemy react to sounds and noise
6. Enemy team work
7. Saying things in different situations
   8 . Ai team work
   And ...

Can anyone help me to make it??
Or at least guide

(Sorry for bad english)

[uneven cloud]

In the pinned messages you'll find a link to the official learning library for the AI with Blueprints course to learn the AI systems.

I do not recommend using bad tutorials to learn how to do anything.

[gusty trellis]

i dint know it was bad sry

[uneven cloud]

Most tutorials are bad. They teach how to do things in very hacky and bad ways. Then people learn from them and make their own very bad tutorials.

[gusty trellis]

yeah but everyone can't learn the good way cause its simply to theorical, see in my case i'd rather watch a guide giving bad habbit than being submerged by super well writed information

[uneven cloud]

There are courses made by Epic devs that do teach the systems correctly.

Would you rather learn from someone who doesn't know what they are doing or by actual professionals?

[gusty trellis]

actualy u right

[gusty trellis]

can u help me on something ?

just tell me if what i want to do is dumb ad not realisable

[uneven cloud]

Just ask the question

[gusty trellis]

ok

nvm i think its dumb im to tired anyway

[dense owl]

https://tenor.com/view/facepalm-really-stressed-mad-angry-gif-16109475

[gusty trellis]

i watched what u recomended in the pinned section its cool indeed even if i like too the guy i linked u are right it dont provide a few thing that are actualy cool in the same scope and it dont seem as sain as the the epic dev, and i rethinked my question so thats way its not to dumb lol

My question is how do u handle an AI behaviour when u want to be able to interupt it at almost any given time with an interraction (dialogue) @dense owl advised me to look at Statetree and it seem to be what i need but i like behaviour tree 😦 but i can't figure out a way to make it work with the behaviour tree

[dense owl]

it's usually either or and my initial recommendation was to use Stop Logic and Start Logic if you don't want to go into state trees

[gusty trellis]

yeah i remmber and i tried and learned the way it work and it basicaly reset the tree in my case so that doesnt resolve my problem or maybe i didnt understand how to use it

[dense owl]

ah right, sec

I can't rmbr if there's a way to pause logic

eyy found Luthage on Reddit, tho I'm not the first one to mention this 😄

[gusty trellis]

it might be pausing but since i have a blackboard based condition decorator on every task with a service to track the dialogue boolean it doesn't work

[uneven cloud]

You don't put that kind of interruption in the BT. You pause the tree when something like that happens. Advising to use a state tree isn't good advice.

[dense owl]

mb, I'd mentioned state trees before he clarified this is about a dialogue system

[gusty trellis]

well thats was actualy interesting to look at so np

[uneven cloud]

Oh no you found me... using the same username.

[gusty trellis]

so i need to do a pause with stop logic and restart logic without boolean condition decorator ?

[uneven cloud]

Putting a decorator on every task is a very bad idea.

[gusty trellis]

look like it lol

[dense owl]

guessing that answer hasn't changed, you can only Lock it with Anim?

[gusty trellis]

thx was bout to ask what lock with anim mean

[uneven cloud]

I believe that's the only way to do it in BP.

[lyric flint]

I've got the following subtree in a larger behavior tree. There is Claim/Unclaim logic tied to the Claim Cover Slot service's Activation/Deactivation AI functions. However, I'm finding that this service is calling Deactivation AI whenever the children fail (i.e., when the tree is reevaluated). I think my mental model is wrong, but is there a way I can set this up so that Deactivation AI isn't called when the parent is reevaluated?

[dense owl]

this is all I can find in regular bp or anim graph

[uneven cloud]

Is the pause function not exposed for the brain component?

[uneven cloud]

Not without making engine source change

[dense owl]

doubt it, checking again ^

[dense owl]

looks like you can only check if it's paused out of the box

[uneven cloud]

The problem with your mental model is that it's not currently in that subtree anymore, so it should call deactivation.

[lyric flint]

Yeah, that makes sense. I was trying to avoid having a Claim and Unclaim task that are run in a sequence since the behavior tree could be interrupted (and I would leave the slot claimed forever).

[uneven cloud]

If you want it to stay in it, then you shouldn't use a selector

[lyric flint]

Ooooooh

[gusty trellis]

ill try to use this with an event tomorow thx

[uneven cloud]

That's the correct logic. If you do a sequence instead of a selector, it should work out better. You can even have a selector further down the tree if you need to.

[lyric flint]

Ah yeah, this all makes sense. Thank you so much!

[uneven cloud]

You're welcome.

[dense owl]

you'd need an anim instance

[gusty trellis]

nice ok i go check wtf this is lol

[dense owl]

it'd prly easier to expose PauseLogic but you'd need to dive into cpp a little

[gusty trellis]

ok

[dense owl]

@uneven cloud you can't really extend the brain component to expose that function right? You'd need to modify the source?

[uneven cloud]

I'd just make a function library function that accesses it.

[dense owl]

oh neat

[uneven cloud]

What I do is have a function on my C++ AI controller to do the locking functionality.

But I do different locking levels.

[dense owl]

oh, you can lock just parts of the tree?

[uneven cloud]

Well you can lock anything. The BT, the path following component and we added locking to our components that we added

So no, you can't lock just parts of the tree

[dense owl]

ah I c

[uneven cloud]

So we have different levels that turn off different things. 90% of the time, it's locking everything, but we have some cases where we only want to lock the BT.

[dense owl]

makes sense

[uneven cloud]

Instead of other systems deciding how to lock it, they just call a function on the AI controller

[dense owl]

and how do you handle the locking level at that point, do they pass a var that decides it?

[uneven cloud]

It's an enum.

[dense owl]

Hehe I had a feeling, sweet

[uneven cloud]

I think the different levels are Logic, Reaction and Pooled. Pooled means that the pawn is also disabled.

Reaction means everything is turned off to react to something. Like being interacted with for dialogue or a hit reaction.

[dense owl]

Nice that sounds very versatile

[uneven cloud]

It's also safe. So if something locks it and then something else does, it tracks both so they both have to unlock.

[dense owl]

Hehe yeah that pretty much covers every scenario

[gusty trellis]

heyyyyyy, i was trying to figure out what an anim instance is and i come back to see that

[uneven cloud]

What I'm talking about is only available for C++

[gusty trellis]

sad face

[gusty trellis]

is that difficult to implement or difficult to reproduce in bp without those both scenario covered ? i just want to pause he bt when i dialogue with the npc

[gusty trellis]

cause i dont understand this animation lock. i need to be in an animation and call that function to lock the bt ?

[gusty trellis]

expose pause logic from cpp ?

[dense owl]

#gameplay-ai message

[dense owl]

Yeah, far as I can tell, so prly not ideal for you

[gusty trellis]

thx a lot i need to dive into cpp and do that thing thx both of u @uneven cloud btw look like a conversation made for me lol

[uneven cloud]

If you can't figure out how to use a basic function like lock AI with animation, it would be incredibly difficult.

[gusty trellis]

cause i never used it i dont know what it does

[gusty trellis]

hmmm can u explain me ?

[uneven cloud]

It locks the AI and plays an animation. It's not complicated

I don't even know if you even have to give it a valid animation

[gusty trellis]

so i can use it in my npc lueprint for example and lock the behaviour tree without giving any animation when the dialogue fire ?

[uneven cloud]

Idk. Try it.

[gusty trellis]

alright

[dense owl]

it asks for an anim instance so I doubt you can just leave that blank but 🤷

[gusty trellis]

im gonna try tommorow i can't right now, i saw on internet that anim instance is a class used in cpp i might have mistaken something

[uneven cloud]

What happens if you just send the anim instance that is on the mesh?

An anim BP is an animation instance

The code base class for an anim BP is an animation instance.

[dense owl]

not sure, I've only barely dipped my toes in animation so far. I can use get anim instance and plug that in but not sure where to get the context struct it's asking for

[uneven cloud]

That function used to just take an anim sequence. I haven't looked at it in a while though

[uneven cloud]

It's not a struct, it's a pointer.

[dense owl]

I was talking about that

[uneven cloud]

You need to get the anim instance from the skeletal mesh.

[dense owl]

from what I can see, you can use LockAI directly from the AnimBP, but I've no idea how it would get access to brain component on its own, without any reference to it

I'm looking at the code for it now

sweet lord, it casts and everything under the hood

[uneven cloud]

It might just get the AI controller from the anim instance's pawn.

[dense owl]

void UAIBlueprintHelperLibrary::LockAIResourcesWithAnimation(UAnimInstance* AnimInstance, bool bLockMovement, bool LockAILogic)
{
    if (AnimInstance == NULL)
    {
        return;
    }

    APawn* PawnOwner = AnimInstance->TryGetPawnOwner();
    if (PawnOwner)
    {
        AAIController* OwningAI = Cast<AAIController>(PawnOwner->Controller);
        if (OwningAI)
        {
            if (bLockMovement && OwningAI->GetPathFollowingComponent())
            {
                OwningAI->GetPathFollowingComponent()->LockResource(EAIRequestPriority::HardScript);
            }
            if (LockAILogic && OwningAI->BrainComponent)
            {
                OwningAI->BrainComponent->LockResource(EAIRequestPriority::HardScript);
            }
        }
    }
}

[dense owl]

yeah, they're doing all the work for us lol

[uneven cloud]

Haha. Weird function.

[dense owl]

ikr, I rarely see built-in functions do so much work for you lol

[uneven cloud]

The one I saw before, it locked the AI and played a specific animation.

[gusty trellis]

thx guys

that look awesome

[stiff gale]

Would this work? I want one of the face location to pass in order for it execute the task on the right

[misty wharf]

It would but is there even any condition when Face Location can fail?

[stiff gale]

So these two uses diff backboard keys. One uses TargetActor and other uses TargetLocation. I need one of them to pass in order to continue

[misty wharf]

Ah, I guess the TargetActor one might in theory fail if TargetActor was not set

[stiff gale]

But is htat how selector works right

[misty wharf]

Yes, selector executes nodes until one of them succeeds

as opposed to a sequence which executes nodes until one of them fails

[stiff gale]

So movetotarget won't execute if both of these FaceLocation fails. Correct

[misty wharf]

Yes, if nothing ever succeeds the selector itself will also fail

[stiff gale]

Ok thanks

[crude fable]

Do you happen to know if there is any standard way to implement waypoint highspeed navigation for vehicles? It looks like a case to use pids, but as long as it is possible i would prefer to not implement one myself

[civic canyon]

Hi! Using move to and use smart object behavior for my smart object, but it keeps failing after I end behavior

Is it not supposed to succeed once I end the behavior?

Nevermind, turns out I needed to make it wait a little bit before ending behavior. Adding a small wait fixed it all somehow

[misty wharf]

iirc there may be a "feature" in the gameplay behaviors where it doesn't work correctly if you end it on the same tick as it was started

which could explain why adding a wait fixed it

[gusty trellis]

@uneven cloud @dense owl That work well thx u very much !

[wooden trellis]

question - how do I make a flying AI move vertically? I'm making a 3d game, and have an enemy that is meant to chase the player in any direction for a few seconds, then explode. Issue is, it just refuses to move up - I have the gravity set to 0, the movement set to Flying, and it can move down just fine, but for whatever reason it can't ascend.

I had the idea to have the enemy always check to see if they are lower than the player, to then move above them - but now they clip through floors and can get stuck pretty easily...

[dense owl]

Depends what you mean by move up. Are you trying to make it climb up a hill/over things or just free float up?

[wooden trellis]

free float up

[uneven cloud]

How are you moving it?

[vocal shore]

how can i achieve this kind of EQS results ? ( both top and bottom floor )

[uneven cloud]

Probably projecting it to geometry. They are also likely using their own generator.

[vocal shore]

thanks

[uneven cloud]

A generator that generated a 3d cube instead of a 2d grid and then projecting them onto geometry.

[worldly crane]

How do you refresh the NavMesh at runtime? I’m using geometry scripts to generate my level which so far looks great, except that the NavMesh doesn’t refresh when hitting play. I have the project settings to force reload and dynamic but its still not updating

Could it be an issue with the load order? Im generating the level at BeginPlay from a BP and maybe the Nav Mesh updates before then?

[misty wharf]

if navmesh is set to dynamic it should update any time something that can affect it gets updated

[worldly crane]

so this is how it looks during play after the terrain is generated, you can see that the navmesh thinks there are holes where there arent and vice versa

what it seems to be doing is still using the nav mesh that the editor has, as here it is correct

is there a setting im missing within my terrain generation BP that i should be checking? do i need a RecastNavMesh actor?

[crimson galleon]

it is a setting on the nav mesh volume. The RecastNavMesh is created from each nav mesh volume. You may also need to enable the same setting in your project settings for navmesh generation - a sort of default settings, it helps if there are mutiple levels or you end up remaking the navmesh

[shy isle]

i created a movement method for one of my AI, when it has to move to somewhere behind it, it first slowly rotates towars it, without changing position first
it works propely, but when I give it a very low rotation speed (which I want to do) the pathing breaks: it stops rotating and stops trying to move to the target
I think there is some kind of timout for the nav pathing, were when the position of the actor does not change after X time, it assumes its stuck, and stops trying to move
is there a way to change this behavior?

[uneven cloud]

That is how it works. The amount of time is on the path following component. You should be able to change the variable.

[shy isle]

mmm, i was looking for it, didnt find it, let me check again

[uneven cloud]

You really shouldn't wait for begin play to generate your level

[uneven cloud]

BlockDetectionInterval or BlockDetectionDistance is probably where I would start with tuning.

[worldly crane]

Whats a better way of doing it then? I'm still new and so far only relying on BPs for my first prototype

[uneven cloud]

You want to do it in the game mode or game instance before begin play happens. There's an entire flow that happens before begin play that you can do some research on.

First thing I would do is look on the learning library to see if Epic has any courses on procedurally generated worlds. That's going to be a lot better resource than some random tutorial on YouTube.

[wooden trellis]

*sorry for late reply, had to go to work - I originally was using Move To Actor, but now am just using Look At rotation mixed with moving forward, which......... eh... well it works, but the new issue is that the enemy gets stuck on just about everything

[uneven cloud]

Well move to uses navigation, which is 2d. So it will keep your AI on the ground. Moving forward without any navigation solution will get stuck on everything, because you aren't avoiding anything.

You have a few options:

1. 3d navigation either made yourself or a plugin from the marketplace.
2. Use the 2d navmesh, but hover above it.
3. Use probes to check for obstacles and do your own avoidance.

[wise sluice]

I recognize that ! This is the DSN 3D Generator from the DSN Plugin

[ruby jackal]

https://i.gyazo.com/8b03a98efe41a9960a9dab78bb0ce018.png
https://i.gyazo.com/b6d926be608c111a38e6df1b97453569.png
Is there a hard cap on nav link proxies? it generates for little gaps not not bigger ones

[rose yew]

Hello, I have quite a tricky situation and I would love to get help on how to solve it. my ideal is for the AI Agent to flee from the player when damaged. if the distance is greater than a certain threshold then the AI goes back to roaming the environment. the problem comes from the environment query (I am not sure how to get the query to be point further and hidden from the players sight only). this is what I have so far. As you can tell, some points are going past the wall which I want to avoid.

[uneven cloud]

What is your problem/question?

[rose yew]

I would like to get a point away from the opponents view and move to that but im not sure how to achieve that. (the EQS points goes beyond the walls, i'm not sure how to stop that from happening)

[uneven cloud]

You can use a context to get the opponent and use a trace for line of sight. For going beyond the walls, you can use a path test.

[rose yew]

Thank you, ill try it out

[cyan sigil]

Hello! I have a EQS that's generating a place for character A to stand so they can hit character B with a melee attack. If character B keeps moving, I want A to keep running the EQS and trying to go to the best location. How can I do so?

[median socket]

yes, there's a limit. It must connect to the adjacent tile, or page, or whatever it's called

Nav Link Proxies can only connect adjacent Navigation Mesh tiles. You can visualize the tiles by selecting the RecastNavMesh-Default Actor and enabling the Draw Tile Bounds checkbox.

[shy hare]

hey devs please help why my navmesh looks like this ? there are a lot of gaps how to fix it ?

[uneven cloud]

If you are using a BT, you can use a service to update the EQS.

[silent hamlet]

How does collision look like on your floor meshes? Is there anything else placed on top of that floor section?

[uneven cloud]

Is there invisible collision?

[naive salmon]

Bit of a random question but have any of you played with integrating an SLM like TinyLlama with unreal?

[cyan sigil]

Awesome. Thank you.

[cyan sigil]

It doesn't seem to be working. Here is the current task config:

The character travels to the place where the target was when the task started, ignoring any additional movement

Found the problem. I had to check "Observe Blackboard Value"

[mild bobcat]

Check the physics and movement plane settings, I believe default some of the movement components don't have Z on. I also have 6DOF and had to toggle some things

[bitter wyvern]

So I have a bug where my AI has s target location to move to and is executing the move to task, but just is not moving

this is the behavior tree, as you can see its going to the Move TO

This is a vid of the guy not moving

[dense owl]

Use the visual logger to find out why

[bitter wyvern]

ah did not know that was a thing! will look into it!

[spark kiln]

I see people mentioning "utility" as another way of handling AI logic, as opposed to behavior trees. What is this? Is this a plugin, or something built into Unreal? Or is it more of a philosophy than a specific tool?

[bitter wyvern]

hey so it seems I was able to capture some info about whats going on. Still have yet to really dive into it, at first glance do you know what might be occuring here with this "ProjectPointToNavMesh" error? the navmesh seems to be calculated fine, and the target location is within the navmesh bounds. (target is y=4226, in the image the blue character is at around y=4632. The enemy is currently at y=3000 and not moving)

I also got it to show the requested point and projected result. I Assume its not valid because it is inside an object? Curious if that projected point comes from "Acceptable Radius" on the move to node?

Holy crap this visual logger is amazing I feel so ashamed i did not know about it sooner!

[dense owl]

Just so you know there’s also a couple of vis log bp nodes you can use, namely a shape trace one. When you use it with your goal location and click on the frame in vis log, it’ll show you where it’s trying to go vs where it’s actually projecting

What’s prly happening is area is too small for your agent to fit in there. You may need to change the nav agent radius but that’s a bit more involved. See what the vis log trace shows first

[bitter wyvern]

So my ai is setting the goal location through a services BP. Should I use the shape trace in there?

[dense owl]

Sure, wherever your MoveTo gets fed the location is prly good

[bitter wyvern]

did u mean the box shape?

[dense owl]

Aye that

Play with the radius if you can’t see anything and rmbr you have to click on the frame in vis log for the trace to appear, last I checked. gl

[bitter wyvern]

do u know of a hotkey to pause or start logging wile in scene view?

so i am now pretty confused lol.

I am only drawing one box with the vislog yet there are 3

[bitter wyvern]

https://forums.unrealengine.com/t/movetolocation-call-fails-to-project-the-location-to-the-navmesh/299553 so this forum also led me to what you suggested. I see the agent radius in the proect settings. I had no supported agents before, but just added one.

[bitter wyvern]

got it working! thanks for your help @dense owl ! visual logger is such a big discovery

[median socket]

Is it bad form (or even possible) to select a new BT from within a BT? I'd like to create some one-off behaviours -- for instance, a character in the game will take you on a tour of the city, and then return to his regular behaviour.

I figured this would be possible by simply having a custom task in the "tour" BT that would exchange return 'em to the previous behaviour.

I guess a sub-BT could also work?

Apologies if this is a stupid question -- any pointers to helpful and relevant guides or docs are also appreciated!

[misty wharf]

You can use one of the run bt nodes to run a different BT as a child, but this does require it to use the same blackboard type, as it will inherit the BB values. Anothe alternative is to have some way of just replacing the entire BT with another one via a custom task

[median socket]

I think it would use the same BB regardless.

I'll do some experimenting I suppose, thanks!

[inner zephyr]

Did anyone got the UStateTreeTask_PlayContextualAnim working with a character and object, that has a skeletal mesh and should play an animation too?
Trying to get a character open a door but my door is not playing the animation for some reason and I am stuck because I am not sure how to debug it

Also trying to figure out where the ContextualAnimationScene is set. Does only the state tree need to set it in the Task and the other actors only need the component?
EDIT: oh well, lacking animation knowledge hits hard. forgot to make a correct ABP setup with the Default Slot

[steady phoenix]

Is there anyway at all to do AI movement on the player character while it's possesed, or is it impossible?

[misty wharf]

Pretty sure it requires an AI controller

You can move the player character around via other methods besides direct player inputs, but I don't think any of the builtin AI Move bits would work unless you possess it with an AI controller first

Eg. you could manually get a path from the player to some location, and then provide the pawn with the appropriate movement inputs via code

(which is more or less how the AI bits work also)

[steady phoenix]

If you Possess the actor, none of the AI nodes work (Even if it has an AI Controller), because it expects direct input.

If it's unpossed it'll work, but was hoping they'd be a way to have AI nodes work when it's possesed, even if it has an AIController.

[misty wharf]

Oh that's interesting it works unpossessed... I would have expected them to require the pawn to be possessed with an AI controller

If you're doing C++ you certainly could just copypaste the node's code and adapt for your own use

[dense owl]

They do the possession for you

[dense owl]

You can’t have 2 controllers possess at the same time but there are some move to nodes that work with player controller, namely Simple Move To iirc

[misty wharf]

Oh it spawns an AI controller if it isn't being controlled by anything?

[dense owl]

Yeah, far as I know

[misty wharf]

Interesting, never tried it

[steady phoenix]

So are you saying "AI" controller and "Player" controller are 2 possessions affectively?

And the "player" one i suppose takes priority.

Sime move to node doesn't work while it's possessed.

I feel like im trying to do something so basic that I see in every single game and it's so hard lol

[dense owl]

Yes, you can only Possess with one controller at a time, effectively. You can either temporarily unpossess your char and possess it with an AICon or use the SimpleMoveTo node I mentioned

[steady phoenix]

SimpleMoveTo doesn't work while it's possessed tho 😦

At least not for me..

[dense owl]

Show your code

[harsh storm]

It's only hard when you don't know what is going on. Can't say I've ever had an issue with the node in the like 4-ish years I've been using Unreal.

[steady phoenix]

Are you using it on a posssed player character?

[harsh storm]

No. Because that isn't an AI character.

I use it on a possessed AI character without issue

[misty wharf]

I think there are a fair number of things that aren't exactly intuitive in various parts of the engine... the fact that you can't move a character along a path easily when it's controlled by a player is definitely one of them lol

[steady phoenix]

So your issue isn't the same as mine 🙂

I would like to figure out AI movement on the Player Character

[misty wharf]

Is it just a temporary thing that it sometimes gets AI-moved or is it always AI-moved?

[harsh storm]

Sime move to node doesn't work while it's possessed.
The issue is quite vage

[steady phoenix]

There is a whole convo above it, might have missed that!

[misty wharf]

Because if it's always AI-moved, it might be easiest to just have an AI controller possess it, and have the player controller follow around

but it depends™️

[steady phoenix]

So I ask this, how do you move a player character during a cinematic? How are you guys doing that?

Spline and Animate and hope it "looks right" ?

[harsh storm]

Unpossess the player, possess it with AI Controller, move it, then swap possessions back.

Add movement input directly. It doesn't have to be actual input. Get the target velocity and feed it into the method.

[misty wharf]

Tbh if it's for cinematics, I'd just move it around using the sequencer if at all possible

[steady phoenix]

Sadly, that is what i originally tried but you cannot reposses during cinematics.

[misty wharf]

Having navigation and pathing in a cniematic is a good way to accidentally make the cinematic not work because of some random glitch

[harsh storm]

Just move them in the cinematic with sequencer

[steady phoenix]

But then i loose like my entire animation setup

all the weights, blend spaces, everything

[misty wharf]

Surely there's a way to move it with the sequencer and allow the anim BP bits to still function? 🤔

[steady phoenix]

Maybe but then I'm thinking about velocity, direction, IK surfaces etc... if im just forcefully sliding it along a spline.

[misty wharf]

Yeah I'm not really a sequencer animation expert but it feels like there would be some nice solution for this

Like I'm fairly sure you can record movement from the level into the sequencer, so you could just walk the character manually and then play that back?

Probably questions for #animation

[steady phoenix]

that would be rad lol i'll see. This is something i see in every game, you run around, cutscene with the player, you run around more... and things like God Of War do it so seamlessly. I got so close but i either can't do AI or i can't reposses 😦

[misty wharf]

I don't think there's any reason you couldn't possess with AI at start of cinematic, do whatever, then possess player back

If it comes to it

[harsh storm]

GoW also uses an entirely different engine.

[steady phoenix]

Once the sequence starts, if you try re-posses, it just doesn't work, it cannot be done during the camera cut because the camera becomes the possession. If you have your character already possessed before playing the cinematic, then you cannot do AI

Yeah i know but like tons of UE5 games do it, heck fortnite does it lol

it's a concept im talking about

[misty wharf]

🤔

I don't think the cinematic/camera/etc. should affect that at all

Camera viewport can be controlled separate from possession

[steady phoenix]

Sadly it does

This is an example

If you call SetDisableCameraCuts it'll work

But you then loose Cut blending

If i want immediate "snap to player" or "fade to black > control player > fade in" i can do that, it'll be all hidden

If you want the camera to smoothly tranisition back to the player, that's where it becomes very tricky.

[steady phoenix]

I have kinda figured it out, at least a workaround

Not specifically AI related but I'll post anyway. Have a dummy cine camera with the same settings as player camera. Possess that --> Play sequence. Perform all your AI w/e you want, then prior to Blend Cut position the dummy cine camera to the world location of the player camera, let the sequence auto blend to the dummy camera, call posses on your player and when you snap to that camera for possession it'll be unnoticable (or you can use viewport blend node if it isn't perfectly lined up).

[inner zephyr]

@steady phoenix did you try using the Simple Move to Location ? It worked when I had my PlayerController on it.

[steady phoenix]

Lemme give that a whirl, for a Auto Posses PlayerController yeah?

Might have success @inner zephyr !

[inner zephyr]

Not sure what you mean by auto posses. You want to instantly after BeginPlay?

I just had an area where the player wasn't supposed to goto, so instead of just placing a blocking box at that spot I used that node to make him walk back

[steady phoenix]

Yeah that is common in games like trying to go back when the story wants you to go forward.

Auto posses is when you either call "Possess" or set Auto Posses

This worked in a test, just trying it in my sequencer

i think i might have used the wrong node last time. Rather than plugging in Players AI Controller i put the Player Controller in

[inner zephyr]

I might have tried move to actor as well and it wasn't working and then used move to location instead

but I was to lazy to look at the code see why the move to actor didn't work 😅

[steady phoenix]

For some reason it's not working in a sequencer 😦

nvm it does..

[steady phoenix]

tyvm for your help, it's working as I'd expect without my faf work around lol using simple move to is more than enough for what i need, sometimes it is just like you say turning them around to a point back on track.

[uneven cloud]

Simple move to I believe also works on the player. It does not require an AI controller.

[steady phoenix]

Yeah! We got there in the end, on my original test when @dense owl told me I plugged in AI controller from player and that didn’t work, realised you need to plug in Player Controller, which makes sense thinking about it now.

So that one was my mistake 😂

[dense owl]

That is what I meant

[steady phoenix]

Simple move to is more than enough, the use cases are very short walks and use priority nav scoring and sometimes splines.

[uneven cloud]

You should not use AI during a cinematic. It's too unpredictable. Sequencer is supposed to take control over the player and AI.

[steady phoenix]

How do you move a character for sequencer?

[uneven cloud]

By moving it. You don't need navigation. Just move it.

[steady phoenix]

How do you get the correct speed and then pass the correct velocity and direction to your blend spaces?

Because moving its location won’t do this

[uneven cloud]

I've only fixed issues with sequencer and have never actually used it. It's usually the animation or cinematics team that actually makes them.

[steady phoenix]

If you’re working with bespoke anim or a mocap then that isn’t applicable here

[uneven cloud]

It's not really an AI question. #sequencer is a better place to ask about it.

[steady phoenix]

Yeah I can ask there but I had AI specific questions

You can make it consistent btw, I don’t have any issues there and have ran well into thousands of takes all matching as I have a lot of other stuff handling that

[uneven cloud]

Sequencer is a very powerful tool that every studio uses for cut scenes. It's a waste of dev time to do a bunch of work to not use it.

[steady phoenix]

I am using it? I don’t think I said I wasn’t

If you’re not going to offer any help then just don’t say anything, you just come off rude and when I ask you how, you then say you don’t even do it yourself

[uneven cloud]

AI Engineers don't make cinematics.

[steady phoenix]

I’ve seen a lot of games reuse their AI logic inside cinematics, sometimes with funny results

[steady phoenix]

Maybe in a big studio, not in a solo project, stop being pretentious

Blocked, not interested in high horse rude people.

[inner zephyr]

@steady phoenix ho there I wouldn't block Luthage, she is pretty knowledgable. she just said what's the best tool to get the job done

you can decide if you want to use it or not

I prefer the simple move to as well because I am not that concerned about some little bugs that might happen

some bugs are fun and should stay in a game

[uneven cloud]

*She. I already blocked them for being rude. I don't help rude people.

[misty wharf]

damn usually it's just #cpp that has all the drama lol

[steady phoenix]

Not unblocking someone who’s rude, ill figure it out

[misty wharf]

I don't think she was being rude

[uneven cloud]

Apparently giving advice you don't like is "rude" and you are free to then insult them.

[undone mason]

Hey friends I have a problem I need help with. I am trying to stop the AI in my game from clumping together. I followed a tutorial and made some Nav Modifier Volumes to make the AI use the platforms in my game more often (2D arena platformer game for context.) The problem I'm having now is the AI won't be able to use the Nav Links I set up to jump off the platforms (they jump but are pulled back to the platform.) Is there a way I can exclude the Nav Links from the Nav Modifiers?

[uneven cloud]

You have 2 separate problems here. The clumping is usually fixed by them not all going to the exact same destination. By using the EQS you can have them make smart decisions about getting close to the target instead of all just going to the target. Then you use RVO or crowd avoidance to get separation.

What do you mean they are pulled back up to the platform? Have you used the visual logger to debug the actual path they are taking? Are they jumping down and then jumping back up? Have you set the area class on the nav link?

[undone mason]

Oooh ok thanks ill look into that. With the platforms the ai will jump when they hit the nav link but will drift midair back onto the platform instead of jumping down. I set the area class of the nav link to one with a really low cost cus i thought that was the problem.

[uneven cloud]

That doesn't sound like an area class problem. It sounds like the nav link isn't actually taking over the AI.

[misty wharf]

I think people sometimes pick up rudeness when the messages are very terse because of how text doesn't contain any actual cues

[undone mason]

Ok, do you have any ideas on how i could make that happen? My understanding was the nav links were connected with the ai's perception, like that's how they know to use them

[uneven cloud]

It's not connected to the perception. It's all navigation. They don't "see" them, all they know is that they have a path from point A to point B. When the path following component reaches a link, it triggers the behavior for the link.

How you fix it depends on exactly what is happening. You need to first debug it. I recommend the visual logger, because it's easier to debug with it.

[undone mason]

Ok wonderful! Thanks for the info. I'm gonna take another swing at it

[uneven cloud]

You're welcome! It might be trying to repath while they are using the link. But you should verify that before trying to fix it.

[steady phoenix]

I just think they're rude, they have absolute no idea about my project setup or why I might be trying to do something some way, they don't offer any suggestions but instead snark comments, many of the ways I would normally do things (eg Anim tracks on the sequencer are not applicable to my project because of how UE works), they confidently state a "you should not do that" then when questioned their excuse is some studio/workplace doesn't do it that way but they don't actually know. It's classic: talking down to people.

[uneven cloud]

Attributing malice to a succinct writing style isn't an excuse to then insult people. I think it's more rooted in insecurity.

[silent hamlet]

Why you cannot use anim tracks?

[misty wharf]

Oh for sure, just seen it happen sometimes

[misty wharf]

You're reading too much into it. She just stated she thinks it's a waste of time which is true.

[bitter wyvern]

If I wanted to find spots in the map where enemies wouldnt be overlapping anything to spawn them, where should I start my learning jounrey for this problem?

[dense owl]

🤔 environment context query? @bitter wyvern

Depends on your game, your map, etc.. prly easier to just place spawn points manually if your use case allows

[lofty remnant]

Is there any way to add navmesh with NavModifier components at runtime (say, on a moving platform) without constantly regenerating the navmesh all the time (i.e. the Runtime Generation:"Dynamic" setting?)

Context: I have only one platform that moves, so I really don't want the overhead of constantly regenerating just for a single lift. I'm trying to see if I can leverage the Runtime Generation: "Dynamic Modifiers Only" mode by adding a NavModifier component to my lift And set the Nav Area Class to Default, but this doesn't appear to be doing what I want -- it doesn't seem to add navigation mesh to the platform at runtime. I'm not sure if this is something that can be done. I'm on UE 5.3.2. Any ideas?

[primal vigil]

yo guys do you know why this is not executing? I am chasing myself with AI and when I get bool that I am at some distance I should shoot

but its not executing idk why

[vivid drift]

@primal vigil well you can watch it execute in the BT editor to get an idea. Off the top of my head, I would guess that FindPlayer is finishing immediately. ?Enemy not being set is of course another option.

[vivid drift]

the obvious answer is to simply not allow movement while on the lift, if that's possible. you can do hacky shit to get around this but its kind of a headache and from the details you've given, I would guess not worth it.

[halcyon briar]

Hello.
I need a tool that will allow me to create smooth paths for my AI agents.
It must support branches and should allow me to convert the path into a graph.

Unreal's Spline tool is helpful but doesn't serve the purpose as it doesn't let me create branches.
Are there any good Marketplace tools that you would recommend ?

[vivid drift]

@halcyon briar can you describe what you want better?

Sounds like what you want to do is to make a series of waypoints with some logic to connect them, or just manually creating connections between them

[halcyon briar]

I want to create a track for my AI agents to follow.
A track might have brances and loops.
I want to be able to convert this track to an array of vectors.

Basically all I need is a tool that helps me build a road map for my AI vehicles to follow.
UE's spline component has some drawbacks :

1. It doesn't allow branches
2. It doesn't give you the intermediate points between two spline points.

I am willing to buy a paid product if it serves my purpose and has a good price.

[vivid drift]

probably splines are fine. just convert them into an array of points and implement some logic to decide if it should switch to a different spline at specified junctions

you can get the intermediate points at distances along splines with provided functions. forgt the name but it something like get transform at distance along spline

[reef shore]

is there a way to read data directly from a BT_Task in for example an AnimBP or an Actor?

[misty wharf]

Probably via some horrible hack. Is there some particular reason you want to do that?

[reef shore]

I think I already did that hack lol

[misty wharf]

heh

[reef shore]

I wanted something cleaner

so I have this task that rotates the enemy towards the player

and I need to access the rotation delta in the anim BP to play the turn right or turn left anim

so I wrote from the BT_Task into the owner and then read the variable from the owner into the AnimBP through casting

[misty wharf]

I would consider putting the rotation logic into the actor itself, so that your BT Task only does something like MyActor->RotateTo(whatever)

Then your Anim BP can read data directly from the actor as it would do normally

[reef shore]

ah nice

thanks!

[reef shore]

still needs a cast right?

[misty wharf]

Well yeah your task would need to cast to your class

[reef shore]

any other suggestions aside from casting just out of curiosity

casting is basically inheritance, I suppose I could also use an interface

[misty wharf]

Yeah interface or putting the logic into a component could be other options

[reef shore]

I could also put the rotation logic into a rotation component and add an IsActive bool on tick where I can just activate it in the task

[uneven cloud]

No you can't add nav mesh with a nav modifier. The ONLY way to add navmesh is with dynamic generation. With dynamic generation you aren't constantly regenerating the nav mesh. It only generates when geo changes and it's only the cells that are affected.

[uneven cloud]

You could, but components already have an IsActive bool, you'd be doing unnecessary things on tick and adding the overhead of a component just to hold a couple of variables.

[reef shore]

how would you approach it, I really don't like doing that on tick

[uneven cloud]

You have the BT task do it. For the animation you either set a variable (I prefer adding gameplay tags) or have the anim bp query if the actor is rotating. You can do that by caching the rotation and checking if it's changed. Then your transition would be is not moving and is rotating.

[zinc sky]

Hi I am trying add a task to increase my enemy's movement speed after completing a melee attack, I really have no idea what I am doing here or if this is even correct

[uneven cloud]

I recommend using a service instead and using the receive activation/receive deactivation. Using a service instead of a task guarantees that it will be unset.

Next up you don't need to cast to your BP character to get to the movement component. That creates a hard reference, which should be avoided. You can just cast to a character.

You need to set the walk speed to change the speed.

[zinc sky]

Thanks I will try that I just got into ai about 2 weeks ago and I'm still pretty confused

[uneven cloud]

If you haven't watched the AI with Blueprints course from Epic, I highly recommend it. You can find the link in the pinned messages.

[zinc sky]

thanks I will give that a watch

[reef shore]

awesome thanks

[maiden cargo]

Hello,
I have a problem with the navmesh when terrain isn't flat. I'd like to get all the points on the paths when using the Find Path To Location node like the image on the right (a simple grid with A* thrown in) but getting the Path Points from the node always give me a straight line with two points like the image on the left.
Is there a setting or property that can give me the in-between points or am I forced to do it by hand?

[uneven cloud]

There are no in-between points in your case when it comes to the nav mesh.

[bitter helm]

My navmesh rebuilds every time I open the level. Is that expected behavior? Is there a way to disable that (I've looked all over)? I'd like to only build the nav mesh when I manually build from the build menu.

[uneven cloud]

There is a setting in the editor settings to turn off auto rebuild of the navigation

[bitter helm]

Thank you! I found it. I was looking all over in the project settings and on the navmesh settings.

[bitter wyvern]

Hi I am having an issue and Id really love some advice on how to debug this. Bassicly my AI dont detect my player, untill they just randomly do. It doesnt seem to be related to touching or shooting or anything its just like for 30 seconds they dont see the player. When I try with less of them they see the player right away. I'm not sure the best way to visualize the enemies detection cones / what its searching for. Is there a console command that displays cone of visions? I was thinking of using visual logger but theres just too many enemies to find which one is which. Any advice would be much appreciated!

[misty wharf]

If they see you right away when there's less of them this sounds like your perception is being heavily timesliced

Have you disabled automatic pawn registration as perception sources, or implemented IGenericTeamAgentInterface?

If you haven't, it's likely that all your AI pawns are attempting to perceive each other, which will cause a lot of linetraces to occur when you have a lot of AI pawns, which in turn will cause perception to slow down because the perception system timeslices it so it doesn't cause your FPS to chug

I've never seen it be as bad as taking 30 seconds though so I'm not 100% sure on this, but this would be the first thing I would look into

[bitter wyvern]

Thank you so much for this explanation!!! I will look into this

[uneven cloud]

In addition to @misty wharf 's advice, you might also have a problem with enemies blocking the sight to the player. If you don't want them to block sight, you can turn off the visibility channel in the collision settings.

[uneven cloud]

The best way to visualize the perception cone is the gameplay debugger, which is accessible by hitting apostrophe (') while playing. Then use the numpad to turn on the perception category, I think it's 4 by default? It's a bit tricky, because it only shows 1 AI at a time, so you have center your camera on a specific AI and then use '. If you want to switch which AI, you have to turn it off and then turn it on again.

The visual logger is actually really great when you have many AI, because each one has their own track. Unfortunately the perception isn't logged there by default.

[bitter wyvern]

so i disabled automatic pawn registration by following this forum post https://forums.unrealengine.com/t/how-to-make-a-pawn-stop-registering-with-ai-perception/368928/2
to put

bAutoRegisterAllPawnsAsSources=false

in my defaultengine.ini filer

I have not implemented IGeneraicTeamAgentInterface yet, going to do some research on that.

But other weird behavior is that sometimes the enemies do detect the player right away. Either they all detect or none of them detect for like 30 seconds. But I cant seem to do anything consistent that changes wheather they detect right away or not. The only thing I cam close to was changing number of enemies. But if I reboot the project with the same number of enemies when its not wokring, sometimes it just works again. Do these symptoms make any other issues come to mind?

[bitter wyvern]

thank you, it seems visibility channel is set to ignore on enemy capsule

[bitter wyvern]

what do the little green faces in the bottom right indicate? also swithcing AI isnt working for me lol. The camera is also unmoveable while playing my game is there a way around this?

So I found I was using a depricated node, FindPointInRadius instead of FindLocationInRadius. Maybe that could have been causing weird behavior?

also, if I disable a perception stimulus source component, will ai no longer detect that pawn? if so, if the component is disabled while the ai is locked on to that pawn will it forget about the pawn immediatly (I want this behavior)

[uneven cloud]

The little faces are AI controllers.

Making your camera movable is going to entirely depend on your game.

[uneven cloud]

What about your enemy's mesh?

[bitter wyvern]

ignore visibility on that as well

[bitter wyvern]

ok thank you

[uneven cloud]

Yes. It will no longer detect that pawn until you add a stimulus source component.

[bitter wyvern]

is there a way to monitor the behvior tree of a specific singular ai?

[bitter wyvern]

this didint help

so the enemies also sometimes cant find the player if the player dies before they see it.

[uneven cloud]

You can watch the BT, but I highly recommend the visual logger instead. The visual logger gives you actual information

[bitter wyvern]

thank the visual logger helped alot

so I was able to capture some info about the last (log I guess youd call it?) before HasSpotted player became true

i dont really see anything wrong with it tho

so even when my player moves out of view, the enemies still seem to see the player

[bitter wyvern]

I added the GenericTeamAgentInterface in the AI_Characters implemented interfaces in class settings. Are there methods I need to override or anything for this to work? The documentation page for it doesnt really tell me much.. https://docs.unrealengine.com/5.2/en-US/API/Runtime/AIModule/IGenericTeamAgentInterface/

So I have a hunch that it just takes longer for enemies to start registering stimulus sources depending on the number of enemies I spawn at the start of the game. If I wait longer before entering the part where the enemies spawn (they are already spawned it but I set them active becasue they are object pooled.) then they register right away. Is there some intialization that could be taking a long time to register perception stimulus sources at the beginning of the game or something?

[uneven cloud]

The team interface is C++ only. All the functions that need to be overridden are in C++. I'm sure there is a guide somewhere on how to set it up.

[uneven cloud]

Don't make assumptions like that. You really need to debug to find out what is actually going on.

They register with the perception system during initialization. There is no delay.

[obsidian star]

Hi guys , i'm new in unreal and i'm looking to make some AI enemies but i want to know where is the best place to code , inside the character blueprint or the controller ?

[misty wharf]

It depends on what you're coding

AI logic would typically go into controller or behavior trees or such, but for example the ability to run, jump, pick something up, would go into the character

[gusty trellis]

hey guys im using a custom blackboard that check a boolean and once its abort it stay abort despite the boolean changeing value, how could i reset the node and reset this abort ?

[gusty trellis]

my brain is fried, context i build an AI for my game wich is being a waiter in a pub and im trying to set up a behaviour tree in a way that i have 1 NPC (customer),

• and i want him to do a Custom Task "Ask" wich is a simple animation where he raise his hand
• Then wait ( custom task with animation)
• then check if he has been served, and drink (i actualy made it a separate task but it can be both same time)

All of this is done and working but then come the question how i handle the loop of this behaviour wich also come to several condition wich are: does he already asked ? so he dont spam "Ask" task if he dont have thing to drink and how do i can add those wait phase (animation custom task) so it dont disurb the loop

[gusty trellis]

Chat gpt seem to have answer

[misty wharf]

Lately I'm kind of moving that kind of complicated conditional logic out of the BT

The BT task could put the NPC into that particular behavior "state", but the specifics of the behavior would be handled somewhere else - potentially just inside the BT task, but just not within the BT "tree"

[gusty trellis]

the specifics of the behaviour tree ?

[misty wharf]

"specifics of the behavior" as in the more specific logic

[gusty trellis]

oh ok

[misty wharf]

the BT just says "you're in the eating state now" and it doesn't really know how the eating state functions

the actual logic of what's involved could be within the BT task, or within an AITask, or whatever works

Since doing that kind of complicated conditional logic, retrying, etc. is kind of annoying in BT's sometimes, so it's easier to handle that elsewhere and just use the BT for the "big picture" coordination

[gusty trellis]

thats clearly what i would love

but

concretely handle the time is what i dont grasp yet

but im a bit lost i mean i know how but to make it work correctly i need to think about it

[gusty trellis]

im gonna try to apply ur philosophy seem the best in my case for sure

[misty wharf]

One of my projects is a video rental store simulator so it has a bunch of those kinds of behaviors on the NPCs where they need to go do something in that kind style

That approach has worked alright so far, so I'd say it's worth at least trying

[gusty trellis]

for sure and tbf i was already doing it that way and chat gpt advised pretty much the same, the action are clearly handle by the task and its a state aproach but i have one real problem

Its that when one of my sequence of task abort it seem i can't reset the abort state even if the boolean changed

[misty wharf]

I've found better luck having custom decorators etc. which query state directly from the NPC, eg. if their current task failed, or if they have any tasks remaining, etc.

[stark tangle]

i'm disable a Run Behavior Tree node in a controller, and dont have a default BT in a project settings
but pawns still runs a BT
where they get it from?
i cant find any settings about it anywhere

[dense owl]

Wdym you disabled it?

And what are you talking about BTs in project settings lol

[stark tangle]

really lol

when i say i disable, i mean disable. disconnect
no logic to give pawns a BT is running
but they still get it somehow

[dense owl]

That says default Blackboard. Not default BT. You can also right click disable node so I had to ask because you’re asking questions that don’t make much sense. So to clarify, show screenshots of your code and explain what exactly happens at runtime, if you want help that is

[stark tangle]

its obvious default Blackboard, and not a BT, but its only one place, where BT and BB is mentioned outside of controller

[misty wharf]

How did you determine the BT is still being ran?

If the BT is still being ran despite disabling it, then you have code elsewhere which runs it

If you have a BehaviorTreeComponent on your controller, it may also automatically start any default BT assigned to the component if you have Start AI Logic On Possess enabled

[stark tangle]

its easy, they dont stand still, when spawned, they doing BT logic, going to waypoints, fighting
this component can run a BT automatically?

[misty wharf]

Nah that one shouldn't

Try doing a search in your project for "Run Behavior"

that should find it if the node is used anywhere else

[stark tangle]

i try, and find it in controller, where this logic is disabled
and nowhere else
i know, its strange, i dunno how its possible

[misty wharf]

🤔

There's no C++ code in your project?

[stark tangle]

nope, blueprints only yet

[misty wharf]

Have you hit Compile on the controller BP lately? There are sometimes weird cases where it doesn't get compiled and you get the old behavior

Beyond that not sure why it would still be running if you disconnected the node... could be some strange case of blueprint corruption

[stark tangle]

i try to fix it about 3 days, so i hit Compile everywhere a lot

[daring locust]

I have this area that's marked with low height nav modifier, but my NPC (who is like half as tall as this space) still won't go trough it. Any way to make the NPC pass under low ceilings?

[misty wharf]

What do you mean "won't go through"? Is it refusing to pathfind through that area or getting stuck on it or what exactly?

[daring locust]

It refuses to take a path trough the area. I programmed the NPC to go directly to the player's location and if I stand on the other side of the low ceiling, it will always take a way around. If I disable the ceiling from affecting navigation, then it passes just fine, otherwise it will see the crawlspace as an obsticle

[misty wharf]

It's possible those are ignored by the default nav filter but not sure. I would try creating a custom nav filter which specifically allows pathing through there, and using that when you do the movement

[daring locust]

How do I change the nav filter?

[misty wharf]

the move to's and such should have an option for specifying which one to use

and I think the controller or character might also have a value for "default nav query filter" or whatever it's called

[daring locust]

I tried generating the nav mesh only in this 1 specific spot and it generated correctly, the NPCs could pass trough it no problem, but when I tried putting nav mesh on top of the bridge, the nav mesh underneath it dissapeard. Can navmeshes not be layered?

[misty wharf]

It should work as far as I'm aware of

[daring locust]

This is the gap if I have 1 huge navmesh bounds across the level and the other one is if I have a very thin layer only on the floor 🤷‍♂️

Trying to add another navmesh bound in the gap doesn't help, it ghets overwritten

[misty wharf]

If you mean you want the navmesh to extend within that "brigde" that goes across, it kind of looks like it might be too small for your agent size based on how it behaves with the mesh on its sides

Or some issue related to that, rather than it just not wanting to generate on top of it

[daring locust]

Under it. The issue is that the small NPC (the little slug thing you can see in the middle) will not go under the bridge even though it's small enough to fit

[uneven cloud]

If you look at the pinned messages, you'll see a link to a course on the learning library called an introduction to AI with Blueprints. That's the best place to start.

[uneven cloud]

Have you set up the supported agent settings in the project settings? You can't do that sort of thing with just nav mesh bounds volumes.

[daring locust]

Yeah, I asked on Unreal forum and discovered that I was changing the wrong setting. I was nudging the cell height instead of the agent height. Changing the agent to be 30 units tall fixed my issue

[gusty trellis]

i followed ur advice and asked some chat gpt it work super fine thx again

[dense owl]

I doubt zomg would’ve recommended chat gpt lol

[uneven cloud]

ChatGPT is certainly not great at architecture. It's not great at most things, but code architecture it's really bad at

[cyan sigil]

Hello! I just discovered that my AI character no longer rotates to face their focus if they are playing an animation with Root Motion. Is there some way around this?

[uneven cloud]

No. At least not without significant amount of engine changes. The root motion takes over.

[cyan sigil]

Crud. So what kind of solutions are available?

I was hoping to be able to steer my AI character somewhat.

I guess I can try making an anim notify state that steers towards the AI's focus.

[uneven cloud]

The root motion takes it over. If it didn't, the different rotations would be fighting. You either use root motion or you don't.

[wide robin]

I have two agents of different size, one big and one small. Feels like that, when I use UNavigationSystemV1::GetRandomReachablePointInRadius(OwnerController.GetActorLocation(),Destination, ResearchRadius, nullptr, nullptr);, it uses the navmeshdata from the small one instead of the big one. There's any way to specify which Recast Navmesh I want to use instead of nullptr?

[pallid mica]

What controls the NavMesh Resolution of a given tile if no Modifier is placed?

I can see some tiles being Default, some being High. What is making that decision? :<

[wide robin]

I can't find a way still to get the correct NavData for the ai controller that is using it

[pallid mica]

If you pass in nullptr, it will use the default one from the NavSystem, which is probably the main, green one.

ANavigationData is just the RecastNavMeshData Actor that is in your scene.
I'm half-sure, that the function you are quoting would only allow movement on a different NavMesh if you pass the pointer in by hand

If you want to dynamically figure out which NavMesh to use, I think you can't use that function

The NavSystem has a function called GetNavDataForProps

Which takes in NavAgentProperties

Which are the thing that tell your Characters apart

It has a Map that it fills when a NavMesh registers. I think you had to map that NavAgent to the NavData somewhere for that to work

Yeah in the Project Settings under Navigation System

There is an array called Supported Agents

[wide robin]

thanks for the answer! But yes, that's what I'm trying to do. I have two agents, I need the second one to use the other NavData and not the default one

[pallid mica]

Where you set the Name and the NavDataClass etc.

[wide robin]

If I do everything in blueprint it works since (i think) it connects in some way the navdata to use to the agent size

what i'm having trouble with is to do that in code

[wide robin]

i can't find this though

[pallid mica]

I can't actually recall where one sets up the NavAgentProps

If that's on the Character or Controller

Actually

I think they sit on the MovementComponent

[wide robin]

for me it's more like, where do i get the navdata? i couldn't find a function for that up until now (i'm using AngelScript, just a disclaimer)

[pallid mica]

virtual const ANavigationData* GetNavDataForProps(const FNavAgentProperties& AgentProperties) const;

Inside UNavigationSystemV1

[wide robin]

Seems like there's something wrong with angelscript then

[pallid mica]

Your GetNavSystem needs a world ()

I have no clue about angelscript

EVerything I researched about alternative scripting languages in UE all let to the same result: They don't support everything.

So I never bothered using them

[wide robin]

that's fair

thanks for the help though!

probably i need to check with AS people and see if they can help

[pallid mica]

In theory your code should do:

UNavigationSystemV1* NavSys = FNavigationSystem::GetCurrent<UNavigationSystemV1>(GetWorld());
const ANavigationData = NavSys->GetNavDataForProps(NavAgentProps, GetNavAgentLocation());

Depends on where you code this

Pawn has a GetNavAgentPropertiesRef function

As well as the GetNavAgentLocation function

--
I'm still a bit stuck on what actually decides on Low/Default/High for the NavMesh resolution when it generates, if not using a modifier.

I can force it to a resolution with a Modifier, but I would love to understand why it uses "High" in some areas, and "Default" in others, automatically.

[pallid mica]

nvm, the collision on the existing nav modifiers were wrongly set up and used the fallback size 💀

[dire jungle]

is there a reason why modifying AI perception radius isnt exposed to blueprints? Sure you can set it on the controller directly, but I am wondering why you cant edit it as a runtime variable like you can with pawn sensing?

replying to myself - seems like they always planned to expose it but never did - just made a helper function in cpp and gets the job done

[misty wharf]

Yeah there's a few bits in there which are possible to change but just aren't exposed to BP's

Same with the vision cone width

[dire jungle]

ah good to know! i was a little worried there was some design issue with changing those bits at runtime, but seems like epic simply forgot to expose them

[stray widget]

hey devs!! I have come across a situation I have never encountered before... So I have this map and I want to Spawn enemies every wave and want it to move along a certain path... and there would be some conditions to it which includes... they would keep some specified distance from each other and also spread around the treasure as you see in the image... how do I approach this thing... some suggestions would be great to move in that direction

[harsh storm]

Is this dynamic? Or set paths?

[stray widget]

I have some experience with using NavMeshBounds and modifiers but am not sure if they are a good choice for this kinda problem

[harsh storm]

Spreading around the treasure, you can use EQS

[stray widget]

its gonna be a set path

[harsh storm]

Just spline that puppy up

Spawn them near the start of the spline and then have them walk the spline

When getting to close to the treasure pile, use EQS to find a suitable spot

Can generate a circle around the treasure location and then pick a spot that isn't already claimed

[stray widget]

I already have RVO in place... is it useful for this case

[stray widget]

actually I have a spawn point where the enemies spawn at random points within the bounds

[harsh storm]

Not really. Because they'll be trying to bunch up to get to the same spot.

[harsh storm]

Then spawn, find closest spline, walk that path

(Can even use EQS to find the closest spline 😅)

[stray widget]

can you suggest me some good resourcesfor this... and does it mean I need to have multiple splines in place

coz I just got started with eqs and I dont have much idea how to use it properly

[misty wharf]

Ugh, I wish this AI stuff wasn't so finicky

Trying to debug why some actors that I'm spawning aren't moving to the goal they're being given

aaand it turns out it's because they spawn in the air, which causes their start position to be invalid, and that just immediately fails the entire pathing request

So I guess I'm going to have to implement some kind of state machine into the AI for this where it would retry the move if it's being told to go somewhere, or maybe make a custom AITask_MoveTo which retries it or something...

[dense owl]

Do they need to spawn in the air?

[misty wharf]

No, but there's no guarantee it wouldn't happen by accident, or as a result of being nudged because partially clipping inside something during spawning

Similarly there's some instances where it might bump into something and that causes the pathing to fail, where it would often be desirable to retry it once or twice if not infinitely

Another thing that comes to mind is that if the AI decides it should move to somewhere in the middle of a jump... F.ex. say perception interrupts the current action and it ends up in a MoveTo. Although this could potentially be alleviated by simply locking AI logic for duration of any action like that, and it seems likely to work correctly

In any case, there's quite a few of these types of situations where it feels the default moveto's aren't quite robust enough 🤔

[dense owl]

Iirc there’s options like checking if your pawn is on the navmesh or increasing the project distance

[misty wharf]

Projecting is for the goal location :)

[dense owl]

Ah right 😅

[misty wharf]

I wonder if @uneven cloud would have any insights on this. I might be approaching the problem poorly since I don't have that much practical AI design experience 🤔

[dense owl]

I’d be surprised if she didn’t

She is the architect

https://tenor.com/view/eyebrow-hmm-gif-25634245

[misty wharf]

Yeah I was just about to post that gif as I saw that message lol

[dense owl]

I miss Mieszko, he’s the actual architect of this system

[misty wharf]

I've seen him post something on very rare occasion

[dense owl]

Used to honour us with his presence here but now I think he’s focused on making #mass

[nimble smelt]

Does a behavior tree tick at the same rate as everything else?

[celest python]

its bound to BrainComponent's tick

[old lichen]

LogNavigation: Error: Navmesh bounds are too large! Limiting requested tiles count (1065800) to: (1048576)

Navmesh is generated dynamically, should I still care about this?

[uneven cloud]

Projecting is also for agent start location.

[misty wharf]

Oh is it? Then that is certainly named very confusingly because in the move request it's definitely named project goal

[stray widget]

hey devs!! I am trying to use the eqs which allows the NPC to find a location which is untaken by the other NPCs... how can I achieve that... I am new to using EQS and still exploring all it has to offer

[uneven cloud]

It sounds like you first need better spawning. Where they spawn on the nav mesh and not encroaching anything. You can use the EQS, but I do it in code since you can get the capsule size when spawn deferred.

The logic is: spawn actor deferred. Get capsule radius and half height. Generate a point (I do a spiral from the spawn point). Project to nav mesh. Move projected point up by capsule half height. Check encroaching geometry. First one that passes is the spawn point that I pass into finish spawning.

[misty wharf]

Yeah that sounds like a good idea. The spawning is in a bit of flux because right now it's a fairly "dumb" implementation where they just appear out of thin air, while in the future we might need some animations and other stuff to occur

[uneven cloud]

It projects both. You can choose to not project the goal, but you can't for projecting the start location.

[misty wharf]

Hm, I don't really see it attempting to project the start point at all 🤔

Looking at the code for the task and in the controller, it clearly projects the goal location if that's enabled but don't see anything for the start point

[uneven cloud]

Use a context to get the other NPC's location they are moving to. Then do a distance test with item location and that context. I usually do a filter test so they have to be a set distance apart.

[misty wharf]

Anyway I think I do need to fix the spawning as you suggested which should fix this

And I need to also update the logic in general since if there's some animation or such being done by the NPC before the spawning has finished, it obviously cannot start a move request until that has fully completed

[uneven cloud]

It's in the AI controller. I don't remember the function, but it's there.

[stray widget]

am not sure I got your point... could you please explain it abit more

[misty wharf]

Yeah I checked in the AIC and path following comp also.. The only case of projecting something is projecting the goal in both of them

[uneven cloud]

What part don't you understand?

[stray widget]

like... do I need to test the distance for every npc... and when you say "context" what exactly do you mean by that

coz if I went onto calculating the distance between the other NPC and the target location... wont it be abit computationally expensive since there are going to be alot of them

[uneven cloud]

Yes you need to test all of them. Here's the documentation on contexts: https://docs.unrealengine.com/5.0/en-US/eqs-node-reference-contexts-in-unreal-engine/

[uneven cloud]

A distance check isn't very expensive. Also the EQS is time sliced.

[stray widget]

aaah that one!! so I got to this point as in the image where I get the provide a single actor for my case

[uneven cloud]

How I do it is that the logic doesn't start until the spawn animation has finished.

[misty wharf]

Yeah that makes sense

I have the spawning system set up to allow the designer to set up a "command" for the spawned NPC's, hence the whole thing with moving somewhere, so I think I just need to update it to accommodate the animations + improved spawning

[uneven cloud]

That's for generating points around.

[misty wharf]

It's a bit meh state trees and BT's don't like to exist on the controller... I have a really basic state machine on the AIC for being able to toggle between "scripted" logic like animations and BT's :P

[uneven cloud]

Whatever works for you. I just use the lock logic functionality for disabling the BT for reactions and scripted behavior

[misty wharf]

Yeah that's what it does actually behind the scenes :) Also gets disabled when the character is falling, but unfortunately it seems the CMC doesn't detect it correctly if it's in the air at the start

But there's probably some workaround for it that I just haven't had the chance to look at just yet

[uneven cloud]

It might start falling before the controller is possessed. Can't you check the move mode when the controller's possessed?.

[misty wharf]

It does actually, but it seems to not report MODE_Falling

[uneven cloud]

How do you know my job title?

[stray widget]

yes... but lets say there were 100 npcs trying to get a spot... where 50 has already aquired their own spots... will the other 50 go through all the npcs checking for a valid location for themselves to go into

[uneven cloud]

Now that I think about it, the controller is possessed during pre initialize components. The movement component might set the starting move mode during it's initialization, which is after the controller is possessed.

[uneven cloud]

Yes. You are trying to prematurely optimize something that isn't even going to be a problem.

[uneven cloud]

Yes. It means your entire area is not getting nav mesh.

[stray widget]

makes sense abit... my next query is how do I check the distance between the point and the npc... so currently the npcs are working something like as follows:

Moves along a spline till they get near the target location

and once they are within a certain distance, they are expected to run an eqs and find their spots

now for instance an npc reaches the area first... how do they decide which spot to choose from... another thing is does the npc chooses a location from one of those generated points

[uneven cloud]

What is the "point" in the context of "how do I check the distance between the point and the NPC"?

Why are they moving on a spline before they pick a point around the target?

How they pick a spot is by using the EQS. When you run the EQS it generates a bunch of points. Then run tests on them which can score the points and/filter them out. Then it sorts them based on the score.

[uneven cloud]

Have you tried using the EQS before asking these questions?

[stray widget]

point as in the location

am using the spline coz my current map looks like this

and the enemies are expected to move towards the treasures whilst following a path... I considered navmesh bounds but somewhere I felt that I am gonna need a lot of them since its a tower defense game and buildings are to be placed at certain locations.

yes I tried EQS but am still very new to this and am still exploring this feature

[uneven cloud]

"point as in the location" is not an actual description.

Once they stop following the spline and move to the location around the treasure, how do they navigate? You really should use a nav mesh.

You seem to be very new to AI in general and trying to overcomplicate things by over thinking. I highly recommend taking a step back and doing the AI with Blueprints course on the learning library.

[harsh storm]

The point of the spline was to make sure they follow the preset path, afterwards, you did want to use a navmesh, in combination with EQS to find the right location to go to. (This isn't directed at you Luthage 😅)

[stray widget]

yeah!! thats right... apologies!! I should have been clear enough🥲🥲

[uneven cloud]

Do they actually need to follow a preset path or is that more overthinking?

[harsh storm]

That's just what they asked for initially. To follow the same path.

So I said to spawn, find the spline and move along it.

[stray widget]

I Initially had a straight path that used a navmesh bound and it was working as expected but then i decided to get a decent map instead of that

now I need something as I mentioned earlier

the demo map looks like this:

[uneven cloud]

They don't use a nav mesh bounds. They use a nav mesh. The nav mesh bounds only define the area where the nav mesh is built.

[uneven cloud]

Do you actually need the spline though? As in you tested it in an actual map and went "they would look better if they played follow the leader"?

[harsh storm]

Something else you could do I suppose, is use like nav classes to have them not be a follow the leader type thing

[stray widget]

I was trying to make something like Bloon Tower defense

[harsh storm]

https://docs.unrealengine.com/5.3/en-US/navigation-system-in-unreal-engine/

The nav docs are actually a pretty neat overview of well...the nav features

[harsh storm]

So they literally just follow a straight line/path?

[stray widget]

not all of them... they have certain turns in some of their maps

[uneven cloud]

What I'm trying to get at is...is this actually needed? If they spawn at the same location and are going to the same treasure location, they will follow the same path.

It would actually look better if they didn't, but whatever it's your game.

[old lichen]

The navmesh works just fine

[uneven cloud]

The places where it did build will work just fine.

[dense owl]

You mentioned it a few days ago

[uneven cloud]

Not surprising. I was just poking fun back at you.

[stark tangle]

what`s the difference?

[old lichen]

As I said, it's dynamically built.

[uneven cloud]

What do you think that means?

[uneven cloud]

One uses the BT to get the blackboard and the other uses the Blackboard directly.

[old lichen]

What's your point?

[uneven cloud]

To figure out how much you understand what building the nav mesh dynamically does.

[stark tangle]

so, "Clear Value" is better, bcz lesser steps?

[uneven cloud]

It's basically the same. If you are within the BT, such as a BT task, the BT version is probably better since it's less BP nodes. But it's unlikely to be a noticeable difference.

[stark tangle]

its in the task in my case, i'll go "Clear Value" then
its a little more clear now, thank you 👍🏻

[distant citrus]

Hey all, can't figure out why my 'Move To' in my BT is failing... the vector 'PointOfInterest' is appropriately set, triple checked with debug spheres... during this time the AI debugger says current Task is None, I just can't see why this is failing immediately. The AI agent just stops moving when this is called. The same node works in other sequences.

Any help is appreciated:

https://gyazo.com/d51369bd983fad2b951eaa163e819b6d

[uneven cloud]

Use the visual logger to debug. The pathfinding logs the errors there.

[distant citrus]

so I hit the ' key to get visual debugger going, I don't see too much but I do see the following:

[distant citrus]

https://gyazo.com/02746a512331cc3665f490f220ee68ef

[uneven cloud]

That is the gameplay debugger and not the visual logger

[distant citrus]

Alright how do I use the visual logger

[uneven cloud]

The tool is in tools/debug. For how to use it, you can Google that.

[distant citrus]

https://gyazo.com/831cd2b5caa31210e7af67b03b787063

So OnPathFInished: Invalid is where the issue is occurring

Oh

[distant citrus]

https://gyazo.com/a10c2b4c09bc6c1f47d549d49b6ff328

Gotcha.

Okay.

That makes sense

Thank you! Valuable tool there that I didn't know about previously

[uneven cloud]

You're welcome

[uneven cloud]

The reason why this channel is so quiet, is that we just teach people to use the visual logger.

No need to ask questions, when you learn to debug it yourself.

[dense owl]

If only the rest of UE had these debugging tools 😀

[fickle vault]

Question: Is it possible to create an AI bot thing that moves, attack and all that shit, all without using NavMesh?

[dense owl]

Sure. Build your own navigation system

Why you would do that is another story

[inner zephyr]

Not sure when to call Finish Task in StateTreeTasks. Because of that I've been debugging my StateTree for a day until I found out it's that and not all the other stuff 😅.
I'd look into that a little more but my brain is mushy after all the debugging so I'd gladly take some some hints while my brain fixes itself

[sage ridge]

Hi everyone, new to the channel, I hope im posting a question about nav mesh in the right place.

Is there a way to make my bumpy terrain use my complex collisions, I adjusted the setting complex to simple in the details tab to no avail! any ideas as it makes my NPC i spawn quite a distance above the ground without it.

[obsidian star]

Hi there, i'm trying to use the AI perception component and it some function doesnt work , the "get sense class for stimulus" , when moving inside the sight zone it shows the name of the class correctly , but when making a noise it shows an empty string when checking for the name

[misty wharf]

Checking the name of what?

[obsidian star]

the return value from the function get sense class

[misty wharf]

Maybe show a screenshot that shows that also

[misty wharf]

Oh sorry I meant the BP logic :)

[obsidian star]

oh okay

[misty wharf]

I would try using the other perception event

On Actor Perception Updated or whatever it's called? The one which only has one actor's info in the event

That one just gives you the currently relevant bits and you don't have to loop anything to get it

[obsidian star]

i was using that one 🥲

[misty wharf]

Oh, why'd you switch?

[obsidian star]

because of the check that doesnt work, i need to know which sense has been updated so i can switch states accordingly inside the behavior tree

i think i found a solution

[misty wharf]

Ah

Yeah that should do the trick. That event gets called with each updated stimulus separately

I don't really know what's the point of the other one besides being a trap to confuse beginners

[obsidian star]

and i fell right into it

[misty wharf]

You're not the first one :D

[wide robin]

Hello! I'm back with a question on collisions this time. I have set the object in the GIF so that collisions are ignored for my BigAgent, for which the pink NavMesh is generated. The strange thing for me is that, as you can see, sometimes that mesh causes a blocking area for my agent and sometime doesn't. Consider that my AI radius is huge, around 1150 unreal units

Any idea about what could cause that and how to avoid it?

[pallid mica]

Does anyone know the Math that is used for narrow passages, regarding CellSize and AgentRadius?

E.g. if I have a CellSize of 20 and an AgentRadius of 30, passages of width 100 don't produce a NavMesh

Like, a 30 AgentRadius Agent is 60 units wide. That fits a 100 units passage.
But instead of calculating it like this, it seems to remove the 30 from each wall, leaving only 40 behind.
But even that 40 would be fine, a CellSize of 20 should be able to fit twice into the 40.
So why does it not like the 100 unit passage.

[worldly quail]

Anyone knows why my AI can't go through the door?

[wide robin]

what does the visual logger say?

[worldly quail]

found the fix, it had to do with the "AI move to" function

basically, it was called on tick, now I do a "do once" node, and refresh it when the "AI move to" fails

[inner zephyr]

Never mind I have to call it but for some reason the OnSucceeded and OnFinished pins will fire instantly and break it. But if I bind an event to the OnFinished using the async Task you get from the node everything works as expected. Anybody had that issue and knows why?

[pallid mica]

It shouldn't be on tick in general

Just fyi

[worldly quail]

where should it be?

[pallid mica]

Somewhere you are triggering the MoveTo

Something logically says "Now you should start moving to this."

That's what should call the event

Not Tick

[worldly quail]

thanks, I'll see what I can use as a trigger

[pallid mica]

What is currently preventing it from calling on the first frame?

Do you have any conditions in your tick before the move to node?

Other than the DoOnce

[worldly quail]

something i need to know, does move to continuously tells my ai to follow my player, or does it need to be refreshed every time my player moves (to update the destination location?)

otherwise, I do have a bool check on tick to see if the zom can move

checking it's status (alive, doing something else, has to follow player)

[pallid mica]

I think move to location would only go to the last location

Move to actor should follow

You could do some kind of tick check, or have the follower AI broadcast an event when it dies or similar

Can't give you a solution that fits all

[worldly quail]

cheers for the help tho

I'll just test stuff & see if it works well

for now, seems to work fine

[pallid mica]

In theory, you as the designer of the game and system should know what events could lead to the Following AI no longer needing to follow its target. And these events should broadcast and the follower can listen to them on the target. Such as OnDeath. Those are event dispatchers, or simply delegates in c++.

Polling something on tick of course also works, but you can probably understand how redundant it is to ask the target every frame if it's dead instead of the target telling follower when it died

[worldly quail]

you got any documentation on broadcasting?

(i'm quite new to ue)

[pallid mica]

Eh don't think so. Blueprints call the EventDispatchers. In C++ and other languages they might be called Delegates

Pretty sure you can find tutorials with those terms

They can be created in a BP at the bottom below functions and variables

[worldly quail]

and that's something I never saw

[misty wharf]

https://docs.unrealengine.com/5.3/en-US/ProgrammingAndScripting/Blueprints/UserGuide/EventDispatcher/

That'll probably get you started :)

[pallid mica]

To bind to them you need a valid reference to the actor that has that dispatcher. Similar requirement as accessing variables and functions of an actor

They are like a mailing list. Actor A subscribes (binds) to Actor Bs Delegate. When Actor B broadcasts the delegate it will call the function that Actor A bound to it

[worldly quail]

somewhat like interfaces?

[pallid mica]

That's also a good design for generally informing a growing number of other objects without having to constantly update code on the caller side

[pallid mica]

Eh no. I wouldn't try to compare that

[worldly quail]

looks interesting, and useful for what i've got planned

cheers

[pallid mica]

@worldly quail An Interface is used when you want to Group an unknown set of Objects (or Actors).
The best example for an Interface is for Interaction. If you have an NPC you can Interact with, and a Door, and maybe also a Lever somewhere.
Those 3 might not share the same parent class, but you want to call "Interact" on all of them (and many more).

If you trace them and get a reference to them, you have two options.

1. Cast it to all possible options (BP_Character, BP_Lever, BP_Door), which is horrible and doesn#t scale well
2. Call an Interface Function on the Actor Variable

With number 2. you would have an Interface that has an Interact function. You would add the Interact Interface to the BP_Character, BP_Lever, BP_Door and whatever else has to react to interaction.
Inside of each BP you can then implement/overridde the newly found Interact function.

When calling the Interface function on the Actor in step 2, it will simply do nothing if the Actor doesn't implement the Interface.
And it will call the specific code of the Door etc in case you happened to trace that.

[worldly quail]

https://tenor.com/view/spongebob-patrick-star-notes-noted-gif-26232003

[pallid mica]

--
A Delegate/Event Dispatcher works in the other direction in guess.

Let's assume you have a BP_Character that can die, and you want to notify the UI of the Player to tell them they died.
You also want to notify all the AIs that the Character is dead, so they don't attack them anymore. And maybe some more.

You again have two options:

1. Inside of some "Die" function/event, you get all the AIs in the game and tell them the Character died. You also get the UI and tell it the Character died.
   This again scales horribly. Imagine you have to add more and more things to this. The Weather subsystem should stop. The NPCs around you should gasp, etc. etc.

2. Have an "OnDeath" EventDispatcher/Delegate in the Character and simply Call/Broadcast that if the Character dies.

In method 2, it's up to the UI and AIs to bind to the OnDeath delegate to get notified. This can be a bit more tricky to setup, but is a lot cleaner, because you don't have to alter your death code anymore.
And it scales better cause if someone else needs to know they also just bind to the delegate.

--

If you want a real-life example, the Interface would be having a room of people, where you don't know anyone.
Now you want to know what each of them studied, but some of them might never have studied. You simply don't know.
The Interface could be a "StudyInterface" with a function "GetStudyTopic". You would go to a person, ask "GetStudyTopic" and those who never studied (don't implement the interface) simply say nothing.

For the Delegate it's for example a newsletter, where you register to and get told the latest news. The news poster doesn't manually keep track of who wants to hear the latest news by asking every one individually.
They send one to everyone who is registered to the newsletter.
Could also see it as maybe waiting in line with a number and someone shouting "Number 242 please". Everyone hears it but you can compare your number with the passed one (delegates can have params) and react to it.

Idk if this helps, but was in the mood to explain stuff.

--

OTHER TOPIC, I still don't get why my Agents can find a path into/through a corridor that is large enough to create it based on the NavMesh AgentRadius, but too small to fit the Agent based on its own AgentRadius.

[misty wharf]

iirc the way those things work is that the agent gets the "most suitable" mesh assigned to it

it won't actually take the agent's real radius into account outside of assigning the mesh for it

[pallid mica]

If there are two corridors next to each other, one 100 units and one 200 units. And the NavMeshConfig is set to an AgentRadius of 30, so producing a NavMesh through both corridors.

Then if there is an Agent with an AgentRadius of 80, it will be able to find a path through the 100 units corrirdor, trying to walk through it and getting stuck at the entrace cause its size is 160 (2x 80).
While it should walk through the 200 units corridor.

@misty wharf Yeah but that's pretty shite

We have lots of different AgentRadius and as soon as there is a Corridor that is large enough for NavMesh but smaller than the AgentRadius, they get stuck

[misty wharf]

I could be mistaken, it was a while since I looked at all this. Have you tried just creating a separate agent config for the big NPC and generate a secondary mesh for that?

[pallid mica]

Secondary Mesh has the same issue

All it does is cut off some options at the bottom

So if I have a NavMesh with AgentRadius 80, to cover that specific AI

And AI with AgentRadius 100 will suffer again

And I can't create tons of NavMesh

2 is already stupid

[misty wharf]

🤔

Yeah I'm not sure how expensive having multiple would be, but feels like there should be some kind of solution for this

[pallid mica]

Well, client has 2 atm

They take up 1.4GB in RAM

[misty wharf]

Wow

[pallid mica]

And with large maps cost multiple seconds to generate

You really don't want more than 1

We couldn't support 3 in The Ascent

[misty wharf]

I suppose you could in theory mark the tight areas with some nav area modifier?

[pallid mica]

We had 2, not sure why anymore

The footprint was big enough that console got problems

[misty wharf]

Oh you worked on The Ascent? Nice, that was pretty fun :)

[pallid mica]

Yeah I did quite a lot on that game

But yeah, idk, feels strange. The AgentProperties reach FindPath

And then are unused

[misty wharf]

What about using modifiers on it? Eg. you'd mark the tight spaces with Area_TooSmall or whatever you'd call it

Then the larger NPC's can use a query filter for those

[pallid mica]

I don't think manually patching up the areas is a solution

[misty wharf]

this way they should still be able to query the same navmesh but not try going through those

[pallid mica]

There is a lot of dynamic stuff being spawned

[misty wharf]

Ah 🤔

[pallid mica]

And that game has tons of islands the player can visit

The NavMesh should just work ^tm

[misty wharf]

I wonder if you could automatically mark them as part of generation if you can determine the cell is too small?

[pallid mica]

I don't know how I would know that the cell is too small

There are quite a lot of agents

Like, different ones with different sizes

[misty wharf]

Hmm

I guess in theory you could do a check of whether the nav mesh poly fits inside a given radius

[pallid mica]

Preferrably the FindPath code should just use the AgentRadius for its pathing

Idk if it even uses any radius for pathing

[misty wharf]

So if it doesn't fit inside the 100-wide agent's radius, it gets marked

[pallid mica]

My current assumption is that it doesn't

All it does is specify the NavData

And then it probably just uses a thin line

Cause it already knows the NavMesh is bound to its AgentRadius

While I feel like it should expand each point that it finds

And check if the poly is even large enough

[misty wharf]

Yeah, I'm not quite sure how the querying of it works, maybe it just doesn't take the radius into account at all when querying and only when building the mesh

Since it leaves a gap between the mesh and walls for the width of the radius so the mesh doesn't even exist in spaces where the radius couldn't possibly fit

[pallid mica]

Right, that's what I assume to

Which is a real bummer and I wonder how others solve this

[misty wharf]

I feel like there was someone here who did some navlink generation during navmesh building, they might have some ideas

[pallid mica]

Like, let's assume you have this map setup somewhere and the red circle is an enemy that is trying to walk to the green circle being the player.
The shortest path is too thin, but its AgentRadius is not the lowest in the list and the NavMesh (any) exists in that path.

The red circle is now trapped there

[misty wharf]

I do recall that during the navmesh generation there is a function you can use to customize the validation, so you might be able to use that to perform a radius check on the navmesh poly being processed

[pallid mica]

Cause it finds a path to it but can't walk to it

[misty wharf]

and then if it's too small, flag the particular poly so a query filter can be made to avoid it

[pallid mica]

I will have to research that yeah

Would be awesome to have Mieszko around ;_;

[dense owl]

Summon him with your mod powers 😀

[pallid mica]

I don't think he's on the server and I don't want to annoy him

Actually he is on the server

I still don't want to annoy him

[dense owl]

Could always annoy Luthage. 😅 runs

[uneven cloud]

Navigation streaming cuts down the memory footprint significantly. On Psychonauts 2 we had 3 recast nav meshes, flying navigation and relative gravity navigation. The memory of it wasn't a problem, because the nav data is streamed in and the entire map isn't always loaded.

The trick is to use buckets for the sizes and prefer the bigger radius when in between, which the system does already.

Having the radius size be taken into account during pathfinding is going to significantly increase the cost.

[daring locust]

Hey, quick question. Anyone knows how to disable a behavior tree after running it? Like, if I wanted to remove it and make the character stationary again

I found this. Is this okay to use?

[uneven cloud]

You really shouldn't be running the BT on begin play, because you aren't guaranteed to have a controlled pawn at that time. Use On Possess instead.

Yes. Stop logic is OK to use.

[daring locust]

Oh, fair point. Will fix that up asap <3

Didn't even think of it

[nimble smelt]

Okay so I know Recieve Execute AI in a BTT is called when that task is called by the tree. But how does Receive Tick AI work? Where do I put that? How do I wire that to a BT?

[harsh storm]

It's just Tick, for the task

So it gets called every frame

[nimble smelt]

Every frame for each associated pawn thats alive? Every frame the task is executing? Every frame the BT is running? Every frame regardless of whether it's assigned to anything?

[uneven cloud]

Every frame that task is running and for every AI running that task.

If you have 5 AI that are running task Foo, the task is ticked 5x per frame.

[nimble smelt]

This is actually the first time it occurred to me that AI tasks dont need to finish in a frame. I mean, obviously they don't, but I'd only ever written ones that do

I think I need to back up and ask some more general questions to get my head right about this stuff

What's the mental model for splitting functionality between a behavior tree and the pawn itself, or the AI controller?
Like in theory you don't need any of this right, you could just have a pawn that handles its own detection and movement and have all the logic just in blueprint functions. Messy, bloated, but perfectly feasible. Obviously splitting it up is a good call, but which parts stay in the pawn? Which parts should be in the AI controller, etc?

I also think I'm not sure what should go in a Blueprint Task. Like, if you take a canned task that comes with the engine, like "Move Directly Towards", there is no BTT for that as far as I can tell. Or if there is, it's just a thin wrapper, the actual implementation of that task is in the movement component attached to the pawn, which I guess is implementing some sort of interface which lets the Behavior Tree call that function?

So when the BTT is just basically "tell the pawn to do something", what's the correct division of logic between the pawn and the task?

[uneven cloud]

The move to doesn't finish in the frame it starts in. It starts the move and waits for it to complete. If you have a task that triggers an ability which plays an animation, you'd want to wait until it finishes before doing anything else.

[uneven cloud]

The mental model that really helps with the Pawn vs AI is that the pawn is the body while the AI is the brain. The AI controller and it's components (including the BT) makes decisions on what actions to take. The pawn then executes those actions. The AI decides the NPC should move from point A to B. The path following component gets a path from the navigation system and handles following the path. What does the actual movement, is the movement component on the character.

A behavior tree is just an AI architecture to organize the behavior in a way that makes it easier to use. Sure you can do a scripted AI, but that gets messy real fast and hard to debug.

[nimble smelt]

If I'm not using unreal's navigation system at all, where would you put a custom pathfinding system?

[uneven cloud]

Move Directly Towards is a BT Task, else it wouldn't show up to use as a task. It's also doing more than just "tell the pawn to do something".

[uneven cloud]

Are you making it in C++? Also why would you make a custom pathfinding system?

[nimble smelt]

BP. And because Unreal's pathfinding system takes a pretty hard assumption on the existence of gravity, which doesn't hold in my game

[uneven cloud]

I think your overall problem is that you are looking for the RIGHT way, when the answer is always going to be it depends.

[nimble smelt]

True enough, but I'm not looking for instructions to blindly follow so much as "best practices" to serve as a starting point

[uneven cloud]

Not writing a pathfinding solution in BP is a more important "best practice" than what should be on the AI controller and what should be on the pawn.

[nimble smelt]

Oh the pathfinding is already done, that was the first thing I did 😄

The prototype was just a pawn that moves towards the player, crawling all over the walls

Look at this guy go! With my excessive number of traces!

[uneven cloud]

The amount of performance issues you are going to run into is going to be...a lot

[nimble smelt]

Surprisingly not! My intention was always to start with a number of traces I knew to be excessive and then tone it down to the minimum that still worked

But I can actually run swarms of like 10 of these things before I even start running into perf issues

And 20 before it really becomes a problem

[uneven cloud]

Until you build the rest of your game

[nimble smelt]

I've got a vertical slice at least!

Everyone over estimates the cost of line traces

Besides i could probably do this with like 5% of the traces I'm currently using. Each pulse is 100 traces, when 20 probably would achieve the same results. I could also preemptively cull about 3/4s of the target points before tracing, by cutting out the ones i know are farther away from the player, or the wall

[uneven cloud]

Nah. Most people underestimates the perf cost of doing that many line traces from BP.

At least from my experience of profiling games

[nimble smelt]

Meh, it's a library function anyways, if it becomes an insurmountable perf issue I can yank that one function out and make it in C++

Last I checked my frame compute time was like 20% of the render time anyways, so I've got plenty of headroom for stupid BP schenanigans

[nimble smelt]

Oh here's a new question, AIPerception seems to just operate by events on the pawn. What's the best way to inform the Behavior Tree of things like "hey I can see a new target"

[harsh storm]

Perception events on the controller

Event happens, update BB key

BT reads from the BB

[nimble smelt]

Oh I thought the AIPerception component had to be on the pawn that had the sense

[harsh storm]

The component is supposed to be on the controller. If it's not, you even get a log message telling you that it should be.

[nimble smelt]

I kinda don't like using controllers if I can avoid it... I understand the idea of them I just find that locating that dividing line between "pawn logic" and "controller logic" is usually tougher than it seems and you waste time and code guessing wrong

It's something I thought i'd get more comfortable with as I used Unreal more, but it's remained oddly opaque

Now, is it possible to go the other direction, and have the behavior tree POLL the perception but only sometimes? Like if I've already got a target i'm chasing, I don't want to abandon that loop when I find a new one. I only want to look for new targets if my current one is dead or lost

[uneven cloud]

I'm not entirely certain why finding the line between the brain and the body is such a difficult concept to understand

[nimble smelt]

If the pawn's the body and the behavior tree is the brain, then what's the controller?

[uneven cloud]

I didn't say the BT is the brain. The controller is the brain and the BT is a component on the controller

[uneven cloud]

Yes. You can get the perception targets from anywhere. I prefer using the EQS, but you can use whatever you want

[nimble smelt]

Is there any kind of system or shortcut to get that, or should I just write a BT service for it? I ask because I sorta feel like I'm working against the system, like I can get a list of detected actors, and then iterate through that list and query the detection strength for each, and then return the one with the strongest detection, but the whole time I'm getting that distinct feeling I'm writing code that already exists somewhere

[uneven cloud]

There's already an EQS generator for getting the perception targets. And a service to run the EQS. Querying a bunch of things, running tests on them and sorting them by a score does already exist... that's the EQS.

[nimble smelt]

I remembered why it was my instinct to have the AIPerception component on the pawn and not the controller... Otherwise, how do you change the values for perception based on the pawn class?

An Eagle and a Sparrow have the same controller, but different sight ranges. "how good eyes are" is body, not brain.

[harsh storm]

Data asset

[nimble smelt]

Wait how do you set the parameters on a sense with a data asset?

[harsh storm]

Have the senses config as a property for it

[nimble smelt]

Oh, then the controller sets its config to the property one on beginplay?

[harsh storm]

Pretty much, yeah.

Can have a property on the Pawn, like "AI Config" or something like that. Then when the controller possess the pawn, it gets that reference and then sets up its config.

[nimble smelt]

That's smart, I've already got a baseclass for all my pawns so that'll work well. Thanks!

[stiff gale]

When using AISenses such as hearing and sight, how do I keep it always activated? Or manually trigger it?

If auto is not possible, then looking for something similar to ReportNoiseEvent for AISense_Sight

[dense skiff]

Hi is there any way, to make RVO avoidance work with agents, that do not move now (guards in doorways etc.)?

[misty wharf]

They are always active, I'm not quite sure what the issue here is. Sight keeps perceiving all actors that are registered for sight always, and hearing will always keep listening for noise events

[feral bluff]

hi, in context multiplayer RTS I need to check character tag to filter only enemy characters, but for different players enemy it's different characters. So I think pass player tag as parameter to EQS Test and check that character not have this tag and it means that it enemy character. How can I use FAIDataProviderStructValue or FAIDataProviderTypedValue?

I tried add some provider values:

    UPROPERTY(EditDefaultsOnly, Category = "Test")
    FAIDataProviderTypedValue TestValue;

    UPROPERTY(EditDefaultsOnly, Category = "Test")
    FAIDataProviderBoolValue TestBoolValue;

but looks like binding not works for typed value

so far the only working option seems to be to use different Gameplay Tags tests for different commands, but this sounds like hard code and I would probably like to avoid this behavior

[misty wharf]

The "typed value" struct is afaik used as the base struct for implementing the other ones. You might be able to create your own FAIDataProviderGameplayTag since this can be marshalled into a number (EQS only supports numeric parameters for whatever reason)

It's also possible to pass in arbitrary parameters directly into the EQS query, but it can get a little bit hacky: https://zomgmoz.tv/unreal/Environment-Query-System/Custom-EQS-parameters

[balmy knot]

I introduced a bug some time ago: my AIPerception system on my AIController is detecting itself (at least the Character that it is possessing) in the event OnTargetPerceptionUpdated. This did not happen before IIRC, any idea how to filter the owning actor out natively? I can of course set a check in the beginning of the event, but have the idea that this can be resolved in an earlier stage.

I have detect neutrals enabled btw, but this behavior did not happen a bunch of commits before

[misty wharf]

Not sure why that would be happening, but you can check the actor against the controller's pawn if you don't find any other solution to fix it

[balmy knot]

Alright, thanks!

[pallid mica]

So in my search for "How to handle different Agent Radius?", I still haven't found a proper answer.
Having multiple NavMesh doesn't solve it, cause the number of different AgentRadius is too big to have a custom NavMesh per.

Following the calls in C++, AgentProperties is never used beyond selecting one of the NavMeshes that fits best.

Deep inside the PImplRecastNavMesh, it doesn't seem to have any support beyond QueryFilter for Cost and some Link filter.

So I guess the size of whoever wants to find the path is not at all considered beyond choosing a NavMesh.
And I don't understand how one solves said issue then, where a unit with a larger AgentRadius than the NavMesh, seemingly finds paths for corridors (shortest path) that it doesn't fit through.

Either I'm missing something obvious or this is simply not a thing?!

Idk if there is a way to dynamically increase the cost of a given cell for a given agent.

I can't set any cost up pre-runtime, cause players could also place buildings that cause any kind of narrow path.

[misty wharf]

You might have to look at the detour bits then 🤔 That's what generates it in the first place

I think you can also extend FRecastQueryFilter which allows you to specify some filtering rules on the code level

But not sure if it'll solve this specific thing

[pallid mica]

@misty wharf I also somewhat noticed that the "Smallest Agent to traverse this navmesh" is kinda weird

If you have two NavMeshes, one with 30 and one with 120

And you have an agent with 31

It uses the 120 one

So nothing larger than 30 can traverse it, so the comment feels inverted

[misty wharf]

Yeah I remember looking at it and finding it kinda weird also how it gets picked

[pallid mica]

Like, I can make the AgentRadius the biggest per type

Instead of the smallest

But that just means that anything larger than 30 would not be able to traverse any corridor between 30 and 120

Cause it picks the 120 one and that will not generate nav there

I'm getting closer to pinging Mieszko

[misty wharf]

Yeah

[pallid mica]

@misty wharf So, I had enough and checked UDN

Funny story, Céleste (my wife) asked the same question 2021 for a client of ours, and she forgot.

In her question she linked this post

https://forums.unrealengine.com/t/how-do-i-set-up-navmeshes-for-characters-of-different-sizes/280867

With Mieszko's answer being

The basic way we support different sized navigation actors is to have a separate navmesh per agent size. The reason is that Recast pre-bakes agent size into the navmesh structure for efficient pathfinding.

You can configure navigation system’s supported agents in Project Settings > Navigation System > Supported Agents

Different-sized agents support on a single navmesh is in out backlog, but it’s not going to be done soon.

–mieszko

Different-sized agents support on a single navmesh is in out backlog, but it’s not going to be done soon.