gameplay-ai

[slow bobcat]

Hummm

[hallow compass]

i guess weird stuff happens when you have a insanely big character

thanks a lot for helping out

[slow bobcat]

Yeah... We use the capsule size to calculate the extent of the projection when moving

[heady silo]

Attempting to write a service to enable multiple AI to move in a snake. Please let me know if this is too heavy for an aitick/completely deranged

[silent hamlet]

Find nearest actor might be heavy on BP on tick but it really depends on how many units are running this and what else is going on when this is ran. For a real answer...profile!

[zenith hill]

How does a task return if it fail or succeed?

Blueprint Service to be exact

[slow bobcat]

Services do not succeed or fail. They just tick when relevant.

[zenith hill]

Ye i have to tick the bool that starts the tick

Otherwise service don't get activated

[unborn jungle]

Great that there are now automatic navlink generations, are there any changes to large world navmesh generation since 4.27? Something like lego fortnite where it's not possible to build the navmesh offline

Would that be using invokers?

[slow bobcat]

Lots. You can now bake the nav in with World Partition using the concept of nav chunks from 4.X for example

[unborn jungle]

Can I generate PCG world at runtime and bake before the player starts?

With generated navlinks etc.

[slow bobcat]

There's no baking at runtime in unreal afaik

That's pretty much dynamic generation

[unborn jungle]

In 4.27 with dynamic navmesh you can delay player start until a navmesh delegate triggers

I wonder if that also does the automatic navlinks

[slow bobcat]

Ah yeah that you can do still in 5

You can still lock nav generation, wait for it etc

The automatic nav links no idea, haven't tested that feature yet (we don't need it)

[unborn jungle]

Thanks! I wonder how lego fortnite does it as they use procedural worlds afaik

So maybe it's all dynamic or using invokers

[slow bobcat]

Probably

Is the world really procedural in that game or is it more a "you can destroy/add stuff to the existing map"? Never played

[harsh storm]

I've heard that the automatic nav link generation can cause crashes in 5.5

[jagged plover]

Bit of a noob question, I'm brand-new to StateTree but I've got extensive experience with BehaviorTrees. One thing that isn't clear to me is if there's an expected pattern/system for monitoring states on the AI, basic stuff like current health or ammo.

For behavior trees you'd usually use the Blackboard for this, in StateTree it seems like maybe the expected approach would be to bind to blueprint properties as parameters? But something surprised me here, which was that I have a Health property on my Vitals component, and though it's marked BlueprintReadOnly I cannot bind to it unless I also make it EditDefaultsOnly or EditInstanceOnly rather than Transient, which makes me think that maybe this isn't the expected method?

Look for some guidance if other people already have established patterns for this. 😄

[harsh storm]

In 5.4/5.5, you can use global params as sort of a "blackboard for State Tree"

But in general, property binding is the expected workflow.

[jagged plover]

How dynamic is the binding? Since it wouldn't let me bind to something marked Transient it seems like it only updates once at the start of the tree rather than anytime that value changes, right?

Which I guess is why you'd want to do something like global params plus a global task that updates them on Tick?

Or binds to delegates that trigger when the values are changed, etc.

[harsh storm]

I don't know how the binding system works under the hood, so I can't answer the first one.

But you don't need a global task to update the global params. You can bind to them or set them from outside of the ST

[jagged plover]

Gotcha, I was thinking you'd need the global task to keep them updated as they changed. But I'll have to investigate to what extent the system updates those values.

[harsh storm]

It is regular property binding or the new set/get value nodes for them. Available in C++ in 5.4, but BP in 5.5

[jagged plover]

get/set value meaning BlueprintGetter and BlueprintSetter right?

[harsh storm]

No. There are some new nodes in BP that you can use.

You will be interacting with the global params, which is going to be in the ST asset

No sane person is making the ST in C++

So you would get a reference to the state tree, then get the global param, then set the value.

[jagged plover]

Ah okay, just to make sure I understand (sorry if I'm being thick) we're talking about making global params for stuff like health or ammo, and then having some other system (probably the ones that changes those values elsewhere) set the value of the global param by getting the state tree reference and updating those params, right?

[harsh storm]

I can't find the doc page, so watch this overly long video. It'll cover what I'm talking about. Probably closer to the 3rd act though https://www.youtube.com/watch?v=AQn8KtyMsSE

[jagged plover]

I miss when people wrote articles but I'll watch this video at 2x, thank you!

[harsh storm]

Yes. This is kind of a quasi blackboard. Personally, I still think the workflow is rough, but w/e.

[jagged plover]

Yeah it makes sense, thank you.

It seems like the other option would be for me to take the Health property on my VitalsComponent and remove Transient from it, if I do that I can bind to it directly as long as I have a param for the vitals component itself that gets set as part of the Context.

But...that seems like an anti-pattern, I don't want to serialize the value of current health that way.

[harsh storm]

I'm not sure why you want your health prop to be transient personally. Seems like something you'd want to serialize, but I'm sure you have your reasons.

[jagged plover]

Only relevant to serialize for savegame reasons I suppose.

It's current health not MaxHealth.

I've spent most of my career doing networked MP so save games aren't something I think about much haha.

[harsh storm]

That's why we make roguelikes. So we don't have to save.

[jagged plover]

I watched the video you were right this is a very gross workflow but I see how it works, haha. Feels like it would be much easier to just...be able to bind a global parameter directly to the output of a task but maybe there's some underlying reason Epic didn't go that direction.

[jagged plover]

I think I actually maybe found a simpler way to do the blackboard pattern? If you make a task with a bunch of Output properties and add it as a Global Task on your state tree, then you can bind to those properties in any state. No need to do the Set By Ref pattern in that video.

Is there a downside to this I'm not seeing?

[harsh storm]

Can you access that from outside of the state tree?

[jagged plover]

Like, in the AI Controller for example?

[harsh storm]

Yeah

[jagged plover]

I assume no, at least not via blueprints. So yeah if you needed that it'd be a downside, but these values are coming from outside the state tree anyway so I think that's fine?

At least for my existing use cases.

[harsh storm]

Then I don't think it fits the bill personally.

As a quasi blackboard that is

[jagged plover]

I guess since you can't easily do shared keys this way, yeah.

[harsh storm]

Being able to access it outside of the ST is the important bit imo.

[jagged plover]

It's just an FPropertyBag under the hood so in theory I could write some utility functions in C++ to expose it.

Not sure that's better than just using those blueprint nodes in the video though.

[coarse smelt]

Hello, I'm using BehaviorTree to get my AI characters to find hiding locaitons near the player, but I was having difficulty stopping them from running in front of the player when navigating. To solve this, I created a "viewcone" on my PlayerCharacter by adding a BoxCollision component and setting it to be a DynamicObstacle (in the BoxCollision>Navigation details). This works well, but is there a "better" way to do this? And does anyone know if there's a way to do this with a cone collision component, instead of Unreal's sphere/box/capsule?

[misty wharf]

I did something where I wanted NPCs to run across the player's view where I used dot product to see which points would cross the player's line of sight and scored those points higher in the EQS query I used... I don't know if it's any better than what you're doing though

[coarse smelt]

Thanks - good suggestion. I'm having an issues with my current approach (attaching a BoxComponent and setting it to DynamicObstacle), because I can't work out how to disable/deactivate the BoxComponent, or otherwise change the Navigation settings at runtime - this is important because the Ai won't approach the player when it's on, and I need that to be a possibility!

Should it be possible to somehow just deactivate the BoxCollision component? I've tried the node "Deactivate", but that doesn't seem to work. I've also tried "SetActive"

[misty wharf]

It might work if you disable its collisions instead

[coarse smelt]

Thanks, the collision is already set to "NoCollision", so I didn't think that would work. Instead, I'm setting the scale of the component to 0,0,0, which seems to work for my purposes.

[zenith hill]

My service only get activate once and the execution is stuck at the failed children 0o. What can I do so it repeat from the root again?

[slow bobcat]

There's nothing in your sequence at the bottom. That's not supported. If the sequence doesn't have child nodes the tee will not evaluate

[zenith hill]

Actually the issue was my service was only activated once. So I had to use Receive Tick to constantly evaluate

but maybe im just misunderstanding

thanks anyway

[cyan sigil]

I have a question about NavLinks. I have a character that is using a mantling ability to jump over a low wall. However, the ability technically takes them past the link point so they turn around and walk back to it. Is there a way to force the character to be marked as having reached the end of the navlink once the ability finishes?

[stable plover]

If you want to simulate blackboard behavior for state trees what you could have is an actor component attached to the controller that state tree is running on. call the component blackboard or whatever, this basically is a storage component then make a global task to retrieve that component in your state tree. That’s a good work around to the storage issue

[slow bobcat]

Yep, that's not needed I think (I would need to check). You are making a bt tick, not making the service tick when relevant. Your sequence at the bottom doesn't have any leave nodes (tasks), so it will never activate and make the services relevant

[zenith hill]

newb question, I am baffled here.

I have blackboard keys and I have set the value.

but the value stay as default when it goes down the tree.

I have printed and confirmed the value to be correct but it's not reflected the same at Blackboard based condition check.

[zenith hill]

Fixed it 🤦‍♂️ pulled from the wrong pawn.

[odd kestrel]

Hello. I was able to successfully use StateTreePropertyRefs in both C++ and Blueprints with structs and objects, but I was wondering if it's possible to use it with Enums. I know that enums are fundamentally integers so in theory it should be supported, but I'm a bit confused on how I would declare the property ref in C++. Does anybody know?

[misty wharf]

enums are represented by UEnum in the reflection system, maybe that's something you'd need to use

[odd kestrel]

I have the feeling they are not exactly supported without some casts to int. Just as an exercise I tried to create a property ref in a blueprint to see what the UI would let me do, but it looks like the type is not resolved correctly

[hallow compass]

if i dont move the AI doesnt count it as a new stimulus, is there a way to make it continuously "detect" ?

but its seems weird, so maybe how i designed my ssytem is just bad

never mind

the issue was a if statement i missed

[harsh storm]

Does State Tree support binding to a subobject's property? I have a data asset on my AI that has a value in there that I would prefer to bind to. I can see that it can look at props from a component though.

[hallow compass]

have you tried ? im curious for the answer to

[harsh storm]

Yes, I have tried. But it isn't showing up in the list.

But there might be something I need to do to get it to show, so that's why I'm asking.

[harsh storm]

But when I try to to hover over the Parameters binding - it crashes. Womp womp

[hallow compass]

it looks like AI perception traces from the capsule center ?

is there a way to change that, for example i would like to trace with an offset (on the head)

[harsh storm]

Override GetActorEyes, or something with a name like that. Don't remember the exact name.

C++ only if I recall

[hallow compass]

i also found CanBeSeenFrom

[hexed copper]

If I had AI that I didnt want to move in a certain room until its unlocked how would I approach it? I was looking at navigation modifiers and volumes but I am not sure if I can ever make the cost high enough to just never try and also I am not sure how to handle changing it at runtime to allow it

[storm sail]

you can just set it as an obstacle

[hexed copper]

/** In general represents a high cost area, that shouldn't be traversed by anyone unless no other path exist.*/
UCLASS(Config = Engine, MinimalAPI)
class UNavArea_Obstacle : public UNavArea

The issue is "unless no other path exist", they walk straight through it in those cases

[storm sail]

Nav modifiers can be changed at runtime, tried that?

[hexed copper]

okay but listen

I was looking at navigation modifiers and volumes but I am not sure if I can ever make the cost high enough to just never try

[storm sail]

Oh

No you can't make the cost so high that they will never try

But what I meant was you can use a nav modifier to completely turn off the navigation area

[hexed copper]

how would I do that

[storm sail]

You should be able to set it to null see anytime. It's one of the options. Like you can Change it from obstacle to null

[hexed copper]

i'll try but I thought null was the same as having a collision thing there so I would need dynamic instead of dynamic modifiers. ill try it

yep that did it. Thank you, weird calling it null for something being untraversable but whatever

[hexed copper]

i found the secret sauce btw, its literally just because it has its flags set to 0 in the constructor AreaFlags = 0;

[hallow compass]

how expensive can be CanBeSeenFrom ?

im running a sweep inside, so it should be okay

[slow bobcat]

As expensive as you want. That will depend on how many fps you need for your game.

[hallow compass]

well now targeting a full 60fps would be nice

[stoic python]

Hi all!

I am learning to create a basic AI system (Not Gen-AI, please) that relies on the built-in AI MoveTo function alone. The idea behind is that I want to create a scenario where the AI will go to a few locations and interact with a machine or switch or anything that location has. It will be used for cinematic purposes while being interacted by the Player.

I have looked online about using the built-in AI MoveTo function but there hasn't been any videos that goes further to utilise it for animations and trigger events.

The instruction goes as follows:

1. Spawn At Location
2. MoveTo Location (Waypoint 1)
3. MoveTo Location (Waypoint 2)
4. Play Animation (Inspecting)
5. Wait to finish Played Animation
6. Player Presses "E" when Overlap Collision Box at AI's Location
7. MoveTo Location (Waypoint 3)
8. Play Animation (Inspecting)
9. Play Sound At Location (Explosion)
10. Play Emitter At Location (Explosion)
11. Destroy Actor (AI)

The instruction I describe is how I would approach it, but I would like to hear your thoughts on what would be the optimised or clean setup in creating this logic with the AI MoveTo without using Behaviour Trees?

I would appreciate to hear your suggestions and thoughts!

[slow bobcat]

Check what Smart Objects are. They have examples where they "inject" animations on the user AI. Also look for Play Montage nodes in BT's

[toxic perch]

Anybody know the best videos or tutorials to understand AI? I’m just trying to figure out what is easiest to understand (blueprint or behavior tree). I’m making a police game and need my AI to run from the player in vehicles and respond to police commands. I have a decent basic ranged AI but want to make it smarter.

[slow bobcat]

Check the pinned messages, there are some tuts therw

[naive flower]

Hey, im working with state tress and im getting this error. I cant see any imidate issues and its not pointing to the specific state that is causeing problems.

Task 'StateTreeBlueprintTaskWrapper': Malformed task, missing instance value.

[hallow compass]

when the AI is moved by an move request, is it controlling the AI controler rotation or pawn rotation ? if yes, how ?
im asking this because i a mhaving weird behaviors between the AI character rotation and controller rotation values

[slow bobcat]

If in your CMC you have "Orient Rotation to Movement", the AI rotation will point to the movement direction

[hallow compass]

From what ive done to my character and controller it should be rotated to the controller rotation

But from what i tested the rotation is fucked up

[slow bobcat]

But why are you using controller rotation on the AI?

[hallow compass]

im working on a smoother look at

also, while the AI runs towards the chased player, i have some custom behaviors that requires rotation changes

ideally, i want to tell the engine to let me control the rotation and dont do anything when it does a move req

[slow bobcat]

right yeah. So... you are overriding the DesiredRotation in the CMC from the Controller right (this what we do too btw)

[hallow compass]

im just doing AiController->setcontrolrotation

thoses are my settings on the character/controller

oh nvm, i think i know the issue

[slow bobcat]

and what's the issue you are seeing?

ah ok

[hallow compass]

let me recode something

okay so as soon as the movement is finished, the AI is looking to the correct value

so my guess is that the movement is overriding something that makes it ignore my rotation

[slow bobcat]

Then I guess you will have to debug in code where is the rotation set and who's setting it. And you don't use root motion for your movement right?

[hallow compass]

im not using root motion, its the default blendspace

allowing strafing didnt change anything

[hallow compass]

okay, its about the focus setting

[slow bobcat]

As in....?

[hallow compass]

there is something call focus on the AI, AAIController::SetFocus and AAIController::SetFocalPoint, it looks like its linked to controller rotation.
each focus that is set has a priority (default, move, gameplay)
when moving it seems that move is used

i need further research to check

this is called on tick

[hallow compass]

what can cause a move req to abort when RequireNavigableEndLocation is false ?

[wispy folio]

Does anyone know why AI_MoveTo Doesn't rotate the character when moving?

[hallow compass]

check your pawn and controller rotation settings

[odd kestrel]

Hello everyone. UE has recently added a native state tree task that allow us to use EQS, which was pretty helpful. One thing I can't figure out though, is how do I pass data from the tree to the EQS query. Let's say I have a query to figure out what is the point around my AI character farthest from its current target. With Behavior Trees, I would set the current target in the Blackboard, then create a Query Context object to retrieve it from the BB and use it to evaluate the points in the EQS query. But state trees normally do not use blackboards and instead I have data stored as StateTreePropertyRefs in the tree itself. So in this case, how could I pass the current target to the EQS query?

[slow bobcat]

Maybe failing to find a path? Not sure. If the movement has started already, you can check the callback for path events in the movement task. If the movement hasn't started yet, check what happens when finding the path

[hallow compass]

Its a repath error

As soon as i have my PC running ill tell you the exact code error

[hallow compass]

in more details, when using an AI move to task, what are bool bLockAILogic, EAIOptionFlag::Type ProjectGoalOnNavigation and EAIOptionFlag::Type RequireNavigableEndLocation for ?

well for the last two they are some info on the docs that werent on the source code https://dev.epicgames.com/documentation/en-us/unreal-engine/API/Runtime/AIModule/Tasks/UAITask_MoveTo/AIMoveTo

[misty wharf]

block ai logic stops BT's (and probably ST's) from running while the move is active

[hallow compass]

okay thanks

im trying to make my AI continuously go towards my player (even if the player is at a unreachable place) but the AI fails at some point everytime (repath failed error)

those are my params

co_await UE5Coro::AI::AIMoveTo(
        InstanceData->AIController.Get(),
        InstanceData->TrackedInGameCharacter.Get(),
        InstanceData->AcceptanceRadius,
        EAIOptionFlag::Type::Disable, EAIOptionFlag::Type::Enable,
        true, false, true,
        EAIOptionFlag::Type::Enable, EAIOptionFlag::Type::Disable
    );

screen for what params is what value

level screen:
here is my table, the AI is the giant pawn on the top of the screen

in red is the navigable nav data of the AI
as you can see, an area is red on the table (because im dumb and the nav bounds volume is to high. But his made me found that the AI will fail going towards me if IM NOT INSIDIE the red area

i think its because the AI cannot project my player position to the nav data

so idk what setting i need to have to make it possible

[slow bobcat]

Make what posible? AI navigate outside the nav mesh?

Can you show a video with that camera above showing the problem? To. Understand what you want to solve

[hallow compass]

no, i just want the AI to use the nav data to get the closest possible

but ofc, it cannot go further since no nav data in the player area

[hallow compass]

as i go to the border (where there is no nav data (red)) for the AI, it fails the move data
so it goes to idle and doesnt move

also, i found out that I cannot make a null area myself
when i place the default NavArea_Null it works
when i place my NavAreaSubclass, and copied over the value of the cost (340282346638528859811704183484516925440,0), it doesnt work
when i make a child of NavArea_Null and dont change anything and place it, doesnt work
(in my BPs both have all agents types maked as supported)

[slow bobcat]

I would expect the human to move towards the rat when you jump to the floor. What's the state of the AI at that point? Why is not moving towards the eat when is in the floor? Is the rest error you mentioned?

[slow bobcat]

Nav are null is not the same as a nav area with a very high cost. Nav area null removes the nav polys under it. A high cost area will be trasversed if it's the only path available

[slow bobcat]

The more I look at it, the more it looks like a bug in your logic. It seems is unable to move after finishing the previous movement request

[hallow compass]

well before i jump down the AI already failed the move and aborted, this happens as soon as im not in the nav data of the AI

if i spawn in a no nav data area and wait for the AI to see me (this triggers the "go to player" mode), it will freeze (because it tried to go to player and failed instanlty)

[slow bobcat]

yeah that's fine. but as soon as you land, it should just move since you are on nav already

[hallow compass]

well no since its aborted

i need to continually try a new move to to trigger it again

[slow bobcat]

yeah that's how it works

Your tree will re-evaluate, get to "you should approach the player" and try a new move request

[hallow compass]

ill check in source code to find how to reproduce that

because it should do the same for null navarea subclasses

[slow bobcat]

are you supporting partial paths? if you do, it will get to the closest point in nav possible

[slow bobcat]

what do you mean?

[hallow compass]

see here for my moveto req params

[hallow compass]

i made a BP that is the child of the engine NavArea_Null class, and it doesnt work like the null one

[hallow compass]

yeah i do, but it didnt resolve the issue

[slow bobcat]

not fully understanding what am I seeing it here

[hallow compass]

what do you want to focus first, null area of AI move to ?

talking about both at the same time is confusing imo

[hallow compass]

screen 1: using NavArea_Null (engine class), as you can see no nav data is generated on the table
screen 2: using NewBlueprint (child of NavArea_Null), as you can see nav data is generated
screen 3: proff that i didnt changed anything and that NewBlueprint is a child of NavArea_Null

thats all is done inside the class

[slow bobcat]

AI Move To:

• those params look fine
• Once the AI Move To finishes (either success/fail/abort/etc), you need to trigger another one to move. By the sound of it, you are missing this part where your AI re-triggers a move-to when the rat lands on the floor (or when it gets back to a navigatable area for that matter)

Nav Area:

• ok image2 is super confusing. Are the area nav polys painted in blue/grey? what's the swuare selected in the middle?
• What is your BP Nav area inherinting from? I'm guessing you are using this?

UCLASS(Config=Engine, MinimalAPI)
class UNavArea_Null : public UNavArea
{
    GENERATED_UCLASS_BODY()
};

Have you re-started the editor after creating your new nav area? they get registered on boot, maybe it's that? long shot I know

[hallow compass]

so for AI move to, i should spam requests until it doesnt abort ?

i was doing that but it felt wrong design

[hallow compass]

ill try that, but my child was a BP only, not C++

[slow bobcat]

no, your Bt would check things like "is my target visible?" and then trigger an AI Move to (a task). Once that's finished, the BT will re-evaluate (or continue if it's a sequence and movement was successful) and do what it needs, which might be to move again

[slow bobcat]

doesn't matter

[hallow compass]

im not using a BT, but i see what you mean.
but in your example, this means requesting on tick, because here the rat (player) is visible for the AI, even when not accessible

[slow bobcat]

what are you using then?

[hallow compass]

even tho, the AI isnt even trying to come close enough, it fails, even if it could walk 100m to be closer

[hallow compass]

state tree tasks

[slow bobcat]

that's a "you" problem, nothing to do with movement logic of the engine I believe. 99% sure

[slow bobcat]

then once your movement task finishes, you have two:

• on failure: transition to a higher state like root to re-consider what to do
• on Success: do something else if close enough (like attack the player or somehting)

feels like you are finishing movement and just staying wihtin the same state

can we see your state tree?

[hallow compass]

on sucess this is what i aim
the issue is that it fails when it shouldnt

[slow bobcat]

why shouldn't it? that's the key question.
If you debug using visul logger, I think you will get some info.
The player is probably to far from the nav to allow projection on it, hence the AI can't find a path. It should find a partial path though (that you will see in the Visual Logger too)

[hallow compass]

going back on the nav area, i just made a c++ class that copies the nav area null c++ class constructor, and it doesnt work
my guess is that the engine hardcoded something

[slow bobcat]

don't copy anything. Just inherit from it and reboot the editor

[hallow compass]

why shouldn't it? -> because the AI doesnt try to use a partial path somehow

let me record a more obvious case

[slow bobcat]

you need to debug with visual logger to get more info about it

[hallow compass]

the AI doesnt move as soon as i try to move to the player

[hallow compass]

i never done that, ill check the VLogger

[slow bobcat]

once the rat is in the desired position and while the AI doesn't move, eject (F8) and select the Top view. Le't see how does it look

[hallow compass]

done that, doesnt work either :/

[hallow compass]

yeah it sounds like the player position cannot be projected

[slow bobcat]

then there's something wrong locally because I can guarantee that works. We have a bunch of custom nav areas. Simply inherithing should just work I believe.

[slow bobcat]

right click on the Navigation category -> disable all but this. Look for the first time it tried to look for a path. Should be one of the starting lines

[hallow compass]

the nav category doesnt have anythiung regarding path following

its just abotu mesh registration and nav data

[slow bobcat]

if while in this view, you click with your mouse wheel on the rat and drag to the nav border, you will be able to see how far from is it. Let's see that distance

[slow bobcat]

you are showing Path following, not Navigation

you want to check navigation for this case

[hallow compass]

as you can see, nothing useful in our case

only reg stuff

[slow bobcat]

what about that warning there

[hallow compass]

roughly 4700cm, the acceptance radius of the request is 100

[slow bobcat]

is that the AI capsule center?

[hallow compass]

yes, roughly

[slow bobcat]

and what projection extent is being used in your moveTo AI requesT?

[hallow compass]

oh i forgot about that param

is it this ?

[slow bobcat]

I doubt that's the problem, since you have your AI moving in other cases but worth checking

[slow bobcat]

check it in code. Debug the code where they project the start onto the nav, just in case

[hallow compass]

it works fine as long as im in the painted area

[slow bobcat]

you will have to debug how they project each location and why partial paths are not being calculated

[hallow compass]

yup, for some reason the query wants a valid end location

[slow bobcat]

I don't think your AI is even getting here, but worth placing a breakpoint just in case

[hallow compass]

when it fails i have a path event fail event

RePathFailed

[slow bobcat]

then you can follow the stack and check why did it fail

I would check all this code (if you have trouble debugging, remember you can suround the function with UE_DISABLE_OPTIMIZATION and UE_ENABLE_OPTIMIZATION and re-compilie the cpp file. That will allow you to see the variables as if you were running the editor in debug mode. Works with Live Coding too. Quite convenient when dealing with engine code)

[hallow compass]

ill do that thanks

RePath isnt always called, in my latest test its wasnt called like 10 times consecutive

[hallow compass]

ill debug from the UEVLOG error

okay i hate myself, some external plugin i was using to set requireendlocation wasnt used

i got it mostly working, still got some weird cases when my player drops of to an unnavigable area, RePathFailed is called

[misty wharf]

Hmm, it really makes me wonder how did Epic intend SO's used for multiple purposes 🤔

Like.. they support multiple activities, but actually not really. You can have multiple activity tags on a SO, but that's where it really ends. If you want a single slot to do this, then you can't use literally any of the other features because none of the other features allow you to work with activity tags in a convenient way.

I have some of my own systems built around this but as I adjust these it really makes me wonder what's the intended usage for this because I'm having to work around it so much :P

It seems like in theory you could just specify the slot in the same position twice, once for each activity type, and it would then let you have preconditions and others for each activity type separately... but then you need to have some kind of system in place to disable the other slot. And it gets increasingly complicated if you have more than one slot where these activities could be performed from

[elfin python]

What is the correct way to report noise from a player that has a AiPerceptionStimulySourceComponent?
Calling ReportNoiseEvent works even if I have unregistered the AiPerceptionStimulySourceComponent. I'd like to set it up so that unregistering the component is a one step solution to not be perceptible by NPCs

[misty wharf]

The stimuli comp is only really required for sight iirc. Report noise just reports it directly to the system

[elfin python]

ohh ok

But the component does have Register As Source For Senses with two inputs; AISense_TrackingSight and AISense_Hearing

[elfin python]

So before calling the noise event I need to check if the player is registered with the perception system for hearing. I can't seem to find any entry point for that to use in a BP 🤔

[dim jackal]

You can do simply get the RegisterAsSourceForSenses array which contains the registered senses

[elfin python]

nice!

no wait, that doesn't tell me if the perception stimuly source is currently registered or not

[misty wharf]

Yeah not sure if that's in BP's. You could consider just setting a bool when you register/unregister

[dim jackal]

If a sense is registered, like hearing or whanot, it'll be in the array, otherwise it won't. Unless I'm not sure what you mean? If you want to check whether the component itself is registered or not, then you'll have to access bSuccessfullyRegistered in C++, unless you wanna modify the engine to make it readable in BP

[elfin python]

oki, I'll go with tracking my own bool variable for now, and if I have more cases like this to consider if I rather want to inherit from the class and make bSuccessfullyRegistered accessible

tyvm for helping me work this out 😊

[dim jackal]

you won't have to inherit anything, just add UPROPERTY(BlueprintReadOnly) on top of the variable

[dim jackal]

No problem!

[hallow compass]

for state tree bindings, is it possible to bind it by ref ?

i got a struct thats binded, and it feels like its getting copied

[hallow compass]

since it seems to works with UObject ptrs, should I make a Uobject instead of a struct ?

[azure bloom]

Hey got a question. How should I do this? So for my games, I am going to have them trigger based, and for this game I'm working on, you go through a doorway, and a shelf falls after you, blocking the door way, and then you hear a huge screech by the monster. Should I have that logic in the BT or just have that play after the shelf fall?

[dusky lodge]

Has anyone come across an issue where Navlink Proxies do not trigger when a Pawn arrives at them?

[slow bobcat]

Completely up to you. Event Driven is always the best way to do things and it's a very tailored approach to each game. In your case everything could be part of the same event or a chain of events (the screech happens because the monster listens to OnShelfFall, the shelf falls because it listens to OnDoorOpen). Up to you really

[slow bobcat]

Yep. You need to check your proxies support the nav agent used by your AI and, if that's all right, look for other info (check visual logger in the navigation category). Let's see if that gives some hints

[azure bloom]

So I can make sort of a "Game Controller" or a handful of controllers that activate after one another that do different stuff, depending on the part of the game. For example, a C++ class deprived from GameController.cpp that for example has a reference to BoxTrigger.cpp and then I can subscribe to the event from BoxTrigger.cpp

[dusky lodge]

I should clarify, it seems to be an intermittent issue, in that the Pawn will run through the proxy but when it doubles back it will end up using it correctly.

[slow bobcat]

check how the path looks like when that hapens in visual logger. Maybe you have some logic somewhere you need to tune a bit

[slow bobcat]

sure. It's really up to you.
If you want to do a very scripted event like the one you described (it always happens), I would just do it in one single place (upon that door opening in that specific door instance I would trigger the shelf and the scretch events).
If, on the other side, you want the scretch happening upon the first noise that happens in a room, you can do something more elaborated like things triggering noise events in the monster etc

[dusky lodge]

Ill verify with that first and reach out if there isnt anything obviously wrong.

[keen crow]

sooo i have the same question atm. is there a short answer to it so that I don't have to analyze the source code of both tests?

[slow bobcat]

the batch uses the batched path check. When you path find in unreal there's an "expensive" (open to discussion) operation in cpu to gather the data and project the points onto the nav. If you need to check several paths, you can gather data once and then project all the points.
It's a way of saving precious cpu cycles when you know you will project down several points onto the nav.
Internally it uses ARecastNavMesh::BatchProjectPoints

[keen crow]

hmm, so when should I use regular EnvQueryTest_Pathfinding over EnvQueryTest_PathfindingBatch? when the amount of EQS items is not that big ||and the "big" meaning here is deduced experimentally|| ?

[slow bobcat]

pretty much yeah. If you were to discard most points and end up with a single location, use the regular path. if you will end up testing several items' path, then used the batched.
The way I work with this is that I always disable the auto-arrange tests in the generator and I order them myself (I use lots of custom tests and I know what they do inside). I run the cheapest ones (distance, dot products etc) and always leave the path checks at the end, so they test with the smallest amount of items possible

[keen crow]

ok, thanks for the clarification. i do the same thing with tests ordering and disabling auto sort, but thanks for reinforcing my thoughts that it matters

[gritty bridge]

Hello y'all. I need help with my AI shooting behavior tree. Basically my Enemy strafes around me and is supposed to shoot me

here is the code for playing the animation montage

and here is the code for the behavior tree task for attacking

However the AI stops strafing after strafing once and doesnt shoot or play the animation of shooting
sometimes it doesnt even focus on my charachter alltogether
in the behavior tree, everything works fine until i trigger the enemy to attack me, the behavior task stays on default attack and doesnt move
Can anyone help me?

[odd kestrel]

I'm a bit curious with how StateTree transitions are supposed to work. I have a condition that is supposed to abort the current state and transition to another state. So I created global task that runs on tick, and upon a given criteria being met, I send a state tree event. One of the outermost states of my tree has a transition set to execute "on event" with a tag matching the state sent. Using the tree debugger, I can see the transition is triggered when the condition is met, but it never actually goes to the desired state. Does anyone know why?

[slow bobcat]

Try setting that OnEvent Transition in the root node just in case

[odd kestrel]

same thing. And we can see the transition was triggered:

[stable plover]

Interesting, mine works fine, I call the send state tree event function when I transition, how I have my tree set up is if I’m in combat it goes Combat ->attack, attack is a child of combat, I bind my event transitions to the combat and works fine, because I’m assuming if you make a transition rule for the event in a state and are not in that state, even though you send it you are not in that state, so I put all my event transitions to my Combat

[odd kestrel]

that's precisely what I'm doing. And I also tried what Bruno mentioned above and put it the Root state just for troubleshooting purposes. But for some reason it never enters the state the transition points to

[stable plover]

Wow

[odd kestrel]

ah, found it. The state that I was trying to transition too was inside of another state, and the parent state had a rule so it couldn't be transitioned too directly. So the transition was trying to go to a state whose parent couldn't be selected

I changed the parent rule so it could be selected, but had an enter condition that requires an event

[stable plover]

Oh wow

[desert kelp]

I've got 4 AIs spawning, and I've kinda got them going into a group follow the leader type thing, where one AI is set as the "leader" and the rest should follow in kindof a wedge type formation, but the follower AIs are getting the position of the leader at a different time from when the leader AI decides to move, so they're not really following properly... I think my move/wait times are set a little different, if I fix that up should that fix it? or am I going about this the wrong way

[slow bobcat]

What do you mean by "at a different time"? Is it taking your logic several frames to propagate the position where they need to go?

[slow bobcat]

Can you break down this a bit? How come the attack is an event? How is that supposed to work?

How do you call finish in your AI bttasks?

[gritty bridge]

its called in my bp_enemy and its run as a task in my behavior tree

[gritty bridge]

for this i created the on attack end event and finishes the attack when the montage ends

[slow bobcat]

So...

• on BTTask_Attck you trigger the event Attack on your owner pawn (the AI) and then register to finish <- I would register to the end event first just in case. I think the event will take 1 frame to execute and should be fine, but the other order will ensure 100% that you always receive the callback
• you didn't connect blend out. That can happen and you will miss the end call event

That said, you mentioned that the problem is with the strafe and then that it gets stuck in attack. So.... Which one is it? Or is it 2 problems?

[gritty bridge]

im not sure i understand the blend out part

[gritty bridge]

the issue is with BTT_DefaultAttack, when i run the tree, everything works fine but whenever i connect the BTT_DefaultAttack, the tree gets stuck to it, not playing the animation or tracing the line

[slow bobcat]

Animations can blend out automatically if you play another mktnage after. If that's not connected and your mktnage blends out, the attack bbttask will never finish because you are not sending a Finish callback call to your bt. Not saying it's your issue (it doesn't look like it) but doesn't harm to connect it

[slow bobcat]

If you set breakpoints in each node within the attack logic, where does it stop?

[gritty bridge]

it would seem it doesnt stop at all

tried for every node in btt_Attack and the attack event

[slow bobcat]

If you set a breakpoint when you call the Event Attack... Could it be that casting is failing? Maybe that actor being passed is the owner of the BT, the controller?

You should use ExecuteAI node, that passes both, the controller and the pawn

Worth checking

[wooden trellis]

not sure if this is the chat to ask but - I am working on a tiny enemy that I want to crawl into a vent, but the nav mesh doesn't cover it. Is there anyway I can alter this specific nav mesh? This isn't a common enemy, so I don't want to affect any other NMs

[slow bobcat]

You can alter params like cell height and some others to try generate nav under surfaces. That vent might be too low (as in height) for nav to generate. On top of that, since you are generating nav on the top surface, there will never be nav below it. What you can do is to add a nav link that conecta both sides of the vent and set it's area to a type that only your crawling enemy can use

[wintry osprey]

Is there a way to check what made the navmesh dirty? Currently having expensive rebuilds when it shouldn't

[desert kelp]

basically I think the "move to" is changing the timing. if it takes pawn 3 a little longer to get to the target from where he was, then he's still waiting the 1s or whatever before getting the new location and moving. So just about all the time they're getting a location based on the leaders position while the leader is still moving, so by the time they get there he's elsewhere

[slow bobcat]

ah ok. Yeah, moveTo when using an FVector doesn't have repath upon Goal Moved. For that you need to move towards an Actor, then you get re-path while the goal actor moves, gaining continuity

[slow bobcat]

One way could be to show the navigation while you play and observe how that happens. You can also record with VisLog, will drop some data and, from that, figure out who's causing the issue. But nothing (AFAIK) will tell you "object X is forcing nav re-gen". I had this problem in our released game and we had to build custom solutions that required engine source changes. We ended up printing info in visual logger where we show the boundaries of the object causing the re-build, affected tiles etc

Or you can just set some breakpoints in code and check (I like visual debug in general)

[keen crow]

is there an easy way to run EQS in true async manner? I mean AFAIU EQS execution is spread over time by limiting amount of operations EQS manager can do per frame, but it is still ran in the game threatd. So is it possible to start EQS on game thread, then subscribe to some callback, and have EQS actually run on some arbitrary background thread? I assume there might be problems with accessing actors in generators, tests and contexts, but maybe someone has already tried to do something like that?

[slow bobcat]

nope. That's been a long pending item in the "cool things to have in unreal" list. There're some plans for a new EQS 2.0 system that allows for it, but that's about it

[keen crow]

eh. a man can dream

[slow bobcat]

I guess you can call the eqs execute from an async task in a background thread and see what happens? If you run the whole thing... Maybe? Tbh I never tried

[keen crow]

maybe I will one day... so far it's more of a curiosity question for me with a bit of hope that someone would say like "hell yea man there's an easy and stable way to do it"

[wise sluice]

Hello! Anyone already modified the FAIStimulus class by adding a FGameplayTag?
Having a FName is pretty bad compared to a FGameplayTag 🤔

Any pro/cons about doing that?

[misty wharf]

Unless you want to recompile the engine it's probably easier to do a conversion on it yourself between tag and name. I'd say there's probably very few downsides to doing the modification if you're so inclined

[harsh storm]

To be fair - async doesn't mean "not on the current thread". So time slicing is still true async. You're asking for actual multithreading (which async is not)

Mieszko has stated that he has some plans for EQS 2.0, but doesn't know when he'll be able to get to it. Based on trends, I'd imagine it'll be somehow tied to #mass though

Just like I'm sure AIPerception will eventually want to be powered by mass as well 😅

[misty wharf]

I hope they give it some method of actually receiving parameters whatever it is using at the end

Coincidentally I'm currently again annoyed by how poorly EQS can be parametrized

[harsh storm]

Yeah, it is kind of annoying. But ohs well 🤷

[wise sluice]

That's ok to recompile the whole engine 😄

Thanks!

[slow bobcat]

What? You don't like passing params with strings to match them?it's bullet proof! XD

[misty wharf]

Not only that but it's also stuff like actually being able to tell it which actors to use because the only way to include those is via contexts and that's super clunky

I vaguely recall it marshals every parameter type into a float which makes it impossible to extend the parameter system for more complex types

[slow bobcat]

Yep. I have about 10k contexts (joking, but pretty sure the number is between 20 and 30, many used in one specific eqs or test and that's it)

[slow bobcat]

Yeah, that even gives problems with integers (I can pass you a udn link about it)

[misty wharf]

Yeah I don't mind having contexts but it's like... you have to store it somewhere like on the NPC doing the query, and you literally don't need the value for anything else than some EQS query, so you just end up having a whole bunch of single-use props in the class just for that

(which is a problem that annoys me with blackboards also)

[slow bobcat]

Not sure I follow this. What do you mean by storing the context?

[misty wharf]

Well for example if I wanted an NPC to do an EQS query where the generation is done around some object the NPC is nearby... There's no nice way to pass it to the query because I can't pass actors to it, so instead I have to store the actor in the NPC and create a context which pulls the value from the NPC

[slow bobcat]

Ah right... Yeah, that sucks yeah. I had a hack ages ago where we could access the generator from the tests etc. Wasn't beautiful but super useful

[hallow compass]

for an AI continuously going towards a target, without a required navigable end location. how can i know when it reached the "end" ?
do i check for the path index ?

when debugging path it seems that the path is constantly updated

what is a Corridor and a String in this context ?

[desert kelp]

Yeah, the problem is I dont want the other npcs to just bunch up around the leader, I'd like them to be in somewhat of a loose formation with offsets. I started a custom moveto task, but I seem to be doing something wrong. going to check some documentation and try and figure what I'm doing wrong

ah, I had goal actor and goal location hooked up. disconnected goal actor and they started moving

[slow bobcat]

A corridor is the same concept as in a flat corridor.
Imagine a flat where the entrance door is on one side and you have different rooms alongside the flat. They are all connected by a corridor.
If you path from the entrance to the last room, your path will start at the door, follow up the corridor until the end, then into the last room.
Same concept in nav, except that the corridor walls are defined by NavEdges. There's also the concept of PathCorridorEdges, which are lines that connect NavEdges sides on both sides of the corridor.
I marked the PathCorridorEdges in green, the NavEdges in red. They all conform data for the corridor.
The corridor will only contain nav polygons/nav edges needed for your path and nothing else

PD: ignore the "path" painted in my image. That's some other thing I do

String you mean StringPulled/WantsStringPulling?
String in this case doesn't refer to a c++ type for characters, but to an actual string made of fabrik, like a shoe lace.
When the detour library initially calculates a path, it will consider every edge towards the goal, like the red path in the image.
Imagine now that the red line is a string.
If you pull from it, it will tense, giving you a shorter path skipping unnecessary edge points, resulting in the green line

[slow bobcat]

hence the term String Pulling

[hallow compass]

Thanks a lot for this detailed explanation !

[reef wharf]

Hey all I'm trying to get hearing sense to work but the AIcontroller is not picking up any reportnoiseevents, even through the breakpoint is travelling through the reportnoiseevent fine

Am I missing something?

I compare it to forum posts and its pretty close to what this is

Its something to do with AI controller perception component

because even calling makenoiseevent in BP doesnt trigger it

figured out the problem

detection by affiliation was not correctly set up

[hallow compass]

im again havign weird behaviors with the recast actors

it looks like it alwasy happenign when I edit nav agent settings

https://issues.unrealengine.com/issue/UE-229996

[slow bobcat]

Target fix 5.6....so....next year if I have to guess

[hallow compass]

:)

[silent hamlet]

Is that a known thing? That new releases will take more time?

[hallow compass]

no, there is always a few months between 5.X updates

the joke is that this bug will probably not be fixed in 5.6

[silent hamlet]

Ah

🙃

[harsh storm]

I'd imagine 5.6 will come in June, around the time of Unrealfest

[slow bobcat]

One can hope. If not there's ankther unreal fest in September. Life at Epic is what happens between unreal fests

[silent hamlet]

That's more time than other recent releases. I feel like since ue5 came out they've been sprinting

Anyway, getting off topic here. I just was hoping for that fix to state tree overrides not loading in ai component to come soon but whatever. I'll just keep my patch on that

[unreal plover]

when getting to the end of the array the AI doesn't do their intended Looping type the AI just walks straight forever ive made sure ive set the Waypoints up in editor and the type of looping i want https://blueprintue.com/blueprint/b53gjukx/ not a perfect site but decent enough to show what ive changed
Error :Blueprint Runtime Error: "Accessed None trying to read property CallFunc_Array_Get_Item". Node: AI MoveTo Graph: EventGraph Function: Execute Ubergraph BT Task Waypoint Patrol Blueprint: BT_Task_WaypointPatrol

[silent hamlet]

Hm, are you removing/changing waypoints at runtime somewhere else?

[unreal plover]

no

[slow bobcat]

Umm difficult to say but I would check if that index always starts at 0. The other thing is to add a "is valid" in that WaipointMamager. About the check with num-1 etc, isn't there a IsValidIndex in BP's when you drag from the array variable?

My bet is that the variable is null, so it can't get the array. If it was an index problem, you would get and Index out of bounds or. Similar error

The other thing is that, if your array is empty, the task will never end

[zenith hill]

What would you all recommend for A.I to detect other pawns?

is pawn sensing the go to method?

I think there's A.I perception too 🤔 and either that one or pawn sensing is deprecated.

A.I perception it is I guess.

[silent hamlet]

Pawn sensing is the system that came before the ai perception one. I think you can still use it but probably makes more sense to use the latter yes

[harsh storm]

Pawn sensing has long been deprecated, just hasn't been "officially" deprecated until 5.5. I plan on removing it in 5.8.

[slow bobcat]

Wait... "you" plan? You work in the AI team at Epic?

Even though it's not what you mean, you might get better support at #generative-ai for what I heard (haven't check myself). Some people here use generative too

[harsh storm]

No. They're just pretty open to community PRs. I've already talked with Mieszko about removing it as well. He is fine with it. I'm the one who gave it the official deprecation 😅

Just got tired of people coming in asking to solve their issues with it, because a modern YT video told them to use it.

I hate that dang thing, lol

[slow bobcat]

I'm going to make my own Unreal! Without Pawn Sensing! And with stable Eqs!

[wide oxide]

Appreciate the response, I'll keep that in mind. I actually just managed to figure it out after watching a few others videos on the subject. It was a simple thing that I wasn't aware of (cause im just new to Unreal stuff in general). Apparently I just had to right click and hit "Create variable 'AgentId' " (as shown here https://youtu.be/387ym_2VLUA?si=8ZkR3JcGBBRP2CEA&t=347) and it compiled up and I'm seeing the "LogLearning" in the console showing the now. Hopefully the rest of the tutorial goes alright lol

[harsh storm]

I'd just like more dynamic EQS honestly 😅

But I don't do anything crazy with it, so I haven't dug into how it even works under the hood.

[wide oxide]

Well I managed to get that Learning Agents setup with the vehicle using that tutorial. Have it training at the moment, pretty neat stuff once you get it sorted out https://imgur.com/a/dVlXqQd

[wide oxide]

Actually pretty impressed at how much it's improved in just about 30+ minutes. Gonna let it run for a few hours maybe just for fun testing it out for the first time. Cool stuff https://imgur.com/a/yIXK1WP

[slow bobcat]

How does the process work on a high level? Can you apply the same example to get a walking mannequin to learn how to transverse a level?

[wide oxide]

I assume it could be setup for something like that no problem depending on what you're trying to achieve. I'm honestly not too knowledgeable as it's my first time messing with the feature itself. Though if you're just looking for an AI to be able to move around a level (possibly enemy/sidekick following type deal), there are a few good videos I saw on that topic as well that didn't involve the machine learning stuff (Decent series on enemy AI : https://www.youtube.com/playlist?list=PLNwKK6OwH7eW1n49TW6-FmiZhqRn97cRy ) I'm just following this tutorial here though: https://dev.epicgames.com/community/learning/courses/GAR/unreal-engine-learning-agents-5-5/7dmy/unreal-engine-learning-to-drive-5-5

[wide oxide]

About an hour or so of training. Already looking pretty decent imo. Much more to explore and test after the first section to get it a bit more realistic and play around with stuff. But overall super cool and honestly can't believe stuff like this is free lol. A lot of potential https://imgur.com/a/kOwISYH

[wide oxide]

Ok last one for now lol. I just think stuff is really cool and happy with how it turned out so far. https://imgur.com/a/tSb30zi

[heady silo]

I was thinking of an upper limit of 12-20 active enemies as a theoretical limit

[silent hamlet]

Really cannot judge just based on the numbers of enemies. As a rule of thumb i tend to optimize when things become a problem, not really before that. Unless I know that something I wrote really sucks. Profile and see how much time your ai agents are taking to do their thing

[heady silo]

They seem to be reacting decently fast. I tried the routine I wrote and it started tweaking them out like Jacob's Ladder

[pine steeple]

Anyone here ever seen a crash on BT regarding simple parallel ? const UBTTaskNode* ParallelTask = ParallelTaskInfo.TaskNode specifically this being nullptr

[harsh storm]

Considering how infrequently I use simple parallel's - no 😅

[grand halo]

I am also looking into doing some hearing sense in Cpp, let me know if you solve it!!

[dim pier]

I don't suppose there is a guide of some sort about reusing BTs for multiple characters?

or a general one that involves multiple characters

[harsh storm]

Just don't hardcode certain things. For example, if you are trying to play an animation, get that info from the pawn instead of hardcoding it in the task.

[dim pier]

Yes that's what I'm trying to do

So far, I created a BB key for the AnimMontage and have each character set the desired montage in the BB then have the BTTask use that key to play the montage (or trying to do this with not much success)

[harsh storm]

Show code

[dim pier]

left one is probably correct, right one I'm not sure what's going on

[harsh storm]

2 things:

1. You're assigning the BB on begin play, the blackboard may not even be valid by that point.
2. You're playing the animation and then immediately finishing the task. With latent nodes, you often don't want to pull the execution wire from the top most pin. Instead, from one of the other pins.

[dim pier]

I didn't get that far yet, I think I'm stuck on the retrieving the animation in BTTask

[harsh storm]

You're setting it with the right node and you're getting it with the right node. But it may not even be valid because of my first point.

Also, make the BB key a variable in the task, so you can select it from a dropdown in the BT editor.

[dim pier]

I added a delay to give time for the blackboard to initialize but I guess I missed something
Blueprint Runtime Error: "Accessed None trying to read property CallFunc_GetBlackboard_ReturnValue". Node: SetValueAsObject Graph: EventGraph Function: Execute Ubergraph BP Humanoid Viking Blueprint: BP_Humanoid_Viking

[dim pier]

Ok, I got that part sorted and the correct montage is showing up but not animating

I tried casting to the base BP and the Viking BP to get the mesh, still nothing

[dim pier]

Finally got it working sort of, the issue was the animgraph wasn't setup properly

[tropic vine]

Hi guys I'm new to this ai stuff can you tell me when do I need to use the agents plugin and how is it compared to the new Nvidia ace plugin

[static flint]

Dynamic obstacle doesnt work as intended
I kill object but area behind it still unpassable

Any solution?

Related 1: https://www.reddit.com/r/unrealengine/comments/uxshac/genuine_bug_the_dynamic_obstacle_system_was/

Related 2: https://forums.unrealengine.com/t/dynamic-obstacle-on-nav-mesh-visualization/1745391

Theres barely any info about dynamic obstacles
I need to make a box that makes ai not able to pass it, then remove that box to make ai being able to pass it

Related 3: https://forums.unrealengine.com/t/moving-dynamic-obstacles-do-not-rebuild-nav-mesh/393243

[slow bobcat]

What type of navmesh do you use?

[static flint]

Just changed to "dynamic modifiers only" and seemingly it started working but Im not sure its thats intended solution

[slow bobcat]

Yep, that's the solution

Nav modifiers only work on Dynamic with modifiers or Dynamic nav

[static flint]

Thanks

[pastel moth]

Is there a way to pass in data to parameters via the blueprint of the owner of a state tree?

I get that I can set a default via the params but is there a way to have it settable at runtime? It looks like I'll just have to pass in information via the context actor 😦

[slow bobcat]

Either that or a global task that sets it. Or you can have a task that runs in the root just once that sets it.
Or it could be data driven entirely and you have a Data Asset that defines your AI peculiarities and you read the required EQS from there

Plenty of choices

[slow bobcat]

GameAIPro.com is down... I hope it's something temporal while they fix issues...

[gentle hamlet]

I want to make a horror game monster... I'm trying to read on the current state of game AI. I think I'm going with a state machine and that's it. Who thought mimicking intelligence is difficult

[stone walrus]

hey guys how can i change sight radius during runtime in BP well in c++ we can but i am making lecture for my course which is typically only BP focused i don't want to involve c++ in my course video , i have checked every where , the solution is only c++ based no Bp solution

please help

[misty wharf]

I don't think you can

[neon basalt]

Hey everyone, I'm starting with AI, I think that need to set up a Behavior Tree and a State Machine together. Are there any tutorials I can follow to learn? I searched on YouTube, but most of them are over a year or two old. Have there been any changes? Or are these methods outdated? Is there any new plugin that helps with AI and related features?

[pastel moth]

You'd want to use state trees. I'd follow the quick start guide with state trees on unreal engine

https://dev.epicgames.com/documentation/en-us/unreal-engine/state-tree-in-unreal-engine

Question on transitions. I have a group state "Idle" that I can successfully enter. My tree is failing to move from there to my child states.

There are no pre-requisites to enter the child states of that group state. What am I doing wrong here? My root is set to try to select children in order. I set the children states to have "Try enter" for the selection behavior. Is there something obvious I'm missing?

[misty wharf]

Try Enter is a bit confusing, I never quite figured how it's supposed to work. In my experience it works best if you have a leaf state that can be selected

[neon basalt]

Perfect! Okay, thanks!

[pastel moth]

Anyone figure out what the group states are for?

Thanks ZOMG

[misty wharf]

As far as I can tell, Group is entirely identical to State but you just can't put tasks directly on it

It's literally handled by the same code other than the UI logic where it prevents you from adding tasks into it

[pastel moth]

I figured it's nice to have things organized (but you can do it with state) but I'm not sure about what their intended use cases were for

[misty wharf]

It seems to be purely for organizational purposes yeah

[slow bobcat]

Wrong channel. Spam or mistake?

[misty wharf]

It's just a shitpost, he posted on multiple channels

[slow bobcat]

Report, blocked and ignored

Also: award to most misleading thumbnail ever

[grand saffron]

StateTree question: Apparently tasks do not run on parent states when the tree is determining whether to move into child states. But I want to move into a child state only if the output of a task on the parent returns true! Is the solution to this to make a dummy state before that child whose sole purpose is to run the filtering task for one tick, or is there a better solution?

[slow bobcat]

You can put that bool as an output parameter that a condition reads as an Input parameter. I think, not sure

[grand saffron]

That was my initial design, but the output is not set because the task doesn't run on the parent

[grand saffron]

"Important note: if the enter condition relies on the output of a task higher in the tree’s hierarchy, there are times where it can fail because the input parameter does not have a proper value. StateTree will not run prerequisite tasks for the enter condition to be able to evaluate it. This could change in the future, but it is not currently planned."

[grand saffron]

I ended up making the parent state "Try Enter", and then have it transition to a dummy child on tick which holds the actual children that have enter conditions. 🤷‍♂️

[spare salmon]

I'm trying to debug a State Tree, but I cannot get the State Tree Debugger to work. I can't connect it to an active ST. When doing a simple test on my personal machine with editor launched from the Epic Game Store, I can use the debugger. But the project I'm having the issue with is built from source. Anyone here have issues like this before?

Edit: Turns out we didnt have the UnrealTraceServer built and the debugger requires it to work

[grand saffron]

I can get the StateTree debugger to work, but it activates a profling trace, creating all of these trace files that I don't want on disk.

[little mulch]

Does anyone know how to get the USplineNavModifierComponent working on an actor? It doesn't seem to do anything and I'm not sure how to set it up properly.

I'm wanting to do a Simple Move To into that tunnel so it seems like a nav modifier for that location is appropriate

I think the collision on the simple collision on the mesh is causing the hole in the nav dmesh (which I want for the most part) but I want to allow the nav mesh to go into the tunnel, at least a little bit where the character will then be put on a spline and not have to nav by itself

[pastel moth]

Has anyone had issues where moveto is ALWAYS failing in your state tree but it works in behavior tree?

Im using the same EQS query for both the behavior and state tree

[slow bobcat]

You are going to have to modify the collision of the tunnel if you want it to affect the nav but still have nav up to its entrance. Have you tried using a nav link? You will need several if the tunes is not straight though. I have never used the spline nav modifier. Not sure what it is for, my guess is that it is like a navmodifier volume but in the form of a spline. That doesn't generate nav, just modifies existing one.

[slow bobcat]

What does visual logger say? If you debug draw the location choosed, does it make sense in both cases?

[pastel moth]

sec gonna use it I forgot there was the visual logger 😄

yeah it makes sense

Would this mean that the location is none 0_o

my navmesh has a bunch of warnings too
LogNavigation (Warning) FPImplRecastNavMesh::InitPathfinding start point not on navmesh (X=-1261.520 Y=-2241.978 Z=28.038)
LogNavigation (Warning) FPImplRecastNavMesh::InitPathfinding start point not on navmesh (X=-2260.000 Y=390.000 Z=198.038)
LogNavigation (Warning) FPImplRecastNavMesh::InitPathfinding start point not on navmesh (X=-2053.083 Y=-2319.803 Z=28.038)
LogNavigation (Warning) FPImplRecastNavMesh::InitPathfinding start point not on navmesh (X=-265.502 Y=-2155.962 Z=28.038)

[slow bobcat]

Your default extension for points projections onto the nav is not big enough for the logic to be able to project your points.
Try one of this 2:

• change the default extension in nav project setting to something that makes sense with those numbers (my wild guess is that the Z component of your extension is too small)
• Or project the points onto the nav before sending them to the MoveTo logic passing a custom projection extent

[pastel moth]

oh okay so probably increase the projection so the nav points appear on the navmesh

[slow bobcat]

you can do that on the eqs directly, but it's cheaper if you just project the one you pick rather than all the ones used by the eqs

[slow bobcat]

I would debugDraw all this points in VisLog to see where they are exactly. Seeing them will help

[pastel moth]

I managed to get the points to project evenly across the ground

Still no dice 😦

[slow bobcat]

do you get the same error in visual logger?

[pastel moth]

Please hold 😮

[slow bobcat]

holding hard

[pastel moth]

hah sorry

I dont see anything in the logs that stands out

maybe it is my state tree thats fucked

LogPathFollowing (Log) RequestMove: Path(complete:3) Location(X=200.000 Y=-1200.000 Z=76.899) Mode(partial path) Filter(NavFilter_AIControllerDefault) AcceptanceRadius(5.0 + overlap) StartsFromPreviousPath(No))
LogPathFollowing (Log) Updated deceleration segment: 1 (MaxSpeed:300.00, BrakingDistance:300.00
LogPathFollowing (Log) RequestMove: accepted, ID(54152) dist2D(1562) distZ(47)

That looks good

[slow bobcat]

If you click the navigation category of the Visul Logger, do you see a path drawn at any point?
If not, have you checked with the state tree debugger what's going on?
Is your MoveTo task running alongside some other task? could it be that the other task is finishing before the MoveTo task and triggering a Transition out of the move state finishing your move?

[pastel moth]

yeah Ive checked with the state tree debugger and debugged through my tree

My moveto task is failing so I have it go to a task to increase my characters Z axis.

[slow bobcat]

What does it mean "increase my character Z axis"? as in "scale the capsule"?

[pastel moth]

sec

Im getting a new location via the EQS query if the actor movement fails because I think they got stuck somehow when moving

I basically copied cropouts code and Im trying to migrate it to state trees

[slow bobcat]

if that's the case, the path following category or the navigation category in visual loger will show the result of the movement (blocked, success etc)

You don't need that increase. You can move to a point on the floor. It doesn't need to be at capsule-center height

[pastel moth]

oh ok so get rid of the add part

[slow bobcat]

For real this time: debug draw your location before you call move To to it in visual logger.
Trace an arrow from the character location to that location.
If you want, show it here. Just to make sure things make sense

[pastel moth]

one of the problems is that teleport is constantly recurring because its boolean is false

ok

oohh they stopped teleporting after I removed the +45 oh nvm

[slow bobcat]

My advice is that you simplify things for debug: do simple moves and only that, disable anything else. See why it fails. Once that works, re-visit if you want to teleport and how.

[pastel moth]

yeah youre right.. just do a moveto on beginplay and just see where it's screwing up

[slow bobcat]

you can try that to a know point in the nav (drop a cube on the scene that can't affect navigation, copy its coordinates and use them hardcoded)
If that works, then do the same in your Move To (skip the EQS and hardcoded your cube coords)
If that works, re-plug the EQS

And, sounding like a scratched record but: debug draw your stuff.
2 hours building a decent debug tool saves countless hours of hit-and-miss debug

specially when it comes to movement, rotation and physics

[pastel moth]

yeah ill do that

thanks man for the help

[slow bobcat]

no worries. Let us know what did fail once you figure out. "through others pain we learn"

[pastel moth]

so I have a cube and I feed it into my pawns move to. The visual logger is saying the endpoint isnt valid even though it is in the navmesh. That doesn't make sense to me

[slow bobcat]

see that red volume? that's your extension used to project points onto the nav

It can't reach the nav. If you don't want to change it, simple set the flag bCanEverAffectNavigation to false in the cube and re-build the nav data

that way there will be nav below the cube and the extension will find it and project the point (technically inside the cube, which will not be acceptable in the game, but this is for testing only)

[pastel moth]

I thought if my navmesh was set to dynamic in the project settings it'd auto rebuild

[slow bobcat]

ah then you don't need to do a rebuild. Didn't know you were using dynamic nav

[pastel moth]

any idea why this wouldn't be working otherwise?

[slow bobcat]

step by step.
Change the flag so there's nav under the cube, try again. Then let's see

[pastel moth]

done and still the same error/warning

[slow bobcat]

which one? the last one?
Can you show the image please?

[pastel moth]

Sure

oh what the hell I turned it off again and now theyre navigating

They're walking to the cube woo

[slow bobcat]

ok cool

so movement works

Now, use that exact same location in your MoveTo task within the state tree.

[pastel moth]

I fixed it

I used version control to reset my all the stuff I changed

Now I'm slow rolling my changes out and more gradually testing..

[slow bobcat]

Classic

[pastel moth]

one last question, is there any trick to getting a delay node to work? Can I just add it as a leaf node and set the delay and it should work?

[slow bobcat]

Are you talking about Bt's or BP's?

[pastel moth]

state trees

a delay task in a state tree*

[slow bobcat]

No idea sorry, never used it. Delays are a big "no no" for me in general. Nothing good comes from them in my personal experience

this might be interesting for some people in the channel
https://www.youtube.com/watch?v=ndvxqIPR_aU

[tidal pelican]

Hi, all. Anyone know under which conditions a state tree's debug info won't be recorded to the Debugger window? My tree is DEFINITELY executing, the Debugger window is open, but nothing records. I had it recording in a different session a few days ago, but now it doesn't seem to ever record. Ideas?

[slim quest]

Hey, does anyone know why my AI isn't playing the animation Montage?

I have set up a print string, before and after the play montage, but only the one before triggers

[slow bobcat]

If you add a debug print to a task you think is running, does it print to the log? As a silly test to ensure the tree runs

[slow bobcat]

If you attach the print to the blend out node, does it print?

I wonder if your anim is blending out and you are missing the end task all together

Same thing with the failed cast

[tidal pelican]

My state tree debug breaks don't even fire off and the state tree is absolutely executing.

[slim quest]

the only time the print string after the play montage runs is if it is through the top, none of the other nodes will print it

[slow bobcat]

If you are 100% sure the tree you have open in the editor is the one running... Then no idea. I would double check with some print that identifies the tree running, just in case something is off

[slow bobcat]

Any logs in the output window or info in visual logger (not 100% sure animation stuff shows in visual logger, but worth checking)

[slim quest]

not exactly sure what I'm looking for within the output window, would you like me to just screenshot it and sent it here?

[tidal pelican]

The tree is most definitely running. The pawn is moving around and performing some tasks in the states, I'm encountering a bug that I'm trying to debug (obvs) but none of the tools are working.

[slow bobcat]

Can you make a silly ST with one state running one task in loop and replace yours with that one? if that one doesn't work, maybe there's a bug in the engine?

[tidal pelican]

Will try that next, just rebooting my project for the 6th time, in case that makes any difference.

[slow bobcat]

Filter errors only first, see if there are any about montage or animation. If not filter warnings and check again

[slim quest]

nothing

[slow bobcat]

@slim quest
if you ignore your tree and you play a montage on begin play, does it work?

[tidal pelican]

Nope, state tree with minimal states/tasks also don't debug.

[slow bobcat]

And if you click the record button on the debugger window does it say or do anything?

[slim quest]

it does not, tried to see if it was just an issue with that montage for whatever reason, it's not that either.

nevermind, I got it, thank you for trying to help, the issue has been sat in front of me this entire time and I just realised it... I'm using the wrong mesh

[slow bobcat]

That should have thrown an error in the log about the wrong mesh being used

[slim quest]

strange, I didn't see it. Maybe I just missed it, I'm not too sure

[tidal pelican]

I restarted my PC and went for a 25km mountain bike ride, started up UE and tried again and it works. So either my mountain bike is magic, or there's an external process that's used for the debugging that didn't get cleaned up last time. I'd like to think my bike is magic.

[slow bobcat]

Definitely the bike

[solemn garnet]

Is there any way to get the AIController context when using a gameplay interactions state tree schema? I am trying to find an entry location to a smart object and then move to that location with the Move To task but it needs an AIController context.

[uncut cloak]

does anyone have a good resource for learning AI for the vr template?

[dim pier]

I don't see how VR changes anything

[honest mulch]

Why the NavMesh not fill the full area

[slow bobcat]

Not familiar with that schema (I only use the AI schema) but, don't you have to set what is the actor? Is that's the case, you could have a global task that, on enter, outputs a parameter that is the AI Controller for the actor in your schema.

[slow bobcat]

Seeing that the tree in the background has nav on top, you seem to have something in the floor that affects navigation. You will have to inspect all your objects and volumes in that area and see what's causing it

[honest mulch]

can EQS fix this?

[slow bobcat]

Eqs has nothing to do with nav. Eqs stands for environment query system, in other words: a system that gives you tools to ask questions to the environment to obtain data from it.
I would do some A/B test first and try removing the 2 buildings there. If the problem persists, remove some other asset etc

Until you get nav there. Then restore the level and just remove the last asset you removed when you finally got nav. If that works, check the asset setup on your level

Also pay attention to volumes and triggers if there's any

Another tests is to remove all nav bounds volumes and just place one covering only the area that doesn't get nav on the ground (to check if you are hitting some nav limitation)

[honest mulch]

I checked the collision, the object not have a weird collision, all correct

[slow bobcat]

What about the bCanAffectNavigaiton? What about the collision going further enough on those buildings to cover that area? Why does the tree have nav but not your floor? What's the difference between them? Plenty of things to test.

[honest mulch]

What features i need to check maybe?

[slow bobcat]

Start here to find first which asset is causing trouble

[slow bobcat]

Could it be the asset you use for a floor with crosswalks? Seems the areas without nav have that in common

Maybe bCanAffectNavigation is set to true for that floor asset?

[uncut cloak]

The follow command doesn’t seem to work because the player character isn’t a static mesh in the vr template. So I’m trying to learn more about it. I can’t get a simple ai to work.

[slow bobcat]

Never worked in a game where the player is a static mesh. What exactly is the error you are getting?

Also not sure I know what do you mean by "follow command"

[uncut cloak]

I’m very new to ai so I’ll try to explain it better. I am using a “AI move to” node to make the enemy ai follow the player when on the nav mesh. But from what I’ve learned is that the the vr character doesn’t have anything in it that communicates with the nav mesh. So the ai can’t see you and doesn’t know where to go next.

[uncut cloak]

I think I got bad info from a video. I just re built a different one and it’s working.

[stable plover]

I’m sure that’s wrong, the Ai needs to be configured to see, you mess with detection by affiliation?

And use the gameplay debugger to debug the perception?

[uncut cloak]

i solved it, it was user error. but since i have you here. Im having trouble figuring out different movement speeds. I have the behavior tree set up with a "move to" node for when the ai sees the player character. Do you know how to increase the ai's movement speed when it detects the player?

[dim pier]

I did it by BTTask_SetCharSpeed with an interface on the character, not sure if its the correct way to do this

but it works

[odd tide]

when i click Add Transition in first State, my editor crash. How to fix it?

[slow bobcat]

Your log have 0 info because it doesn't have any debug symbols (I think, looks like it doesn't have the pdb files needed). Try downloading/install them and then repro the bug again. It's possible the log will contain then more specific info

[odd tide]

This can happen at any time, when adding a Transition in any State, sometimes it happens, sometimes it doesn't

[slow bobcat]

Then download the debug symbols and, if a crash happens, check the log

The crash window might also offer more info with the debug symbols loaded

[odd tide]

okay will make it now

[odd tide]

@slow bobcat

UnrealEditor-DetailCustomizations.dll!FVectorStructCustomization::GetSortedChildren(TSharedRef<IPropertyHandle,1> StructPropertyHandle, TArray<TSharedRef<IPropertyHandle,1>,TSizedDefaultAllocator<32>> & OutChildren) Строка 47 C++

I don't know what information from these windows I need to send out

[slow bobcat]

Ummm difficult to say. So... Are you loading the dump file in visual studio or are you running the editor from it because you compile it?

[slow bobcat]

Look like you have something tied to a vector param property that is failing

[heady silo]

I'm trying to get an EQS for flanking the player. I'm pretty close, I think, but the EQS system is considering some props as visual cover from the player when they are clearly not. Is there any way I can call tags in the EQS and filter out the exercise equipment?

[slow bobcat]

Call tags you mean filter by gameplay tags?

[heady silo]

yeah something like that

I mean I could just set the props to not influence navigation, and then put nav volumes to make the areas null

but I would prefer something more scalable that takes less work

[slow bobcat]

Then you can make your own test that grabs the gameplay tag interface from the item (or the ability system component if you use GAS) and filter using it

[brave citrus]

I was trying to get an ai to attack the player, but the animation keeps repeating before the animation finishes. I'm not sure what I did wrong if someone could help

[stable plover]

thats one way

[stable plover]

where are you calling that event from

maybe dont set the can attack to true every time you follow the player thats a logic error

because every time your following the player you set it to can attack everytime

[uncut cloak]

nice! im going to work on this now to see what i can come up with

[oblique basin]

If the animation is not completing, it's probably being called again before you reach the end of the function. You could set a Boolean gate bIsAttacking which you set to true before calling the animation, setting to false on complete and only allowing the process to start if bIsAttacking is false. Also, you're setting Can attack to true then checking if it's true, which is redundant. There's also a more useful play montage node in which you pass in the pawn's skeletal mesh and comes with async output pins for on complete to save using the delay set up you've gone for.

[severe sorrel]

How does the AI Moveto command work?

For example my character has some custom locomotion, but using the moveto it just linear interpolates to it

[slow bobcat]

What do you mean by custom locomotion?

[brave citrus]

Thanks for your help but I ended up rewriting the attack code separate from movement. This code also lets me slot different animations and fix the sliding problem with the animation.

[severe sorrel]

uh, basically an actor can only go forward, then it steers to face the desired input direction

[bleak drift]

does it make sense and can i implement statetree tasks in c++?

or should i usually opt for bp?

[severe sorrel]

should be able to do C++ if you inheirit from the correct class

[bleak drift]

cool ill try that! 🙂

btw I'm not sure just saw it, do you think it is worth looking into EQS an will i be able to use it with statetrees?

and additionally, is if someone wants to give me some pointers, is there some pointers like eqs, i could look into to make a good ai? ^^

[slow bobcat]

But are we talking about root motion driven movement or it's a non-root motion and the movement is scripted in bp/c++?

[severe sorrel]

Non root motion driven by bp

from what it's looking like I may need to write my own custom moveto

[slow bobcat]

Riiight OK so your problem is that the AI is following a path that you don't want it to follow? Rather move it straight?

[severe sorrel]

Well, it's moving straight to the goal

but with the logic I want to run, it will need to turn first as it has a turning radius and speed

if that makes sense

[slow bobcat]

Sorry didn't make myself clear: I was asking "you don't want it to follow a path, you want it to just move straight forward right?"

Ok so:

• turn on the spot if needed
• then move forward
  No?

[severe sorrel]

https://i.imgur.com/lz92mA0.png

[slow bobcat]

Aaaaaaah

[severe sorrel]

I guess basically I want to just give it a location, then the logic will take that location, create a movement vector, then feed it to the pawn movement

until it reaches its goal

moveto location doesn't seem to do that, but something else instead

[slow bobcat]

You could do this with the path following component. When a new segment of a path is given, calculate that curve to connect the start/end of a segment and calculate points alongside. Then in the Following function make it to move to each of the points.

Another way would be to use the Set Focus and just move forward and control the focus

In any case you need to calculate the green curve for sure

I guess there some other smart way, I can't think of any other right now

[slow bobcat]

MoveTo calculates a path with the nav mesh and follows it

Hence my suggestion of tapping into the path following component and handle it there

[severe sorrel]

That's what I was thinking

[slow bobcat]

But I'm pretty sure you need c++ for that

[severe sorrel]

https://i.imgur.com/XNU2zpW.png

get each checkpoint of the path, then once we get there go to the next one

err, set the desired direction to the next one

then just let the pawn handle the movement itself

[slow bobcat]

Then yeah, calculate the curves (spline, Catmul Rom or whatever you want to use). Curves will be a series of points. Then follow them (you could add them as extra points in the path for example)

[severe sorrel]

I don't need to calculate the curves

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

[slow bobcat]

Well you will need to calculate them in some way? Either poonts in it or rotation rate based on distance walked to now how to handle the focus (what will drive the green path) to pass through the red dots no?

[severe sorrel]

basically just a vector to the goal

that I then feed into the locomotion as a desired direction

[slow bobcat]

2 vectors

One moves away for the red line, the other points to it. A combination of forces

So to speak

[severe sorrel]

https://i.imgur.com/Xa3aHWO.png

[slow bobcat]

One would be your forward, the other the yellow one

And with the rotation rate...

[severe sorrel]

uh basically the pawn is just taking the input vector, getting the length, then the actor forward, and making the pawn move forwards

[slow bobcat]

Sounds like something that would work. You would have trouble with obstacles and going out of nav though

[severe sorrel]

then as they're going forward, I get the control rotation and rotate the actor to it

[severe sorrel]

yeah I'm fine with them bonking, almost encouraging it

and I'll have some quickturn logic for tight spaces

[slow bobcat]

Yeah, gotcha. Move forward pointing perpendicular-outwads the red line. Then slowly turn towards the end of the red path

[severe sorrel]

it's just however the actor is facing

if they're facing along the red line they'll go straight to the goal

[slow bobcat]

But you don't want that right?

[severe sorrel]

I don't want it if they're not facing it

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

if that makes sense

[slow bobcat]

That you can automatically handle with the rotation rate

Set it low

Then just move

[severe sorrel]

that's what I was saying

[slow bobcat]

Ooooook

Didn't get you

I was like "but... You already have that as default just tweaking the RR..."

Sorry for the confusion

[severe sorrel]

but yeah I need to get each segment of the navigation

[slow bobcat]

Not really

Just follow and set a low rotation rate

Los 45 or less

You should get what you show in the last image following the path (unless you want to tweak the acute corner cases etc with a different RR as you mentioned)

[severe sorrel]

the rotation rate in CMC doesn't seem to really work, you still go down the path linearly

it's just a visual for the actor rotation

[slow bobcat]

We use it in our game because we set the Allow Physics Rotation to true. That allows us to override the root motion rotation rate. Maybe something like that?

There's also the flags in the CMC to use the movement rotation or not

That will make the AI to face the path direction or not

[severe sorrel]

if I use that, it faces the path instantly

from what it's sounding like I'll need to make my own move-to function

[slow bobcat]

So setting the rotation rate to any number does nothing to your movement?

0 or 360, all the same?

[severe sorrel]

nah, doesn't affect it at all

I even made a blank actor

[slow bobcat]

Then something is very off

Can you show us your flags for rotation in the CMC?

[severe sorrel]

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

[slow bobcat]

Yeah that need to be false

And the use controller rotation to true

I think

I would also double check bAllowPhisicsRotation

Can't recall the exact name

[severe sorrel]

yeah still snaps to the rotaiton

[slow bobcat]

https://youtu.be/i9v2aRBTRi0

Check that

He sets up what you need

I'm pretty sure I'm forgetting mentioning an important flag somewhere

Aaah but then you will have the problem with the AI rotatimg and walking backawrds etc. That's where the forward vector movement comes into play in your path following component

[severe sorrel]

yeah I have that, not actually sure why it's snapping

[severe sorrel]

Is there a function that generates a path to a location but doesn't actually move the AI?

[bleak drift]

Is it possible, in a EQS context (picture one) to get parans/config i pass in via the query (picture two)?

Since I'm using it in a state tree, I want to pass in the current target, is this a viable approach an if so how could I get a parameter in the context?

[midnight scroll]

Is it possible to allow two AI to use the same smart object at the same time? I have a case where I need multiple(number unknown) AI to interact with the same thing. The claiming stuff is nice but can it be worked around in this case?

[slow bobcat]

Yeah, in the navigation system there's find path and find path async

[slow bobcat]

I theory, if you set the RR to 0, it shouldn't rotate at all. Very weird

[severe sorrel]

found my issue

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

I had this checkmarked

[severe sorrel]

I only see find path sync

[severe sorrel]

Should I use AI tick? or is it better to let the pawn do the ticking

[slow bobcat]

Aaah good find

[slow bobcat]

Not sure what do you mean. Are you asking if it's better to add your logic within the AI Controller tick or the pawn tick?

[severe sorrel]

Yeah

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

basically here's the logic I'm trying to run

Basically I'm asking the pathfinding to give me a path, then I take the points on that path and set my desired movement to that point, then if I get close enough go to the next point

but I'd be constantly updating the pawn on the desired direction

From my understanding it's good practice to not tick things if they don't need it, but wasn't sure if AI tick was an extra overhead

[slow bobcat]

Without seeing the image (I'm on my phone and discord app sucks zooming on images), both pawn and controller tick by default, so the overhead here would be whatever you add to it. You could do all that in the Path Following Component thoug. It has 2 functions that are suited for your goals:

• on path segment: called when a new segment of the path starts
• on follow path: like tick, but only calls while a path exists

Alongside that it has nice utils like get current path segment etc

If you use that, you can also use the default move to AI, which not only handles the get path etc, but also has a path observer that will re-path if the nav changes or if the goal actor (assuming you move towards and actor) moves. It also will stop the movement if your AI gets blocked.

ok I'm on my pc now. I don't get what you do in the BP: you calculate the direction to the next path point and pass it to some other function (AI Desired Input Direction) and then? what do you do with it?

Everything you do up to that point you could do using the default AIMove to and implementing the logic in UPathFollowingComponent::SetMoveSegment but that function is only available in c++ unfortunately (I'm guessing you are doing everything in BP's)

there are workarounds for some of the stuff you do

For example, if you use AIMoveTo, then in your logic you can call this BP function from the AI controller to get the current path segment your AI is following and straight get the direction you look for UFUNCTION(BlueprintCallable, Category = "AI|Navigation") AIMODULE_API FVector GetImmediateMoveDestination() const;

you can also get advantage of the automatic block detection in the AI Controller

    AIMODULE_API void SetMoveBlockDetection(bool bEnable);```

Just mentioning all this in case you can have less custom stuff (it always helps)

[heady silo]

...I'm assuming that's going to require C++ that is well outside of my ability

[slow bobcat]

Not at all. You can do eqs tests in Bp's without issues. And I would assume the Gameplay tag interface is imolementable in Bp's (haven't checked tbh)

[bleak drift]

If I want a AI move to a location that itself changes, should I then call move to all the time (maybe even in on tick) to update the target location, or is it better & more performant to create a actor at said location and target that or is there another option of doing it?

[spice badger]

Can advanced AI ( such as chatgpt gemini etc ) contorl enemy AI in video games?

not literally but like can we have like a LLM integrated in our games and let it control certain aspects of our game ( like ai in this example)

like how does that work

[bleak drift]

Strongly depends on your goals.

If you want to train such a model yourself you need the data for it.

But letting a LLM make decisions based on player input, should be quite easy, for example asking equestion & getting some answers. You probably want to add some context to your LLM query and then let the (game) AI act dependent on it.

[spice badger]

it would control an enemy AI player, patrol, attack the player, seek cover, you know, the usual stuff, but in an advanced way over a regulat ai implementation

so it would feel more genuine

and a little random each time

[bleak drift]

LLM's are quite slow and expensive, but you can query an LLM to make high level decisions/plans.

[spice badger]

expensive in terms of perfomance?

[bleak drift]

in performance and hence also time

If you make a request to ChatGPT you see how long it takes?

Besides that look at the token cost, so you use a fast model you can get a request every 1-5 seconds?
Then imagine you have 1000 players

How many request you would have to do, since you are probably not able to ship an LLm with your game, since it won't run on most of the machines and if it would, there would be no perfromance left for the game itself, you have to run it on a server.

Or pay for a service, whcih is running cost for you.

[spice badger]

arent there cheaper models out there?

i dont want it to be super intelligent

like know every topic

but just a specific one

[bleak drift]

The more frequent your requests are, the more you pay. Also a LLM knows different things, it probably won't translate actor positions and inputs well.
You need to provide higher level context to it.

[spice badger]

requests wouldnt be that frequent either tbh, like if it sees a player it would engage in a combat mode and if not it would like just roam around until the next interaction

or until it finds an item

[bleak drift]

You can have a specific tailort smal neural network for some task, but then you need the data to train it on and you need to know what you are training it for which you probably don't have. Also cheaper models lack the understanding of the world.

[spice badger]

how do you effectively train an ai?

just let it run on your game till it figures shit out?

or how do i imagine that

[bleak drift]

Yea you can make a request to chat gpt, "I have a player roaming around, he has equipped a gun and already killed x of my species, should i hide, run away, ..."

[spice badger]

yeah kind of like that

[bleak drift]

with lots of data, knowledge of AI and money/hardware 😅

[bleak drift]

To complex/to little hardware to do that.

Or you need millions of players and collect all the data and then can train your models on that data.

If you are asking, if it is possible to use LLMs to have less work for better results, the answer, especially in the first degrees is no.

It will cost you and you probably will have running costs for you game.

Also you probably have to implement everything you would do normally + the AI stuff, but you probably could get some nice results if you go the extra mile and pay the price. Like for example the player can have completely custom input and get sensible reactions.

[spice badger]

Do you know any games that have it implemented ?

like iv seen some skyrim mod where they added an LLM to chat with but it was only for chat purposes

like with an NPC

[bleak drift]

Yea thats a really good usecase for it.

But as I said, you will not get around implementing the AI stuff, if you want to use an LLM you probably have to do everything you would normally do with additionally the LLM stuff.

[bleak drift]

You will not get an LLM to control a NPC in real time.

You will just get instructions out, which would translate to higher level functions , that you have to implement.

and LLM is way to slow and expensive for real time.

But you can create a query, describe your environment and situation and then ask if for instructions (which should just be instructions you call in your game)

[slow bobcat]

I would start simple. Some thimgs are achievable.
First I would start researching about the actual state of machine learning in video games. Things like the system used by Forza Horizon or the several videos you can find in this playlist (the list is being updated every week, so keep an eye)
https://youtube.com/playlist?list=PLJ9nFbJFanabyUul0S7idr_ouzSc1z4m3

Then you can try this (link to message below). Cars training in ue5 are something you can achieve. It will depend a lot on your hardware for what I understand.
Once you understand how is the car trained, maybe you can train a walking AI to follow a path or a spline and you would get your patrol.

#gameplay-ai message

[slow bobcat]

There was a game demo tech recently where a company showed a game where you play as a squad leader and you give orders with your voice like "move forward and take cover to flank the door" or "loot the room for me". Can't remember the name or find the video in youtube

[tough nexus]

Hi Everyone i downloaded the Ue5 Motion Matching Project File and i am wondering if i can use/implement easily the Motion Matching of the UEFN Rig for an patrolling AI?

[elfin smelt]

https://www.youtube.com/watch?v=ype12MU8yFI&feature=youtu.be

you speak of this video maybe

[slow bobcat]

Yeah that's the one

[bleak drift]

Thats pretty cool 🙂

[severe sorrel]

Almost anything an LLM does for AI can just be programmed procedurally at a much higher efficiency

LLMs are also a poor fit for it, since they don't understand context

and are text based

Is "Find Path to Location Synchronously" an expensive operation?

[pastel moth]

Is there a way to inject a tag I want to search for into an environmental query at runtime?

[slow bobcat]

Depends: one path that is short, no. Multiple paths in one frame, yes. A Long/complicated path with many points can be

But use async. I have never found a case where I needed a path in the very frame. AI's almost never need anything immediately.

[slow bobcat]

A tag you mean a gameplay tag? Yeah, you can add/remove g.tags at will

[severe sorrel]

I don't seem to have the option to make an async path

in the BTTask

[slow bobcat]

Aaah in BP's... No idea if the async option is exposed BP's. I will jump into my pc in 10min, I can check for you if you need it

[severe sorrel]

it's not an immediate need

I'll just fumble around then refactor it later

I want to figure out some more complex behaviour later

[pastel moth]

even in environment query tests at runtime? I didn't see a way to do that

[slow bobcat]

Yeah. If the Items are Actors, you can manipulate them in your tests I think

[pastel moth]

The only way I saw was this

https://zomgmoz.tv/unreal/Environment-Query-System/Custom-EQS-parameters

[slow bobcat]

ah sorry, I was missunderstanding you.
I thought you want to inject tags into the items, not into the query itself.
Yeah, the hacky way zoomg is the only way I know of too.
The problem with that one is that it will fail if you re-arrange the tests and it will not work if you derive your tag in runtime based on the result of another test (technically you can access other tests from within another test, but it's very hacky).
Another option you have is to define an array of g.tags somewhere (or even an enum with values) and use a regular EQS int32 param. Set the param to the desired value and then translate that int32 into your g.tag using the array (or convert it to the right enum entry).
As far as you don't re-arrange your array/enum and always add stuff at the end, should be pretty solid

[harsh storm]

The main lesson is that EQS isn't that great with data-driven patterns 😠

[slow bobcat]

yep. It's too bad, because they would be perfect for it. And adding custom parameter types requires some engine changes unfortunately.
Duroxxigar, you can be our saviour. It's time for one of your famous engine requests! make EQS great!

[harsh storm]

Riot did a pretty cool EQS PR - nuffin' 😭

https://github.com/EpicGames/UnrealEngine/pull/10874

[pastel moth]

why was it not merged in? 😦

[harsh storm]

And Mieszko has said that he has some ideas for EQS 2.0. But it isn't really on the roadmap to be honest.

[harsh storm]

Auto-closed. Their bot does it.

[pastel moth]

ah ok

[harsh storm]

Most likely because no movement on it for quite awhile

I wouldn't be surprised if part of Mieszko's plan is to use #mass for EQS 2 😅

[pastel moth]

It feels like Epic lacks staff working on engine features

I mean I get we see Mieszko and one other guy working on mass but I'm like "There can't just be 2 people working on mass"

[harsh storm]

Not really. Just the AI side has a huge work area.

[slow bobcat]

doesn't this mean it will be merge?

[harsh storm]

Best we can do is probably ask @crystal hatch (come from thou #mass cave please 🙏)

Just questions about Riot's PR for the dynamic contexts in EQS PR

[slow bobcat]

would be nice to see that demo to fully understand what they are proposign here

[harsh storm]

I haven't drilled into the internals of the EQS stuff to figure out how I could modify it to make it more data-driven to be honest. I just cry every time I want to data drive it and then just create a new EQS query and move on 🤣

[pastel moth]

Yes haha that's what I'm doing right now. I think the juice isn't worth the squeeze right now. I only have 5 things I need to search for and this doesn't need to scale hard. I'll just swap the EQS's out at runtime in my state tree via my Actor on what I'm querying for

[harsh storm]

This is another cool extension that I feel should be native to UE https://www.fab.com/listings/6e618e4f-9ce6-44a3-9d49-242feee8dab9

[crystal hatch]

that is the plan, indeed

[crystal hatch]

There are 2 people in "core Mass" team, but as the Mass gets picked up by engine teams we get more and more improvements from other teams.

[harsh storm]

I've literally been lightly researching mass stuff so I could get a proof of concept going. Because it is something that I was thinking about doing as well 😅

Was also pondering with the idea of figuring out how to get perception to be powered by mass as well.

[crystal hatch]

the problem with this PR is that it's pretty extensive. The functionality it provides is awesome, but it would take quite some resources to pull in, parse and test. The testing would need to be significant due to Fortnite heavily relying on EQS for a number of things, so significant changes to it need to be thoroughly vetted. There's just not enough Mieszko to cover everything 😉

[crystal hatch]

that's another thing on the list of things I'd like to work on 😄 I've designed UE4 AI Perception with data-oriented mindset (as much as it was possible back in the day) so "moving" it over to Mass shouldn't be that large of a task. Every AI sense is independent and in principle all of them could be processes in parallel. Digesting the events AIPerceptionComponent-side could also be done in parallel - event triggering would need to be collected and triggered in sequence though. Still, I don't see major roadblocks with this (although interacting with physics might need some additional care).
I'd try this first before attempting to write a whole new perception system in Mass.

[harsh storm]

I figured it was the physics part that would be the head scratcher part. Because mass doesn't support doing line traces as far as I know. So would have to roll your own. But I recall you mentioning that Chaos is evaluating integrating with Mass. So maybe in the future.

[crystal hatch]

Chaos is under more and more internal pressure to be nice to Mass 😉 It's going to happen sooner or later.

[crystal hatch]

mass cave
I lolled 😆

[slow bobcat]

any chance you can share the demo project from Riot to see how they use the PR?

[pastel moth]

https://tenor.com/view/well-waiting-caddyshack-gif-5385414

[crystal hatch]

Nope.

[slow bobcat]

sad noises

[pastel moth]

Im running an environment query but I keep getting this..
LogEQS (Warning) Query [EQ_ResourceQuery_Stone_AllMatching] can't use test [EnvQueryTest_GameplayTags_1] in option 0 [], removing it

Why would this happen? I just have a test to look for a gameplay tag

[crystal hatch]

without looking at the code, my guess would be that your item generator generates locations, not actors. Locations don't have tags.

[pastel moth]

oh my god... well I decided to read the docs on EQS. That was my bad. Thanks. You were spot on with that guess. 😄

[crystal hatch]

turns out I still remember something about the system 😄

[slow bobcat]

Hello! I'm confused about something:

• If you set a Path Observer in your logic (I have my own AIMoveTo_Task implementation), NewPath event is never called.
• Looking into it, it seems it's only called if you call "UNavigationPath::SetPath"
• UNavigationPath is a wrapper(?) to FNavPathSharedPtr (this last one is the type you will handle everywhere when doing path operations)
• I only see UNavigationPath being used when you find paths synchronously through the blueprint functions (we don't do things synchornously and never use the BP functions).

My question: should I somehow call that event manually like this "OnPathEvent(NewPath, NewPath != NULL ? ENavPathEvent::NewPath : ENavPathEvent::Cleared);" whenever I calculate a path (I use "UNavigationSystemV1::FindPathAsync" and handle everything in the delegate, which uses the FNavPathSharedPtr type)

[slow bobcat]

ok I think I was missing that call after setting the path observer to handle things properly

[pastel moth]

oh nice

[pastel moth]

Am I smoking crack? Not sure what is going on here. TLDR: My visual logger says my EQS query is returning things found but in my state tree debugger it says no results.

My EQS query is successfully finding my BP_Stone actors in the visual logger.

The debugger in my state tree doesn't show a result being returned though and my MoveToActor state never gets executed 😦 even though the FindClosestResource500CM is succeeding.

Attached are screenshots of my debugger and visual logger.

[slow bobcat]

Is it me or does the debug look like your state to run the eqs has a transition to itself? I can't see in the image and I think and I'm reading the debug wrong but... Just in case

How many transitions do you have there?

[pastel moth]

2 one for success and failure

[slow bobcat]

Ummm is that state something you can debug? To check if the task actually finishes?

[ebon shore]

I have my AI storing the attack target when the player enters their aggrosphere. I used to have it update that locale on tick, but I'm wondering if there's a better way? (ie, uses fewer references?)

*resources

[pastel moth]

It's either transitioning back to collect resources (which I'll change right now since it doesn't make sense to check for resources again if the first one failed) or going to moveto on success

[pastel moth]

@severe oar helped me fix it. My objects had navblockers that were blocking navigation so my movement wasn't working. It's weird that it didn't show up in the debugger but we just changed the navblock to obstacle and that did the trick

[ripe torrent]

what is the intended way to have different hear/sight sense configs for each enemy? this is what i have been trying to do:

• each enemy actor has different property values for hear/sight senses
• on the shared AIController constructor I create the UAIPerceptionComponent and each config object and register them on the perception component
• on OnPossess I read the enemy sense values and assign them for each sense

the problem is that it seems that I am not supposed to edit these values at runtime, as the PeripheralVisionAngleDegrees value is never updated to match the value I set on each enemy class

I tried to solve this by creating each sense on OnPossess instead of the constructor but if I do that none of them are registered at all

is it expected to have one AI Controller for each enemy type then set the sense values directly in each bp file?

[slow bobcat]

As a rule of thumb in games, it's better (for both performance and logic organisation) to have an Event Driven approach: only do things when something relevant happens. Sometimes it is unavoidable to periodically check for things. In said cases, two things are usually recommended:

• if you only have Tick available, check if you can lower the tick rate of your actor/component/object. In general, you should avoid using Tick as much as possible and disable it when possible. Even sn empty tick function impacts performance. Remember you can enable/disable tick in runtime (I.e: you could enable tick in your target tracking component when a target is within the colllider and disable it when not. Also you could reduce the ticking rate to 0.5s, so it checks target location that often. That would fit into the Event driven approach)
• another alternative we use a lot is Timers you start/stop when needed
• no matter which approach you use for "constant checking" (tick or timer), it's always good to have a budget for your operations as in "millisecond per frame". Imagine you have an operation you do in tick like "out of the closest 100 enemies, give me the ones that are facing north and play X montage on them". That will take some time and your cpu might spike and not meet your target framerate (60fps usually). In that case, you can measure the time elapsed after each enemy is processed and, if your budget is reached, continue the next frame.
  For example: your budget might be 1ms per frame for it. If each enemy process takes 0.1ms, on your 10th enemy you should skip the rest until the next frame. It will take 10 frames to process all enemies.

As a last note: in AI, things are almost never needed immediately. Ask yourself the following questions:

• do I need to know the exact position of my target every frame?
• why do I need to know it?
• am I using that location every frame too?

On your particular case looks like you could use the Perception system for it. It already have some optimizations in it

[slow bobcat]

Unless you can do c++ there is no way to change sense values at runtime as far as I know. Maybe someone has some way in BP's to do it but I doubt it's possible

[ripe torrent]

I am trying to change them with c++ and that's where I am having issues

this is what I had on OnPossess: ```cpp
const FAISenseID SightSenseID = UAISense::GetSenseID<UAISense_Sight>();
if (UAISenseConfig_Sight* SightConfig = Cast<UAISenseConfig_Sight>(AIPerception->GetSenseConfig(SightSenseID)))
{
SightConfig->SightRadius = Enemy->GetSightRadius();
SightConfig->LoseSightRadius = Enemy->GetLoseSightRadius();
SightConfig->PeripheralVisionAngleDegrees = Enemy->GetPeripheralVisionAngleDegrees();
}

[slow bobcat]

I can check later when I'm at my pc if we do something else. We definitely change ranges at runtime based on gameplay conditions

[ripe torrent]

that would be amazing 🙏 thanks

I am still learning UE so I might be missing something

[slow bobcat]

That said, we use our own implementation of sight, but I could swear we also change hearing and that's the default from unreal

[slow bobcat]

I think you are missing the call to apply the changes to the config of the sense.

Check this
https://zomgmoz.tv/unreal/AI-Perception/Configuring-AIPerception-Senses-in-C++

//The below code should be in your class' constructor

//Create the config object
UAISenseConfig_Sight* SenseConfig = CreateDefaultSubobject<UAISenseConfig_Sight>("SenseConfig_Sight");

//Set default values
SenseConfig->SightRadius = 300;
SenseConfig->LoseSightRadius = 300;
SenseConfig->PeripheralVisionAngleDegrees = 22.5;
SenseConfig->DetectionByAffiliation.bDetectEnemies = true;
SenseConfig->DetectionByAffiliation.bDetectNeutrals = true;
SenseConfig->DetectionByAffiliation.bDetectFriendlies = true;

//Tell the perception comp to use the config object
PerceptionComponent->ConfigureSense(*SenseConfig);

It says it needs to be in the constructor but I think it might work somewhere else too. Will check later

[ripe torrent]

ohh that was it

thanks!!

I had it on my constructor but I didn't know I had to call it again when changing the values

it seems to be working now

[slow bobcat]

Nice

[pastel moth]

Does anyone remember where the state tree smart object epic tutorial was? There was some really nice docs on how to integrate everything but I can't seem to find them..

[harsh storm]

Hold up. Hold the phone. Wait just one dang minute here. How did I miss that the Move To node now allows you to use a BB key for pretty much every property?!?!?!?

As does the Wait node

There is a new struct on these tasks, FValueOrBBKey_{data type}.

Commit was June 6th, 2024.

I was literally just about to go in and make this feature and do a PR.

[slow bobcat]

what? where is this? which file?

[slow bobcat]

ah ok... I'm still not updated enough to have that

[elfin smelt]

I discovered it in 5.5.3

maybe it's from 5.5

[harsh storm]

It is from 5.5

[harsh storm]

ValueOrBBKey.h in the AIModule

Considering 5.4 released in like April and 5.5 released in November, it is a 5.5 thing.

I just haven't used BTs in 5.5 yet. But noticed it just now when I was going to do this very thing, lol

Yup. It was in the release notes too.

Just missed it

[serene fern]

Whats the method again in terms of attaching the AI Perception Sensing thing so it follows Eye point of the AI ??

from what I've read online the GetEyesViewpoint function handles getting the Loc and Rot and AI Sensing uses that ?

[serene fern]

what would I need to change to fix this ??

bassically allow the navigation to fit through the gap

[elfin smelt]

Could be your NavAgent that is too big (maybe the radius of your default NavAgent)
or Could be your Navigation Mesh that have not enough precision to detect the door (maybe your cell size is too big to pass throught the door)

[serene fern]

thats the settings

I've reduced Agent radius down cause the capsule can be as smaller as 25 radius

but the generation of nav through the doorway I'm not sure there what I'd need to change to fix that

[slow bobcat]

What you need to change is the magnet radius in the project settings - > Navigation System - > agents - > Supported Agents

If the door width is 100, your agent radius needs to be smaller than 50 - some extra that will depend on your nav resolution settings

Is also needs to be tall enough

the settings you changed are general/default

[serene fern]

that not right in terms of how to handle the player location in terms of the AI seeing it ?

seems the Stimulus location only fires once in terms of when the player either goes in or out of vision range of the AI Perception thing rather then it constantly firing whilst player is within the range ?

[pastel moth]

Is it possible to define smart object claim handles as output in the blueprint and assign it to a parameter? For some reason my output just shows up as "Smart Object Claim Handle" and doesn't allow me to bind it to anything.

[tacit grove]

If anyone considers themselves very competent in basic NPC AI (locomotion, but complex) and would be up for a short term task that's very well paid, please let me know!

[slow bobcat]

not interested about working on it but curious about what's your problem

[slow bobcat]

yep. If you want to check you need to grab the array with perceived actors and see if yours is in it. I think there's good info in this chat about it

[serene fern]

but the basic line is I have to do the checking myself for getting keeping the info updated minus the target as thats updated when stimulus changes

thats what i gather from online

[slow bobcat]

wait.. .I'm a bit confused.
That Event you have "UpdateAISenseInfo" to what is it bind?
What do you mean by "Stimulus Location only fires...", which function triggers?
I just realized I was assuming many things here

[serene fern]

so got an issue you can see here the AI finding the path to where it needs to go to

seems to get to there

and then to that point and breaks

gets there an just won't go any further in terms of after it looses sight of the player

[slow bobcat]

because you have some logic that says "do not move if you lose sight of the player"? or why does it stop?

[serene fern]

so i think I know why now which is when it stops it's doing it's Hunting which is going to the last location player was seen, it then should do a searching before returning to normal so either roaming or patrolling. At the moment I think when it loses sight of player and then sees player again there is a timer before it actually starts chasing, I think I need to maybe have it so if player is seen within an amount of time since last time they where saw skip the timer and just chase

so i'm getting an issue at moment where it's getting stuck on Move To node and flickers back and forth

I really don't get whats happening here

I've made my own Move To node this time and it's saying it's a success yet I can clearly see it's stuck in that hunting one

[harsh storm]

Would probably get more bites in #freelance-jobs

[placid silo]

https://www.fab.com/listings/343a9376-61f1-43ec-9bd3-6fe8370df064

Is anyone using this plugin and knows whether the flight trajectory can be manipulated to not be completely straight?

[pastel moth]

That's more of a question for that plugins discord 😮

[reef wharf]

Hey ya'll, I'm using EQS with a custom context that finds all actors in the world and does a scoring based on the found actor's locations.
However, if the find all actors returns more than 10 items, the scoring of the EQS nodes turn into NaN?

does EQS not support scoring against more than 10 contexts or something?

[slow bobcat]

Yeah there's no limit in how many items your tests evaluate. Could it be you are doing something that ends up being conflictive?

[reef wharf]

I'll try just a really basic query

[slow bobcat]

Do you have many tests in it? You could try disabling all and enabling one by one

Until it fails

[reef wharf]

Okay thanks for the sanity check

it was definitely because of the way I set up my test

EQS doesnt like it if I use only a multiply scoring

but it works fine for less than 10 items for some reason lol

[pastel moth]

Is there a reason you cant bind outputs in a state tree? Like for instance I have a smart object claim handle and I can't seem to bind it to any parameters. I have a smart object claim handle parameter but nothing shows up in the UI for it for the output.

My type is the same between the param and the output var

[harsh storm]

You can bind them. But they can't be sibling states. They have to be child states.

[pastel moth]

yeah I was able to do it if it was a child

Just thought it was weird I couldnt bind it to a param

is MoveToAndUseSmartObjectGameplayBehavior useable in state trees? I can never seem to get it working. Other ways of moving to and using smart objects work just fine though

[pastel moth]

Can my HarvestResourceNode state transition to the DeliverResource state in this state tree? Or is this not a valid way of transitioning?

everything is try enter BTW

[misty wharf]

iirc states don't really have any restrictions on where they can transition

[pastel moth]

Hmm I think its an issue with UseSmartObject task I made

I noticed it does finish the task with success but then it doesn't follow its own transition

I added a delay node on the HarvestResourceState and tested it with Transition on State Completed. That seemed to work.

I don't understand why finish task successfully is executing in "Harvest Resource" but not resulting in a transition to my Deliver Resource state.

[pastel moth]

yeah that's what I thought :/ I'm puzzled.

[misty wharf]

Try having it transition directly to FindDeliveryPoint

since that would be the one that gets activated if you were to transition to DeliverResource

[pastel moth]

oh but deliver resource is try enter I thought that'd get activated first

[misty wharf]

Yeah I have no idea, those are confusing as hell lol

I never got the one working where it is supposed to enter the parent first and leaf later

So I'm not quite sure how it's intended to be used

[pastel moth]

i like your layout you had in your notes on your site btw

[misty wharf]

Oh you mean the site's layout?

[pastel moth]

ah sorry I meant your notes on state tree where you showed your states setup as an example

[misty wharf]

Ahh right

[pastel moth]

but your site in general is great

[misty wharf]

This one? https://zomgmoz.tv/unreal/State-Tree/StateTree-design

[pastel moth]

yeah

[misty wharf]

Yeah took me a while to figure that out lol

I think there's some new state tree prop reference stuff to alleviate this but it sounds like it's still kinda work in progress

[pastel moth]

i wish epic would provide some more examples

but oh well

[elfin smelt]

Hey
Someone uses Wwise in UE 5.5.3 ?

I have a weird issue where If I have an actor with a Nav Modifier and I add a AkPortal.
when the actor is loaded or when I move it in the scene. I got this error line 93 of UNavRelevantComponent.cpp LogNavigation: Error: UNavRelevantComponent::GetNavigationParent called before initialization of the navigation parent cache for (your object name). This might cause improper registration in the NavOctree and must be fixed
after some search, I found that the AkPortalComponent add UTextRenderComponent inside its OnRegister callback and if the Actor Construct the NavModifier Before its AKportalComponent.
when the AkPortalComponent add its RenderText (for loop line 694 of AkAcousticPortal.cpp), it called an Navigation Update and your NavModifier is updated but the parent was not initialized

If someone can try to Add a NavModifier and then an AkPortal to confirm the problem to me.

and if someone has this issue you can just delete your NavModifier and recreate it. It fixed it because the NavModifier will be added after your AkPortal

[slow bobcat]

Damn. We do use wwise. Haven't heard of that problem but we doubt we add nav modifiers where portals are

[elfin smelt]

The issue arrived on our doors.
I don't manage wwise but the issue is raise on the Navigation system so its for me x)

I''m not sure that it is 5.5 related

I didn't check this particular issue in 5.4 😅

[still eagle]

Hey, I’m relatively new to unreal engine and I’m trying to create a good BT for the enemy in my game. Would anyone be interested in helping/ working on it with me out of enjoyment 😂 if not I’ll pay someone to help me

[serene fern]

Anyone have better idea for this at all ??

what I'm trying to do is have it where as it's a ranged NPC it goes to a the found EQS point rather then to the player which is what Close Combat NPC would do

[serene fern]

so apparently the EQS Query failed even though I when I use pawn tester and stuff it works fine as in I can see failed points would be retireved

[tawdry zephyr]

@crystal hatch do you have any massive creatures in Fortnite?
We do. And on complex terrain, or terrain with props, their massive capsules like to walk on stuff normal characters can't because it lets them traverse huge dips in the terrain
If we make their nav capsules thin so they navigate properly, then they clip walls, so that's no solution
Any ideas for how to handle this?

Feel like we need this kind of shape but not feasible 😄

[dense owl]

I thought Mover 2.0 helped with this capsule restriction stuff

[tawdry zephyr]

There is no way we're going to use Mover 2.0 tho
We use it for critters but it's no good for anything beyond that

[dense owl]

😦

[tawdry zephyr]

We also just don't have the resources to dedicate to a bespoke solution
Hoping for something a lot simpler

[harsh storm]

Time for Unreal to add a preset for agent size and call it "Johnny Bravo"

As for massive creatures - I don't think FN does. Other than during events. Like when they had Godzilla. But uhhh - I don't think the big G is going to be worrying about a navmesh.