gameplay-ai

[harsh storm]

I can easily push it to 100+, but the reality is, unless I'm making like Dead Rising - zombie games don't often need that much.

Just have like 40-ish and reuse 'em

Design the level around it 🥳

[slow bobcat]

Yeah my experience is with games with heavy AI, animation, navigation etc. "proper" npc's

[harsh storm]

Zombies are, without a doubt, by far, my main area of interest for AI 🤣

[misty wharf]

Project Zomboxxigar when

[harsh storm]

It's in the works. But I have other projects to tend to first

[misty wharf]

:D

[harsh storm]

Like, I'm not even kidding either, lol

I make so many zombie game/prototypes and talk about so many ideas - it is a running joke in my gamedev circle

So pretty much every game has some kind of zombie in it

Even conjured up a zombie dating sim game 🤣

Mainly as a joke, but still

[misty wharf]

That sounds like it could be a hit

[slow bobcat]

You should definitely Game Jam that...

[hallow compass]

Common AI Behaviors plugin is shaping nicely

[hallow compass]

https://youtu.be/AI17tIMczyo?si=L4bSSWWx2qUkcxwL

[obsidian delta]

does anyone know why breakpoints on state tree in blueprint dont work?

I feel like im missing something obvious

[hallow compass]

It works but will be after IDE, BP graph breakpoints

Be sure the breakpoint can actually be triggered

[celest python]

Tick aggregation, vblanco trick, significance manager, multithreaded anims and somehow making montages not bottleneck cpu helps you reach ~50 on PS4 hardware i think

[obsidian delta]

wdym

I know that the tree runs those states from breakpoints in cpp

[hallow compass]

Weird

[obsidian delta]

also i see that my state tree completely ignores an event lol

doesnt even evaluate it

[hallow compass]

Show setup and give context

[obsidian delta]

theres nothing crazy going on

it usually cycles the last 3 states

and goes back to root

I know that this runs

I guess i can try tacing what it does with the event in cpp

[slow bobcat]

We use all those tricks (+some more) but not the VBlanco trick. No idea what is that

[slow bobcat]

Do you have the state tree debugger open for the tree? Breakpoints for state trees only work if the state tree debugger is active

[celest python]

#cpp message

[obsidian delta]

ah that would explain it

[obsidian delta]

Does state tree have debug info for events?

I dont really know what its doing with the event i sent it

[hallow compass]

Check debugger

[hallow compass]

Very nice one

[obsidian delta]

Is it supposed to pop up in the debugger that event failed?

i dont see anything

[hallow compass]

Play your game and pause/stop when the event should of been triggered

Playback in state tree debugger to see if it got tried

[obsidian delta]

hmmm it doesnt show me anything

[obsidian delta]

Ok i just figured im dumb

i was supposed to force transition from states in root

cant just expect the tree to transition because i sent it an event'

[slow bobcat]

So... I need context. Where is that function supposed to be called?

I imagine you are discussing something in the thread but not sure what / with who. Also do you have the original source of info? Who is VBlanco?

[celest python]

It does what MoveComponentImpl of UPrimitiveComponent but skipping all the unnecessary steps

Instead of moving something with a movement component which calls MoveComponentImpl, so use that static function to update the location manually

~2.6x faster on megafunk's tests

for example, that MoveComponentImpl is beyond disgusting yet it runs for a niagara particle

a particle which has no collision does not require any of the things that function is doing

[slow bobcat]

Ok so this is not for characters but for other things withiut physx no?

[celest python]

for components that does not require the steps MoveComponentImpl doing, so its for components characters using

but probably not for the capsulecomponent

it could still work for capsulecomponent if you dont need how it slides you once you collide with a wall etc.

unity-like basic movement

biggest problem is marking render state dirty cant be threaded

if you can achieve that vblanco trick can run on other threads

[celest python]

vblanco is an engineer that implemented this to pubg iirc

he shared the code with us through megafunk

he was quite active in early days here

[slow bobcat]

Is that anywhere around I can check? Github?

[celest python]

yep but its the same code just on github 😄

with no further usage examples

[slow bobcat]

I don't know what megafunk is

[celest python]

https://github.com/Megafunk/MassSample/blob/main/Plugins/MassCommunitySample/Source/MassCommunitySample/Experimental/Benchmark/MSSceneCompTransformToActorTranslators.h

megafunk is another fellow here, also author of mass sample

[slow bobcat]

Aaaah I see ok ok

Thanks a lot for this.

Ah wait hahaha this is Vori's repo!

We worked together for a bit in the se company

I talked about this with him but never checked the repo

Nice

[celest python]

ah yeah, mega and vori authoring the repo together 😄

[runic flare]

I have seen Siggi mention the term property binding functions a bunch -- can someone explain what that means exactly? I'm looking for a way to bind functions in state tree nodes (primarily conditions) so I don't need to sync duplicate state (e.g. I don't want to bind to a bIsVulnerable, I want to bind to a bool IsVulnerable() const), I'm wondering if this is what he's referring to and if so how do I do this?

[crude mirage]

how do you make enemy ai? i’m specifically confused about state tree tasks but really any ai in general

[celest python]

likely yeah

it exists in anim BPs but not on state trees

i dont know what is the technical reason it didnt exist at the first place or was it a design decision but it'll improve things a lot

[celest python]

check the pinned messages

there are many inspiring resources to get you started

[obtuse igloo]

Hi all! I was wondering if someone could help me figure out the Move To Actor Or Location node, I want to use the actor goal and have my ai follow a toy ball as the continuous goal. But when I use an actor and the ai is within the acceptable radius, it isn’t firing off the finished event.

So I swapped to see what would be different if I used the Location of the toy ball actor as a goal instead, and it fires off the finished event just fine, but of course the ball would not be in the location the ai went to (because the ball is rolling around). This is why I wanted to use the Actor goal instead and get notified when the ai has reached the ball (and then kick it).

Any suggestions appreciated! I tried explaining the best I could ^_^

[chilly nebula]

yes you can bind to a function like that, its exactly what binding functions are for 🙂 There are a few that ship with the framework you can look at for reference on how to make your own (its easy!).

https://github.com/EpicGames/UnrealEngine/blob/ue5-main/Engine/Plugins/Runtime/StateTree/Source/StateTreeModule/Private/PropertyFunctions/StateTreeFloatPropertyFunctions.h

[green loom]

Heyop - any idea why the EQS appears to work differently run time. The test pawn shows the 0 score on where the character is at and two green ones in the flank. In the in-game view, the character (BP_Valkyrie) moves to one but the point scoring appears to ignore the Distance to Querier filter. 😩

(Test pawn in the first image has the same valkyrie model. The context is player start in editor - and the player in game)

[hallow compass]

task handles arent unique across multiple state trees ?

two state trees on two different AIs will return the state state handle for the same executed state

[slow bobcat]

I'm not sure about this but isn't the handle a way to identify a state in a tree to grab its Execution data?

[hallow compass]

Probably, but it seems it can be deterministic

If 2 AIs runs the same events with same runtime data

[slow bobcat]

Handles are incremental from 0 (not valid) to N. If they run the same in the same order...

[frank trellis]

my AI Moveto isn't working even though i set a location and what happenes on success. Can i Get help? (Yes, I have navmesh)

[celest python]

print this

from on fail

then check vislogger

[frank trellis]

how do i check vislogger?

[celest python]

tools tab in editor

[frank trellis]

seems like they DO move, just for a split second

should i show more code?

[runic flare]

This is great, thanks! Just so I'm sure I understand, let's say I have a UHealthComponent which is part of my tree's context. I have a function bool UHealthComponent::IsVulnerable() const --> in order to use this function via binding, I basically need to create a wrapper where I define InstanceData that has a member UHealthComponent* HealthComponent, then I define a struct that inherits from FStateTreePropertyFunctionCommonBase and the override its Execute function where I'd do InstanceData.Result = HealthComponent->IsVulnerable() ?

[chilly nebula]

the binding function gets the context, and through that you can just grab the health component directly. Pretty much can do whatever you want 🙂

[celest python]

so its just like in the anim graph, right?

[burnt wraith]

Does anyone know how to mitigate delayed animation montage section playing on clients? I have a state tree triggering a gameplay ability on my AI that is a serveronly ability and is supposed to play an animation montage section that is to melee attack. On the Listen Server it's fine, but when I switch to client it lags at the start of the animation and then finishes the animation and calls my OnMontageCompleted function. I tried making a Multicast function on my pawn that is called by the GA to play the animation on the clients, but there is still a lag

[chilly nebula]

Im not familiar enough to answer that 🙂 but pretty sure its not the same tech, afaik we only use our property binding in state tree and smart objects

[celest python]

Are there any plans to add bindings to the utility weight parameter in nodes in the future?

[regal patio]

https://youtu.be/gL-EpVi8HbQ?si=IjOmuxlgbvv43H4m

[blazing flax]

i'm trying to disable RVO collision after an enemy dies, so that other enemies don't yoink around the corpse and will just run through it, but nothing seems to work. I've tried a few recommendations from around the internet:

• detach from controller pending destroy
• un possess
• updating avoidance group mask
• changing movement modes

but none of them seem to work. I'm 99% sure its not a collision thing, because disabling RVO entirely prevents the behavior. it just seems that disabling it when they die has no effect

[tranquil meadow]

This is more of theoretical than something more concrete. For enemy AI design using a ticket system, how do you guys handle using different moves that may have different range requirements with tickets?

[slow bobcat]

Haven't looked at RVO in unreal for ages but... What does it take into account to avoid and what's the function in its logic that calculates what does the AI need to do to avoid things?

If you figure out those two, you will learn how to disable it

There should be a function (probably in the movement component or the AI Controller) where it looks to nearby agents to calculate the velocity

[hot narwhal]

Hello, i have a really weird bug where my state tree execute 4 or 5 times for one controller. I have a state tree ai component on my ai controller, i disabled the start automatically and just called start logic in the possess event. When i play in editor, with one player, i have 4 execution lines in the debugger of the state tree. I created a task that is executed on the root node and that juste print string for one frame, and the print is appearing 5 times. I testes with the ai being spawner as well as being places in the world but i have this bug in both cases. Does anyone have any clue about this ? I am using 5.5

I have an evalutor where i print string that is called only once

[slow bobcat]

The state trees code have a thing that is a bit unsettling: a while loop with a 5 loops limit that will run taks up to 5 times to ensure all the transitions trigger. Which causes sometimes to get your debug and logs 5 times. I always found that difficult to explain and justify. Check the code in Execution Context where the transitions are evaluated.

[hot narwhal]

So it is a normal behavior ? That's so weird

That means my tasks are executed multiple times one the same frame it feels like a bug more than a feature

[slow bobcat]

Yep. I agree. Sounds like a bad hack

[hot narwhal]

Can you telle me where you saw the code of the while loop i'll try to check as well to understand

[slow bobcat]

If memory doesn't fail, State Tree Execution Context class, in its tick, there's a function to evaluate transitions

Ctrl+F "while (" and you should find it

I'm in a car in a road trip so no way I can check

[chilly nebula]

Not currently. What is your use case here?

[hot narwhal]

Okay thanks i will check that !

[hot narwhal]

Ahah no worries, thank you !

[slow bobcat]

The loop uses a variable Calle MaxIterations iirc

Aaah it's not a while, it's a for

#gameplay-ai message

[hot narwhal]

Oh yeah i just saw that, this is a really weird behavior, all tasks will be executed 5 times wtf. I guess i will create my function directly inside the ai controller

But it is weird that on the debugger itself there's 4 executions lines, i think it is a different problem

[celest python]

Using weights as scores and having nodes as goals that can dynamically score theirselves. Another task on the parent node might update bound parameters based on world state to influence the selection

Right now my custom AI system does everything state tree does minus the editor. Each goal returns an array of score struct and utility component does the calculation based on the data it received to select the most important goal. Only missing feature right now to carry this to ST is dynamic weights/scores

[chilly nebula]

Why not use considerations to represent the changing world state? Thats what they are for, but we have some work left to make them better (like multiplying considerations together)

[celest python]

I didnt know about considerations! will check it out asap

alright, I completely missed we can even add float inputs

appearently what I was looking for already existed for a long time

[hallow compass]

anyone knows from where is the stimuli debug trace sphere and age drawn ?

cant find it, thought the perception comp/system still had an dead stimuli but Perceptual Data is empty

so idk where it comes from

[chilly nebula]

If you want to really dive into utility AI, I really recommend one of Dave Mark's GDC talks 🙂 for example "Building a better centaur" is good!

[chilly nebula]

did you check Vislogger or Rewind debugger?

[hallow compass]

this is the legacy system

pressing 4

[chilly nebula]

could also be in gameplay debugger 🙂 (yes, we are working on consolidating it all)

[hallow compass]

looked into it, it draws data to screen on HUD, but no debug draw in worlds (at least nothing with "age" in it)

[chilly nebula]

stimuli is part of perception yeah, you seeing entries you dont know where are coming from?

[hallow compass]

Hey siggi, while you are here im wondering if you know about the function binding layout bug ?
https://media.discordapp.net/attachments/221798778208780288/1424505446210736220/statetreefuncbug.gif?ex=68ee14d4&is=68ecc354&hm=7c24f8d64f81721237abd8bc0ce6108bc3864fd0fa33e4dc5c6a366dc9917e78&=&width=1697&height=1224

w/e i try i ALWAYS get this when i try to repro
#gameplay-ai message

gif is low fps but basically it spams the collapse/uncollapse so nothing is usable

[hallow compass]

there is a debug draw of a sight perception, it ages forever (which is expected since my AISense Sight uses NotifyType = EAISenseNotifyType::OnEveryPerception which doesnt run the aging system of UE, or at least doesnt call the Forgotten delegate)

but after the MaxAge, i dont see the stimuli anymore on the perception comp & system, but its still drawn

[chilly nebula]

you looking specifically for the code where the debug drawing is happening yeah?

@hallow compass I guess you've already seen UAISenseConfig_Sight::DescribeSelfToGameplayDebugger() ?

[hallow compass]

yep, lots of stuff but not the specifc draw sphere, line and text

this

[celest python]

hmm, so my final mental struggle is - in my current system each goal scores itself and a component that manages all scoring system just calls Goal->Score(...);, while in ST its more "top-down", as in, it looks like scoring and considerations only happen through the state tree graph. but it makes a lot of sense since collection of systems form a behavior in the graph instead of tasks creating monolithic behaviors through OOP. Composition over OOP i guess

so if I have a tree that has a sequence of nodes that links two assets: "slide towards enemy (one asset) and shoot enemy (one asset)", that always has a requirement of a same specific condition to get a higher score everytime, I'd need to put that consideration everytime I link the state tree asset combining two behaviors because there is no way for ST to receive this info from the asset itself

am I right or is there any workflow to make this easier 🤔

[chilly nebula]

The scoring is run as a state selector, and yes you have to have those states pre-authored. We are looking into ways to expand on this in the future, for example to score multiple targets (like ranged attack 10 different targets, and score them on distance, hp remainin etc).

For runtime composition, you might be able to have "slots" for activities in your authored tree that you override at runtime (using tags) with external assets if you want to re-use specific states across multiple trees or selector nodes.

Having to score each action separately is a cornerstone of utility scoring, and the re-usable pieces are the considerations (like distance, or remaining health on target).

[chilly nebula]

UAIPerceptionComponent::DescribeSelfToGameplayDebugger() is handling the stimulus

[hallow compass]

thanks!

[digital pilot]

anyone know if you need to add nav data collision to a mesh in order for AI to jump on top of it? Or is there an actor or reusable thing to use you can just drag out and it will add it for you?

[hallow compass]

Any primitive with a collision and nav enabled can do the job

[odd violet]

Quick question, I'm trying to use AI StateTree schema with my StateTree Asset on my Enemy AI Controller class as a component. I know about the "Context" use in StateTree task variables to get the State Tree context actor as a State Tree built-in feature, however this is my first time using the* AI State Tree schema* and I've assigned the Context Actor Class to to be MyChracterClass and my AIController Class to be MyAIControllerCustomClass however, even though in the StateTree Editor UI there's mentioning of both, in one of my tasks if I try to create a "Context" variable of type MyAIControllerCustomClass it's still invalid. Is there a proper way of initialising the AI Controller Context variable in State Tree Tasks when using the AI StateTree schema with StateTree assets on AI Controller classes?

I've attached some quick screenshots to better explain. I know there's the option of using the Context Actor class instead, but I was hoping to use the StateTree asset by attaching a StateTree component onto my AI Controller rather than the Character class. Does anyone have any experience with this? I tried googling and looking for YT tutorials but everyone just seems to use the Actor Class as Context instead 🤔

[autumn grove]

Has anyone tried to train an Open LLM onto the Unreal Engine 5 source code to create an actually useful, correct code helper for "vibe coding"? ChatGPT and others just hallucinate the shit out of everything. And provide at best 70% correct solutions.

[worthy lagoon]

Anyone know why my navmesh works in editor but not when packaged?

[slow bobcat]

There's the AI assistant that is available on both Web and editor in 5.7. It also generates code.
I would avoid "vibe coding" since it's an euphemism for "let's skip the learning process and produce really bad code"
https://dev.epicgames.com/community/assistant/unreal-engine

[slow bobcat]

Multiple reasons.
What's your nav configuration?

[autumn grove]

Your euphemism implies inexperience. I'm looking for an accurate replacement for all the Unreal documentation that has been wiped out. Basically asking for code snippets, etc. But thanks for the link - I had no idea Epic did this

[slow bobcat]

Not sure what documentation has been wiped out. If you mean the old wiki people maintained, that's still available. We tested several Comercial solutions at work and they all trip most of the time as soon as you try to do anything complex

So far, chatgpt5 (lates model) yields the best results and it's still veeeeeery far from good

[autumn grove]

I've had more success with Manus (https://manus.im/) actually. But eventually, the context just falls apart and it starts making up functions. And it gets very annoyingly apologetic.

[lost plinth]

hello, I need a bit of help with my behaviour tree

I plan to just have a simple perception system with the use of sight

I have already got my AI to do a simple idle and walk to a random location behaviour however I have issues when trying to get my AI to move to a specific object

I'll post some helpful screenshots

[hallow compass]

Yes and it was still bad

Just learn the engine

[lost plinth]

In the first screenshot, I have my behaviour tree, the second screenshot shows how the AI itself works, and the last screenshot shows how the AI is spawned

the issue I have is, this specific branch of the tree is failing, which is the perception part and it causes the whole tree to fail

I believe I have the correct stimuli set up as well as my config file

and since my AI is spawning I have set the setting in the pawn to have the behaviour tree control the AI when it is spawned

I'm just unsure where I might be going wrong

if anyone has any idea I really do appreciate the help you can provide!

[slow bobcat]

I would start by checking the gameplay debugger and see if your AI is actually perceiving the thing or not. It usually opens with the comma operator in game IIRC (I have it remapped always so not super sure)

[lost plinth]

well this is the state of the blackboard keys

ok so I've found out two things

when an AI can't see the object as it isn't in it's line of sight, the tree fails

when the AI can see the object, the tree does run, but it still doesn't move because the move to function is called for a split second then immediately goes back to the wait condition

so I think if it is in the line of sight, it does percieve it, but I'm confused why the tree itself fails when it isn't in the line of sight

[lost plinth]

been playing around and I'm still not getting anywhere

for whatever reason it looks like my AI behaviour tree, if the AI can't percieve the object, it does go into the second branch

but the second branch doesn't call the service, which should set a key called idleorrun to be 1 or 2

instead the key remains as a value of 0

and I don't get why, because this service should be running independent from the AI tree

[slow bobcat]

A service will run if:

• the service is on a running task
• the service is on a composite node (selector / sequence) and a task within the composite node (or a children of it) is running

The selector on the left will only run if the task below it runs, and that's not happening because the decorator prevents it

[slow bobcat]

I'm assuming this is the random service on the left

If you set breakpoints in the BT you should have a better image of what's going on

[lost plinth]

I was trying to use breakpoints but they didn't really activate

though now I think I realise that it just was because the service wasn't being called

[slow bobcat]

I mean breakpoints in the BT nodes within the BT editor

[lost plinth]

I've changed my tree and I solved the issue with the service

I still need to investigate the actual perception part though

something interesting I found

it seems that because the stimuli are static, the AI doesn't seem to walk to them

but when I move the location of the stimuli during gameplay, the AI is then more interested to walk to it

[quartz shadow]

Hi ! Is there a dedicated channel for State trees ? Is this in Mass ?

[misty wharf]

This is AI in general, #mass is mass

You can ask ST related things here

[lost plinth]

something I also just wanted to know

how could I make my AI stop a certain distance from an object

so if say they were walking towards it they stop within a radius of 10 units

[misty wharf]

If you set acceptance radius on MoveTo to the number of units, does that not do it?

[lost plinth]

I forgot that was in move to

thank you!

[misty wharf]

👍

[final prism]

Does anyone have an idea what the "cluster links" in recast are? It seems they are used for hierarchical pathfinding, but i can't find any documentation on these anywhere..

[lost plinth]

hi does anyone know the node that does this: it finds the nearest reachable point on a navmesh from the given location

I'm realised that is why my AI is failing

[slow bobcat]

Ignore them. Some attempt was made and it all points to the fact that they are not being used at all. There might be someone somewhere using them but so far... Nothing
#gameplay-ai message

[slow bobcat]

You can project a point onto the nav using a extension (how far form the point you want to look for nav).

[lost plinth]

so just a scalar value?

[slow bobcat]

look for the node Project point on navigation and you will se what I mean

[lost plinth]

hmm I tried using it but my location ended up being 0,0,0

how exactly is it meant to work?

[slow bobcat]

You pass a location and a extension (x, y, z) and the system will look for the closest nav point withing that extent (if any)

Let's say your location is 1, 1, 10. Your extent needs to be at least 10 on the Z axis or it will never find nav

[lost plinth]

ah ok

thank you, that works perfectly now

[odd violet]

Ok, is there a particular reason why AI StateTree component start even though I unchecked the "Start Logic Automatically"

[misty wharf]

Oh they went from it never starting automatically even if you have it checked to always starting even if you don't have it checked?

[hallow compass]

if only spline metadata was nicer to use ... i wouldnt have to have an ugly array to sync with the splie points

[odd violet]

The AI StateTree (not the regular StateTree component) is driving me insane.. I even reported a bug with the AI Controller context binding not working

[runic flare]

This article from last week shows new methods of avoiding tick in State Tree in 5.6. It explicitly mentions, and shows, a version of Run EQS Query that fires a delegate when results are ready. Is it just me, or does that task not exist in 5.6? I can make my own but the article showing me the exact thing I want gives me pause -- am I missing something?

Ah I see the article was mentioned here, maybe you can clear this up for me Siggi?

[quartz shadow]

Hey ! I have a question regarding meta properties on ST Tasks etc.... If we create a linked asset and we expose a variable from a task being processed in this asset. Then when used in another ST all meta specifiers linked to the property exposed seems not to be propagated correctly. Is this expected/known ?

Example: Let's say I have a task with a float that I want to cap to 0 - If I promote this variable as a global parameter of my tree I can now set a value below 0. Solution I found is to have the instance data wrapping another struct so that it works as expected but it's far from being a great solution as it create other issue

Context: I am investigating composition of ST and how we can reuse them easily (I am on 5.5 BTW)

[stable plover]

question all I am having trouble getting world partitioned nav mesh to work, i been following the docs https://dev.epicgames.com/documentation/en-us/unreal-engine/world-partitioned-navigation-mesh, i have the chunks generated but its not loding them if anyone can help that would be appreciated

some people said disable is worldpartitionednavmesh, i did that and no luck

[grizzled schooner]

Hey everyone,

I'm looking to learn more about StateTree's - is there any good written books/documentation on it?

[slow bobcat]

Check the pin messages. There's good stuff there

[slow bobcat]

If this is anything similar to the old system with sub levels, check if you have a recast object in the game when you play

With sub levels you need to add a nav bounds volume to the persistent to ensure the recast object is generated there, so nav chunks can be added/removed at all times

[stable plover]

yea i have the recast obect when I play

[stable plover]

if this is what you mean then yea

Attached 0 tiles to NavMesh
weird I got this in the logs

but my chunk actors exist

[grizzled schooner]

thank you - I'll review the pinned messages 🙂

[stable plover]

yea @slow bobcat the tiles attach in editor its pie

when i build LogNavigation: Verbose: pos: V(X=-64000.00, Y=-89600.00, Z=9650.00) ext: V(X=12799.00, Y=12799.00, Z=3011.00) LogNavigation: Verbose: URecastNavMeshDataChunk::AttachTiles Attaching to NavMesh - RecastNavMesh_UAID_5811228C0404AA9802-Default LogNavigation: Verbose: Attached 169 tiles to NavMesh - RecastNavMesh_UAID_5811228C0404AA9802-Default LogNavigation: Verbose: [NavigationDataChunkActor_UAID_5811228C0404BC9802_1504567100] ANavigationDataChunkActor::AddNavigationDataChunkInEditor LogNavigation: Verbose: pos: V(X=-89600.00, Y=-64000.00, Z=7200.00) ext: V(X=12799.00, Y=12799.00, Z=1741.00) LogNavigation: Verbose: URecastNavMeshDataChunk::AttachTiles Attaching to NavMesh - RecastNavMesh_UAID_5811228C0404AA9802-Default LogNavigation: Verbose: Attached 169 tiles to NavMesh - RecastNavMesh_UAID_5811228C0404AA9802-Default LogNavigation: Verbose: [NavigationDataChunkActor_UAID_5811228C0404BC9802_1504558095] ANavigationDataChunkActor::AddNavigationDataChunkInEditor LogNavigation: Verbose: pos: V(X=-89600.00, Y=-89600.00, Z=9735.00) ext: V(X=12799.00, Y=12799.00, Z=2426.00) LogNavigation: Verbose: URecastNavMeshDataChunk::AttachTiles Attaching to NavMesh - RecastNavMesh_UAID_5811228C0404AA9802-Default LogNavigation: Verbose: Attached 169 tiles to NavMesh - RecastNavMesh_UAID_5811228C0404AA9802-Default LogUObjectHash: Compacting FUObjectHashTables data took 1.60ms

when i play in pie

so my conclusion is that they are not getting serialized

[muted fossil]

Maybe some kind of dump question regarding StateTree, but how do I avoid that there is an infinite loop happening from root to idle and idle to root?
While ActorState.TaskState is idle it will loop infinitly instead of staying in the idle state and thus call STT_Idle all the time even though the idle animation is already looping.
Any tip?
(in 5.5 this was not a problem btw, guess they changed sth in 5.6 in that regard?)

[muted fossil]

is there a better way to solve this except introducing an empty state?

[slow bobcat]

If the condition to enter idle fails, doesn't it evaluate the walk and the carry?

If all of them fail their conditions, you are pretty much trapped in root indeed

[muted fossil]

default needs to be at the end of the state list

yeah

[slow bobcat]

If you want you can add default for clarity (we have the same, we call it dummy state), but gives you nothing aside understanding the tree better when reading it. It's fine if you are just in root until something happens

[muted fossil]

yeah I think I work with this default / dummy state for now

moved the default state to the very end of the state list, otherwise it will never evaluate the other states if default state is at the start obviously 😄

[muted fossil]

Found a better way. Since 5.6 this is toggable and then doesnt go further if completed

[slow bobcat]

But that has nothing to do with conditions to enter a state or not. That's for when the transition should trigger: sometimes you want the first task to finish to trigger the transition, others you want all of them to finish before the transition triggers

[muted fossil]

yeah for more complex stiuations that might be not sufficient

[reef trellis]

Hello everyone. I'm experimenting with State Trees for a project of mine. And have been stuck with an issue that I can't seem to fix, probably due to a lack of understanding. I've read a lot of the pinned post hoping to find a solution, but haven't been able to find something that explains what I'm experiencing.

This is my state tree. I am using it in conjunction with smart objects' gameplay behaivours and it runs on a AI character. The character is able to roam, look for work, start working, re-evaluate and change just fine.

Roam - [OnCompleted] -> Next State :
Move to Random Location - 1st Task: Get Random Location (does not affect); 2nd Task MoveTo (affects completion); [OnCompleted] -> Next State
Wait

Look for work - 1st task: Search for smart objects (does no affect completion); 2nd Task: claim smart object (does not affect completion); 3rd Task: Move to smart object location (affects completion); [OnSucceded] -> Work; [OnFailed] -> Root

Start working - uses the smart object (runs the gameplay behavior)
Reevaluate - search for smart objects with higher priority than the current one
Change work - stops current and switches to new smart object in case it finds one.

When performing one of the jobs (which is to carry an object from point A to point B), while waiting for the Event to trigger in the "Work" state (which should only trigger when move to + playing an animation on arrival finishes), it jumps right into Roam State for some reason, before the event is even triggered, all while playing the deliver animation.

Does anyone have any idea of what I could be doing wrong?

[chilly nebula]

This article created a custom EQS task that inherited the EQS task as an example, I know the article has been updated recently to point that out.
We will most likely update our tasks in the near future, but we are working out a few more things regarding async task patterns first.

[chilly nebula]

the "interface" to linked assets should be tree parameters, so you expose and bind your tasks to those and then where you link/run parallell your sub tree you bind to inputs there.

[chilly nebula]

hard to say without seeing more of your setup, but my guess is you have tasks on that state with signal complete and that is triggering a completion. In the tasks section there are control whether a state completes when ALL or ANY tasks signal complete, and you can also toggle this per task.

[quartz shadow]

Thanks for your answer ! Yep that's what I did but the issue is that meta spécifiers are not propagated. Let's say I create a ST asset and want to expose a param of a task as an input of my tree i.e tree parameters. When promoting the variable from my task to a tree parameters all meta spécifiers (Like ClampMin/ClampMax etc..) are not propagated.

So if I write a state tree task which take a float parameters with ClampMin = 0 for example I could technically set a négative value after promoting this variable to a tree parameter , same issue with gameplay tag filtering etc....

[final prism]

Is there a way to share data from sub state trees to the main tree? Outside of sending events

[chilly nebula]

That might be a bit confusing, but its because they are different parameters. You bind your task input to a tree parameter and you cannot place those input param restrictions on the tree parameter, since another task can also bind to it (and that one can have its own meta specifiers).

[chilly nebula]

You should be able to use property references, but they are a bit of an advanced topic. We are working on some new data binding patterns that might help.
What is your use case? You want sub tree to update a value that is then read by the main tree? or changes on a parent tree value to propagate to linked one?

[final prism]

Yeah update a value on the main tree. This works for behaviour trees because the blackboard is a decoupled component.
An example could be a subtree that ticks down on a timer used in the main tree, but there's many more.
Subtree here is meant as the version that takes its own asset, you can't pass properties to those I think

[chilly nebula]

You can bind to the sub tree parameters yeah, but I dont think Ive tested declareing those as property references (and that works for C++ only)

[quartz shadow]

I see indeed but is there any plan to expose task, Condition, etc param as a tree parameters (Or another category that is tweakable from parent ST) so that it can only be bind to this task and we can propagate spécifiers ?

Because without it, It break a bit the "data validation" assumption we can do using spécifiers.

Or is there a better way of achieving what I want to do i.e create small reusable trees that have input param with restricted value ?

Again thanks for taking time to answer this is really appreciated

[quiet iron]

Theoretically, with the same AI behavior implemented with Behavior tree and State tree. Do they have any differences in performance when scaling?

[hallow compass]

ST better afaik

You can have more controlled ticking, or completely event based state tree

BTs needs to tick to run a state

[hallow compass]

is it normal that the move to fails here ?
im allowing partial path, but it fails to project the location (gizmo here) on the navmesh

[final prism]

Ah right it does create property bindings to the sub-tree parameters - So you're saying if I made those property references I could also write to them from the subtree? I'll have to give that a try..

[merry blade]

hey folks how does one simply make tick work in state trees, my global task and my regular tasks refuse to tick for whatever reason

[chilly nebula]

Yes, quite a lot but it depends on usage (like what your tasks do).
Matrix demo had tens of thousands of State Trees running, and recently we introduced a tick free patterns that scales even further. I have not tested though how far it could scale or how it compares directly to a similar behavior in BT.

[chilly nebula]

You can create a task that clamps the value in the subtree? Now using a propertyref in the task to mutate the parameter, and in 5.6 you'll have a new thing where a task output parameter can bind to/reference a tree parameter to do the same.

[chilly nebula]

https://dev.epicgames.com/community/learning/tutorials/z3km/unreal-engine-tickless-statetree-changes

[final prism]

so I'm actually unsure how this would be possible. I'm not sure I can add the property reference in the property bag of the tree via cpp, if I add it via the state tree editor I don't have the option to bind it to anything

[chilly nebula]

Did you use FStateTreePropertyRef ?

[final prism]

The only thing available in editor is the FBlueprintStateTreePropertyRef. If I make the property a normal property, then bind to it via a property ref in the subtree, im not sure how id propagate to the parent tree, as the subtree state type doesnt allow me to add any tasks

unless I can create a property ref that binds to a property ref?

[chilly nebula]

As I mentioned, this is difficult with BPs 🙂 it needs to be done in C++. Thats part of the reason why we are working on something better for this pattern

But no, I have not tested this with subtrees

[final prism]

I'm happy to write this in cpp, I don't think it's possible to create the state tree global property bag in cpp though

And that's the only thing shared between the sub state tree and the state tree calling it

[chilly nebula]

good point, my head was still in tasks which is where I've used this

that wouldnt work anyway, since you cant reference the parent tree's properties. The parent tree would have to be the one pushing a reference into the parameters

@final prism actually, it might sorta work? I could setup the bindings at least. Just remember to set the underlying type of the property in order to reference it

This is on 5.7 fyi

[hallow compass]

i like the new ST icon

[final prism]

oh hold up, Im on 5.6, but this may actually work, I think i cant bind to anything because my test tree doesnt have any other properties..

[chilly nebula]

I was able to bind the ref to a tree parameter, state parameters etc. I didnt check if the value goes through or can be changed. Im interested to hear if it will work

[final prism]

I would still need to be able to create a task with a property ref that binds to this property ref, not sure that would work

[chilly nebula]

I was able to create a state parameter in the subtree that bound to it, so cant see why not

[final prism]

yeah Ill give it a try.. exciting

[final prism]

Does the state tree debugger have a way of inspecting data state? It seems to only show running states

[quiet iron]

Guess I have to start learning the state tree

[misty wharf]

The one very neat thing about ST's is that you can run multiple on the same actor at the same time if I remember correctly

Eg. you can have a character be controlled by an ST, and it can interact with a smart object, which in turn has logic via a state tree to control the character for the interaction

I've not tried it though so maybe someone can confirm how well it works :) I know in theory the scenario I described is possible with behavior trees too, but in practice it's pretty clunky because it will always restart the character's original BT from zero after using the SO's BT

Probably gonna look at these more complex behaviors with ST's and stuff more when I get around to overhauling the AI logic in my game which is something I'll probably have to do at some point. It's all BT's now and has a few rough edges lol

[chilly nebula]

This above is correct, and State Trees are used for more things than just AI behaviors! Quest systems, dialogs, encounters and even whole minigames have been implemented using State Trees.

[final prism]

Looks like that actually worked! nice

[final prism]

There's also a really odd bug with binding to float parameter types, it seems theres multiple types specified as float? (it seems to be double or single precision but im not actually sure, its very odd)

[misty wharf]

Yeah due to legacy reasons floats are messy in BP's now. It used to be float, but it's now double, and if you have a float, it's... also a float in BP's. The distinction in BP's iirc is that if you hover over it, then it will tell you it's a float or a float (double)

[final prism]

The issue is the property reference type in state trees, I can only specify "float" i think, but it seems arbitrary, sometimes it can bind me to the double precision, sometimes to the single precision type, but never both

Why is the new default for float double precision anyway? Why not have float and double types?

Wait do i have to bind to double in the property reference then? This is so confusing haha

[misty wharf]

I guess they thought it would be simpler to just have a single float type in blueprints so it wouln't confuse people who aren't necessarily programmers

even though it's actually two float types and it's actually confusing now.. :D

[final prism]

Just Unreal things I guess

[final prism]

It seems making it a double property reference works

not a huge fan of having to use doubles in code, but at least it works..

[harsh storm]

Because UE uses large world coordinates now by default. So doubles all around. What zomg said is correct. They did think just making the BP float as a double would be less confusing. It hasn't worked out that way in practice. BP has some really great decisions and some really big head scratchers 😅

[final prism]

I think if they removed float in favor of double, it would've been at least a bit less confusing, since double in code would actually be the same as double in blueprint

[reef trellis]

Why would a state go to the Root, if there are no transitions in the state or any child state to move there?

Is this the default behaviour for when all child states of another state under Root complete?

[warped tiger]

I've tried this and it works. One small caveat: if you run the state tree on the smart object actor then you won't be able to re-use tasks that assume that the context actor is the character interacting with the smart object

[west carbon]

So one of the built in decorators is called Gameplay Tag Condition and claims that it can check if the specified actor has a gameplay tag or not. But I don't see any option to add gameplay tags to an actor in the first place, only tags you can add to them are FNames

[misty wharf]

Your actor must implement IGameplayTagAssetInterface for those to work

[analog mural]

I'm working on enemy hit reaction logic but I'm running into an issue. When an enemy gets hit, it plays a hit montage. During this montage I want to prevent the enemy from walking. I have a bIsStunned BB variable which I set to true when the enemy takes damage and false after the hit montage is done playing. In the BT I have two IsNotStunned decorators to prevent any movement or actions (e.g. attacking) while the enemy is stunned.

However, I noticed that the enemy still moves forward sometimes. In the video after the 2nd time it gets knocked back, it moves forward while the montage is still playing. Why is that?

The following code gets called OnTakeDamage and in the hit montage blend out delegate

void AMWEnemyAIController::UpdateBlackboardIsStunned(bool bNewIsStunned)
{
    Blackboard->SetValueAsBool("bIsStunned", bNewIsStunned);
}

[silent hamlet]

hm im not sure why the enemy is still moving but perhaps you could consider stopping the brain (or pausing it) while you play the stun reaction. then reactivate it

[analog mural]

I feel like that's overdoing it a bit though. There will be things that need to run in the background, even when stunned. I only want to prevent it from moving and certain actions

SetMovementMode won't work either as that prevents knockback which I implemented using LaunchCharacter

[silent hamlet]

hm from your video it seems when it is stunned none of your bt branches are executed (which makes sense given your decorators). is maybe the move logic somewhere else?

[analog mural]

Yes, the first time it gets knocked back it doesn't execute any of the branches, which makes sense. But for some reason the 2nd time it gets knocked back, it still runs the MoveTo with the IsInCombat decorator in the middle of the bt

The IsNotStunned decorator details panel can be seen in the screenshot. Don't think it's related to that

[silent hamlet]

Ah yes I see. Hm it seems the isstunned flag is not set then. Did you check it remains applied during the stun?

Perhaps there is some flickering (apply/remove) behavior somewhere

Nice combat feel btw

[analog mural]

It looks like if I hit the enemy while it is still in the hit state, it stays on false and doesn't set the variable to true again. Thought it was related to the bt but I guess not. Thanks, atleast I know where to look now

[west carbon]

I don't see that interface in the list of interfaces I can add

[misty wharf]

It can only be implemented in C++. If you don't use C++, you will need to add a gameplay tag container variable to your BP, and a custom BT decorator that can read it (f.ex. via your own BP-implementable interface)

[analog mural]

Turns out I didn't fix it. It seems to be a race-condition. Very rarely,BTT_MeleeAttack in my BT gets triggered a split second before the enemy enters a hit state (plays hit montage, sets bIsStunned to true, etc.). This is probably because the enemy gets close enough to the player to attack, but then gets hit right after.
On the 3rd hit in the video, you can see MeleeAttack getting triggered in the output log, OnTakeDamage being called, and montage ending immediately after getting hit. Any tips on how to fix this?

OnTakeDamage removes the attack state, invoking UMWAttackState::OnExit

void AMWEnemyCharacter::OnTakeDamage(AActor* Weapon, float InDamage, EDamageTypes DamageType, UMWAttackConfig* AttackConfig)
{
    CharacterStateManager->RemoveState(EMWCharacterStateType::ATTACK);
    CharacterStateManager->AddState(EMWCharacterStateType::HIT);
    EnemyController->UpdateBlackboardIsStunned(true);
    UE_LOG(LogTemp, Error, TEXT("Take damage"));
}

void AMWEnemyCharacter::PlayHitAnimation()
{
    AnimInstance->Montage_Play(HitReaction);
    HitBlendingOut.BindLambda([this](UAnimMontage* Montage, bool bInterrupted)
        {
            EnemyController->UpdateBlackboardIsStunned(false);
            CharacterStateManager->RemoveState(EMWCharacterStateType::HIT);
            UE_LOG(LogTemp, Error, TEXT("Hit animation ended"));
        });
    AnimInstance->Montage_SetBlendingOutDelegate(HitBlendingOut, HitReaction);
}

The attack state below is entered at the start of MeleeAttack

void UMWAttackState::OnEnter(TObjectPtr<ACharacter> InOwningCharacter)
{
    EnemyCharacter->bIsAttacking = true;
    EnemyCharacter->EnemyController->UpdateBlackboardIsStunned(true);
}

void UMWAttackState::OnExit()
{
    EnemyCharacter->AnimInstance->StopAllMontages(0.1f);
    EnemyCharacter->EnemyController->UpdateBlackboardIsStunned(false);
    EnemyCharacter->bIsAttacking = false;
}

[stable plover]

hello all im trying to bake the navigation mesh using the build paths, I followed the docs and did everything they did, its visible and exists in editor, with nav chunks etc, but in PIE its not there, everything exists in the world but the data isnt getting serialized for the new scene if anyone can help that would be appreciated thanks

these were the docs https://dev.epicgames.com/documentation/en-us/unreal-engine/world-partitioned-navigation-mesh?application_version=5.3

im using 5.4.4

has anyone experienced this issue?

are there other ways to bake the nav?

[slow bobcat]

There were important changes for nav in 5.5 and 5.6. Any chance you can upgrade?

The other thing would be to check the tiles loading, see if their nav chunks contain information. If they do, check where the nav chunks are added/removed from the recast nav data, debug that in code to see if anything is missing

[rancid wing]

anyone ever have a sub bt not working?

as you can see is going into the roam sub tree

but in there

*crickets

[tiny sinew]

try change them place
up/down left/right

and see what going on

[rancid wing]

yeah I did

still not working

its the sub tree

idk

well actually

[rancid wing]

I also changed the sub tree and still has the same effect

[tiny sinew]

Why you place a wait with nothing?

Show the full tree

[rancid wing]

sorry Its gonna be a moment

I now have an issue with cpath volume throwing wrong end location

for now I put that aside

[hallow compass]

seems like the state tree property picker doesnt take into account public BlueprintGetters for wraper protected properties

[hallow compass]

i have "Should State Change on Reselect" set to false and i still have Enter/Exit called when my task is reselected from parent

[robust veldt]

FStateTreeExecutionContext& Context
...
Context.ForEachEvent([&](const FStateTreeSharedEvent& Event) -> EStateTreeLoopEvents

I wonder why this lamda is invalid.

[slow bobcat]

Remove the return type. Not sure but I think that's not a valid sintax in a for each statement

The for each will run the lambda for each element, makes little sense to return things from the lambda

I might be wrong, but it's worth trying

[robust veldt]

FStateTreeEventQueue require lamda with return value.

    /**
     * Iterates over all events.
     * @param Function a lambda which takes const FStateTreeSharedEvent& Event, and returns EStateTreeLoopEvents.
     */
    template<typename TFunc>
    void ForEachEvent(TFunc&& Function) const
    {
        for (TArray<FStateTreeSharedEvent>::TConstIterator It(SharedEvents); It; ++It)
        {
            const EStateTreeLoopEvents Result = Function(*It);
            if (Result == EStateTreeLoopEvents::Break)
            {
                break;
            }
            if (Result == EStateTreeLoopEvents::Consume)
            {
                It.RemoveCurrent();
            }
        }
    }

[slow bobcat]

Yeah but doesn't it mean to call "return yourState" in the lambda without the -> to indicate the return type? Like you would do with any of the Algo:: functions

[robust veldt]

My implement is like:

Context.ForEachEvent([&](const FStateTreeSharedEvent& Event) -> EStateTreeLoopEvents
{
  if (...)
    return EStateTreeLoopEvents::Consume;
});

I remove -> EStateTreeLoopEvents it is still same error.

Maybe this syntax has issue.

        template<typename TFunc>
    typename TEnableIf<TIsInvocable<TFunc, FStateTreeSharedEvent>::Value, void>::Type ForEachEvent(TFunc&& Function) const
    {
        if (!EventQueue)
        {
            return;
        }
        EventQueue->template ForEachEvent(Function);
    }

[slow bobcat]

Ah ummm

Wonder if you can pass a lambda as a universal reference. My c++ is bit limited when it comes to types in templates

It's seems you can indeed.

What's the error you get?

It's weird that I can't find any usages for that foreach either

[robust veldt]

StateTreeExecutionContext.h(227): error C2187: syntax error: 'template' was unexpected here
StateTreeExecutionContext.h(227): note: the template instantiation context (the oldest one first) is

[robust veldt]

I found the issue, Seem the guys write this code never test it. LOL.

[slow bobcat]

Ah... It might not be an issue with your lambda but with the template definition

[slow bobcat]

Yep....

[robust veldt]

Remove the template and cost will make it work.

[slow bobcat]

Yeah but then it will not be a templated function resolved at compilation time

[robust veldt]

All template is resolve at compile time. If I remember correctly.

[slow bobcat]

Yehq what I mean is that "if you remove the template" from it.

[slow bobcat]

You mean from the Epic's function definition no?

[slow bobcat]

Here

[robust veldt]

I mean that remove template keyword in this line:

EventQueue->template ForEachEvent(Function);

and the cost in FStateTreeEventQueue::ForEachEvent

[slow bobcat]

The you will be calling the implementation of the function that is not a template (I guess it has one?), or even worse, think that "<xxxx" is a "smaller than" and not compile

[slow bobcat]

Where is that? I can't find it in my code (5.6.1)

[robust veldt]

Oh it is fixed in 5.6. My project still at 5.5

[slow bobcat]

Ah ok ok

Maybe you can bring over the fix from 5.6

[final prism]

I believe that function is broken. Unreal doesn't use it anywhere, so the linker errors only pop up when you start using it.

[severe kayak]

Hey, is there a way to debug and see the cone or the radius of how far the AI that has AIPerception component has?
I know the gameplaydebugger has nice utility for it but it shows spheres on my player character when I was heard or seen but I want to be able to see the NPCs line of sight etc before they notice me.
Thanks!

[hallow compass]

press 4 (not on the numpad)

you will see a circle (the radius of the sight) and some lines (the cone)

[severe kayak]

no way! it was so big I havent noticed the radius! Thanks!

[hallow compass]

yeah default is 3000

so 30 meters

[final prism]

Has anyone added turning angle to Unreals pathfinding before? Something like reaching a target location while facing in a specific direction requires a modified pathfinder I would think

[celest python]

I did only on 2D surfaces. 3D ones require a math skill beyond my comprehension

[slow bobcat]

What I do is to get the path returned by the engine and then, in the path following component, check if the path has an angle ahead (easily check comparing the dot product between the current and the next path segment). Of that happens, I calculate a Catmull Rom spline curve between the two paths and make the AI follow said curve points until the curve is done

[celest python]

What if its mostly vertical line though

catmull rom acts weird when it happens

[slow bobcat]

we have stairs and inclined ramps etc. All good there. Shouldn't be a problem if it's a very stepped angle
Shitty drawing of the curve in blue, but on the left a very common case for us, on the right what you are describing. It will just work

The main issues for me with Catmull are:

• even though it has several params to tweak, we can only use a very specific set of values to actually get a decent curve, otherwise you end up with things resembling a zig-zag
• I had to add some math to evenly distribute the points alongside the curve because the original formula (the one in the wikipedia article) doesn't have even distribution. It concentrates points at the beginning and sparse them at the end

[celest python]

appearently I used CubicCRSplineInterp

https://gist.github.com/intaxwashere/fda3607bf3dda02059b14e30f49279d9

beware, old and dumb code

and it runs on tick instead of manipulating the path for some reason

[quiet iron]

Hey guys, quick question. I was watching some Unreal talks about state tree and I see that the task UI is different from my UI. Is this something that they custom in their engine?

[slow bobcat]

UI changed quite a lot between 5.5 and 5.6. Talk is probably in older engine versions

[quiet iron]

Ah yeah, makes sense, I watch another clip in 5.3, they have the same UI as above.

[quiet iron]

Also can I ask what's the different between Global Task and Task that is assigned in the Root state? I'm assuming that putting task in root state act like global task but it will eventually call Enter and Exit anytime that the tree is re-evaluate?

[rancid wing]

anyone here ever use the cpathfiniding plugin?
the 3d pathfinding plugin basically

[elfin smelt]

It is great !
Do you had some resources to do that ? I think I will try something like that in my free time one day
the mathematical part doesn't really scare me but deep dive into the recast implementation in Unreal do 😅

[celest python]

#gameplay-ai message
also see Bruno's message

you dont need to dive into recast for this

just manipulate the direction of the movement like I do or manipulate the path array

[final prism]

Do dynamic sub-state trees exist yet?

[reef trellis]

In state trees, if there is a condition or event that should stop execution of any state and transition directly to a specific state, what is the best way to go about it?
Should there be a event transition on every single state (child and parent states)?
Or should the tree logic be forcibly restarted after the condition is set (perhaps outside of state tree logic)?

[static raven]

I'm trying to replace the default Hearing sense implementation, the "config" class have an "TSubClass Implementation" member, with the config specifier. But if I set my custom implementation class as default in DefaultGame.ini the property is unset in editor instead of using my class.

This happened to anyone before?

[slow bobcat]

Global task will evaluate every tick and you should never call finish taks unless you want to terminate your tree.
A root task will run when the root state evaluates, which is not neccesarelly every frame. Also a task in the root state will influence in when/how should that state transition to another

[slow bobcat]

Yes. They are called linked trees and since 5.6 you can change them at runtime I believe. The documentation and code should have some info about it (I don't use linked trees so kit super sure how it works)

[quiet iron]

Is it not supported in 5.5?

[slow bobcat]

I do this adding the transition event to the root node to the jump from there to where I need

[final prism]

I believe they are broken in 5.5, though I haven't found a way to change them dynamically, will investigate further..

[slow bobcat]

Linked yes, not so sure about runtime changes

[slow bobcat]

Why not change this directly in the Perception component instead? That's where I do it and it works. Not saying your way is wrong but not sure if there are hidden implications

[quiet iron]

But I can still setup all of the linked tree before start logic, so it should be fine right?

[static raven]

hmmm I'm not sure what do you mean by change it in PerceptionComponent. I want to set the default class to my own custom one to avoid possible human errors due to designers setting the wrong class when prototyping new classes

[final prism]

Ah nice thres a property for it

[slow bobcat]

If you look in the Perception component in your controller, you will see you have the senses there and you can set which config class to use

Aaah sorry I completely missed the second image!

Ooook yeah. Not sure why that shows up as none

Maybe you can set a data breakpoint in the implementation?

I see you have 2 examples, one with your own class and another using the default from the engine. Does the defualt work?

[static raven]

yup, that one works, that is in BaseGame.ini

[slow bobcat]

Only I could think of are:

• the path to your class is not set right and soemthing is missing
• or maybe your class implementation is not right? You can always copy paste the engine file in a different folder and see if you can make that one work as if it was yours

[static raven]

yeah I think it could be related to the path. Is there a way to actually run any kind of validation over the implementation of a class and prevent it to be a default value of a config variable? because I can manually set it in the editor panel, it just doesn't work when I try to set it as a default class 🤔

anyways, thanks for your time, I just wanted to see if this happened before to somebody else, for what I know could even be a bug in the engine 🤷🏻‍♂️

[obsidian delta]

is it possible to increase nav accuracy to encompass this better?

[hallow compass]

see nav radius on recast mesh actor

[misty wharf]

That's the wrong place. If you modify it there you'll lose the changes as soon as you regenerate navigation. You need to modify it in project settings.

[misty wharf]

This looks pretty good to me, you want it to be more rounded or what?

[obsidian delta]

it worked

actually changing it in project settings didnt do much

idk something with nav settings is fucked i feel like

some settings work some dont

[obsidian delta]

well i have pretty small agents

[somber kernel]

hi, In the state tree, I want to place a task that gradually slows down the AI's speed. However, the task blueprint does not support the 'tick' event, so what is the solution?

[misty wharf]

Yeah make sure the agent radius is set correctly. Having a gap between the edge of the mesh and your geometry is normal, the edge of the navmesh is calculated from the origin of your agent. Also, if you modified it in project settings and it did nothing, you may need to delete the navmesh actor and regenerate. Modifying the actor directly may cause you to lose your settings when regenerating the navmesh

[quiet iron]

What is the best way to retrieve the AIController inside a custom task in C++?

I read that if you're using StateTreeAIComponent (using the AISchema), it would guarantee the AIController for me. But I don't quite sure how to get that. I can retrieve the schema but I see no function to retreive the AIController

[misty wharf]

Typically you would add a variable for it into your instancedata, eg. UPROPERTY(EditAnywhere, Category=Context) TObjectPtr<AAIController> Controller

If you're using the AI schema, this should automatically connect it to the Controller context value from the schema (assuming it was named Controller, I forget)

[quiet iron]

what's Category=Context do? It will auto connect? I only know 3 categories (Input, Output, Parameter)

[misty wharf]

Context will automatically connect things to the schema's context values

As long as their names and types match

[quiet iron]

That's explain

Also can I ask, usually how can you bind to dynamic delegate in a custom task? Because the task is a struct so it won't have the UFUNCTION macro to be able to bind.

[misty wharf]

This is one way to do it: https://zomgmoz.tv/unreal/State-Tree/StateTree-InstanceData#class-based-instancedata

I don't know if there's a better way in 5.6 though, I'm a bit out of date on the details of that :)

If you're unsure about using this, you should check the builtin tasks if any of them have dynamic delegate bindings and see how they do it

[quiet iron]

I'm using 5.5 btw. And yeah, the first thing when I changed from struct to class UObject is the engine crashes from the start, saying something corrupted. I have to delete my state tree and it won't crash again.

[robust veldt]

Delegate is supported in 5.6.

https://portal.productboard.com/epicgames/1-unreal-engine-public-roadmap/c/2028-statetree-delegates-and-asynchronous-task-support

[misty wharf]

I wonder if a finnish person took that screenshot. Parameters.Lurppa sounds finnish even though Lurppa doesn't mean anything lol

oh actually Lurppa is Droopy's finnish name, so I'm like 99% a finnish person took that screenshot

[final musk]

hi folks, i'm having problems with a character going around a higher cost nav area, it slows down and kinda jitters while taking a corner. there are no collisions or anything. Im issuing the move command using the MoveTo task node in behaviour tree
I also tried creating a kind of straight "hallway" with a thin normal cost line surrounded with high cost areas on both sides, depending on where i click in the space at the end of that hallway, for example if i click on the left side, the character walks through normally but if i click on the right side, it goes supper slow and jitters like crazy.
im not really sure whats even the cause or what to change, if anyone has any idea that would be a huge help!

[exotic merlin]

Helloooo, hope you're doing well !

Was wondering if there is a State Tree repository with examples, best practices etc.
If there is something using GAS as well, that would be nice!

Like Tranek's, Narxim's, Laura's etc ... (we have them pinned in #gameplay-ability-system )

I went over the pins, but didn't see to have one.

Thanks for your notes @misty wharf 🙂

[obsidian delta]

tbh state tree is kinda

do whatever you want

theres no forced workflow

epic docs are sufficient I think

[exotic merlin]

Alright!
We'll just migrate our behavior tasks to state tree ones and that's it

[obsidian delta]

just think of it like graph nodes

[harsh storm]

No. We don't need our hand held

https://tenor.com/bqwWF.gif

[exotic merlin]

https://tenor.com/view/i-recognize-you-simon-cowell-britain's-got-talent-i-remember-you-i-know-you-gif-680188867924399285

[harsh storm]

Yeah, this is one of my 4 main channels.

Everything else is secondary

[exotic merlin]

https://tenor.com/view/i-wanna-run-away-lil-nas-x-sun-goes-down-song-saturday-night-live-im-trying-to-get-out-of-here-gif-21910428

[harsh storm]

https://tenor.com/bFDtK.gif

[exotic merlin]

How do you guys manage to handle this guy ? He's the bane of our existence on Gameplay.

[celest python]

By shitposting more aggressively

[hallow compass]

i mean whats good with ST is that it has a decent amount of abstraction

so you can make all your GAS async stuff just fine

[exotic merlin]

Yeah, I am really not worried about it

it's more about having premade tasks (just avoid to recreate them myself if someone already did), and having some examples

[hallow compass]

didnt yet used GAS with STs so i havent made ones in my utils plugin

[exotic merlin]

I'll definitely have to create those different use cases (I wanna attack, and I need the actor close enough, and my attack not in CD etc)

etc etc etc

You could argue that there is 100 different ways of doing it (like in a BT too).

If, if, there was a repo with examples showing one "good practice" one 😉

Thanks for the feedback!

[slim egret]

hi, is it possible that when creating StateTreeAIComponent in the AIController cpp, even when exposed to BPs, its properties can't be modified?
I tried every possible specifier on the UPROPERTY but no matter what i can't modify its internal state tree (or other properties) when selecting it in the components window in my BP AIC that derives from this cpp AIC

[exotic merlin]

https://dev.epicgames.com/community/learning/tutorials/z3km/unreal-engine-tickless-statetree-changes

Might want to pin that ... New stuff in 5.6

[hallow compass]

got passed a few time here already

im adding this to my notes tho

[slow bobcat]

That's odd. Should be enough to have it as public or protected with an EditAnywhere specifier

[slim egret]

Yes, frustrating. I tried also regenerating project files and restarting the editor after compilation and still it didn't work.
When adding a state Tree Component via cpp it did let me modify the properties in the BP, but then I noticed that I need state tree Ai component because my state tree using the Ai schema..
Changing the component to state tree Ai broke that properties visibility/edit ability

But from what I understand from you, it should show correct? What I encounter isn't the intended behavior?

[slow bobcat]

I will check once I'm back to the pc in a bit, but I think I just do that

[slow bobcat]

As I remembered, nothing special.

• Our AIStateTreeComponent inherits from AIStateTreeComponent and nothing special to it

class OUR_API UOurAIStateTreeComponent : public UStateTreeAIComponent```

- we add it to the AI Controller as any other component (it's a protected member in our AI Controller)
```/** component handling the interactions of this actor with State Trees*/
UPROPERTY(EditAnywhere, BlueprintReadOnly, meta = (AllowPrivateAccess = "true"))
class UOutAIStateTreeComponent* StateTreeComponent;```

- the only "weird" thing we have is the meta AllowPrivateAccess because of a bug we had were we couldn't bind to it in the ST's
```Adding "meta = (AllowPrivateAccess = "true")" to AITargetSystemComponent and StateTreeComponent in the BaseAiController so that state trees can bind to it```

[hot narwhal]

hey, is there a way to make a replicated state tree task or do i need to trigger the RPC in the ai character directly. I just want to have a task that play an anim montage but it is not replicated

[misty wharf]

Not really a mp expert, but don't AI Controllers only run on the server, so a state tree can't really replicate as a result?

[hot narwhal]

Yeah that's what i was realising

I will call an rpc on the caracter directly

[misty wharf]

Yeah that sounds reasonable. #multiplayer is probably the best place for mp-related questions since they're a bit more specialized I guess :)

[hot narwhal]

Yeah, my question was more state tree related but yeah i might ask there next time

[final musk]

after searching the entire internet and not finding anything, i just started testing random shit. and im pretty sure its the CrowdMovementComponent that's causing the issue, not sure how to fix it yet

[misty wharf]

Pretty hard to tell what's up with that. Have you looked at the Visual Logger? I think you should be able to see the path that gets generated using it, which should at least tell you whether the path itself is wrong or something else. F.ex. if it was the crowd avoidance, the path should still be correct since afaik avoidance is applied separately from pathing

[final musk]

ye the pathing is fine, havent used visual logger before ill check it out.

the right one is with crowd movement disabled, and thats the behaviour i want, ill check in the code to see what are my options with it, since i still want the avoidance for other characters

[misty wharf]

The first one isn't playing, but if the crowd avoidance is really affecting it, the only thing that comes to mind is that maybe your character is trying to avoid itself, since it doesn't look like there's any other crowd agents in there 🤔

But having it avoid itself sounds kinda strange. Either way, maybe it's worth trying to unregister it from the crowd system so it doesn't count as something to avoid

Maybe it'll put you one step closer to finding out what's happening

Also, If you have RVO avoidance and crowd avoidance both enabled, iirc they may conflict

[quiet iron]

Hey guys, what factor to determine a stimulus is newly added among sensed stimuli? I can't rely on age because if the AI perceives 2 entities, the age variables will always be zero (until it cannot perceive anymore). So I cannot figured out which one is old or new.

[misty wharf]

Can you not use the event which gives you the stimulus that triggered it? On target perception updated or whatever it was called

[wise sluice]

It should be OnTargetPerceptionUpdated from the PerceptionComponent

[quartz shadow]

Hey ! Is there a way for a ST Event to live across multiple frames ? From what I understand the EventQueue is stored in the InstancedData of the component and is getting reset on tick (When we trigger transition) so if I am correct that is no possible right ?

Context: Let's say I have a running State that perform a GA and under certain condition I trigger a ST Event. I would like to wait for my current task to finish go back to root and check if I received some ST event to trigger a specific behavior.

Ideally if it could have a life time that could be useful as well. But maybe I should just create a dedicated system that handle this kind of logic and only have a dedicated enter conditions.

[exotic merlin]

I haven't used it, but my research yesterday showed this:

StateTree Delegates Give Another Way To Limit Work Done Until Data Is Ready

In addition to the changes for scheduling ticks, StateTree has added delegate listeners that can be used by tasks and transitions to respond to a certain bound delegate dispatchers. The transition is intuitively called On Delegate and allows for binding which delegate dispatcher to use much like binding any other property in StateTree.
https://dev.epicgames.com/community/learning/tutorials/z3km/unreal-engine-tickless-statetree-changes

[quartz shadow]

Thanks ! Yeah right now I am on 5.5 not 5.6 yet so I will definitely give this a try asap. I guess I could solve my problem with this but I am not sure this is what I exacly want (Will see ^^)

[runic flare]

I want to use State Tree LinkedAsset states to run trees from my main tree to perform specific behaviours (like Patrol). However it looks like LinkedAsset state type requires that the chosen tree asset have the same schema as the "parent" tree. I was hoping it would accept a subclass of the parent tree's schema but it doesn't look like that's the case ☹️ . Does anyone know if there's a way around this? Maybe this functionality will be added in a future version?

[slow bobcat]

Is there a restrictive check in code like "IsA<>"?

[white pike]

Hello guys, I made an enemy AI which follows the player if the player is in the sight of the enem. If there's Nothing in the enemy sight, then it moves randomly.

But when my enemy detects the player it just stops Moving, and i Don't understand why. Random movement works well. does anyone had a similar problem in the past ?

This is the BP of the enemy chasing action

[slow bobcat]

Chrkc visual logger, see why the movement stops

[hallow compass]

how are you setting the black board key "target Actor"

also, you can use a print/breakpoint in your BT task to check if the AIMoveTo is called

[white pike]

Thank you!

[white pike]

Thank you! I tested and my AIMoveTo is called "every frame" when the player gets inside the sight of the enemy, so it wierder.

[hallow compass]

did you checked its actually called

[white pike]

then it does this

[hallow compass]

well here you are close so it makes sense

what is printed if you are far enough from the AI but seen ?

[white pike]

It does the same 🙁

but actually i think i have a clue !

I will try to solve it Thank you!

My Target Actor is set here, the code is inside this blueprint

But I think I forgot to assign this BP on my Enemy, maybe that's why it's not working

[white pike]

ouch I was wrong I cancel what I said

My target actor is well assigned, it's a problem of movement

[hallow compass]

weird indeed

[white pike]

Yup... I think I will restart from new to see where it starts bugging. But thanks for your help!

[hallow compass]

there is not point restarting

see step by step what is working

[slow bobcat]

Check in visual logger. Chances are you are not projecting the location and its not on nav, canceling the movement...

[hallow compass]

it does seem like its the movemenet faults, you need to debug with the visual logger if the target is correct

[slow bobcat]

Do you have nav mesh? That could be the other reason

[hallow compass]

this wont call sucess ?

thats the weird part for me

[slow bobcat]

Nope. Calls failure because can't find the end goal location on the nav

If you can't move it fails

[hallow compass]

well on the screens you can see "Success" being called

[slow bobcat]

Ah wait, I missed that image. One sec

Ah... Ummmmm

Weird

Visual logger for the win then

[white pike]

I somehow managed to start visual logger, but I didn't understand what i can do with it (first time using it)

[white pike]

Yes I have navMesh and it works well for random point movement

Stops moving only when it comes to chasing the player

[slow bobcat]

right click there -> dissable all but this

and yoiu should see entries there just for the nav. One of them should draw the path calculated. Let's see it

[white pike]

I think the ai think it's already in the goal wherever the player is

[slow bobcat]

could it be you are setting the AI to reach itself?

Maybe TargetActor bbk is set to the wrong actor?

if yoiu look at your BB while running the game and the movement stops, what is the value in your BBK?

well you have it here too

is that the player?

one thing you can do is to grab the BBK actor location just before passing it to the IAMoveTo node and draw debug a spwhere. See where is it

[white pike]

OMG you're right the debug cube is on the enemy and not on the player

[white pike]

So is this code in the enemy is reaching the wrong destination ?

[white pike]

this one here

[slow bobcat]

add a debug draw to the actor location before setting it to the BBK in both cases (perceived and not perceived)

I wounder if the AI is detecting itself or you are passing a nullptr because the BBK is not set and, somehow the move to picks the current location as a safe option

you might be setting the bbk in two places or something like that too

[white pike]

Really wierd. The code seems to detect well the player's location, yet the destination assigned to the enemy is its own location

the blue cube means that target actor bbk is well assigned and green cube means that target actor bbk wasn't well assigned

[quiet iron]

in state tree, can I communicate between tasks with non blueprintable data types?

[harsh storm]

In C++ - sure. But it can only happen in C++.

[exotic merlin]

The bringer of bad news

[reef trellis]

Having a hard time understanding state tree flow and transitions. I've read unreal documentation and some of the links posted hear. The behavior I'm expecting is not happening, for sure I'm missing something.

I have this State Tree I've been working on (posted here before, it's a little bit different now), and it is getting stuck on root.

After it enters "Scared/ Act scared" state, if "IsScared" bool changes to false again, it does not go in "Search for Work" state. Based on my tree and it's transitions, shouldn't it though? I can provide more info if needed

[reef trellis]

Had to force an event transition for it to work. But still confused why it simply does not continue normal execution with the boolean change.

[quiet iron]

Yeah but how can I access it in C++. From other task?

[slow bobcat]

Can you show your transition? Is it an "on tick" transition?

[reef trellis]

Which one specifically?

[slow bobcat]

The failing one from act scared to root

[reef trellis]

They are all default, except for "Look for Work"

[harsh storm]

You'd have to get a reference to the task somehow. Not a route I'd go personally. But where there is a will, there is a way.

[quiet iron]

I guess I just gonna wrap the variable into a blueprint wrapper

[slow bobcat]

That's the thing. If you don't set a transition that checks a value in tick, or reacts to an event, the state will only transition when a task finishes (or the last, depends on your setup). If you want the state to be interrupted, you either have an event that will force another state (ie: an event transition in root) or you have a transition checkign on tick some value in your state (ie: if X bool is false, go to root)

In your example, you don't have any transitions in the image. You are waiting for the tasks to finish and then go to root

Nothing there handles interruption

[reef trellis]

But once it goes back to root, isn't it supposed to try and enter each child state again? At least that was my understanding, since the tree is not resolved or concluded.

[reef trellis]

I guess this is what I ended up doing, in the root state. Added a transition based on an event that also checks the boolean as a condition.

[quiet iron]

How can other task or transition listen for the value changed to trigger the transition logic? For example: I would like other tasks / transitions to listen the PerceptionInfo output changes, then it could react and transition to the correct state. Like Observer in behavior tree.

[reef trellis]

I guess like Bruno was explaining to me before, you could add an "On Tick" transition that checks for that specific value?

[quiet iron]

I'm looking for an event driven method though. Like blackboard, they have a delegate to signal when a parameter is changed.

[reef trellis]

you can use "On Event" as well, I'm using it right now, and it can also check a condition before transitioning.

[harsh storm]

Make the task a root task or a global task.
Have the task listen to w/e from the perception component.
Broadcast one of the new delegates for state trees added in 5.6
Have other tasks bind to that broadcast

https://dev.epicgames.com/community/learning/tutorials/z3km/unreal-engine-tickless-statetree-changes#statetreedelegatesgiveanotherwaytolimitworkdoneuntildataisready

The "observer" concept from BTs did not transfer over to STs. You have to build it yourself.

[reef trellis]

For Perception I'm using a state tree evaluator and on "OnTreeStart" and on "OnTreeEnd" I am binding and unbinding custom events that handle perception stuff like sight and hearing. not sure if that is what you were looking for

Note: haven't fully tested it though

[slow bobcat]

Yeah. Root should re-evaluate but, if it doesn't, something is wrong in the tree. Use the state tree debugger and see what happens, why is it stuck in root

[reef trellis]

Need to learn a little bit better how to use it. It was just showing being stuck on root. Thank you for the replies, I appreciate the help

[slow bobcat]

Feel free to post what the st debugger shows. Maybe we can help unraveling the mystery

[gentle widget]

Anyone know why my AI would move toward the player if I load directly into the level within the editor, but the AI will not move if you load into the level from another level (i.e. pressing start on the main menu)

State tree for reference

Additional details, the AI moving toward the player is the intended behavior. Basically, it works correctly if I play the level directly from the editor, but doesn't work if I start the game from the main menu.

[runic flare]

IsA would be good bc it'd allow for child schemas -- I'm guessing there's a schema class == somewhere. I could hack around it for sure but I'm wondering if A. I'm thinking about the intended use of schemas incorrectly, or B. child schemas will be supported for subtrees in the future. It just seems like a very restrictive design making it harder to make my trees modular. I really don't want all my trees to need to share a giant megascehma when my e.g. Patrol subtree should be the only tree that cares about my PatrolComponent context object.

[gentle widget]

Still trying to work out this issue, I initially thought it might have something to do with load order but that does not seem to be the issue. I've tried delaying activation of the state tree, spawning enemies in based on player distance to the spawn point, etc. Does not seem to fix the issue. Visual logger is not being very useful in this case, only error it's showing me is that the NPC failed to navigate to the player. It's a weird issue, everything works just fine if I play the level directly from PIE.

[crude mirage]

Does anyone have like a base enemy ai blueprint that isn’t the one in the combat template? I’m specifically running into errors with the state tree tasks because I stupidly tried to just duplicate all the template ai ones so I need to start from scratch but all the tutorials are either outdated or hard to understand.

[somber kernel]

The collision of the box in the level is what makes the AI ​​work or not

[gentle widget]

Thanks I fixed it, rebuilt the navigation paths in the level (from the build menu) and it worked.

[quiet iron]

Have anyone tried this in 5.5? Does this node work?

[spiral robin]

Hi everyone! Duplicating question from #cpp here, as it's probably more appropriate place to ask about state trees.
I want to use StateTrees for my quest system, as it sounds ideal for my purposes. However, I'm not sure about a few things:

1. Are there currently any ways to run state tree other than from UStateTreeComponent (or reimplementing state tree running logic based on this component's logic myself)? It would be ideal If I could run them from game instance subsystem, as I'll have multiple maps, and quests must persist on map changes. Of course I can just make some QuestRunnerActor, spawn it on map open and add a component for each active quest, so it's not that big issue, however the approach with subsystem would be cleaner.
2. Can I persist internal state of running state tree? I have a simple save/load system, and I want be able to store/restore quest state transparently, without some ugly workarounds. If I understand correctly, the component stores running state in FStateTreeInstanceData InstanceData, and the documentation says I can serialize it. So, would it be enough to restore state tree's state on game load?

[reef trellis]

Reading posts in unreal forum, looks like it hasn't been working for quite a few iterations of UE5 including 5.5. Haven't used it though. You could try using "SendStateTreeEvent" if what you want is to force a transition to a particular state, and add the transtion "OnEvent" in the root state or something.

Works in 5.6 apparently
https://forums.unrealengine.com/t/is-state-tree-request-transition-broken-in-5-4/1887469

[quiet iron]

Yeah I've read that

[somber kernel]

Hi, How do I interrupt a state tree to run a specific task/function, and then resume execution from the point of interruption?

[harsh storm]

Anyone else have issues with state tree event transitions? I have an event transition on my root, sending the tree event, and then nothing happens 🤔

[hallow compass]

can you see the request in the ST debugger ?

[harsh storm]

Yeah

Just nuffin' happens

[drifting ginkgo]

I was having some issues on 5.6 with some cherry picks from main. Was going to retest in latest.
Setup was just a simple transition in a parent leaf to transition such as this ^^

[hallow compass]

you should see something like "Root -> Transitions -> Your event"

in the tree

[harsh storm]

Yeah - that ain't happenin

[hallow compass]

even if it failed

[harsh storm]

Nope - nuffin'

What I show, is all I see.

[hallow compass]

how are you sending the event

[hallow compass]

[drifting ginkgo]

Is that an "oh the answer is obvious" or an "oh no that should work" XD

[reef trellis]

I'm still experimenting with state trees, however I was having some issues not with an on root event transition, but some other state, and it was mostly due to being stuck somewhere else, so that transition would never be triggered. Could that be happening to you? (Just brainstorming here)

[hallow compass]

its a "fuck idk"

[drifting ginkgo]

I had some problems with certain child states and event dispatchers in a similar way. Might be best to try a global task just to see if it can work without fear of the state getting wiped.

[somber kernel]

[harsh storm]

@slow bobcat Did you ever figure out a good workflow with dealing with State Tree tasks in C++ and binding to a delegate from an object in your instance data?

[robust veldt]

I hear that UE5.6 support delegate and async task. I wonder how it is implement with struct.

[harsh storm]

Yeah, I use those in BP. But I'm talking about when you are binding to a delegate from another object. It needs a UFUNCTION, but structs can't have those.

Right now, I'm just kind of copying what the EQS task does and relying on a lambda.

[robust veldt]

So you should you UObject for task instance data.

[harsh storm]

I don't think you can. The instance data needs to be a struct. If you're saying just do like BTs did with the AI Tasks, then sure. But I'm curious to know if there is a different approach others have found.

[robust veldt]

IIRC, Instance data can be UObject. I see it some where in source code. Let me find it..

[harsh storm]

Please do. Because the GetInstanceDataType function returns a UStruct*

[robust veldt]

StateTreeTask_PlayContextualAnim.h

UCLASS()
class UStateTreeTask_PlayContextualAnim_InstanceData : public UObject
{
    GENERATED_BODY()

};

UClass is extend from UStruct so it will fit.

[harsh storm]

Oh yeah - you're right. I forgot about that.

[harsh storm]

Is that in the smart object plugin or something?

I can't find it and I am guessing because Rider doesn't have it indexed

[robust veldt]

Engine\Plugins\Runtime\GameplayInteractions\Source\GameplayInteractionsModule\Private\StateTree\StateTreeTask_PlayContextualAnim.h

Fullpath here.

[slow bobcat]

Would need to check. Can't remember now if I ended up using data as a UObject or not. How urgent is this?

[harsh storm]

Not very

[final prism]

yeah this is how we do it as well. Make instancedata a uobject and put the delegate bindings in there

[minor pawn]

Hey,

I'm trying to build "rooms" at runtime.
My logic works fine, so I want to go to the next stept, to make my characters look alive.
I want to add at runtime NavModifierVolumes. Which works so far.
What does not work is, that the NavigationSystemV1 will acknowledge the spawned obstacles on runtime and block navigation.
I've built a tiny map to see, if my code works generally and it does.
I've built a tiny square of NavModifiers and assigned the correct AreaClass.

My Characters won't move out of the square and I can see in purple, that there is a modifier.
On my other map, where i build everything randomally, I cannot see the modifiers and the Characters move out of their area.
Since I can see the nav mesh, I guess, I setup the runtime gernation of my navMesh correctly to Dynamic.

What am I missing? Is adding Modifiers at runtime not supposed to work?

[real raven]

could invokers not fix that issue for you?

[minor pawn]

I guess you mean NavigationInvokers?
My Characters have NavigationInvokerComponents on them.
I'm not sure If I know, what I should do "to fix it" 😕

[real raven]

there is settings you need to enable in your config to make use of the invoker system

[minor pawn]

Generate around invoker
That's active. But I still need a navMesh for that, I figured

[real raven]

this although with the standard pathfinding system will cause for some pathfinding to fail, if the target is not close to the navmesh

[real raven]

yea true, you need a nav mesh volume

there has to be a way to get rid of that, but i have not yet found a solution to this problem

[minor pawn]

I also managed to break my setup, so I no longer see a color for my AreaClass 😮‍💨

Ok, reset helped. I can see my NavAreas on static gernerated content, but not when spawned at runtime. Also the AI behaves like that

[final prism]

I'm not quite sure I understand what you're trying to achieve - If you want to spawn actors that block navmesh you can use the dynamic obstacle component or the navmodifier component to have them block navmesh properly. Just make sure the navmesh is dynamic or navmodifier only

[minor pawn]

I am spawning NavModifier at runtime, but I can neither see them nor they seem to be used on navigation.
When they are statically added to the level, the affect navigation.
Navigation is set to dynamic. I already said it AND I can see the nav mesh in my level when playing in editor

Ok, the missing piece was, that you can not use a NavModifierVolume at runtime, because it does not have a NavModifierComponent.
I need to create a duplicate of this actor and add the Component myself.
When spawning the duplicated Actor, I can see the Modifiers at runtime and my Characters behave as expected

[minor pawn]

My next problem is, that I don't know how to update a FStateTreePropertyRef with the WeakExecutionContext.

I'm adding a listener to PerceptionComponent->OnTargetPerceptionUpdated But I don't get a reference to the Property's value:

const FStateTreeStrongExecutionContext StrongContext = GetWeakExecutionContext().MakeStrongExecutionContext();
FGameplayTag* GameplayTag = Motivation.GetPtrFromStrongExecutionContext<FGameplayTag>(StrongContext);

GameplayTag is nullptr

[woven aurora]

Hi everyone, I have this problem where I have 2 states for my enemy AI, passive and combat, and basically I have a key just to switch between them for me to see how they switch, but whenever I press it (it is in passive by default), it says on the bahaviour tree that it transitions into the combat sequence, but the actual actions of the enemy tell me otherwise as he's still doing patrol, it's been a day already that I've been trying to figure this out, anyone?

[minor pawn]

I am wondering, why I can't bind my evaluators output variables in an enter condition. I can see the variable on the left marked as output, but can't bind it. I'm using the blueprint base classes in c++

[somber kernel]

Guys, does anyone understand the golden rule for smoothly transitioning between statuses without problems?

It seems confusing between parent and child statuses and the Selection Behavior options

[harsh storm]

Update - I changed nothing and now it works. Thank you for coming to my Ted Talk.

CC: @hallow compass

[robust veldt]

I have task that have bool input variable, Is there any way to make condition as input. For example, I want to check if length of array output of other task and assign to that bool.
Similar with MVVM plugin, that allow convert type.

[slow bobcat]

Confirmed: I use UObject for all our instanced data classes. Most of our use cases are listening to tags being added/removed, and for that we cache the Ability System Component and subscribe to its tag delegates

[harsh storm]

And if that function is supposed to be what finishes the task? How do you handle that?

Because it is the dynamic delegate that allows you to pass additional arbitrary args. But you need access to the ST execution context to finish a task.

So some delegates just won't let it happen to my knowledge.

[slow bobcat]

Change a bool in the Instanced Data and then, in your Task tick, read the bool to call finish task. Good old tick based taks. But I believe that since tasks now support delegates you can do this in a better way?

• on enter subscribe to delegate with your instanced data function
• On Callback (within your instanced data), change bool bTaskDone to true
• on tick check a bool "bTaskDone". If true call FinishTask
• On End un-subscribe from the delegate

[final prism]

I believe there was a thing to bind to functions instead of variables, but I have never used it

[robust veldt]

It is property function, Look like I have to write one.

[somber kernel]

Transitions in the State Tree are very complicated , has anyone managed to understand them??

[rancid marten]

I have been stuck on this problem for over a week.
https://streamable.com/gz5div

Switched the NPCs in the Arena Shooter variant over to a struct health system and hit scanning. In the Move to Roam Location state, their movement is doing tiny steps to the target location. And ONLY when the game is running smooth, they act normal when frames chop.

[harsh storm]

They're fairly straight forward. What issue do you actually have?

[reef trellis]

I feel you. Still trying to understand them myself. Although I've been managing to make it work

[somber kernel]

@harsh storm @reef trellis
For example here, state A does not transition to state B , what’s the reason?

[harsh storm]

Show the transitions and the tasks

[slow bobcat]

Didn't debug text never call finish?

[somber kernel]

state A does not transition to state B

[reef trellis]

what happens if you remove the transition from root, and put it select children in order?

[final prism]

Definitely an odd setup. I'm not sure if putting tasks on the root is a good idea. Additionally, try enter won't select children, I'm not sure you can transition directly to children of states with the "try enter" type. Try setting all of them to "try select children in order" and see what happens

[harsh storm]

Having tasks on the root is fine. Same with transitions on the root.

And yes, the debug text task never actually "finishes". However, the Delay task should.

And what does the debugger say?

Are you even in state a?

[somber kernel]

@reef trellis @final prism
"Try Select Children in Order" runs all the children at once.
This means I can’t take advantage of transitions or set a delay for one of the states

[final prism]

It selects the first leaf node in the state tree. This is supposed to be a hfsm, you're supposed to select multiple hierarchies of states

Try select children in order will not run all children at once, it will select the first child in order, then run the next child once that one completes

[reef trellis]

should't it select one after the other?

[final prism]

It's important to understand the hierarchy at play. If it selects A as a child of Root, it will have selected both root and child, as they are a hierarchy, thus it will run the tasks of the entire hierarchy at once, i.e. both root and a

[reef trellis]

will it do the same once it runs "b" state? so both tasks from root and "b"?

[harsh storm]

Yes

Tasks on root should be seen as always running if the tree is active pretty much

[reef trellis]

does that mean it will also run enter conditions and transitions on root or any parent state? every time a new child is selected I mean

[final prism]

I believe there is an option for that on the state tree enter condition

[somber kernel]

@final prism
I understood a little, but I’m still confused.
For example, in the image , why can’t I transition from the root directly to state B?
The system is forcing me to go through state A !

[slow bobcat]

Yeah that's what I mean. I think debug text task never calls finish task internally, so your state will forever stuck. Try making your own state tree task to print something and then call finish task

[harsh storm]

The Delay task does and they have the task completion requirement set to "any".

[harsh storm]

Because it works top down. Set an enter condition to prevent from going into A

[slow bobcat]

Ah ok ok. Missed that part

[reef trellis]

I think it's because you are setting "Try Select Children in Order" so it will always go through A. In that case you probably want "Try Transition"

[somber kernel]

@slow bobcat @harsh storm @reef trellis
I’m starting to understand a little, whenever a parent state runs, it must activate all of its child states at the same time and without delay.
Each of those states has its own transitions that only work once its assigned task is completed.

Also, you cannot transition between sibling states if their parent is not in Try Select Children in Order mode.

[harsh storm]

That mode shouldn't have any dictation on how a state transitions.

The state itself has its transition rules.

[somber kernel]

But if the mode is set to Try Enter, it will behave like my first problem and the transition won’t work.
Then Moritz told me that I should use Try Select Children in Order to solve the problem

[harsh storm]

The state's transition logic will allow it to jump to another state. That is a huge reason behind state machines in general. The fact that you can go from state to state.

When a state is selected, it will try to evaluate other states to see if they can be entered as well

The child states of the state that is being selected that is

[somber kernel]

You’re right, it works even if it’s in Try Enter mode.
Hmm, from what I understand, Selection Behavior works in one way, and Transitions work in a different way. This is what causes confusion, and sometimes they end up working at the same time

[harsh storm]

I'd advise you to just leave the selection behavior on the default until you understand more about how the ST works in general.

Baby steps.

[minor pawn]

How can i enable that debug view?
I cannot find the keybinding. I have qwertz layout

[rancid marten]

select the NPC and push apostraphe during play

[minor pawn]

yea, that's not working for me, that's why I'm asking for the keybind option

[rancid marten]

You pushing the one next to the enter key? I was dumb and pushed the top left one when I was first told about it. Beyond that idk.

[minor pawn]

for me it's the # key

left from enter

Maybe you can check your editor settings and see if you can find the key bound to any option.
You should be able to search for the key

[rancid marten]

went through editor and project settings and couldn't find anything bound to the key

I only found out it was even a thing 2 days ago

[minor pawn]

shiiii...
I really need this, because I witness strange senses.
Feels like my vision is tilted to left

[rancid marten]

I'm not doing too much better. I can't even get NPCs to move to a target location unless the frames are chopping.

[reef trellis]

Posted in #blueprint , but forwarding to here since this is happening in a state tree task 😅
I'm at a loss

trying to stop "move to" tasks when entering a specific state.

[harsh storm]

Digging more into the C++ side of the state tree delegates. Looks like you still use a lamba anyway. So - not what I was hoping for.

In order to broadcast the delegates, you need the execution context.

[robust veldt]

Old World /InGameMission/UEDPIE_0_TestMissionLevel.TestMissionLevel not cleaned up by GC! Object BP_Enemy_EtherealGuardian_C /InGameMission/UEDPIE_0_TestMissionLevel.TestMissionLevel:PersistentLevel.BP_Enemy_EtherealGuardian_C_0 is being referenced by StateTree /InGameMission/Missions/ST_SpawnMobAndWaitForDeath.ST_SpawnMobAndWaitForDeath:
 (standalone)  StateTree /InGameMission/Missions/ST_SpawnMobAndWaitForDeath.ST_SpawnMobAndWaitForDeath
 -> StateTree /InGameMission/Missions/ST_SpawnMobAndWaitForDeath.ST_SpawnMobAndWaitForDeath::AddReferencedObjects((Garbage)  BP_Enemy_EtherealGuardian_C /InGameMission/UEDPIE_0_TestMissionLevel.TestMissionLevel:PersistentLevel.BP_Enemy_EtherealGuardian_C_0)
    ^ UE::ReferenceChainSearch::FReferenceInfoSearch::HandleObjectReference() [Source\Runtime\CoreUObject\Private\UObject\ReferenceChainSearch.cpp:1106]
    ^ UE::ReferenceChainSearch::TReferenceSearchBase<UE::ReferenceChainSearch::FReferenceInfoSearch>::FCollector<1>::HandleObjectReference() [Source\Runtime\CoreUObject\Private\UObject\ReferenceChainSearch.cpp:284]
    ^ FInstancedStructContainer::AddStructReferencedObjects() [Source\Runtime\CoreUObject\Private\StructUtils\InstancedStructContainer.cpp:442]
    ^ FStateTreeInstanceStorage::AddStructReferencedObjects() [Plugins\Runtime\StateTree\Source\StateTreeModule\Private\StateTreeInstanceData.cpp:257]
    ^ UStateTree::AddReferencedObjects() [Plugins\Runtime\StateTree\Source\StateTreeModule\Private\StateTree.cpp:404]
     ^ This reference is preventing the old World from being GC'd ^
  -> UObject* UObject::Outer = (Garbage)  Level /InGameMission/UEDPIE_0_TestMissionLevel.TestMissionLevel:PersistentLevel
   -> UObject* UObject::Outer = (Garbage)  World /InGameMission/UEDPIE_0_TestMissionLevel.TestMissionLevel

I am getting this error, when spawn actor from task and cache in UObject instance data. Any idea to resolve it, although I clear ref when exit task

[shadow furnace]

when exit task? is that early enough to be before the cleanup?

[robust veldt]

Yeah, I checked ExistState, it is called before the error.

[shadow furnace]

what does the UStateTree have for the PerThreadSharedInstanceData?

after you supposedly clean it out

hmm... it should nuke this in ResetLinked

[robust veldt]

I am not sure what it is.

[shadow furnace]

you can use your IDE to look at the UStateTree object and inspect its member properties

I am reading UStateTree::AddReferencedObjects

[robust veldt]

Look like instance data of property function keep the ref.

[shadow furnace]

then I guess you didn't clean it out

I'm not really sure what callback to use here

[robust veldt]

But property function does not have clean function.

[shadow furnace]

why would it need to?

just reset it

[robust veldt]

I mean there is no way to reset it.

[shadow furnace]

If there's a more statetree specific thing I am probably missing it

[shadow furnace]

edit: I am wrong, this is an editor-only function!

[robust veldt]

It is instance data of FStateTreePropertyFunctionCommonBase

Could be bug of StateTree as instance data of task does not have this issue.

[shadow furnace]

can you obtain it from UStateTree::GetSharedInstanceData?

yeah it's possible

I'm confused why exiting the state isn't cleaning this out

does one of them add instance data and just leave it around?

I've only really made one simple state tree thingy so I might be asking a weird question

[harsh storm]

@slow bobcat Okay, so I think I have somethin' here that you might even be able to use. No more need for tick based bool check if you don't want to. But you pretty much just BindUObject to your instance data. Then give a payload of FStateTreeWeakExecutionContext. Then the first thing you do in the bound method is make it a FStateTreeStrongExecutionContext.

Then from there you can call FinishTask. I haven't dug too much into the dangers of this approach yet though. But this is how I'm doing it with the lambda approach. And I kind of copied it from Epic's EQS task.

One drawback that I know is that payloads can't be references. And the struct is like 48 bytes.

You may not even need to make a strong execution context honestly. But overall, that's the gist. You will need it if you need access to the instance data though. So, probably more useful in a lambda than the instance data itself 😅

[robust veldt]

I workaround this by change to WeakObject instead.

Damn, ST variable does not support weak object.

[slow bobcat]

So basically capturing the execution context in the lambda?

[harsh storm]

Yeah. That's how the EQS task does it.

And you can use that same idea when it comes to binding to a uobject really

[slow bobcat]

But isn't that super dangerous? The execution context is created anew every tick of the tree. I don't get how is that supposed to work.

• the state tree tick creates an execution context
• the execution context tick is the thing actually ticking the tree (global tasks, events, transitions, conditions and tasks in that order if memory doesn't fail me)

If you cache that context, aren't you keeping a ref to a temporary object of undefine lifetime? Or do you capture it by value (a copy)? But if by value... The context will be incorrect next tick no? When a new context is created.
@chilly nebula i could really use your insights here if you have the time please.

[quiet iron]

I'm still confusing about the usage of State Tree for other purposes beside AI. I hear that State Tree can be used for Quest system or door / chest. But I don't understand in what cases it could help. Seems like it's overkill for something with simple states.

[celest python]

its a tool that abstracts transitions in a way where you can do utility selection, prioritization or state machines and you get to have a nice way to marshall data to your tasks from the editor since it has property binding system

whats normally a complex sphagetti code in a scripting language becomes a structured graph system in ST

[quiet iron]

I know, that's what Unreal advertised, I can see it strength when implementing for AI. But for simple chest (with < 3 states) or quest system. I don't know how state tree can help. Especially quest system, I haven't seen a quest system implementation that using state tree or someone could enlighten me :D.

[celest python]

I'd do a chest in blueprints tbh

https://github.com/MothCocoon/FlowGraph/wiki/Getting-Started and quests in flowgraph

[slow bobcat]

In the end, as everything in unreal (or game dev for that matter), things are good ad they are useful for you. If your quests are simple, it's probably overkill. In my case we have very complicated quests, so we need a system like ST's. There's also things like controlling the world state in persisten environments where it's very useful

[quiet iron]

Can you give me an example of your complicated quest that need the use of ST? Just wondering.

[slow bobcat]

We don't use ST's because we have a plugin to control with nodes all the steps needed in a quest. Most of our quest have branching paths that will be decided based on how you play, outcomes of battles etc. Also based on player level. To that, add dialogs for npc' s and areas that will be accesible or not.

But, at hearth, our plugin is a state machine

We use that because the dialog support was very good

But ST's were second in line

[quiet iron]

Maybe this could be the use for ST for a quest system. Found it on the internet.

[chilly nebula]

We now have both weak and strong contexts, you should be able to store a pointer if you use those correctly.

[harsh storm]

Yeah, but if you look at the weak context, it explicity mentions that it can be used for async operations. So it is absolutely designed for having the context across frames. Which, confuses me as well, because the context isn't supposed to be done that way. But - idk. This is how the EQS task does it. So, they probably bolted it on later on down the road because they needed it as well.

[harsh storm]

I'm assuming the EQS task is using them correctly? Because that's where I copied this pattern from. And the make weak context just creates a struct. No pointer shenanigans there. Maybe under the hood?

So it is expected that you can pass the context around using Context.MakeWeakExecutionContext()

And that is the "safe" way of doing it?

[harsh storm]

Echoing what others have already said. The decision to use a ST would fall in line with the decision to use a full blown state machine really. You don't necessarily always reach for one, because sometimes, just operating off a simple bool or enum is sufficient enough.

That said - there is nothing preventing you from using it as the main driver of say, your main menu. Because those can be thought of as "states" as well. Or as the overall game's flow. Those can also be "states". It really depends on what type of game you're making.

I used it instead of Flow Graph, for example, in a small prototype my wife and I were doing. It served as the overall game manager.

[reef trellis]

Does anyone have an idea why this AsyncTask reference isn't valid during this state tree task execution, but showing as valid in the character tick?

[slow bobcat]

Can you briefly explain the differences? Maybe some example of usage too? That would be super helpful. Thanks Siggi

What I mean by differences is how is each type handled. Is one created when the tree starts and updated at some point? And the weak is the one passed around created on each tick?

[harsh storm]

Tick still creates a whole new context

I only caught this late last night, so I wasn't able to follow the code with the weak/strong context all that well

[slow bobcat]

I'm off today but I will take a look into this tomorrow. Haven't dig in state trees source much since we upgraded to 5.6 and I completely missed this

[harsh storm]

I plan on digging into it more tonight to try and reason about.

Or hopefully Siggi explains before 🙏

[chilly nebula]

Yes, if you are on recent code it should be the right pattenr.

[harsh storm]

Yeah, I'm on 5.6

[harsh storm]

I think it is just a bit confusing to us because the ExecutionContext explicitly says that it shouldn't exist across multiple frames.

But by doing this - we kind of are. Even if superficially.

[chilly nebula]

more changes coming in 5.7 and 5.8 regarding instance data storage, and yes I agree its not easy to understand or use atm

[harsh storm]

Yeah, I'm just going under the assumption that as long as you make a weak context, you can pass around the context. Then need to check if it is valid before using it elsewhere.

[signal island]

best way to transition from an A state to a B state, state tree events, state tree delegates? what you guys use?

[hallow compass]

it depends™

[slow bobcat]

but can you give us a high-level explanation on how the two are handled and what's the expected lifecycle?

[quartz shadow]

Is it also working on 5.5 ? I tried something but it did not work. I promoted a state tree property ref as a ST parameter and linked it to one of my parent tree but it seems not to have worked 🙁

I probably missed something. Context: I wanna modify parent params from a child tree. What I tried is to promote a boolean property ref as a parameter of my child ST and bind it in my parent tree to a parent parameter but this flow seems not to work I did not debug yet

[chilly nebula]

There are a few different types of instance data, most of which exists while the state is active. We've been adding new ones recently that allow you to persist data longer.

[slow bobcat]

is there a recommended place or way to figure out the following?

• which types are there (name of classes and/or structs)
• when are they created / destroyed
• recommended / intended usage of each type

Sounds like a crucial piece of information to know when dealing with state trees

[chilly nebula]

I'll see if our devrel guys can get something together, they have been working on tutorial/doc improvements for State Tree.

[slow bobcat]

it would be fantastic if we could get some good info about. I miss the days where we get graphs like The Actor Life Cycle.
TY

[chilly nebula]

That graph still gives me PTSD

[drifting ginkgo]

The biggest struggle atm I find is the state lifecycle eg.

In 5.6 event dispatchers in a parent state appear to get destroyed/cleanedup and as a user coming in you're never sure if it's intended or an inprogress piece.
But also massive kudos keen for STs future

But I guess alot of this stems from the flexibility of it atm, how do you wanna handle transitions, event/requestTransitions/delegates/enter conditions/etc

[chilly nebula]

event dispatchers in a parent state appear to get destroyed/cleanedup and as a user coming in you're never sure if it's intended or an inprogress piece.
Can you elaborate here a bit? When a state gets disabled then everything belonging to it will get destroyed. Is the confusion about when you re-enter the same state, when things get persisted vs destroy+re-created?

[chilly nebula]

yes, but we need to do a better job of it being intuitive to understand 🙂

[drifting ginkgo]

I've done a cherrypick so maybe my engines just cursed but take this setup.

I previously had a simple OnDamaged binding in the initialize thinking of it as a ParentState, but it appears when I then enter the combat state the binding or the state "instance" is gone somehow

I've used global tasks/evaluators to start offsetting this but I love the mental model of that parent leaf of the tree allowing bindings/setup in this way, although may just be a lack of understanding the workflows for now

[chilly nebula]

thats pretty much what makes hierarchical/nested state machines great. Its how it should work, so Im not sure whats broken in your case. Did you check the debugger to make sure you aren't temporarily exiting Initialize briefly and going back to root for example?

[drifting ginkgo]

Yeh I believe the debugger showed up clean, given the entire binding was prone to disappearing I thought it may be intentional sort of cleanup, like there's only allowed to be 1 truly 'active' state?

[chilly nebula]

To clarify, what type of binding are you referring to? A standard property binding where the Combat state refers to something in the Init state?

[drifting ginkgo]

I have a Vitality component that calls OnDamaged, then I'd bind to that in Initialize to try push into the combat state, but that binding that existed via a Task that was within Initialize would get cleared somehjow'

Sec I'll open it up and can grab any screenshots of use, but my expectation was those kind of things could exist in the tree instead of global task/eval side, again may just be a weird setup though. I am always using a dedicated server with Steam that might be playing odd

[chilly nebula]

A binding to an external component delegate? You need to use a weak context for that afaik. See what the EQS task is doing in latest version of that code

[drifting ginkgo]

There's the theoretical setup, whether Delegate or StateTreeEvent that TestEvent was either Unbinding or the State that contained it "Initialize" was getting cleared somehow.

I'm getting the ropes of C++ and starting to dig into the code though so will check that EQS example out

Have found workarounds but this is the underlying research problem I'm looking to understand as a user atm, if find anything as dig my way through C++ will share but imagine it's likely a lack of knowledge or attempting to use it incorrectly at this point. Appreciate the time and info though

[chilly nebula]

Let me know what you find, its possible we are missing something on the BP side. The inner workings of the component can sometimes be a bit opaque and we need to make it easier to "do the right thing" 🙂

[drifting ginkgo]

This feature is ace and y'all cooking please keep it going 🙏 idk it'd help but have just punched through 2 weeks of rigorous onboarding for it, would be happy to write up a pretty lightweight doc on that experience if it helps the feature/devs. In the lense of a mainly BP dev with basic C++ to look/tweak but not really fully ingest new systems like this easily.

[chilly nebula]

That would be amazing 🙂 I hope you are on UE 5.6?

[drifting ginkgo]

Keen for the days StateTree just outright supplements BPs akin to FlowGraph etc.

Yeh 5.6 with the cherrypick to dodge crashes for Setting StateTree refs at runtime, think that doesn't pollute much, was gonna poke 5.7 but might just hit up ue5-main since seems there's a bunch there. Added you as a friend to DM *** that through in couple days or so if that's fine 🫡

[hallow compass]

love seeing Mass and State Tree getting together

[chilly nebula]

They have always been friends 🙂 City Sample was made with Mass-based State Trees

[hallow compass]

all the "external data" thing is a bit opaque though, how are the structs are queried ?

[drifting ginkgo]

UAF as well if I'm not mistaken

[hallow compass]

because it feels like each task runs for a single entity, is it the mass manager or a mass fragment/processor who links an entity to a state tree ?

[chilly nebula]

UAF was used in the Witcher demo we did with CDPR, but didnt exist in the days of City Sample 🙂

[drifting ginkgo]

That demo made it look like the intended future to offset some of those thicc anim costs, keen to slap together UAF / Mass / ST and see where perf tanks out. Future's looking real sunny

[chilly nebula]

There are many Mass processors for state tree, init and also execution. Much of it can run in parallell too if you set a boolean in config

[chilly nebula]

Definitely 🙂 UAF is quite amazing

[drifting ginkgo]

Appreciate the jam Siggi gotta dip, will send through that ST new user exp doc in coupla days please let team know we lovin this stuff

[reef trellis]

Can someone explain to me how context actor in a state tree gets updated during execution?

[chilly nebula]

It wont be updated after being set, its the owner of the component. Im guessing you might be using the AI component on a controller? There is a bug we are working on now where possess/unpossess doesnt correctly set the actor in time when the tree logic starts.

[reef trellis]

Yes I'm using it on the controller. Shouldn't I be?

I'm having an issue where I can't stop a character from moving because the Async task pointer I'm setting outside of the state tree comes out as invalid in the state tree. And I can't think of a work around, or how to do this properly

Is there a way to force it to be updated?

[reef trellis]

Otherwise how to people update variables within the actor itself? What is the best approach to accomplish this?

[chilly nebula]

No thats fine, I was just saying we recently found an issue in that component where the state tree might start without the controller having possessed a pawn and everything being initialized correctly.

[chilly nebula]

Im not sure I follow. Is the actor pointer null or something else?

[reef trellis]

No. The actor is fine, but I'm setting a AITask variable with an async task pointer in a GameplayBehavior outside of the state tree, and when trying to access it in a state tree task it shows as invalid. Even though in the Actor tick it shows as valid

[harsh storm]

For what its worth - I have a few BP tasks like ArcChar and I haven't ran into the issues they're describing. My ST delegates have always broadcasted fine. 5.6.0

[reef trellis]

That's why I was asking if the context actor ever get's updated, during state tree execution

[chilly nebula]

its still the same actor, its just a reference. My guess is you have a timing issue, probably trying to read the value of an external variable before its fully set?

[reef trellis]

Unticking LockAiLogic fixed this 🫠
I'm not sure I understand why. Thank you for the help though

I thought it would at least let me terminate it at will

[drifting ginkgo]

Interesting, wonder are you using LinkedStateTrees? Or TryEnter and then transitions to move things? Wonder if my setup has something specific

And 5.6?

[harsh storm]

No. I don't use LSTs

And I hardly ever change the child selection logic

[drifting ginkgo]

Legend I'll try without those then to see if I can't find a consistent repo. Thanks

[rancid marten]

Has anyone had this problem before? It is specifically the Move to Roam Location state of the NPCs in the Arena Shooter template.
https://streamable.com/950cjd

[minor pawn]

What is the name for the keybinding to open the ai debugging?
https://dev.epicgames.com/documentation/en-us/unreal-engine/ai-debugging-in-unreal-engine?application_version=5.6
The docs say it's ', but nothing happens for me.
I want to change the keybinding. because i think there is a problem due to qwertz

[somber kernel]

Give more details, is this an AI template or did you create it?

[harsh storm]

#gameplay-ai message

Looks like "Project Settings -> Engine -> Gameplay Debugger -> Input -> Activation Key"

[rancid marten]

I altered the NPCs in the Arena Shooter template to use a struct for stats and hitscan weapons. Move to Roam Location only works if frames are chopping.

[somber kernel]

If the template was working correctly before the modification, create a new project template from this one, then copy the systems from the faulty project to the new one. This should resolve the issue. Also, during the transfer, you might discover which system was causing the problem

[rancid marten]

That's what I'm about to do but I was hoping someone has seen an NPC behave this way when just moving to random locations (only when frames aren't chopping).

[rancid marten]

Right now I think there's a problem with the EQS and I'm going to replace it with some other move to random location task

The NPCs definitely are finding a location to move to, but they're just glitching while doing it and only if my frames aren't chopping.

I don't understand EQSs and if it has anything to do with why they glitch-move to the target location that it finds

[somber kernel]

What exactly do you mean by "when frames aren't chopping"? Do you mean that the V-Sync feature is causing the problem?

[rancid marten]

The Move to Roam Location state works fine whenever frames chop. If you see the video, at one point I alt tab to demonstrate how they will move at normal speed for a brief moment

at :30 in

I suspect something is ticking constantly and refreshing. So that when the frames lower it is able to run that one logic normally.

But I have zero idea what would be doing that because Find Roam Location seems to work fine.

I have isolated the problem to the Searching state and its sub-states.

They seem to be stuck in the Search for Enemy state

[somber kernel]

Compare it to the template and see what it changed in the enemy search state

Perhaps you entered incomplete or incorrect variables, or perhaps you performed incorrect transitions

[drifting ginkgo]

Yo for this specific one I think I figured it out, When I changed the Initialize from
Try Enter >> Try Select Children In Order
The binding worked as expected. But I had it said to TryEnter which I guess although the debugger shows the full leaf chain, I guess it means I break the natural nesting flow?

I'm using TryEnter because I prefer rather than 'The ball rolls down the hill' I feel like I have more choice when the ball rolls to the next increment. Although it seems like that may be causing a few of my problems along the way?

[drifting ginkgo]

Interestingly you can add an 'Base' State with the SelectionBehaviour - Try Enter. (Changing Initialize to - TrySelectChildrenInOrder)
and get the behaviour I'd expect it rests in Base, but the Initialize bindings will receive as expected. But Initialize set to TryEnter and you lose those bindings events

[chilly nebula]

Wrong channel, please try #generative-ai

[spiral robin]

Hey guys, need help from StateTree gurus. Can anyone please tell, what am I doing wrong here? I use StateTrees for my quest system, and I'm planning to implement persistence via workaround using Selection Utility. I've implemented my custom selection utility and assigned it to each child state of Root. However, this doesn't work as expected, the first state is always selected no matter what. And judging by logs and debugger, my custom selection utility's GetScore method never even gets invoked. Why is that?

[spiral robin]

Also, I probably just don't understand correctly what Selection Utilities are. My previous understanding was that it's just the thing that provides you a score, that score will be multiplied by weight, and that result will be used as final Utility (that will be used for selecting child by highest utility). However now I've noticed this:

If you add another selection utility, it builds an expression, and it combines Utility Scores with boolean operators for some reason (despite the fact they only have GetScore that returns float). I also don't understand why is that

[spiral robin]

Also, here is my consideration implementation:

This log is never printed, I've also tried putting breakpoint in this node and in C++ code that invokes blueprint's ReceiveGetScore, the execution never stops there as well

[warped tiger]

Hi, I've noticed that if a state tree's state with children has bShouldStateChangeOnReselect = false and EnterState returns EStateTreeRunStatus::Failed, the next time FStateTreeExecutionContext::TickTriggerTransitionsInternal() will try to find a new state as part of its max 5 iterations, it won't call EnterState. Is this how it's expected to work? Shouldn't EnterState be called again if EnterState failed in the previous iteration? 🤔

[minor pawn]

I'm facing a problem, where the controller seems not to be rotating with the character.
I have a different project, where it works, but I can't spot the difference.
II have a perception on the controller, so I can check the cone to see, if it's rotating.
So the characters look always in the same direction, but the mesh is rotating correctly.
I checked every option in both (character and controller).
The only difference I can see is, that the one with the problem is created with c++ and the perception component is created in the constructor

[warped tiger]

In other words, should this tree print "Hello"?

FAlwaysFailTask::FAlwaysFailTask()
{
    bShouldStateChangeOnReselect = false;
}

EStateTreeRunStatus FAlwaysFailTask::EnterState(FStateTreeExecutionContext& Context, const FStateTreeTransitionResult& Transition) const
{
    return EStateTreeRunStatus::Failed;
}

EStateTreeRunStatus FPrintTask::Tick(FStateTreeExecutionContext& Context, const float DeltaTime) const
{
    UE_LOG(LogTemp, Log, TEXT("Hello"));
    return EStateTreeRunStatus::Running;
}

[heady silo]

I'm really struggling with StateTree. I've been trying to move some of the logic I got working under Behavior Trees to StateTrees. Basically, on spawn for the AI I have them assigned a patrol route with an instance-editable variable based on the actor that guides the patrol (BP_AIPatrol). The patrol route actor has a function in it called IncrementPatrolRoute that will add to a patrol index that varies depending on the length of the patrol, and when it reaches the end, swaps things around to get the actor to move back along the same route to the origin.

This works in BTs just fine (fig 1.), but I'm not sure how to get StateTree to pull the patrol route. I'm assuming I need to pass it via payload somehow for an entry condition? But when it comes to the StateTree task, there's no way to grab what I need from the controlled pawn, STT's EnterState only has transition data and none of the pins will reference the source actor. (fig 2.) How do I get the state in my state tree to fire if the actor has a patrol route assigned to it? (fig 3.)

I tried to add it to the global bindings but I'm not sure how to call that

(plus, that is only setup at runtime, and if I want functionality later where a guard will relieve another guard and take over their patrol route, that's not flexible enough)

[heady silo]

...well shit

[robust veldt]

What does it actually mean, currently I am using soft ref in instance data.

[robust veldt]

When task is finish, then get active again, does instance data get reset?

[drifting ginkgo]

I'm not sure I fully understand the problem but using the Output and Input 'Variable Categories' for variables isn't letting you do what you need?
I'd imagine you just Output the Path variable but think I'm not understanding the issue.

[quiet iron]

no, as I'm aware. I have to manually clear pointers when exit task to avoid keeping reference.

[quiet iron]

I did not know you can insert icons into your state & task xD

[robust veldt]

After some testing. If the next state is child state, instance data will not be clear yet. Otherwise it will be clear.

[quiet iron]

cleared meaning it will create a new instance data for that and the old one will be invalid?

Or they will stay at the same block of address

[robust veldt]

It is same block of address. I checked.

[quiet iron]

Thanks

[hallow compass]

That was in 5.4/5.5, dunno if still the case

[hallow compass]

In code or a param ?
I remember trying this with a global param, using the picker in editor

[drifting ginkgo]

Yeh the age old, can do BP comments as well although the StateTree debugger doesn't display them cleanly all the time it seems

[quiet iron]

how to do BP comments?

[drifting ginkgo]

Emojipedia or whatever website you can find emojis on and you'll find them work in alot of places just out of the box

[quiet iron]

Ahh, just thought you can do comments on the state tree itself.

[celest python]

not a bad idea tbh

[heady silo]

You can SORTA do this. Basically filling out the description of the tree leaf at right

[static cape]

Hello everyone, hope this is the right place to ask this. I am relatively new to unreal. My ultimate goal for a project I am working on (for uni) is to make an AI in unreal 5 that can play an RTS game. However as a shorter term goal i would happily take some pointers on where to start my research. Because quite frankly at the moment i cant even get an AI controller to control more than one BP independently from one another. Any and all advice appreciated, thank you in advance

[magic brook]

Anyone able to assist me with EQS troubleshooting? I an new and attempting to do some very basic EQS. I am just setting up a sphere trace from a location to determine "good cover locations" for an AI basically. If I set up the location to draw the trace from an actor in the world, an AI character, or the playerpawn the EQS test pawn returns all locations as good even when they are not. If I manually set the location to trace from by entering world coordinates in the blueprint instead of grabbing an actors location it also does not work as expected. IF I set the location as the EQS test pawn OR the player start it works as expected.
I think I am missing something of vital importance-
example of it working, useing the player start as the actor of class

example of not working

[slow bobcat]

I would start by debug drawing lines and spheres at your locations to visually see what is actually happening

[slow bobcat]

Also the functions you posted do not help since we can't know where/how do you use said values

[magic brook]

one sec- I will add more detail- I have also been working on trying to get it to work

[magic brook]

linking a video showing everything almost done

I recorded and uploaded to youtube? I recoreded at too high resolution and was hitting file size limit https://www.youtube.com/watch?v=giBaHyGUXMI

[magic brook]

what I was typeing in notepad "Ok I am working on this state tree
I have it going from root to the search part of it (all testing stuff ect)
it goes from search to test eqs
under test eqs it runs the task I set in there
this is the task
enter state- it then runs the hide part
thats just a custom event
it pulls the character running the tree, runs the query, then prints that it ran query (for troubleshooting)
on query finished- it gets the results (should only be on result)
the result is location, i also print location to view it for troubleshooting
move to location then finish task after delay (im not sure the delay is working, but thats a later problem)

this is the EQS that runs

checks path exists, filters?
then does sphere trace
thats just the default one atm - offset so it doesn't get stuck in floor when it looks
looking for the test dummy sorry
howed he get there rip
the dummy has the test in it- so it is traceing correctly i think
currently useing self location of scource of trace-

the state tree is being used by this actor

he should do trace from self, then go to another location that is not seen from self trace i think?
but he is not
observe

its running the query, and returning a location it selects, but does not appear to be actually filtering the results/doing the sphere trace.
"

[magic brook]

alright I discovered where the issue stemed from, the EQS trace scoring basically, I switched to filter and score and now it works as intended, it appears it was not Appling score values to the locations ; time to figure out how to get scoring to work

[rich wolf]

what's up with blackboard having struct type possible, but you cannot set or get them?

or am i missing something?

[rich wolf]

apparently it is so

[slow bobcat]

You can use it for custom bbk types (c++). Is what Vector uses

[heady silo]

would this be a realistic ask for a perception system in UE or is this going to be impossible?

PITS means Pie in the Sky, basically preemptively being ready to cut it if I don't have time to do it

[regal patio]

https://youtu.be/illvibK_ZmY?si=k5YRW5F32dPlbbQX

[harsh storm]

Not for this channel. This is game AI using the tools in Unreal. Like Behavior Trees and State Trees, EQS, Perception, etc... - not the investors wet dream AI. Try #generative-ai

[mighty geode]

Any good AI tutorial for UE with C++?

That teaches in deep so I can understand how to implement certain behaviour like investigate but when the other soldier arrive, the alerted soldier alert him and he also starts to investigate but instead he provide cover in case the investigating soldier find anything

[mighty geode]

Tutorial to learn the AI

[drifting ginkgo]

Anyone have suggestions for good C++ common tasks they have built for StateTree that aren't possible in BPs?
Looking to extend stateTree Tasks/Components as a starting point to learning more about it 🙏

[drifting ginkgo]

Went with exposing up some simple functions/types to better understand the relationship of elements, anyone learning this is going quite well for better understanding the relationship of things.

[warped tiger]

<@&213101288538374145>

[quiet iron]

I don't understand why the Context Data needs to be added with a fixed FGuid. So it's persistence when serialize and deserialize?

[heady silo]

I dunno if anyone's tried this, but I wanted to know if maybe someone knows.

Is it possible to add SmartObject slots to dynamic objects or moving actors like NPCs? I was thinking about doing this to allow for specific formations for NPCs (like a wedge or a snake formation). Or are SmartObjects baked into the level at runtime and cannot move/change?

I guess if it doesn't work that way I could always try and grab a followed actor's position and just do the vector math for where they should aim to follow, but I was hoping to make things easier on myself

[cyan sigil]

Smart objects should be spawnable. It would be absurd for that to be not the case.

[heady silo]

I don't mean spawnable, I mean attached as a component to a dynamic actor

[ornate onyx]

Anyone know why **** NAVMESH NEEDS TO BE REBUILT **** is showing up at runtime, when everything in project settings and elsewhere is set to "DYNAMIC"?

for whatever reason, moving the actor that has dynamic navmesh on it at runtime, forces a refresh and it works, but not before doing this... and I don't see anything in blueprint to recast navmesh manually

[slow bobcat]

So you nav mesh is set to dynamic and you get the nav mesh need to be rebuilt? That doesn't add up.

[ornate onyx]

what's weird is, the blueprint actor spawning at runtime (dynamic mesh actor) works fine, and I have an actor with a navigation invoker spawning on that spawned mesh, which works fine... (except the nav invoker part), but no navmesh

until... I move the spawned mesh actor at runtime, THEN the navmesh appears

also works if any options are toggled on the spawned mesh actor, which forces some random refresh on it, which navmesh sees apparently

@slow bobcat I guess the theory is that navmesh does an initial pass at runtime looking for already existing meshes, does not yet see the dynamic mesh created..., then via bp mesh is made, navmesh is not signaled that this happened

[slow bobcat]

Ummm that sounds odd...

[silent hamlet]

hello! I am writing my first State Tree Task in C++. I took an existing state from the Gameplay Interactions plugin basically as boilerplate. Now, the source compiles fine, but I cannot see the task in the editor.

The code is in a Loaded GameFeature Plugin, and I can correctly see other stuff from that plugin. Not sure if i forgot something obvious? Any help would be great!

(attached source because discord wants me to pay to write long messages) 😢

[slow bobcat]

Not sure that is how it works. I would imagine that an actor invokers would not create nav unless/until it needs to move

Or there's a path check

[ornate onyx]

&*(#%$

so it looks like this is going to be another one of those, spend 3 days trying 200 different things, and it ends up being a box you need to uncheck somewhere in a sea of 600 other options

pain

[silent hamlet]

ha, there is a mandatory other layer to the system. i dont understand it really but the thing is, I had to inherit from FStateTreeTaskCommonBase

[slow bobcat]

Explain please. I'm interested

[stark pulsar]

Hey everyone i am trying to make a metahuma chatbot but i am soo lost right now in how to do so i want the chatbot to response even questions like what is the current weather in New Orleans i wanna know if there is any step by step tutorial or something or maybe like what ton use how to do so so far what i know is audio2face nvidia ace and metahumansdk has something to do with the process maybe even TTS bot something but i am absolutely lost please help a friend out.

[heady silo]

that would go in #generative-ai most likely