gameplay-ai

[hallow compass]

from code, can i set the value of a parameter in the state tree ?

in my case its a global event so i cant use a output task parameter

[stable plover]

is anyone else experiencing an issue with state trees where it wipes data related to state tasks

or is it just me

[stable plover]

there is a way in 5.5, a guy called Three Ways talks about a potential approach that is cumbersome but does it https://www.youtube.com/watch?v=AQn8KtyMsSE around 16 min is where you will be interested

[hallow compass]

Thanks ill check that, but ill rather find a solution for 5.4

[fresh cove]

With pawn sensing there is like a way to see the range and size of ai senses. Green for sight, yellow for hearing, etc. is there a way to see it with ai perception?

[misty wharf]

afaik only via the gameplay debugger during actual gameplay

you probably could create a custom component visualizer for it if you wanted to though

[fresh cove]

How do I turn on the ai debugger?

[misty wharf]

See here https://zomgmoz.tv/unreal/Gameplay-Debugger

[fresh cove]

Nice shameless plug, lol

[stable plover]

yea I dont mess with parameters, there is just not that much flexibility for them, what i have is a actor component that acts like a blackboard but for state trees, have a global task that gets the component and boom you have all the data you need

[halcyon briar]

Are you talking about this setting?
We run automated stress tests. Can we pass this as an argument?

[slow bobcat]

you have all the flags and everything you might need here for your tests (link).
When I profile, I launch the build from rider using EzArgs (same as if you use VS Unreal plugin) with the arguments -statnamedevents -trace=cpu.

You can do exactly the same with your automated tests

https://dev.epicgames.com/documentation/en-us/unreal-engine/unreal-insights-reference-in-unreal-engine-5

[tough nexus]

Hello Somehow using AiPerception Sight and want to read the stimulus the tag it dont work iit gives me always none at the aicontroller

[misty wharf]

How are you changing the stimulus tag for sight?

because I'm pretty sure you can't set it and thus for sight it would always be empty

[stoic rose]

Hello everyone.

Is there a Begin Play-esque function for BT services? I'd like to just get some information once and then can deal with that in the rest of the service, but I'm not sure whether InitializeFromAsset or anything of that nature would be the 'Begin Play' in C++

[tough nexus]

i was justthinking that a normal actor tag can be used

[harsh storm]

Should be some function like, "Enter Node" or something like that.

Can't remember the exact name

[stoic rose]

OnSearchStart?

[harsh storm]

Yeah

Or is it the activation one?

Can you list out all of the ones it lets you override?

[stoic rose]

Well that depends on the parent I go to lmao

[harsh storm]

What do you mean? It is a service node

[stoic rose]

Right but there are functions you can override from AuxiliaryNode, Node, etc.

In UBTService, there's OnSearchStart

In UBTAuxiliaryNode there's OnBecomeRelevant

[harsh storm]

That's the one

[stoic rose]

OnBecomeRelevant?

[harsh storm]

Yeah

Then there is like OnCeaseRelevent to do cleanup

[stoic rose]

Is relevance determined by the execution flow of the BT?

[harsh storm]

Pretty much

[stoic rose]

I'd like something that ideally only fires once, i.e like a BeginPlay

[harsh storm]

So once the node that the service is on becomes active

[stoic rose]

Okay, so that'd be close, then. That's still event-based

[harsh storm]

The nodes don't have a BeginPlay like thing at all.

[stoic rose]

Gotcha, okay.

[harsh storm]

Service/Decorator/Task - none of 'em do

[stoic rose]

Thank you very much for your help!

[harsh storm]

If they do - it'd be news to me!

[misty wharf]

That doesn't work with it. Stimulus tag can be specified with some perception things, like if you use report noise event, but sight doesn't use it

[stoic rose]

If I want the node to act differently based on whether the AI is on the player's team or not, do I need to instance it or do anything in particular with the service outside of the logic to make the team determination?

[misty wharf]

(You could read the actor tags from the sensed actor though if you wanted to)

[harsh storm]

Hard to say. Depends on what you're trying to do specifically. But all instancing does is creates an instance of the node per BT. This allows you to store state on the class, instead of relying on the memory.

[stoic rose]

Is the BT instanced per AI controller?

I'm sure the answers to my questions are blindingly obvious but I rarely work with AI and am being tasked with this for work.

[harsh storm]

BT is shared, but the nodes themselves might be instanced or not.

So, each AI controller will use the same BT

But they each have their own Blackboard

And their own nodes if the node is instanced

If the node isn't instanced, then it is a shared node throughout.

[stoic rose]

Right, so then to get this node to execute differently based on team, say by setting a different tick time, I need to instance the node itself

[harsh storm]

It is one way, yes.

But you can also use AI Tasks.

Look at how Move To works

The slight advantage that AI Tasks have over instanced nodes is that it pretty much delays the object creation

So, if your tree never hits that branch, you'll never pay the cost for creating a new object

Instanced - you always pay that cost

This difference can matter in certain situations.

[stoic rose]

Yeah well it's a fire-rate service for a shooter MP game so I have no advantage to gain there

Okay, is making the node instanced just a matter of setting a flag in the constructor?

[harsh storm]

Yup

Here comes Bruno's acktually 🤓

[stoic rose]

Just that, I assume

[harsh storm]

Yeah

[stoic rose]

Fantastic, thanks so much

[slow bobcat]

The node instance is shared but each node in each tree for each controller has its own memory, which is where you would put its state control variables.
They way it works is this:

• when you run a tree, the behavior tree component gets an instance of that tree. Within the instance. There's an array called Known Instances, which will contain all the memories from all the nodes your AI Controller executes within the tree. There's another array for Instanced nodes.
  So: ai controller owns the behavior tree component. Bt comp owns a tree instance. Tree instance owns the Known instance executions of your nodes, which contains the memory of each.

So while the node instances of each node type are shared, they get their memory passed to their execution functions.

You can write your own memory struct to save whatever you need the node execution

[slow bobcat]

Yes

[slow bobcat]

Just trying to add to your answer.l XD

[harsh storm]

Yeah - this is the long form proper answer of what I said. This is what you would put in your college essay answer. Mine is what you tell someone at the water cooler at work.

[slow bobcat]

Oh man I miss water cooler conversations... Work from home took that from me

[stoic rose]

I am so much more confused than I was before

[harsh storm]

Which is why I answered it how I did the first time 😅

[stoic rose]

Which, honestly, is the thrilling part.

[deep bramble]

I want to have my AI disabled by default and only begin running the behavior tree and tick once the player reaches an overlap box collider which calls an event. I made an actor with the box collider to handle the overlap detection and made a variable reference to the AI and setup logic to enable tick and posses during the begin overlap. The problem is that when I use GetAIController it returns null and I can't possess the pawn. The AI pawn is level placed and has the correct AI controller assigned to it on the details panel. It also has auto possess and tick off (to save on performance). What am I doing wrong? Why is the AI controller null when it's assigned but not possed yet?

[slow bobcat]

But when you said the nodes are shared, it sounds like, if you don't make it Instanced, they share their values across the tree, which is not true. 2 decorators "do whatever" with int32 A as a parameter can have different values each while them being Not Instanced, as far as you out that value within each's memory

[harsh storm]

Yeah - I didn't expand on the node memory stuff because it confuses a lot of people.

[stoic rose]

I'm presuming that's related to the uint8*

[harsh storm]

Yeah

[slow bobcat]

Yep

[stoic rose]

Yeah okay so uh

W h a t

[slow bobcat]

You cast that uint8* to your memory type

[stoic rose]

Right so the memory is instanced and not the node, so the node being shared means what you do is pass in a pointer to the memory for a particular instance of that node

Cast the memory to your type, get and set your stuff?

[slow bobcat]

Then store stuff there. Otherwise, you will have to store values as properties of the node and, that's when you need to instance them so the properties have different values per each use of the node within a tree

[stoic rose]

R......r........ri...right.

Okay so rather than instancing the node, I can create my own 'memory'

[harsh storm]

Yes

[stoic rose]

Follow-up, why the ||fuck|| is it like that and how the ||fucking goddamn bullshit|| do I do that

[slow bobcat]

Yep. You need to create a struct, override couple functions in your node and off you go

[stoic rose]

Wow

[harsh storm]

Look at the delay task

[stoic rose]

That's insanely cool

[harsh storm]

It is probably the simplest example for the node memory stuff.

[stoic rose]

Okay so the struct I create is the 'memory' that will be passed through to some of those virtual functions

Ah, explains why it's a pointer and not a reference

[harsh storm]

References are pointers

[stoic rose]

:headdesk:

[harsh storm]

🙃

[stoic rose]

I promise I'm competent at this

[harsh storm]

It's not you. It's C++.

[stoic rose]

Okay so which functions do I override in the node for this?

I mean this sounds like the right way and the cool way to do it

[harsh storm]

I'd check to see if the auto possess actually creates the AI Controller and then possesses. The controller most likely hasn't been created.

Because if you have the auto posess turned off, it'll never create the controller

[slow bobcat]

Can't recall the names, you will have to check within the engine. You have to override one that says how big is your custom memory struct and some other

The node Durox recommended is a good place to check

[stoic rose]

oh it needs a sizeof

GetSpecialMemorySize?

[slow bobcat]

And you will see there's a special cast called cast_memory

[stoic rose]

oh wait

virtual uint16 GetInstanceMemorySize() const;

[harsh storm]

I wonder where @uneven cloud has been 😭

She was always super helpful

[slow bobcat]

Check where's that used in the spruce code, you will see examples on how to handle the memory

[stoic rose]

Yeah I'm not entirely sure where to find that node @harsh storm

[deep bramble]

This makes no sense... its there in the on the pawn AI Controller Class = AICon_Shark but on the GetAIController node its null.

[harsh storm]

@deep bramble Yup - that's why.

This is what gets called inside of PostInitializeComponents

Yes. The class is there. But not the instance of the class

[stoic rose]

Ahhhhhh

[deep bramble]

hmm okay so what am i missing? I need to make the instance?

[harsh storm]

Yeah

[stoic rose]

So you give back the sizeof of your custom memory struct, then when you need it, cast to it

[deep bramble]

by posessing it?

not sure how to do that

[harsh storm]

Or just spawning it

You have to manually create it if you are not auto possessing

So create it like any other actor

Then possess the pawn

[deep bramble]

ohh I spawn the AICon_Shark controller

yea okay makes sense

[harsh storm]

My bad, it is called "Wait" not "Delay"

So look up BTTask_Wait

[slow bobcat]

Yep. I think there's another thing somewhere. Check the header of the class where you saw the function implementation,jist in case. Look for usages of the struct FBTMoveToTaskMemory

[harsh storm]

But looking at MoveTo is a good one too

[stoic rose]

FBTMoveToTaskMemory* MyMemory = CastInstanceNodeMemory<FBTMoveToTaskMemory>(NodeMemory); like this?

[harsh storm]

MoveTo will show you pretty much everything with the nodes. The memory shenanigans and the AITask shenanigans.

[slow bobcat]

Yeah. Check if they do declare soemthing within the class header with that memory struct. I recall a typedef but I might be mixing things with state trees

[stoic rose]

No typedef that I can see

Yeah just the memory struct declared alonside the class in the header

This is wildly cool.

Dude this is so cool!

Thanks for your help @slow bobcat and @harsh storm ❤️

[slim quest]

hi all, i have an issue with my nav mesh bounds and I'm not sure what's causing it. Essentally, I am making a dungeon crawler, and as such my nav mesh needs to have dynamic run time generation. I have turned off the fill collision underneath, as suggested by another friend of mine, and when that didn't work tried to see if it was the collision and turned them to have no collision, and I have made sure to build the level, replace the nav mesh, nothing has been working. Any ideas?

[misty wharf]

If you mean the problem is that you don't want the characters to affect the navmesh - turn off "can affect navigation" on the capsule and if that doesn't fix it, also on the character mesh and any other primitives on it

[slim quest]

thank you so much, this fixed it

[stoic rose]

Now if I wanna make a Blueprint of this, am I able to just make it as Blueprintable, or is dealing with Blueprints a whole other kettle of fish?

[harsh storm]

All BP nodes are instanced

Honestly never made a BP from one of my C++ tasks.

[stoic rose]

okay so then this memory stuff is totally irrelevant to a BP service

[harsh storm]

Yup

[stoic rose]

Fantastic.

I mean at least I learned something

Okay I imagine, then, that inheritance with custom memory would have to be pretty well-thought-out.

Wow this is presenting some interesting mental challenges.

I suppose if you wanted to make a memory struct for a child of a service node, that memory struct would have to inherit from its parent's struct and then you still override the GetInstanceMemorySize to sizeof your new child struct

So that any calls in the parent node that cast to the parent struct will still be correct

Yeah the <= here does imply that would be the desired way to implement child nodes with child memory

oh jesus that's exactly what that comment says, I didn't even read it lmfao

[bronze birch]

What needs to be done for blueprint task in behavior tree to tick?

[slow bobcat]

I do everything in c++ but, isn't there a "start with tick enabled" flag? Or maybe "set tick enabled"?

[bronze birch]

I found the problem i had, it was because of move to location or actor node running in the task, for some reason it stops execution of behavior tree and nothing can tick while its running, i'm not sure why

Simple Parallel doesn't work as well

[misty wharf]

iirc there's a "lock AI logic" pin on it, just turn that off

[bronze birch]

u right, i didnt even notice that, thx

[ruby shard]

is it possible get effected nav mesh polys of a nav modifier volume ?

[north oriole]

is there a way to add a NULL modifer around a character (so that other actors will avoid it), but make the character itself ignore?

[misty wharf]

Possibly via using nav query filters, but I would consider trying RVO or crowd avoidance

[slow bobcat]

Yeah, in code only. There are functions in both, NavigationSystemV1 and Recast Detour to obtain information from the nav. IIRC there's a get nav polys in area where you pass a AABox

[ruby shard]

i know those kind of query functions but i think the nav modifier object has already kept the poly ids somewhere 🤔

[slow bobcat]

No, that's what we all think at some point but, unfortunately, that's not how the nav system works. The nav volume is nowhere stored, but the octree (which stores transform and bounding box). Then the nav data for a tile is generated based on that octree (I'm skipping several steps here for the sake of explanation). There's no direct link between objects in the scene and data in the nav data, only a functionality layer in between

The nav volume is just a nav relevant object with a flags' config for polys and areas that is read to write info in the nav mesh.

[ruby shard]

so nav objects do not know what polys are effected by them, they just send tile update events when created, removed ? 🤔

[misty wharf]

Essentially yeah

[slow bobcat]

Yep
The objects pop up in existance with a note "I affect navigation as null". The object doesn't really care about the note. But the nav system is checking who has notes on them and maps them creating the nav data.

[hallow compass]

is there anyway to change EAISenseNotifyType NotifyType (in UAISense)

its protected, and there is no setter function, do I have to make my subclass ? its not exported so idk how is this possible

[hallow compass]

nevermind, it seems you can subclass it

[coarse smelt]

Hello, I have the basics of the AIPerceptionComponent and StimuliSource working in Blueprints, but I'm keen to do a bit more research into how the PerceptionSystem works. For example, is it basically creating and invoking delegates? If so, is the StimuliSource broadcasting and the PerceptionComponent listening, etc...does anyone know of any existing documentation or tutorials that cover this?

[zenith hill]

As soon as I add a test, the Grid never comes up again. Any idea?

[slow bobcat]

As in many instances in unreal, the source code is the documentation. But pretty sure there are tutorials in YouTube about it. To clarify a bit, it's not an event driven system really, more a "iterate over stimulis, run functions on tick, flag the ones that can be perceived (or not)

[slow bobcat]

Happens with any test?

[zenith hill]

ahh nah, that was a mistake on my end, was spawning it on the wrong AI and got the querier mixed up.

I ran into anotehr issue though

Trying to test dot product with the character forward vector, but didn't get the result that I want.

The test is
Line A -> Character's forward vector. Mode is rotation (probably wrong of me to do it?)

Line B Two points
From Player Loc to query Item

but as you can see from the image, it doesn't respect the character forward vector

[slow bobcat]

Draw your forward vector with a debug draw line just in case. Looks like it's pointing to the right

You shoudl also be able to see it if you eject in runtime and select the players capsule with World cords selected in the editor widget

[zenith hill]

No matter what direction the character face it's giving the exact same result

I probably give incorrect rotation?
Isnt that supposed to be the forward vector?

it's asking for rotation though 🤔

[slow bobcat]

The forward is already normalised

Aaaah yeah

Sorry

You can't put forward there

Pass the querier

It will grab the rotation of the actor

[slow bobcat]

This is also wrong. A direction is not a location. Forward is a direction

[zenith hill]

So what can I use?

[slow bobcat]

Not sure what's the intent with that BP, but if you want to provide a location for soemthing, you can do something like PlayerCharacterLocation + (forwardVector * 100.f)

That will give you a location 100cm in front of the player

But try passing the querier context to the rotation

It should grab the querier actor and, from that, the rotation

[zenith hill]

the querier can only return Actor, location and Single actor as far as I can see

[slow bobcat]

Similar issue from this message onwards

[zenith hill]

and the intent of the bp is to simply scoring Query Items based on the dot product

[slow bobcat]

Return the actor

[zenith hill]

k

@slow bobcat you are right, thanks

[zenith hill]

Can i view the EQS while Unpossessing the controller?

[slow bobcat]

What do you mean? You mean at runtime while you un posses?

[zenith hill]

Yeah

[slow bobcat]

But if there's no controller... You can't run queries I think? Not sure, never tried, but I believe it's tied to the AI

[storm sail]

if you mean see the EQS running ingame at runtime, while flying around the map on your own AI. yes you can do that

[zenith hill]

the momment I unpossess the debug sphere dissapear

any idea why this is reversed?

i want green on item that can see my character

[storm sail]

I'm a little confused, are you using EQS within your own character

or is that an AI Character

[zenith hill]

The EQS is generated around the player character.

the idea is to find the best point for the A.I to go to

2 vanguards, one at the left of the player and one on the right

with support at the rear

[storm sail]

I think you can check for Errors. There's like an EQS log

is the player character still being returned when you unposess. I would imagine it returns your spectator pawn, which there is no AI Navigation for

[slow bobcat]

I'm a bit confused @zenith hill Do you mean while unpossing a character due to logic or while you eject in runtime editor (Pie) while playing to move the camera around?

To see the points, debug etc?

[zenith hill]

Yeah it's being returned but I am not worried about visualizing while possessing anymore

trying to figure out how to do a trace test that gives me a good socre when it can hit the context (the player in this case)

right now it's inverted

the items that don't hit the player is the one that get score

[storm sail]

there's usually a boolean in the default tests that invert the result

[zenith hill]

you are a champion, it was the bool match

I can sleep without crying my self to bed now

cheers

[quartz forum]

Hello, In an AI Decision Making context. How do you handle event-driver Behaviors (patrol, heal, hit reaction, ...) and "on-tick" Behaviors (attack target X if it is close enough, change target to closest one Y, ...) ?

[misty wharf]

You would probably need to determine the priority of an action in some way

Ie. if we assume a hit reaction would supercede other actions

[stable plover]

just as zomg said also there is multiple ways to go about it

the sight the handles the closest target should not be dont in state trees, use an eqs percieved actors query and every stimulus update run that query, thats what lyra does kinda

the result of the eqs sets a target variable

[full wing]

Any recommended videos for state tree tutorials?

[slow bobcat]

https://dev.epicgames.com/community/learning/talks-and-demos/yj09/unreal-engine-exploring-the-new-state-tree-for-ai-unreal-fest-gold-coast-2024

[full wing]

thank you

[buoyant pebble]

Hi! im creating a behavior tree using Unreals behavior tree entirely from C++ and am currently trying to run a UBTTask_RunEQSQuery and store its result to a blackboard key; but the only reference i can find to it is the BlackboardKey from BTTask_Blackboardbase, which doesn't seem to work. Has anyone else encountered this issue / knows a potential solution?

[simple crescent]

I'd like to change the vision cone in AISense_Sight to add different checks for vertical angle and offset. Is there a recommended way to do this. It doesn't like like it is simple to override AISight_Sense. Very few functions are virtual and it relies heavily on FDigestedSightProperties and AISenseConfig_Sight, both of which appear to be difficult to replace with derived versions. Is there an easy way to do this (besides doing all of the changes in AISight_Sense itself)?

[misty wharf]

Bit hard to say - You can do some things via IAISightTargetInterface but it might not be what you're looking for. You could consider creating a custom sense.

[simple crescent]

It looks like I could do it in IAISightTargetInterface, but I'd end up with two very different locations for the existing and new vision cone parameters.

[crimson karma]

Hi guys, I ran into a problem with my AI where it slows down before reaching the target actor. If the target actor is also moving (but at a lower speed) then the AI never reaches the target actor. (Video 1)

I am using the AI MoveTo node in blueprint.

I managed to solve the problem temporarily by setting the "Fixed Braking Distance" to 0. That way the AI reaches the target actor (Video 2), but it also causes a weird "swaying effect". (Video 3, edited, see the message below)

Any advice?

These are the settings atm. Does anything look wrong?

Video 3:

[near condor]

Hey all, packaging question relating to State Trees, do we have a fix for packaging when using State Trees in 5.5? Specifically for the crash that occurs because it loses references?

[misty wharf]

Didn't they address that in 5.5.2?

[near condor]

I had thought so but a packing test on 5.5.2 is causing the same issue

[slow bobcat]

Set the Use Path Acceleration flag to false.

[crimson karma]

Alright, I'll try this!

Thank you, that indeed did fix it, although it also creates another problem. By unchecking that setting. The velocity resets everytime the AI recieves a new MoveTo instruction. Is there a way to have the AI not reset the velocity?

[crimson karma]

.

[near condor]

Here is the crash report snippet:

[2025.02.03-16.15.57:841][833]LogOutputDevice: Warning: 

Script Stack (3 frames) :
/Game/TDRPG/Assets/Blueprints/AI/StateTree/Evaluators_Gen_04/STE_IG_HealthCheck_Gen_04.STE_IG_HealthCheck_Gen_04_C.ExecuteUbergraph_STE_IG_HealthCheck_Gen_04
/Game/TDRPG/Assets/Blueprints/AI/StateTree/Evaluators_Gen_04/STE_IG_HealthCheck_Gen_04.STE_IG_HealthCheck_Gen_04_C.ReceiveTreeStart
/Game/TDRPG/Assets/Blueprints/AI/AI_Gen_03/BP_IG_AIController_Gen_03_StateTree.BP_IG_AIController_Gen_03_StateTree_C.ExecuteUbergraph_BP_IG_AIController_Gen_03_StateTree

[2025.02.03-16.15.57:850][833]LogWindows: Error: appError called: Assertion failed: Result [File:D:\build\++UE5\Sync\Engine\Source\Runtime\CoreUObject\Public\UObject\Stack.h] [Line: 416] 


[2025.02.03-16.15.57:850][833]LogWindows: Windows GetLastError: The operation completed successfully. (0)
[2025.02.03-16.16.57:863][833]LogWindows: Could not start crash report client using ../../../Engine/Binaries/Win64/CrashReportClient-Win64-Debug.exe
[2025.02.03-16.16.57:863][833]LogMemory: Platform Memory Stats for Windows
[2025.02.03-16.16.57:863][833]LogMemory: Process Physical Memory: 1622.82 MB used, 1878.49 MB peak
[2025.02.03-16.16.57:863][833]LogMemory: Process Virtual Memory: 5195.05 MB used, 5196.97 MB peak
[2025.02.03-16.16.57:863][833]LogMemory: Physical Memory: 72347.41 MB used,  58655.68 MB free, 131003.08 MB total
[2025.02.03-16.16.57:863][833]LogMemory: Virtual Memory: 113161.28 MB used,  26033.80 MB free, 139195.08 MB total
[2025.02.03-16.16.57:863][833]LogWindows: Error: === Critical error: ===
[2025.02.03-16.16.57:863][833]LogWindows: Error: 
[2025.02.03-16.16.57:863][833]LogWindows: Error: Assertion failed: Result [File:D:\build\++UE5\Sync\Engine\Source\Runtime\CoreUObject\Public\UObject\Stack.h] [Line: 416] 
[2025.02.03-16.16.57:863][833]LogWindows: Error: 
[2025.02.03-16.16.57:863][833]LogWindows: Error: 
[2025.02.03-16.16.57:863][833]LogWindows: Error: 
[2025.02.03-16.16.57:863][833]LogWindows: Error: 
[2025.02.03-16.16.57:875][833]LogExit: Executing StaticShutdownAfterError
[2025.02.03-16.16.57:876][833]LogWindows: FPlatformMisc::RequestExit(1, LaunchWindowsStartup.ExceptionHandler)
[2025.02.03-16.16.57:876][833]LogWindows: FPlatformMisc::RequestExitWithStatus(1, 3, LaunchWindowsStartup.ExceptionHandler)
[2025.02.03-16.16.57:876][833]LogCore: Engine exit requested (reason: Win RequestExit)

[near condor]

I see someone asking if they are bugged for packaged builds back in December for 5.5.1

So I am rebuilding the blueprints for the state tree and evaluators from scratch. Putting the updates here in case someone else runs into this issue as well.

Will update to see if it worked.

[slow bobcat]

That looks like you are using move to to a locations. If you move to an actor, that doesn't happen. When the move to task is created to follow and actor, and the actor chased moves, the AI move to task updates the end goal. When you chase a location and the location moves, there's no update, but an end call to the task (which causes a stop) and a new task is created

Try spawning an actor (just an aactor with a scene component to handle its transform) and make your AI move to that actor

[crimson karma]

Oh, I see, I'll give that a shot 😊

[near condor]

Did you find a solution for this

[near condor]

As of now, my understanding is that Get Property reference does not work in 5.5.0 or 5.5.2 (in packaged builds for both, Editor is fine).

[naive oar]

No I didn't, overall i wouldn't consider State Tree production ready. Overriding Linked State Trees also doesn't work in shipping

But they have so many advantages over BT that i don't want to go back

they'll fix it, eventually 😄

[near condor]

I agree with you. I submitted a bug ticket to them

[near condor]

I hope, but the thing is, that bug has been there at least since 5.5.0 and I did not find any bug reports other than what you said about it here in the discord.

Nothing on git in pull requests or the ue bug site

[naive oar]

I think property reference was just exposed to blueprint in 5.5, and many people are still using BT, so it's quite fresh

they'll fix it when Fortnite needs it 😄

[misty wharf]

From what I can tell the property ref is not that useful anyway unless you want to do something like write data out from a state tree task

[naive oar]

Yeah but that's quite essential, because if you using any sort of async stuff in your tasks you need to wait for it before going in another state, so the normal out properties are not an option

[misty wharf]

Oh yeah I guess that's one usecase for it 🤔

[harsh storm]

I'd imagine you just write the async stuff the same way you do it in a BT.

Just return that you're still in progress until you have what you need.

[naive oar]

You can't "write" to a state tree without Property Reference, that's the point

[harsh storm]

I guess I don't understand the problem enough. I haven't had much issue with async state tree tasks yet.

[misty wharf]

Basically if you have State A which does something async and outputs some data, and state B which needs to use data from A, there's no way to connect them in a nice way

Only way to bind a property in B to A is to make A the parent of B which then doesn't really work

I think you can work around this via property refs but I've not used them myself yet so don't know the specifics of that

[naive oar]

For example, all my nav stuff is async. So if find navigable location in radius is async, you need to wait for the location in the task that actually moves the character.
You can't use out properties, because it won't wait for async.
The only option is to have 2 tasks which are siblings, and you transition from one to another when the async stuff is done. And only Get Property Reference can write to a shared variable in a higher up state which both tasks can access

[harsh storm]

Oh, okay. So it is a structural problem

[misty wharf]

Yeah it's almost as if they forgot that asynchronous tasks are incredibly common

But it's good that the prop ref thing works now (except if it's defined in BP's I guess?)

[naive oar]

I found that it works in shipping if you use a source build, but not an installer build. Who knows whats going on 😄

[misty wharf]

Could be your source build isn't from the same tag as the launcher build?

[naive oar]

I use the regular release branch source, so it should be the same, i just need it for better debugging. You can't set breakpoints in engine code without engine source

[crimson karma]

I found the problem, I had a Stop movement node that caused the behaviour. My bad

Anyone know how to abort an active AIMoveTo operation?

[stable plover]

sounds like someone else had the same issue, I actually figured out a way to bypass that

[stable plover]

you can still use the use acceleration for paths, I had the same issue yesterday if you want to keep using it

[crimson karma]

Yeah, I'd like to keep using it, since the movement becomes very jittery with it off. (It becomes too "snappy")

Any ideas?

[stable plover]

these are the current settings that im using currently and help with what im working with, but my current setup for what im doing is different from yours so not sure if it will work for you

[crimson karma]

^ These are my settings

[stable plover]

my ai move to has a small acceptance radius though

your first pic is too high

the friction factor, i use a seperate braking function as well and make the braking distance smaller

[crimson karma]

Hmmm, alright

I'll try copying your settings

[stable plover]

i mean not sure if this is the result you want but this is what actually happens for me

[crimson karma]

Seems correct

As long as it manages to complete then I'm fine with it

[stable plover]

should just be stop movement no?

[crimson karma]

Yeah, that's what I'm using at the moment, was just wondering if there are any alternative ways.

[stable plover]

ah im not sure tbh

[crimson karma]

Seems like my character overshoots if I use your settings.

[stable plover]

its prob the friction factor

set it back to 2

but keep the braking distance small

actually you may want to mess with the braking friction settings

it will prob be braking friction that you need to mess with

[crimson karma]

This is the problem I'm facing

[stable plover]

he stops when he reaches the target but the friction value is low and thats what causes the overshoot, it works for my case but not yours

[crimson karma]

He slows down before reaching the target

[crimson karma]

gotcha

[stable plover]

whats your moveto acceptance radius set to

mine was set to 10

[crimson karma]

These settings didn't work neither

Seems like the same

[stable plover]

ah

[crimson karma]

The behaviour I want is for the AI Move To to move as close as possible to the target without stopping until it finishes.

But it keeps slowing down before it calls On Success and On Fail,

Thus never reaching the moving target

[stable plover]

im not sure tbh the zombies can get away with the lower braking friction but since you dont want to overshoot but it will work when its chasing though, maybe if you are going to a point dynamically change the friction factor to 2 and when you are going to a target change it to .5

[crimson karma]

Good idea

I can try that

I was wondering myself about just turning on and off the "Use Acceleration For Paths"

So that it's On by default but I turn it off once the MoveTo starts, and when it finishes I turn it on again 😅

[stable plover]

yea I tried but thats not blueprint readwrite

[crimson karma]

Is this not the one?

[stable plover]

no the other one

use acceleration for paths

bruno was saying that you would have to override the function and expose the variable

its under nav movement

[crimson karma]

Btw, do you know what the difference is between these two?

[crimson karma]

Yeah, I did that before, and it became VERY snappy haha

[crimson karma]

Haven't tried this though

[stable plover]

I think the differences are minor tbh

[stable plover]

yea unfortunately I cant do that because my project is in bps

atm

[crimson karma]

Yeah, turning "Requested Move ..." off did nothing for me😅

[stable plover]

yea, i thinkk that is if you are not using navigation

[crimson karma]

Imma try your approach with the friction value

[crimson karma]

Aha okay

[crimson karma]

Is this the correct way of getting the friction factor from a Pawn btw?

[stable plover]

i just get the character movement component thats all

so yea

i mean how else are you going to get it?

[crimson karma]

Wohoo success. Your solution fixed it!
I saved the initial "braking deceleration walking" on begin play. Then right before the AI MoveTo i set the Braking deceleration walking to 0 and after the MoveTo fails or succeeds I reset the Braking deceleration walking to the initial value. Works like a charm.

Thank you BilliamsFluster 🎉

[crimson karma]

I tried the same with the braking friction factor, but that didn't work, but the normal Braking deceleration Walking worked great :D

Nvm, correction, I have to set both to zero for it to work 😅 (both braking deceleration walking to 0 and braking friction factor to 0)

[stable plover]

it works as intended?

[stable plover]

beautiful glad I could help

[crimson karma]

Seems like it, yeah. When I zero out both then it reaches the target, and once the target is reached the values are reset.

[stable plover]

its weird because you would think someone would have talked about this problem on forums or youtube but no,

I had to think for a few on a approach that worked but I haven’t tried what you just did

[crimson karma]

No more slowing down :D

[stable plover]

good stuff

looks nice

[crimson karma]

I have to give you all the credit though for leading me in the right direction :D

[stable plover]

Appreciate it and wish you best of luck

[full wing]

does anyone know if i can set those parameters outside of state tree, like in aicontroller, not it the details panel?

[stable plover]

you can in 5.5 but its cumbersome, but other versions not dynamically i dont believe

what I did to get around that was to make an actor component, attach it to the player controller, I called the component StateTreeAIBlackboard, then I made a global task to fetch the blackboard

[full wing]

Do you know about default way of changing variables in statetree in actors or controllers?

you know like in behavioral trees i could just get blackboard and set its variables by name

[stable plover]

its exposed in the editor you cant change the values at runtime unless in 5.5 but even then its cumbersome

[stable plover]

thats exactly what my actor component approach mimics

and thats what im using for my current Ai project

because state tree's storage system sucks right now

the system is still technically new if you think about it

[full wing]

ahhh i get it, you attached other component to your actor with state tree, then somewhere is state tree you get it, and communicate between tree and actor through this

[full wing]

am i right?

[stable plover]

if you have a custom ai controller I attach it to that

then i have a global task to actually get and return it

[full wing]

thats smart

really good work around

[stable plover]

and here is it being accessed

its a good way to get around the storage issue

because you can just go into the component just like a regular blackboard and get the data

as well as access and change the data dynamically

[full wing]

And in your get state tree blackboard you just do get component by class

alright

will do it then

big thanks

[stable plover]

you can do that if you want to but i just do this

make sure to use the correct categories

[lyric flint]

hey I know this is the wrong chat, but I’ve really been struggling trying to get this right for the last two weeks. Do you mind helping me in https://discord.com/channels/187217643009212416/846520322642411570

[stable plover]

why are you in this chat if you know its the wrong chat?

[lyric flint]

Cuz I have been to 7 different servers, across the entire internet and even went as far as to ask ChatGPT and nothing had given me proper results

And this is the only chat right now that’s actually active

[stable plover]

what type of issue is it?

[lyric flint]

A VFX issue. I’m trying to make a tornado, a have a reference in https://discord.com/channels/187217643009212416/846520322642411570 but all the tutorials look… less than pleasing

[full wing]

you gotta experiment with that stuff, try to make one and look at examples of other games that have tornados

[stable plover]

im trying to make an attack behind cover query for my soldier ai and its not working as expected. the first image is the actual query and the second is what im actually doing. Im basically casting 2 line traces, one at the bottom of the enemy which needs to hit some piece of cover and the second line trace starting at the eyes and needs to not hit a piece of cover. This should work in theory and it does some times but all of a sudden the enemy just camps outside of cover for some reason. if anyone can help me that would be appreciated thank you

[stable plover]

Nvm I got it wasn’t the eqs lol

[full wing]

what was it?

[stable plover]

I was running another query called reposition if they couldn’t see the player which gave that result, the reposition should only be used for regular attacking

[full wing]

good catch, btw you using ai stimuli detection method?

[stable plover]

Kind of, I use the stimulus to fire individual events/functions like sight and etc. but for sight specifically I run an eqs perceived query that biases the closest actor

Then I implement my own forgotten functionality for each sense

[full wing]

Nice, i was asking because i was wandering if people use other things than just stimuli which after my last project I realized that it is much better to do in c++ at than blueprints. btw nice skin in your lil game xd

[stable plover]

Haha, she’s a classic

[stable plover]

Yea it’s much better in c++ you have more flexibility

But I don’t like to just use the stimulus check functions, and the forget actor

So you branch out and find what works best for you

What you use?

[full wing]

Right, do you think just making simple line trace with multiple lines would be better for example? just finding alternatives

[stable plover]

For what exactly?

Oh I’m not sure tbh

One of my friends did that, it’s all what you want to do

[full wing]

i used stimuli and my code got so messed up, i just left it like that i called it a day, right now storming through state trees, i find out that in 5.5 you can use something like this to make the parameter go from task to set global variables

[stable plover]

Yes that’s the approach that I said was cumbersome

[full wing]

hahah

[stable plover]

ThreeWays talks about it on YouTube

That’s how I found out about it

[full wing]

i think we watched the same guy, he knows his stuff

[stable plover]

Yes I like watching his videos

[full wing]

If you really want a good yt guy who literally makes videos on everything you should watch codelikeme

i made display settings, keybinds, and graphics settings based on his vids

the guy talks slow, is mostly boring to watch but the guy is really good

[stable plover]

He’s got some good stuff and some interesting ideas but some of his ideas could be implemented better. But I watch him to see how he approaches stuff, so I can get familiar how it works

[stable plover]

Indeed he is

What Ai are you making? I’m hoping off so I’ll answer tomorrow

[zenith hill]

quick sanity check, do you guys have AffectNavigation on your character pawns or A.Is?

or do you guys implement detour A.I so they don't bump into each other?

[misty wharf]

If your AIs have affect navigation that's going to break their own navigation usually

Assuming you use dynamic navmesh generation. If you don't use dynamic navmesh generation then that does nothing

Plus navmesh generation can be expensive so it might not be a good option from that point also

[zenith hill]

Right, that does happend before.

I never worked with A.I before, how would I address A.I trying to get around each other?

previously touched detour A.I , I guess that would be the go to method?

[stable plover]

Exactly the problem I had yesterday, I have a zombie melee Ai with box collision attached to the hands and each attack broke the navigation then all of a sudden the zombie would glitch and jitter all around

[stable plover]

There is a setting in the movement component called RVO I think

But if they are chasing a player then that won’t really work, you will probably need to create your own navigation algorithm or use a proxy system @slow bobcat helped me with mine. But I think the setting in the movement component is called use RVO avoidance

[slow bobcat]

Crowd control and rvo do not modify the path to go around. What they do is to handle the situation where AI's collide. They both do the same: push each other out of the way, but they calculate how to do it in very different ways.
With crowd control there's a radius value you can tweak to make the push happen with more distance in between (so they don't touch) but I personally never got a good result (depends on the game you are doing)

I helped put @restive loom with a similar problem. He's working on a solution which can be summarized as "detect which 2 AI's will collide 0.5s in the future based on their trajectory, then pause the movement of one of them so the other passes, resume the paused movement"

Maybe you can work together in a solution that, instead of doing the pause, you re-calculate the chunk of the path needed to go around each other

I helped @stable plover with something similar, but it was not so AI's could go around each other, but for AI's chasing an actor to not collapse in an single line behind the target

Check this message. I posted couple images that might help playing when around with Crowd Control

[stable plover]

ahh gotcha

[restive loom]

Wow it's you with the same problem

Wanna dm and see what I got done?

[zenith hill]

Well the problem is there by default. In my old project I used AI detour

[upbeat hamlet]

Does someone have any insight into how AI Sense Hearing works? I can't find any documentation and ChatGPT is giving me some mashed up bullshit which doesn't exist 😉
So, my AI Controller has Hearing Range 10000 cm.
Which values do I use in Report Noise Event Loudness and Max Range?
Currently, my AI can hear noises which are a lot further away than the 10000 cm. Maybe because I'm using loudness 500? But what should I use?

[zenith hill]

Sorry, class is starting soon. I'm taking a nap atm. Thanks for the offer though.

[restive loom]

Yeah true

[zenith hill]

You solved your issue?

[restive loom]

Yeah np just hmu if you wanna try to solve it with me or if you find something

[restive loom]

Not yet, working on it

Kinda stuck rn I got some bugs

[zenith hill]

I haven't go there yet, if I got it working I will let you know.

I would probably gonna go with roi avoidance.

But not sure what the result will be

[restive loom]

Rvo?

Or roi

[zenith hill]

Still trying to improve my diamond formation follow player logic.

Rvo*

[restive loom]

Rvo only works with agents

[light trail]

I'm populating 4 positions around the player by doing an EQS circle but find no easy way to assign different positions to different AI because the result returning in behaviour tree is best match or 25% etc. Right now, I do the EQS in blueprint and manage the position result array myself but wondering if there is a better (or easier) way? 🤔

[plucky knot]

has anyone found a way to find out why Behavior trees keep breaking sometimes? The debugger doesnt help

[mint terrace]

that presumes behavior trees are breaking, and they certainly aren't for me

there are plenty of things you can do that make it look like it's BT's fault though, maybe share some more details on what you're doing

[storm sail]

I think you're referring to Unreal engine's crowd control, RVO is different

[restive loom]

Well do you know how rvo would work with other things?

And also rvo is just really basic avoidance so that's kinda mehh

[slow bobcat]

Yeah, they don't. 100% it's something in your game.

[slow bobcat]

When you say breaking, does the game crash, the BT gets stuck or what are we talking about?

[stoic rose]

Hello all! Quick question, the UBehaviorTreeComponent reference passed into ScheduleNextTick is a component that's placed on the AI Controller, correct?

So GetOwner would return the AI Controller in question? Or should I use GetOuter for that?

[misty wharf]

GetOwner returns the owning actor of a component

So yes that's the function you can use to get it

[stoic rose]

Right, I just wanted to be sure the component was indeed on the controller, too.

Thanks! 😄

[lunar cedar]

im having an issue that im not sure is a possible bug. long story short, i have a state tree on my ai pawn (call it ST1). the first state is a linked state that runs another state tree (call it ST2). ST2 finds a smart object, then calls UseSmartObjectWithGameplayInteraction on it which runs another state tree (call it ST3). ST3 makes the AI move to the smart object and perform an interaction. this all works as expected and is super simple. once ST3 completes, ST2 moves to the next task and the AI moves to a random location, then tries to find another smart object and claim it again (constantly doing this). im trying to send a event to break out of this so when an event is received, ST1 transitions from state 1 (that runs ST2) to another task. my issue is, while the UseSmartObjectWithGameplayInteraction task is running none of the state trees can receive events until after the UseSmartObjectWithGameplayInteraction finishes for some reason. so when an emergency happens and i need the AI to stop what their doing and go do something else (ie go from the first state in ST1 to the second one)

[misty wharf]

How are you sending the ST event?

I vaguely recall that the gameplay interaction actually holds and processes the ST associated with it... so I'm not quite sure how the events would function in this case 🤔

It does seem like if you were to transition away from the task running the interaction, this would in turn abort the ST within the interaction

[lunar cedar]

right now im testing like this in the ai controller

[misty wharf]

How is the transition set up in the tree for it?

[lunar cedar]

that is likely the case given if i dont run that specific task it works as expected

[misty wharf]

Ah okay, sounds like the transition is set up correctly then

Unless someone here is familiar with how gameplay interactions work, you might have to dig into the source code to figure out what's going on with it. If you're lucky maybe the visual logger might have some info on what's going on with it, but I've not used gameplay interactions myself so can't say for sure

[lunar cedar]

heres ST1. essentially im trying to move from PassiveInteraction to Murderous

[lunar cedar]

yea ive skimmed through but not deep enough to figure out what all is happening. its a shame to as youd think the events would at minimum be passed along since it has a handle on the original state tree

[lunar cedar]

from digging the GameplayInteractionContext's tick is calling tick on the state tree's context which does make its way down to call and process events returned from GetMutableEventsToProcessView and GetEventsToProcessView so now im even more confused as to why the events are not seeming to be received

[heady silo]

I'm trying to figure out how to get my AI to stop moving and stay stopped unless other input is passed in. I kludged together a holdup mechanic where aiming at an enemy will play a line and relevant animations, as well as ordering the AI to stop its movement and execute a unique task that holds the enemy in place, but the patrol behavior keeps taking back over despite putting the holdup task call and the relevant branch of the tree at the highest priority task. Additionally, units that die still move and play dialog lines after ragdolling. How do I make these guys stop?

the holdup triggers by doing a overlap check and a line trace from the player's gun barrel and triggers the behavior if both overlap on a nearby enemy

I hesitate to find a way to completely disconnect the AI from the AI controller, because I want them to be smart enough to break the holdup if another enemy spots the player and engages them

and I might try having held-up AI eventually be commandable by the player to move around or something

https://youtu.be/p5YDra9ieF4?si=jmSohsXmiUffWAh5
the bug in action

[tough nexus]

Hello ! i want to make a behvaiour tree task where an ai chracter follows a pre computet spline in another event, one problem is i cant make the ai follow the spline smoothly? i was thinking of using a timeline but unfortunenatly its not available in the behaviour tree task blueprint

[harsh storm]

Just use tick

Timeline isn't available because it is a component

[stable plover]

I think it works like this, if there is a hearing event within the hearing radius it does a distance check from the event to the ai, so if you use a make sound with 500 or etc it will check the distance, and if the distance is greater than that it wont hear it even if its in the radius, thats how i think it works tbh i could be wrong, but thats what my tests show

[stable plover]

use the visual logger that may help

[tough nexus]

alright its working but how can i then implement that the character is using the locomotion? @stable plover

[stable plover]

are you telling the state tree how to handle the event?

[stable plover]

stop movement function for the controller

makes them stop

[stable plover]

ahh i see nvm

@lunar cedar events work fine for me I basically do the same thing that you are doing, you try using the debugger?

[stable plover]

why not just have a simple boolean that says is held up?

[stable plover]

what do you mean by smoothly?

are you just trying to follow a spline path?

[tough nexus]

so basically it should just move on the spline because i tried a setup where its using steps

yes#

but also using the ai move to so teh character use the legs so the locomotion

[stable plover]

im a little confused because if its moving you want locomotion to play right? and smootly im not sure what you mean by that

just get the spline and get the first spline point, assuming your spline has multiple points move to one point, then move to the next point in a loop

[tough nexus]

so this setup is working that the character follows the spline, but its not using the locomotion so the walk animation

[stable plover]

instead of using location and rotation why not just the move to?

just a question, even if you use the location and rotation it should play the locomotion though

[tough nexus]

ok i dont know its working now with ai move to but the last 2 days it wasnt working xD

no its not using it

because its not bypassing the ai state

[stable plover]

so problem solved?

[lunar cedar]

just checked the debugger, once the state tree enters the smart objects state tree it stops recording, then once that state tree finishes and it goes back to the normal one the debugger kind of backfills everything. the first screenshot is after the ai claimed the smart object and started to use it, it walked over to this bench and stayed there for its idle. the second screenshot is immediately after it exited the idle (finished the tree of the smart object)

[stable plover]

how about the visual logger if that will help

the visual logger gives more information than the debugger

@lunar cedar not sure if you know about three ways but he is using events with smart objects in this video https://www.youtube.com/watch?v=OZIhkmTZyp8&t=29s 43:50 he is using the event, hopefully this helps

[lunar cedar]

visual debugger shows about the same steps. starts the linked state which claims a smart object, that new tree runs, once that tree is finished the event is received and the state changes to murderous

[lunar cedar]

heard of him, ill check that out thank you!

[stable plover]

it should solve your problem, apologies i havent used smart objects yet but his channel is a gold mine for state trees

[lunar cedar]

all good. this i smy first week going into them and so far they are awesome but quirky

[stable plover]

i know they offer their advantages and wish i could use them but im in 5.4

😦

[lunar cedar]

oof

[misty wharf]

At least the state trees seem to work with SO's reasonably well other than this

There's a behavior tree thingy for it also, which has the capability of returning to the previous BT... but it's also completely incapable of returning the BT to the state it was

so it's... pretty useless and bad

Although it can be useful I guess if you design everything with that in mind

[harsh storm]

I don't like how you communicate with the ST from outside of the ST personally.

[lunar cedar]

or get deep in the woods before knowing about the feature and try to use it as a great solution to your problem only to realize its lacking out of the box lol

[heady silo]

If I issue it to the controller, will it stop everyone using that controller?

like that?

[upbeat hamlet]

For me the Pawn is hearing the noise even outside of it's hearing range for some reason.

[lunar cedar]

it shouldnt let it move, for example adding movement input is ignored

[stable plover]

no just cancel the movement

[lunar cedar]

derp i was thinking of something else

[stable plover]

when ever i use the stop movement with the ai move to such as in the middle of the patrol and it sees an enemy i want it to stop and wait a sec, i use the stop movement and it cancels the move to operation

[stable plover]

wow thats interesting , im not sure tbh but im absolutely sure it does a distance check from the make sound location to the actors, it seems like the most logical thing to do

[heady silo]

so this needs to go in the ai controller.... with an event dispatcher?

[lunar cedar]

ignore what i said, but StopMovement is used on the controller class, so get your AI's controller and call StopMovement when you want them to stop

[lunar cedar]

well darn in his example he didnt actually use the event, he just showed "you could do it this way" and went a different direction lol

[stable plover]

crap I apologize, I thought it would be useful but it isn’t

its weird how state trees wont work in your example

[lunar cedar]

eh im going to post in his comments and just see lol

[lunar cedar]

its weird, it works, but not if its running the smart object state tree with the GameplayInteractions schema

[stable plover]

weird

maybe a problem the devs need to fix, i love state trees but they have a lot of problems and features they need to implement

[heady silo]

...I wired up a tick in the AIC from OnPossess to check for if we're held up every tick

probably the worst possible way to go about it

[lunar cedar]

ok now im getting some weirder stuff happening. in the state tree (smart object one) if i just set a delay in it so it just chills there idle, the event is received and works

[lunar cedar]

agreed

[lunar cedar]

do you have some sort of event or anything that fires when you want them to be held up?

[stable plover]

possibly a timing issue?

[heady silo]

Basically my check for a holdup is if the player's HOLDUP_RANGE collision sphere overlaps the enemy's capsule, and the player aims down sight at them, the player character speaks a line and an event is fired in the enemy to react and put up their hands and stop moving

but what seems to be happening is, if the player isn't actively aiming ADS at them, it cancels the logic

I want them to stay frozen for awhile

[heady silo]

https://youtu.be/p5YDra9ieF4?si=jmSohsXmiUffWAh5
this is how it currently looks

the guard will play the animation, but then continue patrolling while having their hands up

and even though I tell them to stop paying attention to their senses, it still works and they can go to a combat state anyway

[lunar cedar]

no idea at this point. i replaced my move to task with epics and it partially worked once, now it wont work again at all

[heady silo]

I guess what I could do is swap behavior trees for the enemy to a special subtree where they are captive

and be like CaptiveIdle to keep them frozen, and then if I add a function later for the player to give them orders to move somewhere, it would do a simple MoveTo

like this, maybe?

or should I use a StateTree for this

[solid wedge]

Is Spawn Default Controller reccommended for "repossessing" the old character when i possess another?

I see ai controllers piling up in my level. Do i need to destroy them?

[heady silo]

With a bit of additions, this DOES at least stop the combat behavior and patrols from happening

[lunar cedar]

wouldnt you just unpossess and possess the old pawn you had possessed? if the ai controller still exists there shouldnt be a reason to create a new one when you can reuse it unless you need a new one

[heady silo]

nvm it's not quite working

[solid wedge]

If i possess another character, 1 node only, what should I do to make sure the character i just lost control of, gets possessed by the ai?

[silent hamlet]

You can get the ai controller and possess that character

[lunar cedar]

ok, figured it out. i put the event on the root of my main state tree and i switch from my move to task to epics, now it works. it still continues to run the smart object's state tree though, thats the last part im unsure on is when that event is received how to get it to stop that tree so it also frees up the smart object without having to run it through completion (the AI still moves towards the smart object and such and does its stuff while in the Murderous intent)

[silent hamlet]

There should be a possess function on the aic

[lunar cedar]

so you want to have your ai controller thats possessing a pawn, possess another pawn, then have that previouslly possessed pawn have a new ai controller possess it?

[heady silo]

apparently the output only fires when I hook it into tick @.@

[solid wedge]

Basic character switching. 2 characters in level. I press button to possess the other one. But no AI happens on the character i just left.

[harsh storm]

It needs an AI controller

When you unpossess it, spawn a new AI Controller and then possess it.

[solid wedge]

But then ai controllers start flooding my scene in the list.

Spawn default controller works. But my outliner keeps showing them pile up

[harsh storm]

Either delete the AI Controller when you unpossess it or cache it when the player takes control and when the player leaves, give it back to the pawn

[harsh storm]

So?

It is a whole different actor

Each AI is going to have 1 pawn and 1 controller

I'd be concerned if I spawned an actor and it didn't show up in the outliner

[silent hamlet]

There is a nice example of this in the content examples, blueprint section (can't remember which map)

Uh wait it does not involve spawing an ai controller though to replace the currently possessed character though. Nevermind

[solid wedge]

So what happens when my game has thousands of unused ai controllers just sitting there? Is it not intended to keep the same one per char. Instead of spawning a brand new one?

[harsh storm]

That's why I said to either destroy them or cache them for reuse

[solid wedge]

Im caching and repossessing, but thats not enough. Going to try the reset behavior tree node

[harsh storm]

So you're saying by calling Controller->Possess - you are creating a whole new controller?

[solid wedge]

No

[harsh storm]

Then how are you creating a new controller?

Because that is all you need to do if you're caching the pawn's AI controller before swapping.

[solid wedge]

I saved AI controller as a variable and then on event Unpossessed, I'm taking that AI controller and possessing the character with it.

I would like to avoid creating a new controller, if there's already one just sitting in my level somewhere

[harsh storm]

Use GetAllActorsOfClass. Search for a controller that doesn't have a pawn. Then possess the pawn.

If you get through the entire list without finding one, spawn a new one and then possess.

[stable plover]

interesting issue

thats weird about the event placement

[stable plover]

isnt there a function to exit the tree?

[solid wedge]

This does not work either. I guess spawning a new ai controller and destroying the old one is the way. Annoying!

[lunar cedar]

not sure, usually i just transition to Tree Succeeded or Tree Failed but neither of which i can seem to figure out how to do manually how i have it now

i dont have access to the actual state tree afaik for the smart object thats running

[harsh storm]

You need to show code or something. Because possession does not spawn a new controller.

[stable plover]

oh crap i thought you did

[harsh storm]

Something you're doing is weird.

[solid wedge]

Ok ill start fresh

[lunar cedar]

closest thing i have is the async task that gets returned which i can abort but doesnt seem to work

[solid wedge]

Ty

[stable plover]

really so its not finishing the task when the event gets fired?

[lunar cedar]

nope, even weirder, if i try to abort the task like this that kicked off the state tree starting before i send the gameplay event, i never get the gameplay event

[stable plover]

i have an approach and this may not be what you want, but in your state tree bind to a delegate and once you call the delegate it will execute the finish task, calling the delegate will occur when you send the murdous event

so when the delegate gets called the state can finish and the event will then get sent

[lunar cedar]

your right, thats not what i want but i might not have a choice lol

[stable plover]

just my theory but when the smart object executes it pauses the state tree events and shifts its focus so even if you send an event it wont work because it not in the smart object, before you send an event there is no way to stop the tree right?

can you destroy the smart object reference?

not sure how it works

[lunar cedar]

correct, the non smart object trees can receive events, it just seems theres certain cases where they cannot for some reason (such as the move to and the abort example above). i cannot seem to get the smart object state tree to receive anything though

im about to ditch the smart object setup and just try to do it manually by claiming and unclaiming

[stable plover]

why did you want to use the smart objects, mind me asking?

im not really familiar with them exactly so i dont know

[lunar cedar]

to learn and it seemed like a good case in my position. so right now as an experiment i have essentially a mall. the ai will pick a smart object that is categorized by tag. so i have browse and sit tags. if they select sit, they will find a smart object that has that sit activity tag (a seat) and move to it, then sit down and chill for awhile before getting up and moving again. likewise browse they will walk up and browse at clothes and such in various shops as if their "shopping" or walk up to a scenic part of the mall and hang out before moving to a new task

[harsh storm]

They're pretty simple in concept, so not too bad to write your own.

Pretty much - they're just an object that holds the logic on how to interact with it.

Think the game "The Sims". Especially because they're the ones that popularized it.

[stable plover]

oh then what Cody is doing should work then

[harsh storm]

So the character doesn't know what animation to play. But the blender does. So the blender tells the Sim to do that correct animation.

[lunar cedar]

it would moreso be piggy backing off of it. in this case instead of using the GameplayInteractionSmartObjectBehaviorDefinition id either use the normal behavior and run the logic in the object it creates or honestly doing it in my existing state tree would probably be dirt simple to

[harsh storm]

I'm just saying that you could entirely avoid UE's SO system and build your own fairly quickly to be honest.

[lunar cedar]

this is where its convinent. since my ai are using the ability system, i set up each behavior to give the ai once it gets to its task a gameplay effect which adds a tag that the pawn is listening to to replicate the state such as sit/browse/hide and what not

[harsh storm]

I haven't heard many good things about UE's SO system.

[lunar cedar]

could yes, its just a shame it was workin so well until it wasnt lol

[harsh storm]

https://tenor.com/bmXnr.gif

[stable plover]

im sure there has to be around it, are the docs helpful?

[solid wedge]

Straight up does not work unless you spawn a controller

[harsh storm]

What "does not work"

[lunar cedar]

not in this case unfortunately. just got a response from someone who mentioned "Smart Object State Trees should exist on the items themselves and so they are not tied to the NPC's State Tree and therefore can be ran independent of them" which explains some of the behavior here. also mentioned apparently if you unload an active state tree for any reason while in use it will crash. bout to test that one when i get back

[stable plover]

ahh thats a shame

[solid wedge]

Must create + destroy default/ai controller in order to possess a character that was recently left unpossessed, like when i possessed another ai

[silent hamlet]

You don't need to destroy any controller. You can simply swap them out. Share your code maybe

[solid wedge]

I figured it out. I just decided to manually spawn cash and manage ai controllers on event begin play

I was having difficulty caching and possessing them when spawned as a default controller

I couldnt simply cashe the AI controller if the character spawned already possessed by me. But It was still in the level, I jusy didn't want to loop through, find it and cache it. It puts my mind at ease, manually, spawning it and caching it per character

[solid wedge]

Now, to make camera look at the same location when possessing another, I tried SetFocus and C++ override of the aicontroller but failed. Need to research into cameras

[stiff ibex]

assuming your camera/spring arm uses control rotation, you can just pass that variable around when possessing other characters

[solid wedge]

Cant. Must move actor and setfocus service

Cant simply send rotator to ai controller.

[lunar cedar]

welp switched out for normal gameplay behavior and im facing the same problem again lol. when the behavior is active no state tree events are received. guess at this point i just need to ditch unreals smart objects entirely

[stable plover]

that sucks and you are in 5.5 too with the latest improvements to state trees

i atleast expected that to work atleast

[misty wharf]

That's bizarre

Because I use state trees and gameplay behaviors and smart objects and events sent to the ST work perfectly fine

You're not using something that blocks AI logic during the behavior execution are you?

Eg. a moveto with that flag enabled?

[near condor]

I am starting to wonder about some "best practices"...

I built a State Tree Evaluator that overrides On State Tree Start and creates a timer for a function to execute (and loops until it detects the Actor is dead).

But after running for a little bit (about half a minute, with this timer being triggered every 0.012 seconds), I start to see this in the logs:

LogStateTree: Error: Trying to GetMutablePtrToProperty while node is not active.

Which has me wondering, is it improper to have a timer executing code within an Evaluator for a state tree, and said code is regularly accessing properties set within the State Tree (Get Property Reference node) and updating it?

I have it set up in a way where it verifies if it is executing first before trying to get the property, but I am not sure what is tripping it up or if this is just a bug

[wind summit]

so I had a chest object that used to block AI pathing, and I turned that off, but the holes remain, and rebuilding the pathing doesn't fix this. Help? lol

I've tried deleting the navmesh volume and the recast thing, and it just comes back

nvm, fixed it

having this on and then just reloading the level fixes it, I turned it back off after

[slow bobcat]

Hello! I don't think I have a solution, but you can start by moving the logic from the evaluator into a global task. Evaluators will be deprecated

I wonder if the evaluator is not there (it's instance?) when the timer runs somehow. Just guessing here, no actual facts

[near condor]

Interesting... Thank you for the tip. With evaluators, the intent was to have something happen on a timer for updating the conditions of what state to switch to. Do we no longer do that?

Or what is the alternative?

[slow bobcat]

you do that in a global task. Global tasks are the replacement for Evaluators

Not saying that's the solution, but doesn't harm to switch. You get the new approach in your code base and, if you are lucky, might even solve the problem

[near condor]

I understand that they are the replacement, I suppose my confusion is that I thought evaluators just ran and could update properties, where Global Tasks can still update properties but they are ran before state changes?

[slow bobcat]

Global tasks and evaluators are ticked at the same moment, just before everything else in the tree
first events, then Evaluators and Global tasks (keep in mind I'm in 5.4, so the code might differ from yours if you are in 5.5)

[near condor]

Thank you for that insight. I am working right now on migrating it over to the GlobalTasks. I really really appreciate it.

[slow bobcat]

I hope it helps. If not, you will have to show us your callstack etc

[solid wedge]

Solved. Silky smooth with no ai logic at all.

-cache control rotation
-possess
-set control rotation
-set view target blend (previous character, 0 seconds)
-set view target blend (possessed character, .5 seconds, lock outgoing)

[near condor]

Hmm I am stumped, how would you convert this from an evaluator to a Global Task?

Global tasks don't have an On Tree Start override.

I would just move it to the AI Controller where the perception component resides, but the UpdateTargetActorSeen function updates values within the state tree, which is something I cannot do from the AI Controller.

[slow bobcat]

You use EnterState

And never call finish task, because that will stop the tree

[misty wharf]

Some of these design decisions with how these things work really feels like a limiting factor in flexibility 🤔

[near condor]

Ugh yeah you are right, I am still getting used to the definitions of states and tasks. Thank you again

[misty wharf]

But I suppose it would be easy to work around by providing a bool to select whether it should finish or not

[near condor]

weird, having migrated to the global tasks now vs the evaluators... My FPS went up around 3 times what it was lol

[lunar cedar]

that LockAILogic yes i had that enabled without thinking -_-

i am finding some limitations here such as passing params to a gameplay behavior with how i want so im going to give the state tree SO another shot since ive learned some of what issues i was having were caused from

[lunar cedar]

so i reverted back to using gameplay interactions for smart objects to test further. heres my setup, the state tree starts, 10 seconds later 2 events are sent, one which is supposed to abort the smart object state tree while the Murderous event it supposed to change the main state tree to do something else. so far its half working, the Murderous event is received but the Abort is not. ST_AI_Civilian is the main tree. it starts at PassiveInteraction which runs ST_AI_PassiveInteraction, when Murderous event is received i want it to go to the state Murderous, which it does just fine. ST_AI_PassiveInteraction simply finds a smart object, claims, then uses it which makes ST_PassiveAction run. ST_PassiveAction simply moves to the smart object slot, once there it performs the action i set on it which is applying a gameplay effect and interpolating precisely to the slot location. i setup this tree to receive the Abort event, problem is it acts like its never received and does not abort the state tree. my goal is when the murderous event is received, i want to stop doing the smart object logic/tree entirely and just do whats in the murderous state

[sacred tide]

Hi, I’ve been working on implementing a state tree for my simulation type ai, I am trying to learn unreal’s state trees and the best approach for my use case.

For the basic loop I want my animal to leave their home during the day and return at night. I created a Task that I originally included in the root, it queries the game state and sets the Instance Data enum, TimeOfDay. (I’ve also considered this as a general task, bound to DayChange to set a parameter, but the conditions states still seem to struggle) The state Day or Night don’t ever seem to meet the condition.

Anyone have any pointers? Is this even the best approach, an alternative thought was 2 state trees for day and night and swap on the actor during sunrise/set.

I was going to use the root node to calculate all the animal’s states (health, hunger, sleep) and then use those as conditions for the states, but so far the basic one is struggling.

[solid spire]

What are the common solutions to detect when an AI gets knocked off the navmesh? Does unreal cache the last valid position, or detect when an AI is off the navmesh?

[coarse smelt]

Hello! I'm looking for examples of interesting things that people have made with BehaviorTrees and the Perception System in Unreal...I can find a few videos that cover similar things (e.g. chasing the player or a stealth system, some combat behavior), but I'm wondering whether anyone has links for more interesting or advanced tutorials or examples? I'd also be interested in ideas for things I could try to implement with Unreal's AI features now that I have the basics down...I think squad mechanics could be interesting...any other suggestions?

[stable plover]

use events simple

you can send state tree events and define how you want the tree to handle them

[upbeat hamlet]

I'm having problems with FindLookAtRotation and my AI Pawn.
To make FindLookAtRotation work, I need to spawn the AI with Rotation 0,0,0.
As I need an initial rotation, I added a scene component with that rotation and then I'm attaching the AI to this component.
Now the problem is, that the attach happens after the AI is possessed by the controller, so AI Sense Sight is looking the wrong way.
How to fix this? 🤔

[azure bloom]

Probably the wrong channel, but yolo. Do nav mesh volume's update? For example, let's say a bridge is up so the path is inacessible, but the bridge comes down later in the game, can new navmesh update there where the ai can walk on it?

[reef flax]

Heyo, was on a version of UE4 that I have recently taken all the way up to UE5.5, we used to have a slight source modification so that in the RecastRasterization::RasterizeTri function we had a check if a tri was above/below a mesh marked as RC_PROJECT_TO_BOTTOM, there was a limit to the height that the mesh would effect if the new tri could be added. Essentially allowing us to have things like bridges that went over top or beneath meshes with the "Fill Collision Underneath for Navmesh" enabled on the mesh; then only have that affect within a certain height, it looks like a lot of that code has changed now and there is something similar there with a few new variables, I was wondering if this rcSpanData::smin and rcSpanData::smax would allow this now and if they are a value exposed in ini/settings? Currently recompiling source so I have an hour or more of downtime, before I can test but just curios if things had changed to do with that Fill Collision Underneath for Navmesh flag to be a bit more forgiving now?

[reef flax]

It can do if you set your navmesh to rebuild dynamically in your project settings

[tight crag]

What's the best way to setup navigation for a large, world partition, mostly procedurally generated map?

Placing a NavMeshBoundsVolume per LandscapeStreamingProxy?

Like if the Project Titan sample were to have navigation, how would they likely set it up?

[slow bobcat]

Nothing out of the box. You have to track the last location on nav yourself

[slow bobcat]

Check what a nav mesh modifier volume is. You don't need dynamic nav, it will be enough with Dynamic with nav modifiers only (it's like static, no new tiles will be created/destroyed but the existing ones can be modified at runtime)

[slow bobcat]

Unfortunately, nope. You still get the same functionality as in ue4 for fill collision underneath. Quite limited. No idea about the params you ask for, but my bet is that you will need to implement the same changes

[slow bobcat]

I think it does have navigation. I have seen ai and other stuff shown during unreal fest in project Titan. The best way would be to use world partition with dynamic nav generation around nav invokers

[tight crag]

Epic actually didn't include it due to some technical issues otherwise I would just use the same setup.

Bad news: some concerns were raised about NavMesh and performance impact this close to the end of the jam. It's possible we'll still get NPCs, but it's been pushed beyond the end of the jam.

Sorry guys, I tried :frowning:```

The main issue I'm running into is the `NavMeshBoundsVolume` has a max size ( unless you adjust the tile size iirc. )

So just covering the map with a `NavMeshBoundsVolume` and turning `dynamic navmesh` on won't work.

I could just add overlapping `NavMeshesBoundsVolumes` per `LandscapeStreamingProxy` but there isn't a good way to spawn them programmatically so it'd require hand placing them, which is doable but annoying, and makes me think there's probably a better way.

[slow bobcat]

No no, placing a massive volume and dynamic nav is completely unfeasable. Even if you didn't have the volume size limitation, that will kill your performance down to 0 frames.
You could do the programwtically thing within a commandlet or using things like python in unreal to manipulate levels automatically.

But those are "the bad" options. Your best approach is to use World Partition. Open world is the reason it exists

Try World partition with dynamic nav generation around nav invoker actors

[tight crag]

I have a world partition map, just like Project Titan. You still need to have the entire map covered with NavMeshBoundsVolume(s) even with dynamic nav generation / nav invokers.

[slow bobcat]

Ah yeah, sorry. I completely misunderstood the problem. Do you really need nav everywhere in your map?
If that's the case, maybe place a volume per world partition tile so they stream in/out when needed. That should be easy enough to program since each tile is Axis Alligned and with a fixed size

[loud flame]

Hi guys. With the arrival of UE 5.5 I've started to look into state trees. I know that it is a new tool to implement AI, and I know that it supposed to be more flexible than behavior tree, but is there a website or someone who can point out the advantages and disadvantages of using state trees instead of behavior trees?

[slow bobcat]

You can find a bunch of links about it pinned in the channel

I personally like this one, it has a good pros/cons
https://dev.epicgames.com/community/learning/talks-and-demos/yj09/unreal-engine-exploring-the-new-state-tree-for-ai-unreal-fest-gold-coast-2024

[loud flame]

Thank you 🙂

[rigid tulip]

Is there a point in using dynamic obstacle in fully dynamic navnesh?

[slow bobcat]

I would need to check more, but for starters, a dynamic obstacle will not get nav on top. Also with dynamic obstacles you can set AreaClass, which you can't otherwise

[rigid tulip]

Thanks! Another quests- if I want that every actor that affect navigation to rebuild nav if it changes (move, spawns, removed) is there a reason in working with "Dynamic with Modifiers" instead of fully dynamic?

[slow bobcat]

You add a nav modifier component to it. We use that for walls that show up enclosing the boss fight for example. You enter the boss arena, fog walls show up that "cut" the nav with their nav modifier, impeding outside enemies from walking into it. When the fog walls goes away, it takes the nav modifier with it

Same thing for bridges for example

[rigid tulip]

No no I mean:

[slow bobcat]

When the bridge is down, we disable the nav modifier and nav is generated. When the bridge is up, enable it and a hole in nav shows up

[rigid tulip]

If there's any advantage using Dynamic Modifiers Only, if I anyway need every naviagtion actor to rebuil nav

[slow bobcat]

Yeah sorry, misread. But kinda the same answer

The advantage is that dynamic will recalculate the nav and add new tiles if needed

Or remove some

Dynamic with modifiers only will rebuild the affected tiles and only for nav modifiers (whether they are components or volumes in the map)

Much cheaper

Nav with modifiers is baked, then updated. Dynamic is re-calculated every time. That's why they created the Nav Invokers, so you can just rebuild around them

Even though dynamic with modifiers is called that, it can also be called a "static nav that allows you to modified existing tiles with nav modifiers"

[rigid tulip]

But with dynamic I also bake the navigation in the editor, and during gameplay only whenever I change an actor it rebuild the tile, which is the same when I'm working with dynamic with modifiers so I still dont get it

[slow bobcat]

Nope. Dynamic will create the navigation data when loading the game for everything loaded. Then, when you stream in sub levels (or tiles if using world partitioning) containing nav volumes, the nav will be calculated for all new nav volumes

Nothing is baked in dynamic

It's a trait: convenience for cpu

This also applies for nav links btw

[near condor]

Bruno, quick question, seems like you know state trees pretty well, is it just working experience with them that you know them well? Or another reason?

[slow bobcat]

@rigid tulip I'm just having a conversation with a colleague from another company and he's telling me that Dynamic (pure dynamic) in World Partitioning works exactly the same as Dynamic with modifiers before with level streaming: the nav is bakes into nav chunks own by each WP tiles. So you might be right on that

which means Epic took a very weird decision (smart) but weird

[slow bobcat]

I have been working with them for a while now. Our previous game used them (the first version they released) but in this one I use them heavily. That said, I barely use tree params, I don't use linked trees at all (that was not an option when I started this project, so I went with a custom solution)

[near condor]

Understood, I appreciate the response. I really love these over Behavior Trees for my uses.

[slow bobcat]

it's a matter of taste and need. I use both. So far so good

[rigid tulip]

Thanks for verifying!
I noticed 1 different between them: when rebuilding, the fully Dynamic will rebuild everything in the tile while Dynamic with modifiers only rebuild obstacles/modifiers in the tile so it's still better to go with the modifiers if possible

[slow bobcat]

yeah that makes a lot of sense. It's the baked part that was new to me (I think that was introduced recently). Very nice change

[novel pilot]

has anyone here found a way to use "Launch Character" on AI? it works on player characters, but not AI

[misty wharf]

It should work, but if your AI's are actively trying to navigate when you launch them, this can prevent them from launching

[novel pilot]

I've tried stopping logic, stopping movement, setting movement to falling and flying, nothing works

[misty wharf]

Is your CMC set to NavMesh Walking or regular Walking?

[novel pilot]

I believe regular walking

[misty wharf]

Hmm, hard to say then... Maybe try having the AI just jump (by calling Jump on it), and see if that works. If that doesn't work then it could be indicative of some other issues, but if it does jump, then it's kinda hard to say what could be up with it

[novel pilot]

ok I'll try that

[slow bobcat]

I would also suggest to change the movement mode in the CMC to Flying, so the CMC doesn't try to stick it to the floor

[novel pilot]

before or after I launch it?

[misty wharf]

Yeah flying might affect it, but I do recall if you launch it, it should automatically detect it starts flying as it leaves the ground

[slow bobcat]

Ummm could be? I can't remember if the find floor sets flying mode automatically tbh

[slow bobcat]

Ideally before but I would first check what Zomg says

[novel pilot]

`bool UCharacterMovementComponent::HandlePendingLaunch()
{
if (!PendingLaunchVelocity.IsZero() && HasValidData())
{
Velocity = PendingLaunchVelocity;
SetMovementMode(MOVE_Falling);
PendingLaunchVelocity = FVector::ZeroVector;
bForceNextFloorCheck = true;
return true;
}

return false;

}`

yeah it sets movement mode to falling automatically on launch

[misty wharf]

Ah right it was falling, not flying. I think setting it to flying effectively makes it float

So if you were to launch a flying character, it most likely would go up but not come down

[novel pilot]

I'm still not able to get the AI to launch, I can get a player to launch, but not AI

using navmesh

[misty wharf]

have you tried setting your launch velocity to some ridiculously high number?

[novel pilot]

yea

[slow bobcat]

Wait... Is the AI following a path when the launch happens? I think that doesn't work. You need to stop the movement, then launch and, while in lunch, do not process any paths. If not that... Maybe something in physics? The weight or something?

[novel pilot]

I use Task Graph, so the AI is usually either attacking the player or moving to the player

I'm trying to make heavy attacks knock them backwards

[slow bobcat]

Not sure what you mean by "using the task graph" in this context. I have dealt with knowckbacks in 2 ways: adding a push force and/or using root motion

[novel pilot]

I meant the AI is following its task graph (not using blueprint for logic)

[slow bobcat]

Ah ok, so you are using a MoveTo task. Yeah that will not work with launch beacsue the path following component will make sure the AI follows the path and override the rest I think. Take this with a pinch of salt since I'm talking from memory and that's feeble at times. Try to call stop and then launch

[novel pilot]

ok, this makes no sense, but I finally got it to work. I had to call the Launch Character function from inside the AI blueprint instead of from within the player blueprint after getting the ai bp character. It makes no sense, but it works now

I was casting my hit actor to ai bp then calling launch character on the casted valid reference, now I made a separate function that I call instead which simply passes the velocity into launch character

it makes no sense

but it works

[slow bobcat]

Eeeeh.... OK.... Weird yeah. Maybe you check in code what's going on in each of those cases

[misty wharf]

Task Graph for AI sounds like trouble tbh since it works outside of the usual AI mechanisms 🤔

would probably be better to use BT or ST if you want some kind of graph for it

[novel pilot]

is that a trap

[slow bobcat]

Isn't Mufasa just meaning that the AI moves using move to task?

I'm confused with the task graph part. Is that a replacement for Bt's or what are we talking albout? Is that something you can show me @novel pilot?

I'm genuinely curious

[misty wharf]

I assumed he was referring to this https://dev.epicgames.com/documentation/en-us/unreal-engine/tasks-systems-in-unreal-engine

[novel pilot]

I meant behavior tree graph, sorry

[misty wharf]

Ah okay :)

[novel pilot]

got my graphs mixed up

[misty wharf]

You can potentially use the claim resource node to lock ai logic or movement to avoid issues with launching during ai execution

[slow bobcat]

Aaaah OK ok

[novel pilot]

they got em

[wispy nexus]

is there any free tools to use to be able to use a text generating ai in my ue5 game? I want the npc to be able to output text, or better yet text and speech, but I'm not familiar with any tools you can use for that in ue5, and preferrably thats free and run on the local client (this will be for solo player).

[vital heron]

Longshot but I'm wondering if anyone has encountered the same problem : I have a project that uses recast navmesh and have added navigation data to some static meshes that are in my map. They have navarea_obstacle so that a navmesh exists around them but they are very high cost. This works perfectly in the editor but once packaged it doesn't work (it's as if they had the same cost as regular navmesh). Any ideas on how to fix/debug ?

[vital heron]

Ok I found out what the problem was, it has nothing to do with the fact that the project is packaged. I had unchecked "Create on client" and it looks like that also applies to a ListenServer.

[nova fiber]

Hey, for ai piloted spheres to fight enemies would you use pawns or characters for them?

[vernal owl]

Probably character so you can make use of the character movement component.

[coarse smelt]

There is a resource intensive option to turn on Navmesh runtime generation in the ProjectSettings. Then Navigation will update automatically, but it's experimental and will be processor heavy.

[lunar cedar]

if your willing to do some programming work id recommend the pawn class. it was pretty simple to extend the floating pawn movement component to support gravity and orienting the actor to movement so youd end up with a cheaper cost with them. for easy of use/ready to go use the character class

[little mulch]

Is a EnvQueryContext the correct thing to use if I want to query for actors whose mesh is a certain distance from my querier? For example, I have actors that are placed in a level but their mesh moves around relative to their actor location. I only want to "select" the actor if the actor's mesh is < some distance to my character.

[slow bobcat]

You could either do that in the context or in a custom test. If it was my call, I would tag those actors that have that special mesh property and use that tag to filter in the context

[rugged vapor]

Hello,
Is it possible to tweak the time between each loop in the loop decorator of BehaviorTrees ? This is my sequence for my ennemy shoot player. I want the AI to shoot the player until no ammo left but I would like to be able to tweak the fire rate

[misty wharf]

iirc only tasks really can choose how long executing something (itself) will take

so the solution here would probably be either to make your attack task not end until the next attack is valid, or add another node which waits until the next attack can be taken

[rugged vapor]

ok i see thank you

[vernal owl]

If I want to unset a blackboard value that is set by the OnTargetPerceptionUpdate event, is it best to just set "bForgetStaleActors = true" in UAISystem? I can't see any other way to make the AIController forget the target value once that target leaves its line of sight.

[nova fiber]

should i use the character movement or floating pawn movement for a hovering ai piloted orb that will move and attack enemies? orb doesnt have to move up or down.

[ebon shore]

Looking for a tutorial on making an enemy ai pace back and forth like a koopa in mario - anyone have a good rec

[stable plover]

just look up ai patrol

its the same thing as patrolling

if you dont like that approach you could basically move in a direction and when you hit something do a dot product to turn the other way and repeat

[stable plover]

well i think the forget stale actors only fires when you forget everything about an actor, for my use case I wanted to forget individual senses and that function couldnt cut it so, what i made was a timer that basically works like this, when you see a target first clear the sight forget timer, then make a timer called sight forget(same timer just clearing it before we make it) set it to execute in 5 secconds or whatever time you want to fire the forget event. inside that what i do is try to see if i can see the target again then if we cant see anything i try to simulate an auto suiccess range with a sphere overlap, for example if the ttarget is just around the corner i may not want to forget about him

in the third picture the false on the branch indicates that the target isnt in the success range and we cant see him

[nova fiber]

the behavior tree / blackboard does get the correct target as varriable value but the sphere (Character) is not moving at all

[loud flame]

it seems the sphere creates a hole in the navmesh. I would suggest to check colliders on sphere and uncheck "Can affect navigation"

[nova fiber]

thanks this seems to be it, it now moves 😄

[stable plover]

make sure your affect navigation option is off that will most likely create that issue

[abstract vessel]

Hello, I’m trying to get the Behavior Tree Quick Start setup, and I have an issue where the enemy character doesn’t move at all. This includes when it has line of sight on me and should chase me as well as when it is supposed to randomly patrol. In the behavior tree this appears as it skipping the chase player sequence entirely, and very briefly flashing through the patrol tasks before waiting the duration. It then rinses and repeats.

Worth noting, I did redo a good part of this because originally I was using my own custom enemy character. But when it wasn’t moving I figured I would use the 3rd person character like suggested in the guide just in case that was the reason. After switching a lot of it over, the not moving issue obviously still persisted. But, I say this just in case I didn’t switch something over properly when redoing the setup and maybe something is still referencing the old character. I checked thoroughly and couldn’t find it, but, worth pointing out. The current enemy is called “Hatman_Character” while the old one was just “Hatman”.

I ran a debug by using breakpoints and found that the chase player task never even triggers the initial “Event Receive Excutive A.I.”. When I step in front of it, however, the “Has Line of Sight” is triggered in the A.I. Controller.

On top of this, the enemy doesn’t move to new patrol locations. I ran some debugging on this and found that the Find Random Patrol Task runs all the way through the last couple of nodes.

Attached are some screenshots to help show what I am talking about.

[coarse smelt]

Suppose that I have the following setup...(see attached image)

From what I understand, this Event will get fired whenever any Stimuli is sensed, and this includes

• Start Seeing ActorA
• Start Hearing ActorB
• Stop Seeing ActorA
  So if things happened in that order, the actor set as PerceivedActor would be ActorA, even though I've stopped seeing A. Is that correct?

If I didn't want to set PerceivedActor on a "stop seeing" stimuli event, how could I do that? Is that what the "Successfully Sensed" bool is for? i.e. does this return false if I get a "stopped seeing" stimulus

Are there other kinds of Stimulus that can be recieved aside from "start sensing" and "stop sensing"?

Lastly, is there a way of setting the "Tag" of a stimulus in Blueprints?

[misty wharf]

Your understanding is correct. successfully sensed will be false when sight is lost, so you can use that to choose what to do with the information. As far as I know there are no other builtin senses that use successfully sensed as false. The tag function is possible to use with the hearing sense, since the report function for it takes a tag parameter. As far as I know it's the only one.

[coarse smelt]

Thank you!

[vernal owl]

Interesting. I ended up just setting bools for each sense and then branches after the "Perception Forgotten" event that lead to clearing the appropriate blackboard keys. For example, if the target was spotted but not heard, it sets my blackboard key to the target and bSpotted = true. bSpotted is checked after "Perception Forgotten" fires and the "true" branch leads to a "Clear Value" node that clears my blackboard key; bSpotted is then set to "false". Seems to all work as I need it to right now but I guess I'll see if bugs start cropping up.

[stable plover]

if it works good for you then dont worry about it, im using state trees and the process is a little different

[vernal owl]

My AI with EQS works nicely for areas with obstacles but the same AI will have very plain and unnatural looking patrol behaviour if in an open area. Is there a good way to add randomness to when an EQS picks a winner or is it normal to just swap out the AI for a different one that uses the "get random location" BP method if you know an NPC won't have to deal with obstacles?

[stable plover]

when you run the query you can set the arguments for the scoring

like best of 25%

bet of 5%

etc

what i do in state trees is have a task that runs a query then bind on finished to it, get the locations then do a random on them

[vernal owl]

Yeah, I've tried altering scores but it seemed to always pick the same kind of result every time. I haven't looked in to state trees yet though, they somehow haven't come up until now.

[stable plover]

just create a custom task in the behavior tree that calls run eqs query

then bind the finished delegate to it

then call get results as locations and get a random location

[vernal owl]

Will probably do that then and look at state trees too.

[stable plover]

i like state trees tbh, the way that they are structured allows for better performance, you can do so much with them but the downside is that they dont have many features as BTs do so you will have to get clever and implement your own little additions, for example state trees doesnt really have a storage system like blackboards, you can get around this with a unique approach but you can’t give eqs queries property bindings to my knowledge

[misty wharf]

The fact ST's tick makes me wonder how much of an impact that will have 🤔

[heady silo]

Getting a weird error with my navmesh. I have an object with accurate collisions that is absolutely walkable by the player, but the navmesh refuses to generate within its bounds except for specific areas. Adjusting step height and slope heights don't appear to affect anything. I tried additional volumes and a modifier volume but nothing seems to work. Do I need to regenerate collisions for the mesh or something?

[slow bobcat]

Looks like a problem with that ground/asset. Have you tried checking the "can ever affect navigation" flag, check if there's a nav modifier affecting it etc?

[heady silo]

turning off "can ever affect navigation" does this

collision for reference

god rid of the modifier, it didn't change the mesh at all

the combined fuel silo has a similar problem, though at least there is a path across it

would reducing the grid size help?

It seems to have helped with the combined area

ngl wish there was a way to manually paint on a navmesh like counter strike or half life

I'd rather it worked than have it be shit and take less time

[harsh storm]

You could always build that tooling.

[heady silo]

that would imply I have any degree of skill at programming

[ancient wedge]

How do you guys usually handle ocean navmesh in your games?

I currently have a nav modifier on the ocean level. But if I want navigable areas below ocean level it also blocks it. :/

[slow bobcat]

You mean as in having navigation on the floor and on the bottom at the same time?

[slow bobcat]

Pretty sure that floor has some weird settings. Have you tried placing a plane or boxes or soemthing on top? Do they get nav as the stairs do?

[ancient wedge]

Yes, pretty much. So if I have a building that goes below the ocean level it should be navigable.

Since I'm working with an island, I would love to only have the water nav mesh where the ocean is.

[heady silo]

I added a box, it maps over the box fine

since it's just a test anyway I'll just roll with this

[heady silo]

thanks mate

although even with the navmesh fixed up, none of my agents are even bothering to move

maybe it's all the assets in that pack

shit's weird

[slow bobcat]

Try with visual logger, you might get some extra info. As a general advice, never trust anything you buy from the asset store

[rugged vapor]

hello,
Do you know how i could modify this behavior tree to make the AI shoot 3-6 bullets, then stop for like 1 seconds, and shoot again ? Right now it just shoots continuously and I cant picture how to do this

the first wait in the first sequence is a delay between aiming at the player, and begin shoot.

[misty wharf]

I would consider implementing it as part of the attack

[silk fulcrum]

Did you ever find a fix fro this?

[rugged vapor]

so after your comment i tried to move the fire rate into the BTT_DefaultAttack but for a reason the ennemy only shoot once and stay in the task without doing anything. I dont understand, isnt Sequence supposed to keep playing it until it fails ?

[stable plover]

both have their upsides and downsides but I would say that state trees gives you more flexibility rather than the sequence and selector approach, im sure you have tried state trees and know what i mean

[wise yarrow]

Question, I am creating my own decision tree using C++. However, in my own AI i use entropy, alpha ,split values (to choose left or ride node) and features as a way of online supervised learning. To my knowledge UE doesn't incorporate such a thing in their decision trees, am I re-inventing the wheel or am I expanding on something here? The reason why I'm using real machine learning techniques is so that the decision trees aren't strictly rule based, i.e., "if this happens, do this"

[vernal owl]

If I have multiple NPCs that use the same AI controller and Behaviour Tree, how do I specify which NPC I want the Behaviour Tree live visualisation to track when PIE is turned on?

[rose canyon]

F (great question want to follow)

[vernal owl]

Thanks, got that to work when already running in PIE but is there a way for it default to a specific NPC before running?

Actually, it seems to show the NPC I need on the Behaviour Tree when I just open the debugger.

[heady silo]

when running PIE there's a drop down menu to pick from agents using the tree

[heady silo]

this I don't know, and I'd dearly like to

[zenith hill]

Anyone actually implememt butter smooth detour crowd?

Mine jitter sometimes

When it's about to reach the destination

Or in narrow spaces

I'm at the brink of just disabling A.I collision.

[slow bobcat]

If you select the AI in the outliner, it will show you that one. If your AI is placed on the scene, maybe you can select it before playing? If not... Maybe there's some functionality to select it programatically?

[zenith hill]

is RVO avoidance not working for player character?
My A.I avoid each other but not the Player character

for context, both the player and the A.I have RVO bool ticked

[solid wedge]

Player character isnt possessed by the same parent AI Controller at that time. I'd assume you are SoL, or try the "affects navigation" settings in char move component? Guessing

I always have trouble sending my AI to a world space location, they always run to 0, 0, 0. But invisible actors work perfectly. Is the default AI Controller intended to nav to actors instead of world space locations? (In my case, i'm using the basic MoveTo function, and passinga vector BB Key, instead of an actor key.)

I for sure have some data mamagement issues, as it works half of the time. But using empty invisible actors feels like I could implement / iterate much faster.

[misty wharf]

Moving to location should work just fine with vectors. Try manually using project point to navigation before using it as a move target and see if that helps

[mild spire]

Anytime my navmesh is updated, my pawns stutter/slow during it's rebuild. This happens even when it's far away using my NavigationInvoker component attached to my player.

How is the NavigationInvoker supposed to work if this always happens when moving around?

[stable plover]

What’s your nav invoked tile properties?

[nimble smelt]

My AI pawns can "see" and target actors that they shouldn't be able to see, even purely procedural pawns like the anchor that gets moved around when when the player scrolls the world map. No element of these things should be interactable to any other pawn, how do I stop the AI from interacting with them?

[slow bobcat]

Bit confused. No idea what you mean by the anchord etc. Is the problem that your AI's can perceive all this actors through their sight sense?

Are all those actors pawns?

[nimble smelt]

Yeah, so my map screen is actually a scenecapture with wild filters on it, attached to a pawn that moves around with the map controls like pan or zoom.

It's actually a pretty common way to do it. Skyrim's map works like that.

Problem is, players could strategically position that map pawn to distract otherwise-hostile AI

It's obvious not registered as a stimulus source, but it needs to have a collider (so it knows what room it's in and can populate info based on that)

[slow bobcat]

Pawns are automatically registered as stimulus sources. If you want it to not be a stimulus source, you have to manually un-register it. Maybe that's the problem?

[slow bobcat]

you mentioned this It's obvious not registered as a stimulus source but not sure if you actively unregistered it or you asume it's not since you didn't manually registered it

[keen crow]

It seems there are 2 dozens or parameters in UCrowdFollowingComponent, but I can't find any example or explanation for use cases of those. Can anyone suggest me an article/video to look at?

I mean I can guess what they are for from the names, but what I would like to find is something like "check this out, if you set these to true, these to false, then NPCs would walk this way, and if you do Y, then they will do Z, but if W then D" and so on

[slow bobcat]

#gameplay-ai message

Bit outdated but a good start

[slow bobcat]

This you will have to do yourself and document. I haven't seen anything detailed about it

[low stump]

Is there a way to get all behavior tree nodes/tasks of a type?

[misty wharf]

Get from where?

[low stump]

I have a behaviortree reference in an asset action utility blueprint

If I drag from the behaviortree reference I can't find anything about getting its tasks

[misty wharf]

🤔

Kind of sounds like something you might need to do in C++ to be able to pull out that kind of data from it if it doesn't expose it

Iirc there is a node you can use to pull properties from objects in editor utilities even if they're not directly BP-exposed, but you would still need to know what the property is called

[slow bobcat]

What are you trying to do?

[ocean wren]

Why not get the recast & detour codebase and look at how it works there? I think Epic kind of fudged it when they got a junior to implement it or some such 🙂

Recast was created by Mikko Mononen back when he was at Crytek.. there was a github with a maintainer last I recall

But it had a load of demos and stuff in his library codebase

[misty wharf]

I've looked at those and they're not exactly easy to understand either

Forget what I was looking for at the time though :D

Oh right, it might've been that I was wondering if there was some way to make it so the navigation system would only navigate in the cardinal and ordinal compass directions

[slow bobcat]

I'm knee deep in that code today for the Nth time in my life. It's never easy to read and understand.

[nimble smelt]

Oh shit that's it for sure! How do you deregister a pawn as a stimulus source?

[harsh storm]

There is a config file you have to alter

I forget what it is

By default though, all pawns get registered as a source.

DefaultGame.ini
[/Script/AIModule.AISense_Sight]
bAutoRegisterAllPawnsAsSources=false

[nimble smelt]

Huh, I've added that but it's still getting seen

[harsh storm]

Did you restart the engine?

[nimble smelt]

Yeah

Maybe there's something funky in how I'm using these? First I get all known perceived actors (Awarness level by actor is the output of previous run) then I iterate through current perception info by sense and find the strongest to use. So if it cant perceive this actor it shouldn't ever be on the list, and if it can't see the actor there's no way sight gets onto the list. Yet it's there.

[nimble smelt]

FUCK i put it in the wrong config file lol

Ignore me

[vernal owl]

Would anyone happen to know how, in Lyra, the AI Controller (B_AI_Controller_LyraShooter) is able to detect damage sources? I can see the controller has AISense_Damage implemented on the AIPerception component but I can't seem to find anything with both AIPerceptionStimuliSource and AISense_Damage actually registered. I checked the character BP and that just has AISense Sign and Hearing stimuli sources.

[wicked surge]

I have been following this tutorial here https://dev.epicgames.com/community/learning/tutorials/lwnR/unreal-engine-your-first-60-minutes-with-statetree

My GetRandomLocation task keeps runing non stop. I have all the arrows on the state tree pointing the exact same way.

In the tutorial he is saying it should only do it once and done. But he returns everything back to the root and his root is set to run all children in order. So it does sound like it would just run forever. Makes NO SENSE

[supple hare]

in the ai controller how does the get blackboard node know which blackboard asset to use?

i know there is a run behavior tree, does it just use the blackboard from whatever behavior is currently running? What if you are running multiple behaviors on the same ai controller

[stable plover]

Do you ever finish the task?

That sound like behavior from forgetting to finish a task

[slow bobcat]

Damage works by sending a damage event to the damaged character. Look for that, there's probably somewhere where they send the event. I never looked at the Lyra project but I would imagine they have some logic for the shooting, either with a projectile class that handles the damage or with a line trace or similar approach upon shooting

[slow bobcat]

• each ai controller runs 1 behavior tree instance
• each behavior tree instance has one blackboard associated that you defined in the behavior tree asset
• you can't run more than 1 bt per controller at once. It simply not possible out of the box

[hollow dock]

my ai 5v5 offline multiplayer in progress

[vernal owl]

Found it and the "Report Damage Event" code. It's on "B Shooter Hero Mannequin". I was looking at the wrong character BPs like "B Hero Default".

[wicked surge]

I tried finishin the task and same thing as well so I decided to leave it exactly how shown in the guide

[hallow compass]

it seems that my AI isnt using its own nav agent radius since it tries to pass though small areas

if i edit the agent radius param on the navmeshraycast actor it works

[slow bobcat]

If you delete the recast object and build nav again, do you get the same nav? Do you have more than one agent specified?

[hallow compass]

whatever i do with these ill have two raycast (one human and one rat), and neither working on PIE, and if i try to display the tiles i get nothing

[slow bobcat]

Ok so:

• if you have more than one agent defined in your project settings, each agent will generate a set of nav data.
• unless things changed in the past few years, I think the editor can't show one or the other. I remember to have to hack ue4 to be able to show the different nav meshes
• when you have several nav datas, you need to pass the querier to the move request, otherwise the engine will not know which nav data to use and will grab the defualt

I would delete one of the agents from the settings and leave just one that matches that AI capsule radius. Then generate nav and try.

[hallow compass]

well maybe the issue is because the AI capsule size is small, but im scaling the actor in the level

should i directly scale the mesh and up the capsule size in the BP ?

[slow bobcat]

Yeah, that will not work. When you specify a radius in a supported agent with a the nav settings, what you are saying is "build a nav where a capsule up to this size will fit". If then you use a bigger capsule, that will not work. The AI assumes that the nav data used to calculate its path is ready for a capsule matching its size (at least)

[hallow compass]

well i increase the size directly in my BP, so the capusle and nav agent data for my AI character MATCHES the supported type agent i added in the nav settings

but i still cant see the nav

i deleted all nav actors and added a new one with the only one possible supported type

every time it added the Raycast-Human actor, some properties looks saved, because stuff like "enable drawing" is ticked.
can it be a serializeation issue ?

looks like i had to update Nav Mesh Resolution Params

[slow bobcat]

Debug nav is famously buggy in unreal. Wouldn't surprise me if you run into a visualization bug too

[hallow compass]

now i have the Navmesh need to be rebuilt error message x)

while im using a single player that is a listen server

also, the AI debugger says the AI "is to far from navmesh"

[slow bobcat]

Yeah that could be. If your capsule is very high, the center would be further than the projection extension used by the CMC to navigate. You will need to tweak that (can't remember where)

[hallow compass]

I tried to raise everything one by one and didnt solved it. Some params description are vague

Ill try more

[hallow compass]

setting Nav Agent Step Height to 999 makes my AI walk again, idk why since the floor is flat

either way it doesnt take into accaunt the capsule radius

[slow bobcat]

That has nothing to do. You need to look on code for the projection extent

[hallow compass]

its probably not high enough

[slow bobcat]

Look for Project Point on Nav function in code (or the node in BP). You will see you can pass a projection extent. That means "how far from the point should I look for to find the nav"

If yor point is at 0,0,60 but your extension is 0,0,50, it will fail. You are 10 units too far

That same logic happens when things request for a path

You capsule center is further away from the nav than the Z component on the projection extent

That is something you can tweak and pass to the movement query in code

There's probably some para in dome settings on config file you can tweak for it, but I don't know it from memory.

[hallow compass]

i guess that doesnt mean shit

[slow bobcat]

Nah that has nothing to do

That's to use that param to match the capsule size with the best fit among your different nav datas (for you that's human and rat)

There should be a default query extent in the navigation settings of the project I think

Tweaking that (if that actually exist and I'm not hallucinating like bad AI) should work

[hallow compass]

yeah i increased it

thats my human

[slow bobcat]

And you still get the "point to far from nav"?

What's your AI capsule size?

[hallow compass]

Nav Mesh Resolution Params

i think my cells were to small

[hallow compass]

with increased cell size it now walks

[slow bobcat]

That's weird. That only matter for how the nav is generated. If you have nav below the AI, resolution doesn't matter, there's nav and that's all that matter. I'm not sure why that works if the nav was already generated with the previous params