gameplay-ai

[slow bobcat]

My guess is that your points are too far from the nav for the projection extension you are using in your move to node

On result changed: if the result is different from the previous value. Ie: before it was false, now it's true
On value changed: whenever the value changes, even if it changes to the same value as before. Ie true and now true again

[winter moon]

Is there a good approach to make a state tree state have a random chance? Right now my state tree selects in order, but I want one of the states to have a random chance of being selected - I havent found anything that is a good options for what Im looking for other than a custom solution

[slow bobcat]

Maybe through utility?

[slow bobcat]

If not that, you could have a task generating a rnd number in the parent state and have conditions with ranges? Like a rnd between 0 and 1 and conditions like > 0.2, > 0.5, >0.8

[winter moon]

That would basically be a custom solution

You would think this would already be built in functionality, but im assuming they want you to implement it on your own or use weights

[slow bobcat]

Yeah that's why I mentuined Utility. It has a built in utility system. It's a matter of handling the weights

Not sure if 5.6 has something else. It now has ticking management for tasks, options for how tasks completion should be considered etc

[harsh storm]

I'd imagine this option would work. Just have all the utility scores the same.

Or this one

[slow bobcat]

Is that new or didn't I notice it in 5.5.4?

[harsh storm]

I believe it has been there for awhile. But can't confirm. I was team BT for forever.

[royal hemlock]

#gameplay-ai message . Hi, I have a similar issue. It seems Parent.GetThing is not available of child to evaluate? and it force BOTH condition to be false? (the original logic is A = true, b = false). Isn't parent active when it entering its children?

[woven bobcat]

hey folks so update, the only way i can think of if just create abilitytask/abilityasync foer this as there no way of doing this for the chjaracter.. make sure to change the AAIController to APlayerController.. im currently writing abilityasync and abilitytask based on UAITaskUseGameplayBehaviourWithSmartObject

im using GAS so i need to create abilitytask/abilityasync

if anyone wondering tyo use smartobject with playercharacter instead aicharacter..

i heard from twitter and some messages from here that there going to be playercharacter version of this but its coming in the future. no ETA. they have this working for lego fortnite just not in engine yet

so make tasks based on UAITaskUseGameplayBehaviourWithSmartObject with abilitytask/abilityasync

i'll update when i finish writing my version

[obtuse dome]

Is there any way to integrate navmesh with terrain layers or physical materials so I can paint on unnavigable areas to the landscape rather than manually making a bunch of navmodifiervolumes over the river?

[slow bobcat]

Are you using static or Dynamic with modifiers only by any chance? Also, is it a c++ project?

[obtuse dome]

Dynamically generated, and yes I have access to C++

[slow bobcat]

Umm dynamically might make tings expensive but you can post process the generated nav, detect the material on the ground (with a trace), if an unwalkable material, detect the nearest poly in nav (find nearest poly is a function you can use) and mark the poly with a flag 0 in both poly and area

I do something similar in our project but we use dynamic with modifiers only, which allows us to do that post process during cooking of the build. I could imagine it's quite expensive in runtime but maybe there are workarounds to skip unmodofied tiles etc

[obtuse dome]

Is there a cheaper way if the navmesh is dynamically updated for player buildings, but the obstacles like rivers I want to block are known statically?

I've blocked them off with modifiervolumes at edit time for now & was mostly interested in painting on the modifiers for ease of iteration, but if there is a big perf hit I guess I can just eat the cost to iteration time

[slow bobcat]

Maybe? You should be able to generate nav chunks pero sub level (or World partition cell) and load them? Never mixed nav concepts like that

[slow bobcat]

I asked a friend and he told me that they do have baked data in the nav chunks of the cells (WP) and then they generate nav dinamically around agents

[woven bobcat]

update: its workss now the player can claim the slot with my custom UseGameplayBehaviourWithSmartObjects abilitytask 🥳 its working.. now just need to cleanup it a bit..

works on ym bench if a player claim the slot

tested if an ai already claim the slot, the player cant claim it

now , want to replicate ho movetoand usegameplaybehaviourwithsmartobject since i want the player to move to the locaton.. current implementation doesnt do that

[static crystal]

I'm trying to make some custom state tree evaluators in cpp but I can't get them to show up in the editor. is there something I'm missing? The tasks and conditions I create show up

USTRUCT(DisplayName="Perception Checker")
struct FSTE_PerceptionEvaluator : public FStateTreeEvaluatorBase
{
    GENERATED_BODY()

    using FInstanceDataType = FPerceptionEvaluatorInstanceData;

    /** External context actor (the actor with the AI perception component) */
    TStateTreeExternalDataHandle<AActor> ActorHandle;

    //~ FStateTreeEvaluatorBase interface ------

    virtual const UStruct* GetInstanceDataType() const override
    {
        return FInstanceDataType::StaticStruct();
    }

    virtual bool Link(FStateTreeLinker& Linker) override;
    virtual void TreeStart(FStateTreeExecutionContext& Context) const override;
    virtual void TreeStop(FStateTreeExecutionContext& Context) const override;
    virtual void Tick(FStateTreeExecutionContext& Context, const float DeltaTime) const override;

protected:
// rest of function

am I meant to subclass the UStateTreeEvaluatorBlueprintBase instead?

[slow bobcat]

Use global tasks. Evaluators will be deprecated

[static crystal]

wait so what's the difference? or is there not much hence the depreciation?

and where did you see that they'd be deprecated?

[static crystal]

I got the evaluator working by subclassing UStateTreeEvaluatorBlueprintBase but I can't bind to the outputs.
it works fine when making a blueprint evaluator. but I can't get the cpp one to work

UCLASS(Blueprintable, BlueprintType, meta = (DisplayName = "AI Perception Evaluator"))
class USTE_PerceptionEvaluator : public UStateTreeEvaluatorBlueprintBase
{
    GENERATED_BODY()

public:
    USTE_PerceptionEvaluator(const FObjectInitializer& ObjectInitializer = FObjectInitializer::Get());

protected:
    /** Currently targeted actor (output) */
    UPROPERTY(EditAnywhere, BlueprintReadWrite, Category = "Output")
    TObjectPtr<AActor> TargetActor;

[warped tiger]

Hi, does anyone know how can I change StateTree parameters at runtime?

[slow bobcat]

Commented several times here, originally stated by epic members in the chat

[slow bobcat]

Consistency. You make taks for everything, then some run globally (before states are evaluated) to achieve what your did before with evaluators.

[static crystal]

oh I think the properties needed to be public. they work now

[slow bobcat]

Is the pointer to bind in your taks also a T Object Pointer?

[slow bobcat]

Ah yeah, that too

[static crystal]

fair enough. I'll swap mine to tasks

ty for the help

[slow bobcat]

State tree param refs (some example in the chat I think and online) or binding to global taks that modify them

FStateTreePropertyRef

[static crystal]

interesting that enter conditions can bind to parent task fields

[slow bobcat]

And to values passed through the payload of a State tree event.

[static crystal]

yeah I've set up my perception component and Ability system components to fire off state tree events

I really like listening to using events/messages rather than hard coding connections

its nice that unreal is moving towards that

I'm remaking all my ai with state tree stuff. I've never liked the BT visual scripting

[warped tiger]

Thanks! It doesn't work with structs, right?

[static crystal]

if I want to have a time limit on a state. Should add a delay task, and have "any" task complete the group. Or should I add a transition on tick, with 10 second delay?

[slow bobcat]

No idea mate. Never used them

[warped tiger]

Nevermind, I was doing something wrong. It works with structs and the header file states so in a comment (StateTreePropertyRef.h):
- Supported types are: bool, byte, int32, int64, float, double, Name, String, Text, UObject pointers, and structs

Also found another way of doing it without FStateTreePropertyRef. I'll post it just in case someone stumbles across this problem in the future and can't use FStateTreePropertyRef, which I think it was introduced in 5.4. Source is: https://forums.unrealengine.com/t/modifying-statetree-parameters-during-runtime-in-unreal-engine-5/1216696

template <typename T>
EPropertyBagResult SetStateTreeReferenceParameter(FStateTreeReference& StateTreeReference, const FName ValueName, const T& Value)
{
    auto& InstancedPropertyBag = StateTreeReference.GetMutableParameters();

    const auto PropertyBagResult = InstancedPropertyBag.SetValueStruct(ValueName, Value);
    ensure(PropertyBagResult == EPropertyBagResult::Success);

    // set overriden flag
    if (auto* Desc = InstancedPropertyBag.FindPropertyDescByName(ValueName))
    {
        StateTreeReference.SetPropertyOverridden(Desc->ID, true);
    }
    
    return PropertyBagResult;
}

[minor solar]

Does anyone know why my zombies lose interest in the player when there is a crowd of zombies surrounding the player? It will attack forever if the zombie is alone although. Is there any way to ignore the other zombies? They clearly see the player but for some reason switch to patrol as if they cant? I am using behavior tree by the way and ai perception

[mental knot]

hello! I have a big AI and i made a larger nav mesh agent to accomodate it, but I'm encountering an issue where if my player stands near an obstacle, the AIMoveTo will just entirely fail (due to player standing on no nav mesh). Is there a way to instead move as close as possible? Instead of entirely failing?

[slow bobcat]

@minor solar
If you look in your video, the AI's are blocking each others rays for sight. Notice how the age of the perceived actor starts counting and then they loose interest. You should make. Changes in code so they ignore each other (use a physics profile that ignores other zombies for your sight trace or add the other zombies to the array of Ignored Actors). Another solution is to propagate the target: if a zombie sees or targets the player, any zombie in radius of X meters gets the same target. You could also create another sense that simply checks if the player is in radius of X meters. It they are attacking the player and close enough, that sense will always detect the player. Make sure you don't set the radiua too big or thyw will detect the player behind walls etc.

[slow bobcat]

There's a function to find the closest point on nav. Upon failing move to, you could use that with the player location and a extent that is at least as big as the agent radius used to generate nav

[mental knot]

thanks! i think i found it. Would I use this like

putting a selector between my current AIMoveTo (target actor) and a new AIMoveTo (target location determined by ProjectPointToNavigation)?

[slow bobcat]

Yeah, that's sounds reasonable.

[minor solar]

Thanks! I'll try that!

[final prism]

What's the best way to pass data out of state trees? I understand it's meant to be quite encapsulated, but there must be some feature to pass data outwards?

[royal hemlock]

I usually put it in task and access MyActor->GetComponent<UMyClass>()->SetMyValue(InValue);. In C++ you can acess GetMutableParameters, but I don't think it is the best way though.

[final prism]

Yeah that's how I've been doing it too, though once state trees move to being threaded accessing an actor directly like this would probably be unsafe (plus having a single way to do this might be nicer)

[slow bobcat]

a friend sent me this.
I think this plugin might help a lot if you can define your river with a mesh or a spline
https://www.fab.com/listings/c20533c5-8f2c-4718-b801-350aef942c9d

[mental knot]

in case anyone is looking for reference to this issue

I had to create this BTTask in C++ (in BP you can't get nav data for a particular agent): https://pastecode.io/s/emdbh0we
then I replaced a regular MoveTo with whats in the image attached

Note: for this to work, you still can't have an AI with a capsule that exceeds the max agent radius/height. AI will still get stuck. This mainly fixes the issue of a big AI failing to move to a target when the target is near an obstacle

[royal hemlock]

Hi, anyone has an working example of setting up linked asset (LinkedStateTreeOverrides) in statetree properly? the context is invalid when entering the linked tree.

[simple crescent]

Are you using 5.6? I think the condition is incorrect in UStateTreeComponent::AddLinkedStateTreeOverrides. I think it should exit if nullptr or not IsChildOf(UStateTreeComponentSchema)

It looks like there was a fix submitted last week in ue5-main
https://github.com/EpicGames/UnrealEngine/commit/f602e71240d38286955b13232cdbb436f3186ada

[pastel moth]

Do delay tasks actually work in state trees?

It seems like when I add a delay task it never fires off

[hasty turret]

They work,
but in 5.5, a delay task may be ignored if another task finishes first.

[broken vine]

How can i get my BTS_FindPlayer to always fire every ~1second. If it is in the middle of moving to the location it wont fire until it reaches that. I want it to check if there is a different player closer and then switch directions

[misty wharf]

How did you determine it isn't firing?

The service is independent of tasks so looking at it, it seems it should fire roughly once a second since you have it ticking at 0.9 and 1.1

My guess looking at this is that it does fire. But your task will not update anything within it if it's already in progress, and your decorator doesn't seem to be set to abort anything so even if the TargetActor value changes, nothing will happen until the task finishes.

[broken vine]

It wasn't swapping players as I ran past which made the enemy closer to them. Also watching it while playing it wouldn't light up as if it were running that service.

Ah right, at the start of the service I clear the blackboard value so I thought that would make it abort as its no longer set

[royal hemlock]

What does the little diamond mean exactly in statetree debug?

[slow bobcat]

Marks between states

It think?

[woven bobcat]

hey folks anyone here use state tee, i have a dialouge system in my state tree. it goes like this:
Player Interact with the StateTree->Play the state tree-> in the dialouge,have options to play minigame->play minigame->if we done playing the minigame->play another dialouge->dont want to play minigame-> -end

now how can the possessed minigame actor, when its end, when i repossess back to the player, it play the state tree but it goes to the next dialouge in the tree

i have three stages where in this dialouge, where it asked if you want to play minigame

in my minigame, i have an event that call send gameplayevent to actor when it end which call gameplayability to repossess back to the player character

BUT i want it to go to the next dialouge

since after the minigame end, it should go the next dialouge

currently, after the minigame end,we just repossess the character and the dialouge is lost

[woven bobcat]

found soemthing called FStateTreeEvent and have a node inside statetreecomponent called SetStateTreeEvent

To listen to State Tree state events in Unreal Engine 5, you need to use the "Send State Tree Event" node within a Blueprint or C++ class, and then create an event listener within the State Tree asset that corresponds to the event tag. This setup allows for communication between external actors and the State Tree, triggering state transitions or actions based on events.
Detailed Steps:
1. Create the Event:
In a Blueprint (e.g., an actor) or C++ class, use the Send State Tree Event node. 
This node requires a StateTree component reference, an event tag, and an optional origin. 
The event tag identifies the specific event being sent.
The origin can be used to pass additional data with the event.
2. Create the Event Listener in State Tree:
Open your State Tree asset.
Create a new Event transition within the desired state.
Set the Event Tag to match the tag used in the Send State Tree Event node. 
Define the transition's behavior (e.g., transition to a new state or execute a task). 
3. Example:
In the example provided by Epic Games Developers, a "Hit Reaction" event is sent from a Blueprint actor to a State Tree.
The State Tree has an Event transition with the tag "Hit Reaction". 
When the Send State Tree Event node is triggered, the State Tree detects the event and can then trigger the corresponding transition.
4. Additional Notes:
You can use multiple Event transitions within a state, each listening for different event tags. 
You can debug State Tree events using the State Tree debugger (see the Epic Games Developers documentation). 
You can also use Blackboard variables to communicate data between the State Tree and other actors. 

from google ai

this is probably bullshit but maybe the statetreeevent is what i need to use after i end the minigame, when repossessback, trigger the dialouge again but at the state it was before?

i dont know how to do this ngl or if this event the right solution

[slow bobcat]

You can have event based transitions in the Root node to any node in your tree. Send the event needed to force the tree to any state. Since the root always run, your events will always be considered

[woven bobcat]

is it using sendstatetreeevent?

[slow bobcat]

Yeah

[woven bobcat]

i'll try that

[slow bobcat]

• in root add a transition On Event
• in the Required event fill up the tag section and the "transition to"
• in the state to transition to set the "required event to enter" flag in the Conditions if you only want that state to run because the event happened

[woven bobcat]

so on repossesion after the minigame end, how do i send the sendstatetreeevent node? in my current setup, after repossessiong, the dialouge is lost and you kinda have to interact with the interactable again

am i misunderstanding?

this is bad practive but i asked Chatgpt for idea on how to approach this:

To handle your StateTree dialogue flow correctly when switching between player character and minigame actor, you need a way to preserve and update the StateTree's state so that upon finishing the minigame and repossessing the player, the dialogue advances properly.

Here's a general approach to achieve this:

1. Preserve StateTree Context Before Playing Minigame
Before you possess the minigame actor, save the current state or the next dialogue node to continue from after the minigame ends. You can save this information in:

The player character,
A globally accessible game state or manager,
Or in custom StateTree context properties.
2. Trigger Minigame and Signal Completion
When the minigame starts, you possess the minigame actor as you currently do. Upon completing the minigame, send a gameplay event back to the player character or the controlling system.

3. On Completion: Repossess Player and Restore StateTree
When you receive the event signaling minigame completion, repossess the player character and restore the dialogue StateTree's saved state to ensure it starts at the next dialogue.

the idea that it give is what i want to do technically but just the implementation i kinda dont know if that the right way.. might try the statetreevent thingy

[slow bobcat]

Not sure what you mean by all this. Hard to follow the logic without seeing it.
Let see if I get this right:

• you control the player
• player interacts with some object
• a dialog opens asking if the player wants to play a mini game (this is your state tree)
• the player says yes, you un posses the player, posses the mini game and plays
• mini game ends and re possess the player

And then? The dialogue is not here anymore or what's the issue?

[woven bobcat]

yes, after minigame end, after repossession, it should open the dialog again but play the next dialouge instead of closing entirely

[slow bobcat]

Ah ok... The interaction is lost...
Inguess you could keep somewhere a value to check if you need to re-engage with the interactable object.

Keep track of that to Re open the interaction, keep track of the state you need to jump straight into after re-interact.

Then trigger an event to jump to it

Basically you need to code a way to keep track of the state of your interaction

[woven bobcat]

the flow like this:
Question:want to play minigame?
A: yes->we play the minigame->play next dialouge(dialouge 2)
B:no->end the state tree

(Dialouge 2)After playing minigame:
Statement: wow, you pay that minigame impressively, want o play minigame again?
A:Yes-> play minigame again->go to next dialouge
B:No->end the state tree

something like this

aftet the minigame end, it should go to the next dialouge

[woven bobcat]

yeah, thats the idea that chatgpt gave me, but tbh im not sure where to approach this tbh

😦

[pastel moth]

You could just make this simpler and not use state trees for your puzzle flow IMO. Not really the answer you're looking for though.

I think Bruno's answer is on the money though. You need to keep track of the state of the puzzle

[slow bobcat]

you could try this:

• when you interact with the object, keep a pointer in the player character (or the interact component if you handle things that way) called CurrentInteractionObject.
• When the player decides to play the minigame, assign it a gameplay tag like "interaction.minigame"
• When you re-posses the player:
  * If the CurrentInteractionObject pointer is valid, start the interaction again automatically (as if the player would have clicked on it)
  * if the tag interaction.minigame is present on the player, send the state tree event to force the dialog state you need and remove the tag

[woven bobcat]

so for the keep track, what things do i need to keep track? the statetreestate?

i remember each state tree state have a tag on it, can we use that somehow?

i also see in the sendstaetreeevent have tag,payload on it..

[slow bobcat]

• if you have a tag in each state, you could keep that recorded to identify later which state you need to go back to.
• The payload is to pass objects and data with your event (I don't think you need that)

what things do i need to keep track?
this I already answered here but I'm just guessing. You are the only one that knows which data needs to be loaded to recover the same situation you had before unpossesing the player
#gameplay-ai message

[woven bobcat]

thank you, that is insightful! let me try it on my end and see how it goes

🙂

[robust veldt]

Why it is hitch when first time call this method. Any idea @slow bobcat

Next call only took few pico second

[slow bobcat]

I guess it needs to gather the data for the querier/nav and later is already cached?

does it look different if you call it while the ai is in a different place the first place? It might be the first time you are out of nav and takes longer to look for it?

it could also be you are hitting some ensure. Does the log spit anything that first time?

[robust veldt]

Oh, it could be ensure. I will check the log.

[robust veldt]

Error: Ensure condition failed: AsAIController  [File:./Runtime/AIModule/Private/Navigation/NavFilter_AIControllerDefault.cpp] [Line: 20] 
Error: 
Error: Stack: 
Error: [Callstack] 0x0000006E8FDBA37C libUnreal.so(0x00000000185AA37C)!UNavFilter_AIControllerDefault::GetSimpleFilterForAgent(UObject const&) const  []
Error: [Callstack] 0x0000006E8FB81BE0 libUnreal.so(0x0000000018371BE0)!UNavigationQueryFilter::GetQueryFilter(ANavigationData const&, UObject const*) const  []
Error: [Callstack] 0x0000006E8FB82D4C libUnreal.so(0x0000000018372D4C)!UNavigationQueryFilter::GetQueryFilter(ANavigationData const&, UObject const*, TSubclassOf<UNavigationQueryFilter>)  []
Error: [Callstack] 0x0000006E8FBA2A6C libUnreal.so(0x0000000018392A6C)!UNavigationSystemV1::K2_GetRandomReachablePointInRadius(UObject*, UE::Math::TVector<double> const&, UE::Math::TVector<double>&, float, ANavigationData*, TSubclassOf<UNavigationQueryFilter>)  []
Error: [Callstack] 0x0000006E8FBA2868 libUnreal.so(0x0000000018392868)!UNavigationSystemV1::execK2_GetRandomReachablePointInRadius(UObject*, FFrame&, void*)  []
Error: [Callstack] 0x0000006E8B499AC0 libUnreal.so(0x0000000013C89AC0)!UObject::execCallMathFunction(UObject*, FFrame&, void*)  []
Error: [Callstack] 0x0000006E8B4A3A9C libUnreal.so(0x0000000013C93A9C)!UObject::execLetBool(UObject*, FFrame&, void*)  []
Error: [Callstack] 0x0000006E8B49AF28 libUnreal.so(0x0000000013C8AF28)!ProcessLocalScriptFunction(UObject*, FFrame&, void*)  []
Error: [Callstack] 0x0000006E8B49A300 libUnreal.so(0x0000000013C8A300)!void ProcessScriptFunction<void (*)(UObject*, FFrame&, void*)>(UObject*, UFunction*, FFrame&, void*, void (*)(UObject*, FFrame&, void*))  []
Error: [Callstack] 0x0000006E8B49B6B8 libUnreal.so(0x0000000013C8B6B8)![Unknown]()  []
Error: [Callstack] 0x0000006E8B49B618 libUnreal.so(0x0000000013C8B618)!ProcessLocalFunction(UObject*, UFunction*, FFrame&, void*)  []

Yeah, it is ensure

[slow bobcat]

That means your querier is neither a pawn nor an AI Controller

that's your ensure

[summer basin]

Hey, any advice appreciated! Is the capsule component of a default Character actor is the one driving collision? I can't seems to get a character's mesh physics asset to be a hit box whatsoever, unless capsule also blocks the projective, mesh would just ignore the collision. Is that common? Is that how collision hierarchy works in unreal?

The simple goal is just have a correct hit box for the ai characters where they have a proper physics asset as part of mesh, but it's so far was far from simple

I have a Projectile that would block pawn and the mehs of object type pawn that would block the projectile. When the capsule collision however ignores the projectile, no collision happens 😢

[slow bobcat]

You can set different physics profiles and channels to capsule and mesh(es). Also you can attach collision shapes to your skeletal mesh through bone sockets and use them for colision. If the capsule ignores the collision is because your physics setup for it and the projectile is not correct

[summer basin]

So there is no a known thing like capsule drives the mesh collision and mesh responses are ignored then, due to a hierarchy of collision shapes?

[slow bobcat]

No. That's entirely up to you, your physics channels/profiles configurations and how you handle them

[summer basin]

Ok Thank you, this nonsense was bothering me since yesterday. I tried to resist to believe there is a hierarchy. Seems like character's mesh component only responds to overlaps on it's own if it is simulating physics, otherwise something about physics asset config of bodies that makes one mesh to respond to hit events and another mesh won't respond

Eg I have one AI that uses epics skeleton and physics, and a custom ai mesh, the second one doesn't register collision even though it's all exact same with epics one. Except configs and setup of physics asset itself ...

Oh... Ok solved, thanks Bruno for doubting nonsense I imagined. Apparently it helps, if one won't override physics asset on a parent AI class and then inherit it with different mesh in a child class..

[little bridge]

I believe this is the right channel that can help me.
I'm having an issue with my game's nav mesh. It will always work in the first level, but never will in any of the other levels. If I load the game from level 2, let's say, it would work, but if I load from level 1 and then get to level 2, it no longer works

It used to work fine, and then all of a sudden it just stops

[little bridge]

I got a video of the bug if that would help

[slow bobcat]

Is level 1 a persistent level and level 2 a sublevel?

[pastel moth]

Im trying to do a utility AI and I have a state tree with two tasks to consider but the tree only seems to enter one of the tasks no matter how high I set the constant. Could I get some help?
Could the issue be with my selection behavior? The debugger doesn't even say that cleanup trash was considered.

My two tasks are "Chill" and "Cleanup Trash".

I've confirmed my constant is being set for cleanup trash.
**
Selection Behavior**
My utility selector has select children with highest utility. Chill has select "children in order" and cleanup trash is "try to follow transitions"

Chill has a constant of .5
Cleanup trash has a constant of 0 otherwise its set to 5000000 by my utility selector.

Ive removed my chill state and even then it fails to consider cleanup trash. Here's what my tree looks like.

[hasty turret]

Note that Score is normalized
(0,1)therefore your 5000000 will be 1

[pastel moth]

Ah okay I got you

Maybe I should just do all this in C++. It feels like a black box in blueprints where if I twist the wrong knob its not super apparent and things break. I know part of this is my own ignorance lol

I got it to work. I had finish task in my utility selector being checked off once it finished calculating all the utility scores. I removed the finish task and voila. Not sure why it was causing the tree to freak out.

Maybe Im not understanding how the states are evaluated..

If the utility selector has select children with highest utility, and the children have their own selection behavior such as try enter etc..

Will the utility selector select the child with the highest utility score? Or will the children with their selection behavior overrule that?

I thought it was the former.

oooo

Instead of having 5000000 I put it back down to 1.0

and voila

it considered the state properly

maybe there's some wonky going on the BP side

[hasty turret]

hmmm

[pastel moth]

I got rid of the variables and tried it with just the constants. Set chill to 0.0 and set cleanup to 1.0 then it selected cleanup as expected.

Then I bound the constant to my weight variable and set the weight variable to 1.0.

I'll try to screw it up again just to verify.

[hasty turret]

You have to squint really hard at the State Tree debugger or dig through the source code to understand what's happening. I ended up writing my own utility selector.

[pastel moth]

Okay nevermind that wasn't it. Some knob I must have turned got it working. Jesus I feel like a monkey with at typewriter.. next game Im doing all CPP for AI...

That's probably more wise 😄

Thanks Snake

I appreciate the help

https://tenor.com/view/colonel-codec-mgs-metal-gear-solid-gif-25391892

[hasty turret]

np bp is fine it is just we are not familiar at how state transitions

we are so used to using bt

[pastel moth]

Im not use to either haha

[hasty turret]

hhh

[royal hemlock]

Within a tick how to stop ChildB being entered twice? The condition for ChildB binding to Count which is evaluated with global task. On same tick even the AddCount is performed, the evaluated value still remain til next tick.

[graceful wind]

guys, is it common to use smart object as interaction manager for character to character interaction?

[slow bobcat]

I want to help but I can't make sense of the logic. Can you explain it in more detail? A step by step guide of what happens?

[royal hemlock]

Let's say EnterCondition of Child B is Count == 0.
Count is evaulated by a global task.
ChildB has a Task incrementing Count (Task_AddCount).
Then I noticed within 1 tick, Child B is entered twice and Task_AddCount did made Count = 1.
However since it is still at the same tick, the evaluator still return Count = 0.

And next tick Count = 2 where evalator now trigger properly

[glass onyx]

There’s not really a precedent but it seems reasonable

[little bridge]

Fixed the problem, I had to go in each and every level and hit "Build all levels"

[slow bobcat]

Global tasks (evaluators) tick BEFORE the rest of the tree. When you say "Count is evaluated by a global taks" what do you mean? That the value is exposed by the global taks or that the global task does count++ on each tick?

If child B is entered twice in one tick, shouldnt count be 2? You said "enters twice, count = 1".

[royal hemlock]

the value is exposed by the global task. Yes the value is 2. But just because the value from global task is 0, causing ChildB being entered twice.

[slow bobcat]

I can't make sense of your explanation.
START OF TICK Frame#1

• Global task exposes a value through an OUT param -> Count == 0. It only exposes the value as a param, it doesn't manipulate it
• Tree runs and selects states in order
  * Child A does something and transitions to child B
  • Child B has a task that binds to the OUT param Count and does Count++ in an attempt to change the value of the param. Transitions to Root.
    END OF TICK Frame#1

START OF TICK Frame#2

• Global task exposes a value through an OUT param -> Count == 2. It only exposes the value as a param, it doesn't manipulate it
• Tree runs and selects states in order
  * Child A does something and transitions to child B
  • Child B has a task that binds to the OUT param Count and does Count++ in an attempt to change the value of the param. Transitions to Root.
    END OF TICK Frame#2

Three questions:

• Am I missing anything there? What does Child A do? Asking in case it interfeeres somehow
• the task in Child B, does it try to change the param value directly or is it accessing some variable in some component/actor?
• I have seen cases where conditions/tasks execute up to 5 times in a single frame. It usually happens due to a bad transition planning. What does your State Tree Debugger say?

[royal hemlock]

Count is a gameplay attribute, Thing B is the task to add a gameplay effect which Add 1 to Count.

[slow bobcat]

I'm bit confused with the debugger captures. Aren't them all from a second run?
I can see transitions from previous states etc

Also: you are adding a GE to change the value. Is it a one time GE with limited life or infinite or what is it? That could be the issue too

[royal hemlock]

all the screencap are within a tick.

[slow bobcat]

and it's the first tick? Just to confirm

[royal hemlock]

no, it is in the middle of the game

the ge is infinite, a simple one

[slow bobcat]

no what I mean is: is it the first tick of the tree?

[royal hemlock]

no, it is in the middle of the tree alive time.

[slow bobcat]

It would be easier if we could see the very first time the tree runs.
But looking at that screenshot things make sense (except for Child C, no idea where that comes from).
But is as expected. Child A runs. On complete transitions to child B. Child B runs, runs its task because it passes the condition

Not saying there's nothing wrong, just saying I can't see it in that debug info

You picked the frame where the child A enters again, so we can see the transitions, exit nodes etc

To do a sanity check: if you print logs with the current time stamp (or even better, current frame in the game) from within the task in Child A and the task in child B, do you see the same number several times?
Do you see them printing several times per frame? Just to corroborate you are not being lied

At this point I question hard that the tasks are running twice in the same frame but need confirmation

The other test I would do is to add logic to the Child B task to add +1 to a variable in the owner actor. Some variable in a component. See if it matches your Count value. Just to discard weird things happening due to GAS

[royal hemlock]

yes, serveral times is printed. Yes, the behaviour makes sense, what bugs me is that, since the condition is referring to an global task and the task does not update after any actions happened in that tick but til next tick.

[slow bobcat]

, what bugs me is that, since the condition is referring to an global task and the task does not update after any actions happened in that tick but til next tick.
But this makes sense. The global task writes the value it reads at the beginning of the tick, writes it to a parameter that you use for the condition. You might be changing the original value but the parameter value is not changing until the global task ticks again, because that's the only thing writting to the parameter.
If you make a custom condition that, instead of reading from a parameter written in the global task , directly reads the value from the Attribute Set of the character, does it work?

[royal hemlock]

Actually I found the solution, I move the GetCount to Parent and make sure the update value happened in EnterState instead of tick. yep the global task make sense too, but i think there should be a feature to listen to value change in global task level? or we need to be careful with the global task getting value.

[slow bobcat]

If you want to change the value of the global task you need to look into FStateTreePropertyRef

#gameplay-ai message

[royal hemlock]

I'll look into it. Once again, thank you.

[graceful wind]

yeah, looks neat to utilize the slot availability, position, and behavior feature. but I've seen it utilized just for static environments

[eager trellis]

Hey all! Beginner question. I have tow actors that share the same logic as per image. The first one work as intended and it perceive every actor in the level instantly. The second one it sense other actors but when i ask him, inside a task, what has perceived, he says nothing (it means that the branch goes to the false route). What can prevent my second actor to give me the name of what he has perceived? I've tested directly inside the AI controller that he can sense the surroundings

[hasty turret]

Its empty

[eager trellis]

nope, its not the culprit, tried to select ai sense, doesn't work, and the first actor has no class selection and still works, indeed the description itself says that if no sense is selected, automatically it store data from every sense, if i read correctly

let me give it another try selecting the right sense, anyway

nope, doesn't work

[hasty turret]

Maybe there's another problem lying around

[eager trellis]

fun fact, if i make a loop of the state tree, it start sense after 5 or 6 cycles

[hasty turret]

I was wrong

[eager trellis]

i can't figure out, i'v tried to isolate the task, tried to debugging with classical print string here and there, using visual ai debugger.....

The logic in the pic comes from a tutorial of the epic documentation about state tree

[hasty turret]

Try Start the StateTree at later time, maybe after calling RequestStimuliListenerUpdate

[eager trellis]

ok i've delayed the initialization of the state and seems to work

@hasty turret thank you, you're a life saver

[slow bobcat]

sounds like you are initializing your tree in Begin Play rather than OnPossessed and after initializing the senses

[eager trellis]

yeah, the component has an option to start automatically on begin play

oddly, my logic is inherited from this tutorial https://dev.epicgames.com/community/learning/tutorials/lwnR/unreal-engine-your-first-60-minutes-with-statetree

and the tutorial actor works fine, with state tree firing at begin play

[agile widget]

Hi everyone, I have a problem where I Spawn multiple NPC’s and assign them a position based on some EQS Tests and the available positions for each reach, there are three as of now but I could add more based on reach, right now I have Melee, Long Distance and Wait zones has shown bellow:

Right now I’m testing two zones, the Melee zone and the Wait Zone, each of them are getting assigned correctly but for some reason the NPC’s assigned to the Wait zone stop moving to their assigned spots after some time, even thought there are a lot of open positions:

Has can be seen here, there are a lot of open spots for the NPC’s to move

But for some reason they simply stand where they land and won’t move

Even though the Behavior tree is executing and the movement node is executing

But they still won’t move for some reason, here is a photo showing where the NPC was supposed to move to.
Where he should have moved to

And here is a Gameplay Debugger data:

and here is the EQS and the Behaviour Tree with added location reachable validations:
Location EQS

There is a Path

I use a custom Decorators that checks if there is a valid open position for the NPC to fill, which is working has expected, and another one that checks if a position has been opened when a enemy dies, if not, it turns off the behaviour tree and waits for an NPC to die and then gets activated.

Thank you for any information and let me know if there is any other information that I might need to add.

[neon sparrow]

so im using mass ai plugin to creating a mass civilian npc for my game so i impliment mass ai is roaming in my desired path but a also wnat one thing that some the the npcs from the mass ai can do some work or like talking together and one more thing i am using state tree in it

[misty rune]

Hello!
I am attempting to add a UStateTreeAIComponent to a AI controller C++ class located in a custom plugin. I have defined a UStateTreeAIComponent variable in the header and created the component in the constructor of the AI controller class. After setting the the State Tree AI asset in editor, I closed the editor and restarted the project. Upon restarting, the editor fails to load and gets stuck at 73%. The only working fix so far is to get rid of the UStateTreeAIComponent and the State Tree asset. This does not happen if I set things up in Blueprint.
Is anyone else running into this issue?

[grave sequoia]

How to build async MoveTo in AIController

[simple crescent]

Not sure I understand the question. Do you mean like what is done in the UAITask_MoveTo?

[misty wharf]

Launch from IDE, it will probably be much more useful for debugging what goes wrong

If you added the component and compiled (hot reload or otherwise) while UE was running you could have broken something.

[glass onyx]

Up until recently smart objects were not really meant to be used on movable objects

[bleak igloo]

Hi, I have a task that refuses to stop even after I stop logic, I made a bool a decorator so it aborts self on change but its still going Because I assume if the tree is lit like on the screenshot it means its running, right?

[slow bobcat]

Does your task call finish task? Also, could it be your tree re-evaluates and runs into the taps again?

[bleak igloo]

It does when a response/button on a widget is closed

So when I click a button it finishes and goes through rest of the tree start again from begining and pauses on this task again to wait for a button click

okay I see my problem

When I start the tree it goes too fast and is not able to get the variable from the service on time, so If I want to see the new options added I need to restart it again

I can 'solve' it by adding a wait node but that makes the options take time to load obv

[pine steeple]

@bleak igloo surprised your using BT foe a dialogue/quest system

It's rough doing this, trust me

There's better options available

Flowgraph and common conversation are two plugins that exist

[bleak igloo]

yes I am aware that there are other options

found the solution to this problem is working perfectly now

[signal olive]

So I'm having trouble making an AI agent respond to sounds not directly made by the player

[pine steeple]

is it not triggering the ai stimulus code?

are you in the range?

Gameplay Debugger can show all stimulus ranges

and will show what events happened

[signal olive]

It's a projectile that triggers a noise when colliding with something

For some reason even when the collision is within the hearing range the AI doesn't pick it up

It does pick up noises like footsteps, though

[pine steeple]

did you verify

in the gameplay debugger

[signal olive]

As far as I can tell, yes

[rugged garnet]

https://forums.unrealengine.com/t/fstatetreepropertyref-is-not-working-on-macos/2585730
I have the same problem as this person.

It could have worked at 5.5, but it didn't work in 5.6.

[signal olive]

Cripes I feel like I should either find an alternative to hearing perception on Fab or roll my own system because it's jank and an unintuitive mess that feels like it'd be incredibly hard to design stealth games around despite it being one of the few cases where you'd want a hearing sense for AI in the first place :/

[glass onyx]

Another thing you can check is if your agent is configured to hear things from neutral and friendly sources as well, that might be filtering stimuli. That said the AI perception system is old, from UE4. Im using the hearing stuff in my game but it’s limited for sure

[signal olive]

The annoying thing is that there is a fab plugin for explosing the remaining perception functionality to blueprint, including the 'faction' stuff, but it hasn't been updated to 5.4+

[misty wharf]

I get the feeling it's probably going to be fairly trivial to update if you know even the tiniest amount of C++

Those APIs haven't changed I don't think.

[signal olive]

I eventually figured out that the noise report event requires a persistent instigator, so I put the projectile's owner as the noise instigator, and it works now, which doesn't make any sense but whatever :/

[vast relic]

Just starting with state trees, any insight as to why my move to wouldn't use the output of the find random location?

Are tasks within a state executed in order or how does that work

So I added a print of the target location in the move to and now it works? WTF? Are there some sort of compilation shenanigans with ST?

[slow bobcat]

Tasks run in parallel. In 5.6 there is an option to choose how they run I think? Usually, if you want to do that "gather data - use data" pattern in ST's, you make them nested states
Gather data
Use data

Check the pinned messages for the video about state trees and the link to "your first 60min with state trees"

[slow bobcat]

Not a persistent instigator but any actor that is registered as sense stimuli. For that you need a component and register it, handle the teams part (friend/hostile) etc. People usually use the AI actor or the player since those work out of the box. And that's fine.
https://youtu.be/w0t75DdkazU

In your case, what does the gameplay debugger show? Does it not detect anything? Could it be you are trying to use the projectile location later (trying to grab the projectile actor) rather that the stimuli location and that failing because the projectile is gone (actor is destroyed)? I have seen someone with a similar bug not long ago

[slow bobcat]

The Perception system is a Peeve Pete for me. Next project I start at from scratch, if given the decision, I would simply not use it. It's waaaay too convoluted for the simple things it's supposed to do

[harsh storm]

....I...like it 😅

[hybrid crypt]

I'm trying to understand why my bot is not moving, I made a simple service to display the "MoveGoal", and it shows the correct location vector, but only with the "node1" prefix, not the "node2" one. The navmesh looks fine and the bot has a simple path to reach the goal

What really strange is that after doing a bunch of these debugging changes (adding services, removing services, adding logs, etc.), all of a sudden the bot starts moving again (as if an asset was corrupted and fixed when saving?)

(force success isn't supposed to be there, just tried it for debugging)

[slow bobcat]

It's OK for simple use cases but... As soon as you need to tweak performance, management of senses, extend it.. It's a nightmare

[graceful wind]

In order for your bot to move, you need to make sure

1. your bot is a pawn
2. has ai controller running that behavior tree
3. MoveTo task need reachable location or actor, make sure that you already have nav mesh on your level

[hybrid crypt]

yes these are all ok, the nav mesh is good, the target location is quite close, and in BP I see that my nodes are firing (only the node1 debug string is showing up though)

[graceful wind]

how close? could possibly it works, but the acceptance radius is too big

[hybrid crypt]

5 - 10 meters, while acceptance radius is 10 cm

[hybrid crypt]

I found the issue, there was a small exclusion zone in the nav mesh right around my goal target; I was only looking at the nav mesh from far away to make it was ok

[graceful wind]

hmm, strange, how you set that MoveGoal?

[hybrid crypt]

move goal is the location of a Flag mesh/actor, there is something on the same location, maybe a volume or something, which is creating a hole in the nav mesh

[graceful wind]

ah, i see, it makes it unreachable

[hybrid crypt]

yes!

moving that blue pad on the bottom a little bit fixes the issue

(even when putting it back on the initial location, it works now)

[graceful wind]

you can try to disable this on that mesh, just to make sure it works, if it has collision, the ai will keep trying to reach that location forever, usually I play around with acceptance radius for that

[hybrid crypt]

cool thanks for the help!

[signal olive]

Nah, it was the instigator that was the issue

[slow bobcat]

If you want the projectile to be the instigator, you can add the stimuli component for it I believe

[signal olive]

I've tried, it doesn't work

[slow bobcat]

Umm weird. I will check how we handle it in this project put if curiosity. I can't remember

[signal olive]

When self is used as instigator right before being destroyed it seems the stiimulus immediately disappears

[slow bobcat]

That's probably because by the time the stimulus is processed, the self is already marked for destruction. Stimulus are not processed immediately. At best it will take 1 frame. Destruction is maked in the same frame

[agile widget]

Hi All, has I asked in this question I have an AI that should be able to move to their designated positions, but for some reason they don't I implemented a small debug to see if the locations they where getting are inside the Navmesh or not, but for some reason it's still failing even though the assigned locations are inside the NavMesh and should work, can anyone help me with this? i'm going to do a step by step debbug in one that fails, but I would still appreciate any help.

[slow bobcat]

Hey. Let's simplify things a bit for debug. If you use one character only, does it fail? Is it a bug that happens when you use multiple character only?

Then, use Visual Logger, record, repro de bug and check the navigation and path following categories

[agile widget]

Hi @slow bobcat, the issue only arrises after a lot of the characters have been spawned and they are looking for their respective position, some time start to fail after the 30th spawned enemy and sometimes is after the 20th, it varies from execution

[slow bobcat]

Umm ok. Chances are that they are getting stuck against each other. Of you spawn them in a line next to each other with enough space and make them move in a straight line forward (so the don't cross paths) does this happen?

In any case, record with visual logger and check navigation and path following categories for a stuck enemy. Let's see what the logs say

[agile widget]

Perfect, let me see what Visual Logger says

It seems the path is blocker for some reason, even tough in the video I don't see anything that might be blocking it

@slow bobcat

[slow bobcat]

And if you show the nav while ejected pressing P in runtime?

Does it show anything ususual?

It could also be that you have an issue with your movement logic and no velocity is being applied (remote chance, but that would cause a block too).
That exact same test (same start end etc) with just one enemy works?

[agile widget]

I'm not using any custom movement, and the AI is using basically the standard Move To logic, I only change the acceptance Radius dinamically, in the PIE the Nav data doesn't show anything unusual, I did fix some small issues a couple of days back, but it still doesn't work

[slow bobcat]

I would try to repro the exact same example with just one enemy

See if that works or not. It should not work

I think soemthing is off with your AI because it's not even rotating to face the path

[agile widget]

give me a sec

[agile widget]

@slow bobcat has you can see they move at the beggining, but later on the new spawned enemies, just stop going to their designated positions

[slow bobcat]

Are we taking about the ones in the background or the ones around the player?

[agile widget]

the issue is with the ones that need to encircle the background zone

they are all the same and use the same things for everything

@slow bobcat can we talk more about this tomorrow, It's pretty late here and I need to try and put my baby to sleep

sorry and thank you

[slow bobcat]

Sure thing.

Night night

[agile widget]

Thanks I'll ping you tomorrow

[tawdry zephyr]

Oh did Miezkoz leave the server

[harsh storm]

No

He mostly hangs out in #mass anyway

[agile widget]

Good morning @slow bobcat , I'm back and ready to continue debugging the error if you are available

[slow bobcat]

Working but will check here whenever I have time. Can your record a video showing the AI at runtime while the bug happens?

[final prism]

Are you using crowd or normal path following?

[agile widget]

yes, give me some time, I'm on the same boat has you also working but I will do this whenever I can

I'm also using UCrowdFollowingComponent in my AI

[slow bobcat]

Ah... Important detail. This might be causing issues on its own. Will see

[final prism]

the [blocked] flag is an unreal fallback, I think it triggers if the NPC hasn't moved within a time frame. My theory would be that at some point you've spawned so many npcs that they completely block each others path

Or the spawning logic is somehow borked and the move component isn't able to move the NPCs properly anymore

Different question, out of interest, this may be more of a generic question..
How do you guys link designer-facing data to AI-facing systems? Typically you'd have some DB, a data asset or something like that where all of the values are tweaked. How would you link those?
I've been working with Data Providers with okay-ish success, though they are EQS specific and seem to not be used much at all in the engine. Any alternatives people use?

[slow bobcat]

We have data assets with all the data. Different ones for behaviors, animations, character customizations (mesh, weapon etc etc). We spawn enemies deferred and, before calling Finish spawning, we read their data assets and set whatever is needed in their different components.
For the AI ones are mostly read on runtime on a need base since they are things like "what bt should I run for combat" etc

[final prism]

Hm yes that might be a better idea, I'm not actually sure of the performance issues of reading from a data asset at runtime anyway

[agile widget]

Thanks I will try to also look into this, I do have a lot of Spawning enemies right now really fast for test purposes, and I also hope to make them spawn Faster at some point, but at least not for now

[slow bobcat]

We also have very big excels (Google calc) sheets with all the stats for character balancing. Designers tweak the numbers and a tool inside the editor imports the files into data assets. For them it's much easier to handle and track changes that way for stats

[final prism]

that makes sense, we were playing with unreal data tables but they are kind of tricky to work with

[agile widget]

By the way I can create the Video when they start spawning, but if I wait until it starts getting stucked, it might be to heavy for Discord, how do you want me to upload it for you?

[final prism]

Ended up using Data Assets too

The batch editing on those is actually quite good

[slow bobcat]

Nothing you will notice if you are careful with soft references. If you reference assets (animations, Bt's, meshes etc), make sure they are soft pointer objects that you load asynchronously when you need them. If you use hard references, you will be opening a door to performance hell

[final prism]

We've actually noticed the opposite, where soft references would take up a bunch of loading/editor time if they were overused

[slow bobcat]

Maybe you can just record the few second it takes for the failing one from spawning to fail? I'm your previous video it happens fast

[slow bobcat]

I would imagine that's because you guy load all of them when loading the editor synchronously instead of on a need-base and async? You are probably flushing the buffer constantly

[final prism]

The best method would be to put a breakpoint in the "IsBlocked" check on the path following component and just.. see where it leads you

[final prism]

yeah maybe

[agile widget]

Well it's not that fast but here is one from start to error, basically it always starts to fail after the 25th enemie is spawned

[slow bobcat]

ummm as soon as one fails, all the rest spawning after fail too.

[final prism]

but if you need to set something like a timer value on a state tree, how do you do that? Blackboard keys?

[final prism]

Aren't all the slots occupied here? I'd assume they couldnt move there if somebody else is already standing on it

[agile widget]

no, let me take another one showing all the available slots

[slow bobcat]

If the time value is something that will be different depending on which enemy type runs the tree, then same approach:

• data asset has the value (ie: attack cooldown)
• spawn deferred, read the value from the data asset and store it in a component before calling Finish Spawning (in our case many of the state values are transformed into GAS attributes). Or if your prefer to read it from the data asset, have a function to return the value in your preferred component.
• when the tree needs the value, you can have a task that reads it fork the component and exposes it as an put parameter for the next task that needs it or you could have a Global tasks that exposes all the values as out parameters OnEnter (no need to expose them on tick since they are static).
  In our case we have lots of custom tasks and conditions that read the values internally. We don't rely much on exposing thug s through params because it's way too much and will make our trees dirtier for no good reason

[agile widget]

Here is tha Ranges where they are supposed to go, there are a lot of open spots in the Wait Zone, I'll take a video now

Here is the Video and it always seems to fail on the 25th character, is there maybe an Agent that indicates the amount og Characters can be moving at one time?

[agile widget]

Yes this seems to be the case

[slow bobcat]

Umm not that I can think of but I do recall Crowd Manager having a setup for home many characters could handle at once. Maybe you should take a look into the crowd control configuration

[agile widget]

Yes I have it at 50 Max Agents at once and I disable them when they have stopped moving

[robust veldt]

If you destroy some of AI, does other will start to work?

[agile widget]

only the newly spawned will start to move, but the ones that are stucked, they will stay there

[slow bobcat]

Ok so you are definitely hitting the limit somehow. But you don't have 50 spawned AI's... Are you changing that limit anywhere else? Or maybe you are spawning more than you think?

[agile widget]

no, I don't change it and the limit should not be reached, there are 25 actors and thats all

I'm going to look where the blocked flag is getting set to see if there is something else happening

[agile widget]

@final prism and @slow bobcat I have found that the error happens in the UCrowdFollowingComponent UpdatePathSegment, but I'm not sure what is happening since I can't debug the Unreal Engine code

[slow bobcat]

I think you can download the engine source simbols (pdb's) to debug it. Most values will be optimized though

The block detection is just checking if the AI hasn't moved more than X seconds

[agile widget]

I do have the pdb's downloaded but like you said almost everything is Optimized so I can't really follow what the issue is

[slow bobcat]

yeah... that sucks.
I would be very interested in see this objects and corroborate your limits are not being hit

[final prism]

You can deoptimize using the pragma macro and running in debug mode

[agile widget]

Give me a moment

[final prism]

Crowd manager is probably the next spot to check, maybe where it sets the velocity or updates the agent state

[agile widget]

do you want the values when the error starts?

@slow bobcat, @final prism I really don't see anything strange

plus I haven't been able to actually make the Crowd detour work correctly at all, so I think I'll disable it later on

[celest oracle]

How do I store a result in my state tree global parameters?

Lets say a vector location that i made in one of my state's task.

How do i get that to either the global parameters or stored in a global task?

See this float for example, the change in my wandering state task does not go to the global parameter

[agile widget]

I Think the issue is related to the Crowd Manager, I'm going to bump up the max agents and see if it works

@slow bobcat and @final prism I bumped it up two 200 max Agents and now 100 of them have no problem reaching their respective locations, so I need to fix the issue with them not getting removed correctlly when they arrive so that I can keep the agents in a lower value

but It seems the mistery is solved, I'll start working on this and let you know

thank you for all the help

[final prism]

needs to be a statetreeref type in order to modify the value

[orchid marlin]

As Mortiz mentioned you could use FStateTreePropertyRef. Take a look at FStateTreeRunEnvQueryTask, it exposes its result to configured property.

[slow bobcat]

Only if you compile the editor yourself

[final prism]

Oh I've done that for so long I forgot there's other ways

[celest oracle]

Yeah I noticed that the provided env query task was able to store its results in the global parameters

Thanks guys 😃 for the help again

[final prism]

Does anyone know if the child blackboard stuff is functional in 5.6? Last time I tested it didn't work at all

[neon moth]

hi everyone!

I wish to get a grasp of all possible ways to make enemy behaviors
I've already used Blueprints and Behavior Trees. I'm aware of GOAP and HTN, but still need to gain hands-on experience.

Which other decision-making frameworks should I know about?

[final prism]

Hierarchical Finite State Machines, MCTS and other tree searches (minimax etc). There's a ton of ways depending on the type of decision making you want

Heck even Reinforcement Learning is a way to create behaviours if you really want to, or a combination of algorithms, like Unreals State Tree is a mix of HFSM and BTs

[simple crescent]

Add Utility Systems, Decision Trees (Rule-based systems), and Case-based reasoning to the list

[celest oracle]

Im growing to like Unreal's state trees which kind of combine FSM, Utility AI (GOAP) and Behavior trees.

It has a visual graph so designers can see more easily how the AI behavior flows. It makes a web like FSM more palatable for me.

[celest oracle]

May I ask if I can receive state tree events with their payload inside a global task?

So far I only know how to get it from On Event Transitions and the enter condition to use its payload inside a state's task

[celest oracle]

I wish to get the event payload data inside my task BP like this, but how?

Or is the only way now what Bruno said a few days back to :

#gameplay-ai message

[slow bobcat]

I don't think you can. State tree events are used through transitions and there aren't transitions to global tasks. They simply run at the start of every tick for the tree and that's it.

Ah wait

There is a way

The instanced data has the event queue

@pine steeple didn't you end up doing something similar? manipulating the EventQueue in a global task or something like that?

[spring jetty]

Hey any clue on how can I make the walkable areas limited for big AIs in a game ?
there is an array called **SupportedAgents ** under NavigationSystem in project settings that you can add to, that has some variables like agent radius, agent height, but I have seen the same variables in a few different places like in CharacterMovement, and in RecastNavMesh actor, which rly confuse so I dunno how to start, didn't find much on google.

[celest oracle]

For future reference, this how it looks like in BP to set global parameters from one of your state's tasks.

[strange plank]

Is it needed to build paths for World Partitioned Nav Mesh Volumes? if im working in a procedural level can I just set the mesh to dynamic and check world partioned in the recast nav mesh?

[slow bobcat]

Supported agents are to create different sets of nav data that will be used by matching the AI's capsule radius with them.
The most important value is the agent radius.
If you have a regular enemy of 42u radius and a bug one of 125u radius, you can set a nav agent for each. When your nav builds, it will create a nav data for each.
A corridor of 90cm will get nav data for the regular AI, allowing it to navigate through it, but will not get nav data for the bug ai because it's too narrow to fit

[slow bobcat]

Yeah, you can generate nav dinamically. Initially that was the only nav generation supported for WP. Ideally you would do it around invokers

[graceful wind]

Guys, any idea why the OnSmartLinkReached event is not triggered? the nav link is working though. My ai can walk through separated area. I tried to remove it, my AI cannot walk through the separated area. But this event is not triggered. I am using UE5.6

this actor extends NavLinkProxy

[spring jetty]

So The matching happens automatically ? what if the values I provide are different ? so the NavAgent may have a radius of 40 but what if the AIs capsule has 70 or 80 ?

I also found this checkbox "Update Nav Agent With Owner Collision" in CharMovement.
so I thought if it's checked it will use the CapsuleComponent's radius as the agents size, and then I tried unchecking it and giving a really big number to both height and radius as you see, but the AI I have in the game is still able to move around.
idk this is confusing... is there any tutorial that covers these in depth that you know of ?

[simple crescent]

Just checking that you checked bSmartLinkIsRelevant. I believe that if it isn't checked, the navlinkproxy will still work as a nav link for pathfinding, but you don't get the additional functionality of the smart link.

[slow bobcat]

I believe it uses the closest one. You can always provide the nav data to use (in c++ at least is possible, no idea in Bp's)

About the tutorial yeah, plenty of them in YouTube form ue4 (hasn't changed)

[spring jetty]

Wdym by Nav Data here ? like I can set which Agent each AI uses ? or you meant something else ?

[slow bobcat]

Exactly that. Nav agents are settings to generate nav data. The nav agent radius tells the nav data how far from objects/walls it should be. In code (I would imagine in BP too) you can pash which nav data to use for your move to request. If nothing passed, the engine will try to finde the right one I believe

[spring jetty]

Thx for the help ❤️

[slow bobcat]

Any time

[rare hemlock]

Hey guys. Has anybody tried to implement a queue system with the help of smart objects and the state tree? Lile a queue in front of a vending machine.

[iron roost]

I'm having issues with my Navdata not showing up in packaged game in 5.6? Any help?

[stable plover]

wdym by not showing up

[iron roost]

Yeah everything works fine in editor but if I try standalone or package I get a NAVMESH needs to be rebuilt error. but it's built.

[iron roost]

In editor and PIE it's perfect. But in Standalone and Package I get an error regardless of what I do.
Using Static Nav Mesh, 5.6 is suppose to include navdata with the map. But isn't.

[iron roost]

If I make a new map it works fine. So it must be something in this map. I start making the new map now its broke again. I don't get it.

[graceful wind]

Thanks man, still not triggering, I will try to make base C++ class for this and debug there

[graceful wind]

I will drop this, and use overlap event instead, I hope someone has some solution though

[celest oracle]

How does one get the current state again? 😅 Im trying to make a conditionBP that checks if current state is not X state.

I need to do this in C++ ? 😦 No BP Function yet ? Seems like a critical function for a State Machine 😦 ```cpp

TStringBuilder<256> stringBuilder;
if (StateTreeRef.IsValid())
{
const FStateTreeExecutionContext& stateTreeExecutionContext = FStateTreeExecutionContext(*GetOwner(), *StateTreeRef.GetStateTree(), InstanceData);
for (const FName& stateName : stateTreeExecutionContext.GetActiveStateNames())
{
stringBuilder.Appendf(TEXT("[%s]"), *stateName.ToString());
}
} ```

[hasty turret]

Yes. I created my own bp function library for things like getting current state, get parent state, get child states etc. Not sure it's the intend way to do it, but used it with no problem.

[slow bobcat]

Describe step by step your settings for nav

[slow bobcat]

Yep... And this is new in 5.6. When we started using ST's in 5.4 there wasn't a single way to do it but to have a state tree task in each state for it

[slow bobcat]

How are you setting the nav link points?

The nav link proxy has two sections in the details panel, the regular one and the Smart Nav Link. You need to set that

[graceful wind]

Looks like this is the problem, the smart link was not bridging the nav areas, it tried to fix it, now the event is triggered. But somehow the character sometimes uses the smartlink or the regular one. if the character uses smartlink, somehow it cannot move to another area

still play around with this, thanks lot man, I am planning to make this link dynamic though, thats why I need this

[simple crescent]

You can also look at https://github.com/phildunstan/GameAIUncovered_UnlockingDoors
It has some example code for making a door actor which contains a UNavLinkCustomComponent instead of using a separate NavLinkProxy actor.

[graceful wind]

I see, so it is actually not common to use NavLinkProxy, yeah, it is more reasonable to use component for this, I can add it to all my doors. Let me check that, thanks a lot!

[slow bobcat]

It has some example code for making a door actor which contains a UNavLinkCustomComponent instead of using a separate NavLinkProxy actor.
we do this in our game too. Link si enabled/disabled when the door open/close

If you happen to have a Statetree condition where you store data from the world that might change (like a timestamp), don't.
Turns out that, for some unknown reason, someone at epic decided to make data instance in conditions Shared.
This is like if you wanted to store data in the memory of a decorator but epic told you "nah you can't do that. All decorators of the same type use the exact same memory block".
I'm surprised by the decision and confused too, can't think of a single good reason for it.

Epic message I got about it:

/--------------------------------------------------------------------------------------------/

Hi Bruno,

Sorry about the extended delay. Here's the bug tracker number: UE-302300. It should become accessible once the engine devs mark it as public.

One of the reasons it took me a while is that I did some more research on this issue, and I ended up finding the following commit/CL:

* StateTree: shared instance data
- Separated condition instance data into a shared instance data (mutable but no persistent state)
- Made BP TestCondition() const, conditions should not hold state
- Added macro to define runtime data POD for faster init
- Added custom serialization version for UStateTree, older assets need recompile
- Updated memory reporting to be a bit more accurate, separated shared and unique data
 
#jira UE-147508
#rb Mieszko.Zielinski
#preflight 627b657d2d608c533b5cde15
 
[CL 20134929 by mikko mononen in ue5-main branch]

So, it now seems clear to me that this commit consciously made State Tree Conditions share a single "InstanceData" between all actors and PIE sessions. The assumption expressed on the commit message is that "conditions should not hold state".

I also found this on the StateTreeConditionBase.h header:

Note: The condition instance data is shared between all the uses a State Tree asset. You should not modify the instance data in this callback.
This is a comment above methods EnterState(), ExitState() and StateCompleted(). It is not repeated above TestCondition(), but I believe it applies there too.

So, based on my findings above, I believe there is some chance that the bug report will be flagged as "by design" or "won't fix". I did mention your use case on the report, though, so we'll have to wait and see.

/------------------------------------------------------------------------------------------------------/

[agile widget]

Hi @slow bobcat and @final prism just wanted to thank you again for your help the other day, thanks to you I found the issue and fixed it, at the end of the day I disabled all the Crowd Following Component since I couldn't make it work has I wanted, I'll look into another avoidance solution

[final prism]

be wary of that function, it is a gameplay debugger only function, that will not work in a shipped game. Getting the state names externally is classified as debug functionality so yeah

[final prism]

The navlink proxy is an actor containing a navlink component. It's more of an example of how you can setup a navlink contained in an actor. If you have a separate actor like a door, its best to put the component on that

[final prism]

Not quite sure what this means. The smartlink is just a wrapper around the actual navlink tied to the navmesh. It adds the callbacks so you can run custom logic when the navlinks are used, but it always ends up going down to the "true" navlink on the navmesh. It probably doesn't move when you use a smartlink because you didn't add any logic to move from point a to b in there, that's what the smart links are for, to add custom movement logic

[graceful wind]

Yeah, nothing wrong actually, just my lack of understanding. I create custom navlink component now for custom event to handle the navlink usage action and continue the navigation

[celest oracle]

Seems like we still have to do it that way ourselves for now.

[celest oracle]

Thanks for letting me know, I guess we have to do it ourselves manually for now

[simple crescent]

I'm trying to use linked state tree asset overrides with parameter overrides in 5.5. I can run the overridden state tree, but I haven't been able to get the parameter overrides to work. Are there any limitations on the parameters.
The linked state tree has a global gameplay parameter called "AbilityTag" (FGameplayTag).
I'm setting the override like this:

    FStateTreeReference StateTreeReference;
    StateTreeReference.SetStateTree(LoadedStateTree);
    FInstancedPropertyBag& Parameters = StateTreeReference.GetMutableParameters();
    Parameters.SetValueStruct(TEXT("AbilityTag"), AbilityTag);
    StateTreeComponent->AddLinkedStateTreeOverrides(StateTreeLinkedAssetTags[AbilityIndex], StateTreeReference);

I'm not seeing the overrides at all. The parameters still have the original default values.

[final prism]

What's the typical way to get context parameters from a state tree in code?

I know you can bind to them easily, but what if i want to do so in code?

[rich marsh]

What would be the best way for me to debug why this EQS Query starts to fail after a certain point (it's the far left Run EQS Query Node)?

I can see that both my pawn and the enemy pawn are on a navigation mesh still

[slow bobcat]

Check with visual logger, it has an EQS category. Will throw some clues

[static raven]

Hello guys, I have two quick question about Visual Logger.

one, should I expect a significant performance impact? and the other one, Is there a way to start recording on server initialization?

[stable plover]

Performance impact depends on your scene etc, I haven’t had any issues, it’s just a tool, for the next question idk, I think you may need to write custom code for it

[slow bobcat]

Drawing shapes with visual logger is extremely Un-optimized, it can tank your editor framerate to show a frame if you draw many things. The recording part is fine aside creating/saving the file and its size

About the second part, you can start/stop VisLog programatically. There's a command you can call from bp and code functions from c++

[static raven]

yeah, I was asking about the recording part, since I'm gonna be using it in build.

thanks @slow bobcat

[cyan sigil]

Anyone have recommendations for having crowds of NPCs spread out when following the same path? It seems like when going around a corner, every NPC crowds around the closest point to the wall instead of taking a wider path.

[static raven]

you are talking about regular AI? (no mass entity)

[cyan sigil]

That's correct.

[static raven]

I will have the same problem in the near future. I'm planning about implementing the team sense or a custom way to detect nearby friendly AI and implement some kind of restrictions when generating the actual path. My idea is to create some kind of pseudo flock behavior rules

[cyan sigil]

From what I was seeing, they're all generating the same path because the world is broken up into very large navmesh polygons, and they all choose to cut the corner as tight as possible and bunch together.

That's... not a bad idea. If I can get the NPC to wander across the nav polygon boundary they should move on instead of backtracking.

[static raven]

yeah, the navmesh works exactly like that, a bunch of polygons generated based on the resolution parameters. But you can modify the path generation algorithm if you want, I'm doing it for my flying mobs for instance

other approach, I think this could be my way to go, its to use the same data to desing one "leader", so every X mobs one will be the one who calculates the path, the other ones will somehow follow him in some kind for flock relationship, one they are close to an interest point, maybe the player, they can think by themselves again

I'm just sharing some of my very early ideas, sorry if I can not give you anything more tangible right now

[cyan sigil]

I can adjust the height and step of the virtual agent, but that doesn't make the polygons any smaller.

[graceful wind]

I use spline as guide for this, can use random right vector for each AI to move along the spline. but need to calculate the path several times. for example every 300 distance and need to have new path before the AI reach the destination to make it seamless. Might not the best approach though, but at least it works for me

[cyan sigil]

Do you generate the spline at runtime?

[graceful wind]

no, it is on editor, it is just for the traveling ai actually that the path already decided

[graceful wind]

have you tried this?

and also this, might help a bit for randomness ADetourCrowdAIController

[cyan sigil]

I am using both. That makes characters repel each other but they still try to pick the same point

[static raven]

making it smaller would not help either, they will still pick the best path, will be the same for everyone

[cyan sigil]

Not exactly. My project has tech where it can tell how many other agents are in a polygon and use that for cost.

[static raven]

oh I see, interesting aproach

so its kind of like increasing the cost of the polygon by a factor depending on agents inside it

[hasty turret]

by doing that, the path will be evaluated (and maybe recalculated) during move though.

[slow bobcat]

We do it all the time. You need to hack the engine a little but though.

[slow bobcat]

We used this approach (17:17) in a game. Works quite well.
https://youtu.be/O66mmqOjEy0?t=1037

The way I did it, instead of the angles for movement, was by using proxy actors for movement. Each AI has an assigned invisible actor (no mesh, no physics, just a scene component for the transform)

That actor is positioned to the side of the path segment (with a simple nav trace to ensure it's in a reachable spot from the path). Each AI following the path will place its proxy actor on one side at a specific distance, each distance is different. This we handle in a proxy actor manager. Every time a new actor was enabled for movement, we swapped the side (1st left, 2nd right, 3rd left) and place it further from the path (1st follows exactly the path, 2nd is AI radius * 2 from the path on the left, 3rd is ai radius *2 on the right, 4th is AI radius *4 on the left) and so on. Then, when they get close to the destination, we shorten that distance progressively

Thus works quite well when chasing a character (like in the talk I passed)

the main idea is to calculate just one path for your group of AI's

PA: proxy actor
the traces are nav traces

When the AI's are close, the proxy actors will match the chased character or the path goal

The trick is to pass the location to the PA. You can have couple functions, one that retrieves an Actor and another that retire es a path segment. For the actor you calculate the location in the PA actor tick (enable and dsable tick on a need basis). For the path you can pass the end point of the path when the new segment is assigned in the path following component. Then the PA location function can calculate the right location, asking for the assigned offset (L/R + distance) to the PA Manager

[slow bobcat]

About this: you can't use both. They contradict each other. If you use crowd control, RVO is not used and the other way around. At least outside mass. No idea if mass uses a novel approach that can combine both somehow

[cyan sigil]

Precisely. However that doesn't help as much if the polygons are huge.

[cyan sigil]

Ooh this is cool. However, it seems like it's for converging onto a particular point, when I want them to be able to go to several points (like filling a hall)

[cyan sigil]

Oh yeah? I know our project is using Crowd Control. I'll look more closely into RVO.

[slow bobcat]

ah I see. Your problem is that you have several enemies going from A to B (each a different A and B) but all paths go through the same point. LIke this no?

[cyan sigil]

Almost. I have several enemies starting (spawn point) and ending (player) at the same places. I have some tech that lets them spread out on straightaways (via a wander behavior), but they are forced to converge around corners. I'd like to be able to have them take longer paths.

[slow bobcat]

so... this?

[cyan sigil]

Yes.

We need the npcs to spread out as they travel so the player has more interesting thigns to shoot at. In the above setup they wind up going single-file, even with crowd avoidance, because they all have to path to that corner

[slow bobcat]

if this is more what you need, this is what my proposed solution does for paths

Because you only calculate 1 path (red) and deviate the movement to the rest

In our game we have coordinated patrols where the PatrolLead follows the patrol points and the rest jsut follow in pre-defined locations using the same concept of ProxyActor
The lead (Red) calculates the path and follows it in between patrol locations and his Followers (blue) get they PA's updated to be at the location of the Follow Locations (the black dyamonds)

designers can link spawners and place dyamonds wherever they want to create different formations etc

[graceful wind]

interesting, we need proxy actor as guide and let the follower ai recalculate the path to it right? this PA is not necessarily to be an actor, it might be arrays of relative location to the main actor

[slow bobcat]

The key thing of using an actor is that the engine will use the same movement request to re-path to the actor when it moves without causing a start/stop like when you use a location. There's nothing in the engine tracking if a point moved but there is for actors moving

If you use points, when you move the point you will need a new movement request (move ai to). If you use actors, the same request (move ai task) will use the path observer event On Goal actor moved and re-path without hitches

Hence the Proxy Actor usage

@stable plover used this exact same approach recently and he's happy with the results. The first 2 images I posted are form my explanation back in the day. I can vouch for simplicity and results when using proxy actors

[elfin smelt]

Hello guys
I encounter an issue when creating custom BTDecorator in C++.
I override CalculateRawConditionValue and it works fine when the Decorator is not EBTFlowAbortMode::Self
I have a case with the decorator on a Abort Self and isInverse to true.

The decorator return false all the time and it abort the Behavior Tree
I tried to find some sources on Decorator in C++ like good pratices and tips and didn't find much.

I tried to manage the abort self with the TickNode (found that in the BTDecorator_TimeLimit from Unreal) but it didn't manage the IsInverse in the tick and it was weird to manage manually the IsInverse in the tick.
If someone can give me some tracks to follow or tips ?

I think that overriding the CalculateRawConditionValue is fine when the Decorator is never Aborting self.
but I miss something to manage the Aborting part of a decorator

I found something, the UBTDecorator_BlueprintBase implemented the TickNode with the function EvaluateAbortType
It seems to be the functionality that I have miss

Edit: Yeah it was that, Custom decorator in C++ need to manage manually the different abort mode and the IsInversed in it

[neon moth]

hi guys, anyone know how to make the state tree select an instance to debug?
I have multiple enemies on scene and want to debug one of them

[harsh storm]

There is probably a dropdown somewhere in the ST itself - if it is like some of the other tools in UE

[neon moth]

I thought that, but didn't find anything like it (pic.1 - state tree)

normally would be at the top (pic.2 - common blueprint)

[harsh storm]

Look at the debugger itself

[neon moth]

debugger is empty, and the state is not highlighted,

even if there's only one enemy active

also tried the gameplay debugger (' key), since I use it a lot with BTs, but I guess state trees are not integrated with it yet

[celest oracle]

Usually for me it shows all instances of AI using the state tree in the debugger window. Not sure why yours is completely empty. there is supposed to be some debugger ui too

Maybe you need to use a different schema, for me i used statetree ai schema

You are using "statetree" schema without " AI "

There is supposed to be AI Controller in your context tab, i think the debugger connects through the AI controller

Edit : Lol nevermind just needed a unreal restart

[neon moth]

aaaa new features flow

the good old "close every window, restart unreal and reopen everything manually" solved it

thanks for the help, guys

[celest oracle]

I have a bigger problem right now actually lol, All my includes got messed up and I cant open my project got messed up

Edit: Fixed by deleting guilty template c++ files

[graceful wind]

lol, man, I thought it is a bugs and accept it, usually in btree the debug is seamless, thanks to you too

[maiden cargo]

For a TPS open worldish game. Is AI Perception the component to use currently or is there a better free alternative?
It works okay with low AI number, but does it scales?

[harsh storm]

Define "low AI number"

Because open world games typically don't have a bunch of enemies on screen anyway

[static raven]

this is really cool, but why actors? why not just handling the location in runtime?

[maiden cargo]

In this case, low AI number is 1-5, high would be 50 at most

[harsh storm]

I have it running fine on like 40+ ai

Have you actually hit perf problems?

[static raven]

it depends of how much AI activly perceiving stuff you have

if you have 50 AI roaming around but only 10, 15 actively seeing the player or hearing stuff, you should be fine

even 50 activly perciving something could be fine.

The most important part about performance when using perception system is YOUR implementation when processing the stimulus

try to do an event based implementation rather than just fetching information in a ticking component

[graceful wind]

I am more worried about ai animation cost than this

[static raven]

definitely, the "oh damn, you again" problem

[maiden cargo]

No, it was just a check to make sure it's what should be used or if there are better alternative already
Thanks!

[slow bobcat]

#gameplay-ai message

[maiden cargo]

Yeah, VAT/TuboSequence is what i'm looking at for this problem 🙂

[slow bobcat]

Duroxx has a good number, in a project I was it stopped working for any AI after the 20-ish. They never updated due to how perceptionamages its budget times. Depends a lot in what you do within your senses updates

[static raven]

The key thing of using an actor is that

[warped tiger]

What's SetValueStruct() returning? is it succeeding?
Can you try this instead?

FStateTreeReference StateTreeReference;
StateTreeReference.SetStateTree(LoadedStateTree);
FInstancedPropertyBag& Parameters = StateTreeReference.GetMutableParameters();
FConstStructView AbilityTagView(Abilitytag.GetScriptStruct(), AbilityTag.GetMemory());
const EPropertyBagResult PropertyBagResult = Parameters.SetValueStruct(TEXT("AbilityTag"), AbilityTagView);
if (PropertyBagResult != EPropertyBagResult::Success)
{
    // something went wrong
}

[simple crescent]

Thanks for following up @warped tiger. I figured it out. There is an additional step to call SetPropertyOverridden on the StateTreeReference.

StateTreeReference.SetPropertyOverridden(Parameters.FindPropertyDescByName(AbilityTagPropertyName)->ID, true);

[warped tiger]

Ohh, thanks for letting me know. Glad you got it working 🙂

[static crystal]

I'm confused at how to set up a custom state tree schema. I want to have my perception component accessable from context.

[stable plover]

Yea proxy actors are actually really great just like @slow bobcat said, plus they take some time to setup at the start but have them pooled and it’s really efficient

[stable plover]

What are you trying to do

[graceful wind]

Yeah, actor can make path calculation more efficient than point I guess. If I use point, I need to always recalculate full path everytime the main actor moves if I'm not mistaken

[slow bobcat]

Path calculation is exactly the same. What you do with actors is to get advantage of the path observer within the AI MoveTo task. It will seemesly re-calculate a path without causing a stop

[graceful wind]

yeah, this part, utilizing the observer is neat. otherwise I need to keep recalculating the path to the last known point, or stop first on complete then recalculate, which is not good

[static raven]

I have a obstacle avoidance function inside my PathFollowingComponent used by my flying mobs that adds new points to the path if there is a obstacle, so mobs can fly around.

You can totally manipulate the generated path with extra data.

In my case, to keep the system simple, I'm moving the mobs directly to the target point, so my path its just two points. Then I add one or two in the middle to fly around any detected obstacle

Sometimes the decision about using one thing vs another its just asking what fits better to the game and the already existing code.

[neon moth]

hi all, me again with state trees again
is there a way to update a variable within a state tree (like a for loop between states)?

I have a list of waypoints and the routine is simple: moveToNextWaypoint, shoot, wait, repeat until the last one. Each one is a separate state
the MoveTo state completes when the unit reaches the waypoint. On the next cycle, I want to input the next waypoint

With BTs I would simply update a "NextWaypoint" variable in the BlackBoard when MoveTo is done. How do I store variables using State Trees?

I bypassed storing it in the enemy actor and changing directly but what I really wanted was a way for the state tree to update the value of a bound variable

[graceful wind]

I see. i always want to play around with path modification, i know graph algorithm a bit, but I don't know where to start. looks like this path following component is good start to look into. I'm sure I will reach the point that i will have to do this next.

[static raven]

I'm having an issue I neven had before, I have a tall AI character, not too much, but it seems its over the limit by something like half a meter.

I cannot find a way to change the default AgentHeight. I tried in ProjectSettings>Navigation, tried overriding the default value from DefaultEngine.ini file

[/Script/NavigationSystem.RecastNavMesh]
AgentHeight=300.0

Nothing seems to work

[simple crescent]

The agent height lives in the nav agent properties defined in the project settings - engine - navigation system - Agents - Supported Agents

You can then override it in the RecastNavMesh object on the level that is placed when you add a navmesh bounds volume

I think the value in the project settings (stored in DefaultEngine.ini) is copied into the recast navmesh object when you place the navmesh bounds volume, so changing after you add the volume won't have any effect.

[static raven]

I'll try adding the "Default" supported agent then, but not having any defined agent and changing default values in DefaultEngine.ini shouldn't be the same as creating the "Default" agent?

[simple crescent]

Do you have multiple supported agents, or only the default?

What do you have for the Agent Height in the RecastNavMesh-Default (or whatever other supported agents you have)?

Actually, I was wrong. Changing the height in a supported agent setting in the project settings (DefaultEngine.ini) does change the respective RecastNavMesh object the next time the level is loaded.

[static raven]

only the default one

still nothing works

default one = default values from ini file

[simple crescent]

Is there any chance you have the Mark Low Height Areas option checked?

[static raven]

nope, false

wait, reading the code, that variable doesn't do anything if bMarkLowHeightAreas its set to false, right?

[proper pawn]

navigation system question for y'all:

I'm trying to cost out a feature where AI agents can jump onto/off of a spline-driven zipline for faster movement over long distances

I figured the way to do this was to make a NavLinkProxy every X meters along the zipline but the more I dig into that it doesn't look like it'll give me anything more than connectivity between two navmesh points, even though it does have the interface for taking/returning control to the path follower

What's the thing I should be looking into so I can actually write the bits that the pathfinding can use to cost it? I've looked through the code for NavigationSegmentLinks but there's also comments warning about them being broken

[slow bobcat]

You can try something: set a straight spline and put a nav modifier volume to generate an area alongside it where you reduce the cost (not sure now if the default cost is 1 or 0. If it's 0 maybe you can use a negative value? Not sure about this part) and then add the nav links as you said.
The path calculation will favour using the spline because it reduces transversal cost over walking on the ground

[proper pawn]

How would the modifier volume work? The zipline itself has no navmesh to modify, unless there's something about modifier volumes I'm missing.

Conceptually I want something I can give the nav system where I have nodes with costs my code has calculated that it can consider when pathing (which are then linked to the navmesh)

[slow bobcat]

Ah.... There's no nav... OK I misunderstood that part.
The easy way would be to have an invisible mesh in the spline that can affect navigation, there are examples where people generate pipes and cables from splines in unreal. You could do the same but with an invisible mesh. Then you will get nav.

Whay you mentioned is also possible but not sure how. You would need to check into the detour code. Maybe you can add them as small nav polys in a fake tile? Not sure how. Nav links are simply references in detour polys to other polys, there's nothing specially added

[solid pond]

I found a section of the ai perception where it says that the delegate for when our target perception is forgotten should have the update stimulus yet how it is define it only contains the actor in the delegate. Is this just incorrect docs or a mistake from the code or is there something I am missing here?

DECLARE_DYNAMIC_MULTICAST_DELEGATE_OneParam(FActorPerceptionForgetUpdatedDelegate, AActor*, Actor);


    /**
     * Notifies all bound delegates that the perception info has been forgotten for a given target.
     * The notification get broadcast when all stimuli of a given target expire. Note that this
     * functionality requires the the actor forgetting must be enabled via AIPerceptionSystem.bForgetStaleActors.
     *
     * @param    SourceActor    Actor associated to the stimulus (can not be null)
     * @param    Stimulus    Updated stimulus
     */
    UPROPERTY(BlueprintAssignable)
    FActorPerceptionForgetUpdatedDelegate OnTargetPerceptionForgotten;

[slow bobcat]

it broadcast when all the stimuli expires, not sure why the comment says otherwise
There are two delegates, one for forgotten (The one you posted) which will not need stimuli passed because they are all "invalid", and there's update, that will call when each stimuli update. That one passes the stimuli updatinge

[slow bobcat]

it seems they copy/pasted the coment and forgot to edit the variables part

[elfin smelt]

Hello
Do you have tips, things to search/look at when you want to optimize Behavior Tree ?
I thought that passing BTNode in C++ without instantiation would optimize the Behavior Tree Launch but it didn't do much.

When AIs starting their BT, the BTComp could take in the worst case 7 ms on the frame (GameThread)
I'm using the task RunBehavior but maybe it is something that I should avoid ?

[slow bobcat]

that sounds very very off.
What in RunBehavior is taking so much time when using Unreal Insights with -trace=cpu -statednamedevents?

[elfin smelt]

each small part is a node (BTDecorator_Blackboard, BTComposite_Selector, CustomTasks, CustomServices, etc)

[slow bobcat]

ok so you have a massive BT?

[elfin smelt]

It is in Editor so maybe I'm just way off because I'm not in the right environment 🤔 (but I think 8 ms won't be 0.5 ms in a build anyway)

[slow bobcat]

ah yeah

do not measue in editor these things. If there are ensures or debug...

Check in a Development build at least. Or even better, a Test build

[elfin smelt]

It's a BT with 23 nodes but 4 nodes are RunBehavior
and the biggest RunBehavior has 248 nodes :/

[slow bobcat]

ufff yeah. But even with that, many would be excluded due to decorators and branching, it's not that you run the tree fully constantly

248 nodes is... a nightmare for me. We tend to split the trees in sub-trees when possible (which has other issues)

[elfin smelt]

Ok I will go profile that in build
Maybe in Editor, it load all the Node and BT in the same frame for debug purpose
Because I don't have anything that take so much time in the behavior and it appends only on the first instantiation of the Behavior Tree

[elfin smelt]

You have totally right
I wanted to split more in state (FSM launching BT) AI but my DT said "More State equal more bugs and more transition." soo I have an yggdrazil that I will be happy to burn when the game will be ship x)

[slow bobcat]

yeah your DT will have a hard time if at any point someone says "ah but we need this new behavior that should interrupt this other behavior" and then... tree balancing.
Been there, done that, nightmare

you can achieve the same splitting in subtrees if you don't want more higher level states

[elfin smelt]

thats why I have 248 nodes and just the BT is not the right tools for the AI

[harsh storm]

I'd cry if I had to deal with 248 nodes

[elfin smelt]

I used Sub Tree for reusing same sequence of action in BTs but Yeah I have not splitting just for better understanding of behavior

[harsh storm]

Having 248 nodes is what gives those incorrect talks, "Behavior Trees are bad"

[slow bobcat]

but... it is. Or... it is most of the time.
I see the problem as: "no no no, don't split the code in clases. More classes means more bugs. Better to have the entire game in one single bug class. And also don't use functions, just one. More functions, more bugs"

[harsh storm]

Do you really need anything except int main()?

[slow bobcat]

not me. I just need Java Script and main

and off we go to the races of Gaming

XD

[elfin smelt]

Yeah, I have already tried to talk about it but hey... I have not found the right ear soooo I will ship this yggdrazil and then try to move (already tried but no success for now)

[slow bobcat]

Anyway Bapti, I would check in a Test Build with -trace=cpu -statenamedevents and, if that's bad still, then jump to a development build and see where can you cut off things

[stable plover]

But a test build is only available in source right?

[slow bobcat]

you can use the project launcher to build any type of build you want

[stable plover]

Didn’t know that thanks

[static raven]

I finally found why I was not able to set my navigation system to use bigger mobs.

UWorld::InitWorld calls the function FNavigationSystem::AddNavigationSystemToWorld that grabs the navigation class used for that particular world (by default NavigationSystemV1).

inside NavigationSystemV1::PostInitProperties the function "ApplySuppertedAgentFilters" get called, which populates the SupportedAgents array with the CDO value.

If your array its empty (by default it is) it calls the function FNavigationSystem::GetFallbackNavDataConfig, which uses two HARDCODED values to initialize the default RecastNavMeshData.

So, setting the values for AgentRadius or AgentHeight in DefaultEngine.ini has no effect, since were being overriden by those hardcoded values when the world was initialized, (even if you had change the values in the ReacastNavMesh actor).

Adding a new element to SupportedAgents from ProjectSettings was not working either since I had my custom class so inside DefaultEngine the array was being populated only for my custom class, not for NavigationSystemV1.

So the solution is to set the NavSystem to the correct subclass in world settings or add the element to SupportedAgents to NavigationSystemV1 base class inside DefaultEngine.ini.

[rich marsh]

I have a MoveTo node in my BT. I was expecting that, when a service modifies the MoveGoal's value, the AI character would change direction, moving to the updated value. But it doesn't seem to do that. Isn't that the whole purpose of TrackMovingGoal?

If my custom service updates the move goal in the middle of the character running to a far away point (for testing, my first point is deliberately far away - ~4k distance), the character keeps running to the originally requested position.

[slow bobcat]

Move to runs AI MoveTo task. Inside the path is calculated and a path observer listens to path events. There's an event called OnGoalActor moved, that will then recalculate the path within the same MoveTo (no stop/start happens). For locations there isn't any code to do it. I'm unsure if the track. Moving goal bool does anything. If I were you, I would check the code. Might not be used

[rich marsh]

Ok thanks. I will do that then.

[simple crescent]

Thanks for the update @static raven. Glad you got it to work

[slow bobcat]

Does it say anything if you hover the variable with the mouse?

[rich marsh]

The comment says:
/** if set, path to goal actor will update itself when actor moves */

It looks like the values are copied to the move request. So, at least when using a location vector, it wouldn't be able to react to changes.

I did try to change the key type to be an actor and use the updated actor instead, but that also did not work

I'll dig into the code some more. Maybe there is a way to brute force what I want

Basically, I have different goal actors I have the enemy character run to. If a goal actor becomes no longer navigable (a player shuts a door, blocking the only path to the originally selected goal actor), then I want the enemy AI to run to the next available one.

But I don't want it to have to run all the way to the original goal actor before it tries to go to the next one. That would not work with the current design very well.

At a glance, it looks like maybe if bUseContinuousTracking was enabled, and I used an actor as the key type, then I might get what I want

But bUseContinuousTracking is set to false at the constructor level of UAITask_MoveTo, so maybe I can override that with a child class

I should be able to. The property is protected. I'll try that and see what happens

[slow bobcat]

That's not to swap goal actors.
If you want to detect whether the target is reachable and interrupt the chase, you have 2 ways:

• you can use a service that runs while your move to task runs and checks now and the if the target is reachable and store the result of said check in a variable. The same move to node also has a decorator that aborts itself if the variable "reachable" is false (set by the aforementioned service).
• you do a move to without supporting partial paths. When the target moves to a unreachable area (not in nav), the move to will finish because a complete path can't be calculated.

Whenever that happens (any of the two I described), the tree will re-evaluate, pick a new target and move to it

Correct me if I'm wrong, but I have the feeling that you think that those variables are for "whenever I change the bbk I use to move to, it will then correct to it", but that not the case

[rich marsh]

Thanks for your assistance and yes, you are 100% understanding my intention.

[slow bobcat]

I'm not 100% sure but I believe continuous tracking is used to ignore the offset to know when the actor moved to far from the original location to recalculate and instead recalculates as soon as it move 1cm

[rich marsh]

Oh ok

I was trying to find a way not to leave the current sequence. It sounds like what I want to do, in the way I am wanting to do it, is not supported.

I would, in one way or another, have to abort / fail the current task and then start over, using a new goal / actor.

[slow bobcat]

A bbk is just a way to point to things for the BT to access easily. The BB is container of variables to have a centralised way to access variables for the BT.
Your bbk is pointing to an actor and your move request is saying "chase this actor", but it knows nothing about the bbk. That's between the BT and the BB

[slow bobcat]

Exactly

[rich marsh]

Or, maybe I could build my own wrapper to do what I want, making it an instanced task and track the initial values with the current.

[slow bobcat]

Honestly, not worth it. You are better reevaluating that re-inventign the wheel

Great exercise though

[rich marsh]

Essentially, I have an enemy char that will, when aggroed, run up and steal the item the player is holding. The enemy then will try to flee (if the last time it tried to flee was greater than the allowed time since last attempt) to return the item back to the hive. If the enemy runs out of available hives or gets stuck, then it will try to attack those around it. So I was trying to simplify the graph a bit.

So right now, I have things cleanly separated between the "Enemy is fleeing / trying to drop the stolen item off at the hive" from the "I'm hostile and I am going to attack you" tasks

Having to abort / fail that task makes things a lot more complicated, because now I have to somehow implement a retry and specify those conditions with the graph. Maybe I am overcomplicating things in my head here, but it seems like that is going to really make this graph ugly.

[slow bobcat]

To me is a tree with flee, steal item, drop item etc.
If you enter the drop item branch, pick a hive. If something fails, abort, tree re-evaluates, ends up in "drip. Item" again, pick a hive... No hive? Do the next thing to do while having an item

But you know better what works for your game. If you think having a tasks that handles the te-selection etc is better, go for it. It might as well be a sub tree

[rich marsh]

If it fails, and the fleeing time limit has not been yet passed, then it will stop trying to flee all together and just be stuck trying to attack.

I'm more thinking out loud here. Obviously I don't spend all of my time making behavior trees, so you are probably 100% correct in warning me that what I am suggesting is not a good idea, I am just trying to think through how I can retain my clean separation while also achieving what I want with the abort / fail approach.

[slow bobcat]

My suggestion is that you separate each behavior in a tree and you create a high level tree that decides which of those behaviors to run. That will help a lot to encapsulate behaviors and see how revaluations will go

[rich marsh]

Additionally, when I looked over the UBTTask_MoveTo source, it didn't seem like it was doing a whole lot inside of there. It looked like it was mainly prepping / initializing things, with the main bulk of the work being done in the UAITask_MoveTo class.

And ok, I will put some thought into that, thank you.

[oblique basin]

I think you could do what you want if you're willing to make a child of your moveto ai task. When you call it, you create an object to that task that you could reference inside a delegate that's called when the state you're describing changes.

Or more simply just deactivate the move to task being run and recreate with a new target

[static raven]

just to add something to this:

when using a vector as target for movement, UBTTask_MoveTo listen to that blackboard key value changes and trigger re-path on value change.

you can set ObservedBlackboardValueTolerance to control how much change is needed to trigger the re-path.

sometimes this is very useful so you only have one "interestingLocation" that can be modified by multiples places

[oblique basin]

So I'd handle it like this:

• create a move to ai task with your initial params and call activate
• set up a delegate that you broadcast whenever you hit a condition that requires a change in target
• have your ai task bind a function to that delegate that restarts the process with new parameters

In reality it'd be far easier to just re-run the sequence and make sure any logic that is independent of the sequence entirely external

But fun exercise if you fancy getting your grubby hands all over c++

[rich marsh]

Thanks. Yeah I am weighing the different options. When I looked at the source code of ubtmoveto_task, it seemed like it wouldn't be too hard to achieve what I wanted to do.

[slow bobcat]

Look into AIMoveto_Task, not the BT task, but the async task

[rich marsh]

Ok

Yeah that's what the BT creates from what I saw

and also where the bulk of that work is done

[oblique basin]

My general advice would be to not fight the engine. You can curse it all you want for not having a specific application you want, but there will always be a way as long as you colour inside the line.

[celest python]

How can I debug a stimuli component in an actor doesnt make perception component of an AI perceive the owner?

TL;DR Base character has stimuli component, auto registering sight sense, AI characters have perception in controller and has sight sense config

Perception is active but perceptionupdated is not executed ever

[graceful wind]

the last time I encountered this was because of detection affiliation, by default I think it only want to detect enemies. or implement GenericTeamID on ai controller if it is the problem

[obtuse igloo]

Anyone know a workaround equivalent of this unity nav? https://www.youtube.com/watch?v=LBh8XZHTnLg

I'm specifically trying to make it so that my AI can start climbing anywhere on a climbable wall instead of just at the start of a nav link. (ex. if they're in the air from jumping and they hit a climbable wall, they continue their navigation to their destination by climbing toward it)

If all else fails, might just make a custom movement mode for wall climbing and make functions for climbing a wall when the AI detects they're running into one by tracing climbable walls, then temporarily pause navigation on the Nav Mesh to their destination until they get to the top of the wall they're climbing.
Although, this would make it so that they have to climb up the wall no matter what before deciding what to do next after getting to the top.

For reference I'm trying to get something similar to SA2 Chao Garden climbing that the Chaos do if anyone knows! ( timestamped https://youtu.be/bUWuGzRuySE?si=OjMEOFd66aYobmSm&t=2708 )

[slow bobcat]

Gameplay debugger and/or visual logger will give you a good overview.

[slow bobcat]

I think you can use the same approach I used for covers.
My logic was inspired by this article https://horugame.com/real-time-dynamic-cover-system-for-unreal-engine-4/

The approach will not work in areas where nav is dynamically generated due to it being expensive, but if you use Static or Dynamic with Modifiers Only (or you use dynamic but you load Nav Chunks for certain sublevels or tiles) it will help.

On a high level, you can use the nav mesh data to define the entry points of a wall.
What you do is, just after you generated the nav (during cooking your build or in editor) , you have a proces that loops all the tiles in the nav. In each time you look for the nav edges. A nav edge is the darker-green line you see next to the bottom of the wall, where the nav cuts out, the nav limit - > nav edge. If there aren't any nav edges, go to the next tile because that tile is continuous (doesn't have objects in it affecting navigation).
A nav edge has a start and an end point, which gives you the direction of it. Using that direction you can split the nav edge in parts, let's say a point every 30cm.
Now, member that Edge direction? If you cross product that with FVector::Up, it will give you a vector that points inwards the nav. If you do a cross product with -FVector::Up (down), it will give you the direction towards outside, which will point (potentially) to your wall. If you line trace from the point you calculated towards that direction, you can check against the object next to that nav edge. If it's a wall, record the point in the nav edge (more on this later).
You found a point, move to the next: another 30cm towards the nav edge direction and so on.
At some point you will have checked all the tiles in the game and found all the connection points to your walls.
Now things to consider:

• I recommend storing the points in an octree. When you need to find one it will be super fast and you will be able to do things like

"check if there's a wall entry point in X meters radius" from an EQS

• when analysing the tiles, you can use async tasks in background threads, one per tile. That will make the process super fast. You just need to make the taks report back the points found (if any) to whoever triggers the tasks and, once all the tasks are done, add the points to the octree

Something like that (ignore the blue line, I'm reusing an image for a different system that can be solved in a similar way)

That red hit on the wall will say "yep, the edge point is good as a wall entry point"

If you want to do this dynamically... I guess you could monitor the affected tiles and check5of they have nav edges. If they do, launch a background thread task to analyse it. Your AI will not be able to jump into the wall until that's done but... Sounds doable

The blue and red traces are yo check the heigh of the wall. You might want to record different points (2 octrees) for low and high walls. Try blue, hit? It's a high wall. No hit? Try red. Hit? Low wall. No hit? Something too small to be a wall

Or you could use the blue to define walls that are way too tall.

The good thing is that you don't need to label assets as walls or anything. If the measures check... It's a wall.
The for the put point... I guess you can figure that one out when you get there

[hot hollow]

Does anyone know what I am missing? Right now one of the AI is stuck and wont move, this goes for all the levels

[slow bobcat]

If you print the failure reason, what does it say?

[celest oracle]

Whoa @slow bobcat thanks for sharing in detail.

[rich marsh]

@slow bobcatAs I am reading through more and more of this source code, I am a little confused on how my scenario isn't already supported. The code really is giving me the impression that it is meant to dynamically handle position changes via OnBloackboardValueChange

Specifically, this portion of that method:

// this means the move has already started. 
if (MyMemory != nullptr && BehaviorComp->GetAIOwner() != nullptr)
{
    check(BehaviorComp->GetAIOwner()->GetPathFollowingComponent());

    bool bUpdateMove = true;
    // check if new goal is almost identical to previous one
    if (BlackboardKey.SelectedKeyType == UBlackboardKeyType_Vector::StaticClass())
    {
        const FVector TargetLocation = Blackboard.GetValue<UBlackboardKeyType_Vector>(BlackboardKey.GetSelectedKeyID());

        bUpdateMove = (FVector::DistSquared(TargetLocation, MyMemory->PreviousGoalLocation) > FMath::Square(ObservedBlackboardValueTolerance.GetValue(Blackboard)));
    }

    if (bUpdateMove)
    {
        // don't abort move if using AI tasks - it will mess things up
        if (MyMemory->MoveRequestID.IsValid())
        {
            UE_VLOG(MyController, LogBehaviorTree, Log, TEXT("Blackboard value for goal has changed, aborting current move request"));
            StopWaitingForMessages(*BehaviorComp);
            BehaviorComp->GetAIOwner()->GetPathFollowingComponent()->AbortMove(*this, FPathFollowingResultFlags::NewRequest, MyMemory->MoveRequestID, EPathFollowingVelocityMode::Keep);
        }

        const EBTNodeResult::Type NodeResult = PerformMoveTask(*BehaviorComp, RawMemory);
        if (NodeResult != EBTNodeResult::InProgress)
        {
            FinishLatentTask(*BehaviorComp, NodeResult);
        }
    }
}

return EBlackboardNotificationResult::ContinueObserving;

So then, with the character positioned at 5,50,1, moving to the original location of 111,111,1 and having the MoveGoal vector change to 777,777,1 why wouldn't that cause the move task to change the character's movement / final destination?

[slow bobcat]

Yep, that's is made to work. No idea why it doesn't. You would need to debug that code

I never realised you could call abort with a NewRequest + the requests ID to do that

[rich marsh]

Ok, cool. That's fine. I can try to debug. I just wanted to make sure that I wasn't going crazy

[celest python]

hmm, lets see. thank you

[celest python]

I only see the perception shape debug, not why it doesnt perceive stimulis

[rich marsh]

So... I can see that bObserveBlackboardValue is 0 (false)... but it defaults to 1 (true) and I don't see anywhere where that is being set to 0

[slow bobcat]

No that is not default to true, that's the bit, it means the uint32 will use just one bit, but the default value will be 0

[rich marsh]

Oh, ok

Untested / incomplete funcationality then? Nowhere does that value change.

Yet there is code around it

Well, inheriting from the UBTTask_MoveTo and setting that to true explicitly in ExecuteTask made it work

[slow bobcat]

Ni idea tbh

[rich marsh]

So, shrugs

[slow bobcat]

Let me know how it goes

[rich marsh]

Sure. So far it is workign.

*working

[obtuse igloo]

Holy hell, thank you so much for this! I’m a beginner at this but this gives me an idea and direction on how to get what I want, and I appreciate it!! 🥺

I’m gonna try to figure this out the next time I have time!

[slow bobcat]

Really? Niiice.

[slow bobcat]

Let us know how it goes

[obtuse igloo]

🫡 😤

[celest python]

Setting a different TeamID to perceptioncomp's owner didnt work, any other ideas?

Gosh I hate AIModule so much

I guess I'll deep dive into source

[slow bobcat]

The whole perception API is a fucking nightmare for me.

[celest python]

At my first job they made me do a simpler version of perception and after that never worked on perception systems so I'm also starting as someone who is completely alien to API, its 2x more brutal to work with that way 😦

[slow bobcat]

I'm on my 4th profesional game and I still think it's torture

[simple crescent]

You can eliminate the team id/attitude being a problem by checking all of the team attitude boxes in the perception component (hostile/friendly/neutral).

[lucid ingot]

Is there an easy way to make it so that an AI is moving to a location, the AI will wait to continue executing any nodes until it reaches the location, but if the player has moved, the AI will restart this chain if the player is outside a certain radius of the last location? I'm struggling to map out how this would work in behavior trees

So far this is what I've come up with, but my problem is that the AI will attack the player before it reaches its destination

Oh I guess this actually works, partly

But if there is a better way, I'd love to hear it

[graceful wind]

lol, I think the last time I implement the generic team id the hearing is working, but the sight not, or hearing I forgot. Ended up disabling faction filter for one of them and do team checking on the triggered event it self. and dont dig any deeper on this since i still need to learn other things. but now I am glad I didnt

[hot hollow]

Didnt know I could do that, but AI sais "aborted"

[slow bobcat]

Ok and to confirm: there is nav right? If you eject (f8) and press P while playing, you can see the green nav mesh right?

[hot hollow]

Yup this is nav

[slow bobcat]

Ok, can you debug draw a sphere at the movement location that is failing and show it here please?

[hot hollow]

Wasn't sure best way to do this, I couldnt spawn an boject but I could do a sphere trace

[slow bobcat]

Ummm does it move it you tell it to go to some random location? Let's say 000 or something?

[hot hollow]

Ill test

[slow bobcat]

Btw you have functions like DrawDebugSpehere and the likes

[hot hollow]

Ah good to know

[hot hollow]

Still stuck like glue, he does see me and he does hit me when I am close, its just his movement it seems like

[slow bobcat]

Ummm
You could try not run the BT and run an AI MoveTo node in begin play to some random location, see if that works

Let's discard BT shenanigans

[slow bobcat]

The other thing I could think of is soemthing related to settings in the velocity. Is path calculation OK? If you look in visual logger, does it show the path calculated in the navigation category?

[hot hollow]

hmm, sorry do you know where the visual logger is?

Ill google it

[slow bobcat]

Tools - debug - visual logger.

Check a video on how to use it, but pretty much hit record, play, do the thing, stop recording. Right click the navigation category - disable all but this

And check the marks for each frame for the character. The very first one would show the path

[hot hollow]

LogBehaviorTree (VeryVerbose) BT(94633) schedule next tick 0.000000, asked 0.000000.
LogBehaviorTree (VeryVerbose) Suspending branch actions.
LogBehaviorTree (VeryVerbose) Resuming branch actions.
LogBehaviorTree (Log) Task BT_Enemy_Melee::BTT_MoveToIdealRange[20] finished: InProgress
LogBehaviorTree (VeryVerbose) Resuming branch actions.
LogAINavigation (Error) MoveTo request failed due missing PathFollowingComponent
LogAINavigation (Log) MoveTo: Actor(BP_ThirdPersonCharacter_C_0) Mode(partial path) Filter(None) AcceptanceRadius(100.0 + overlap)
LogBehaviorTree (VeryVerbose) Suspending branch actions.
LogBehaviorTree (Log) Execute task: BT_Enemy_Melee::BTT_MoveToIdealRange[20]
LogBehaviorTree (Verbose) Testing node: BT_Enemy_Melee::Get Close[17]
LogBehaviorTree (VeryVerbose) Suspending branch actions.
LogBehaviorTree (Verbose) Ticking aux node: BT_Enemy_Melee::BTD_IsTargetAttacking[10]
LogBehaviorTree (VeryVerbose) Resuming branch actions.
LogBehaviorTree (Verbose) Deactivate node: BT_Enemy_Melee::BTT_SetMovementSpeed[19]

This error line is highlighted for me
LogAINavigation (Error) MoveTo request failed due missing PathFollowingComponent
LogAINavigation (Log) MoveTo: Actor(BP_ThirdPersonCharacter_C_0) Mode(partial path) Filter(None) AcceptanceRadius(100.0 + overlap)

[slow bobcat]

You don't have a path follwoogm component?

How?

Did you make the AI Controller yourself and forgot to add that?

[hot hollow]

There is a chance I have deleted it

Most likely since everything stopped working on friday

[slow bobcat]

Yehq you can't path without a path component

[hot hollow]

Do you know what the component is called? I cant find it

Or it might be this

[slow bobcat]

Yeah that's the one

[hot hollow]

This was here, to be honest Im still a bit lost on how to get it working again

Its been awhile and I never learned this deeply

[oblique basin]

Is the AI controller possessing the pawn (check if get pawn is valid)? Movement component? Any attached static meshes with physics collision on that could be blocking movement?

[slow bobcat]

If that's there and it still complains...

• check your character is pointing to your AI Controller
• check what Sean mentioned above about the controller actually possessing the pawn

[hot hollow]

Ill check

Checking movement component next

Is there a way to delete and re-add or reset all settings on movement component?

[slow bobcat]

Is the log not complaining about the path following component anymore?

[hot hollow]

I should be still complaining as I havent changed anything

[oblique basin]

I think it's the lack of movement component

[hot hollow]

Except for the prints,, but the aic already had the path following component

[oblique basin]

Pawns usually use a floating pawn movement component. Try adding one to your base class and see if they start trotting

I've almost exclusively used character classes for ai which comes with a movement component

[slow bobcat]

Also delete all AI's but one, having more than one only complicates things

[hot hollow]

Just adding floatingpawnmovement didnt work, and still stuck when using as a world context object

[slow bobcat]

A character is a pawn. CMC should work just fine.
@hot hollow what's the base class of your AI's?

[hot hollow]

Parent class is Character

[oblique basin]

I think it's a pawn no?

[slow bobcat]

Yeah then don't use floating pawn.thats a but outdated (although it has its own usages nowadays)

[hot hollow]

Goddam I cant delete it

[slow bobcat]

Umm weird. What are the settings in your movement component for accelerarion, max speed etc?

[oblique basin]

What was this screenshot of then?

[slow bobcat]

Make sure you are not playing and save first, just in case

[oblique basin]

Oh sorry that was your controller right

[slow bobcat]

The AI Controller

[oblique basin]

Sorry my bad

[hot hollow]

Yup sorry a bit fast in the swings, was playing

[hot hollow]

I am not seeing anything broken here

[slow bobcat]

project the point on nav before sending it to the AI Move To

[oblique basin]

And if that doesn't work, try adding a 2 sec delay before calling move to. May be that some dependency hasn't fully initialised when called. Just to check it off the list.

[hot hollow]

we trying directly in the character bp?

Hey I checked out autosaves, seems like I have an autosave before it broke, might be easier to just check that one out

[hot hollow]

Tried replacing the ai character to no success but replacing the AIC seemed to work. Thank you so much Bruno and Sean for being so helpful!

[slow bobcat]

Sorry didn't get the solution. What was going on then?

Also: please set up a github locally for version control so you can diff with previous changes etc

[oblique basin]

It's possible the characters begin playing was being called before ai controller on possess, which is why I suggested the delay. Would've suggested moving the logic over to the controller.

Glad you got it sorted anyway.

[hot hollow]

No clue what the issue was (just overwrote using an autosave from two weeks ago), but definitely going to setup a version control. Thanks guys:)

[neon sparrow]

For Ai Rotating Toward a Character how to Calculate Delta Angle Between Character and Ai (It Was Animal )so i can play it left and right Rotate Animation

like what should i take from Ai to calculate delta Rotationo

[final prism]

What's the typical way to expose behaviour info from state tree externally? I've been thinking of setting gameplay tags on the character, but not sure if there's a better way

[obtuse dome]

If you want a Rotator, use Find Look At Rotation. You probably want to take the world location of your character and the world location of the animal with the Z axes zeroed out so that if they are at different heights in the world, you don't rotate the character so that they flop over unbalanced (you would need to implement headtracking or something else if you wanted them to look at a higher/lower location). If you just want a float angle, then you can use atan2 https://math.stackexchange.com/questions/707673/find-angle-in-degrees-from-one-point-to-another-in-2d-space, again discarding the Z position

[neon sparrow]

ohk 🫡

[slow bobcat]

has anybody ever got this error in visual logger from the Behavior Tree Component?
I'm having it and I have no idea what is this about or if I should be worried

    if (bTickedOnce && NextTickDeltaTime == 0.0f && bUseWorldDTForCurrentFrame)
    {
        const double CurrentGameTime = World->GetTimeSeconds();
        if (CurrentGameTime - LastRequestedDeltaTimeGameTime - CurrentFrameDeltaTime > KINDA_SMALL_NUMBER)
        {
            UE_VLOG(GetOwner(), LogBehaviorTree, Error, TEXT("BT(%llu) expected to be tick next frame, current deltatime(%f) and calculated deltatime(%f)."), GFrameCounter, CurrentFrameDeltaTime, CurrentGameTime - LastRequestedDeltaTimeGameTime);
        }
    }```

[graceful wind]

guys, how usually you pass state tree event transition payload to the next state? the payload automationcally become output, but cant bind it on the next state

this next state is the child of Root, the transition happened in Root

[slow bobcat]

    unsigned short downDirPolyFlag = 0;

    /// User defined flags assigned to upwards traversal of the off-mesh connections
    unsigned short upDirPolyFlag = 0;

    /// User defined area id assigned to downards traversal of the off-mesh connections
    unsigned char downDirArea = 0;

    /// User defined area id assigned to upwards traversal of the off-mesh connections
    unsigned char upDirArea = 0;

has anybody used this new flags in navlinks from 5.6? what is the intended use?

[slow bobcat]

you need to tick the tick for the Event enter condition in the new State.
Then assign the required payload.

Then in a task you can have a Parameter value with the type of the payload and bind to it

[graceful wind]

It works, thanks man!!

[nimble smelt]

I've got this AI pawn that's using FloatingPawnMovement with just a real simple MoveDirectlyTowards command set. Trouble is, while I can change the max speed of this thing, it acts like it has infinite acceleration.
I have the acceleration set to a really anemic 0.01 with no turning boost, but look at how it still manages to change directions instantly

[static raven]

It seems that claiming AILogic resource does not prevent PerceptionSystem from tick, anyone has deal with this?

One idea is to make my own implementation of resources into perception system but it means a lot of code, anyone has solve this with less heavy modifications to the in-build implementation?

[slow bobcat]

You can enable/disable senses at runtime. You can do a simple function that handles all of the senses you use

[static raven]

hmm, good idea, I'll try going in that direction, thanks!

[static raven]

could be that you actor is not actually turning but bouncing?

[nimble smelt]

Nah, it's got zero collision, these things fly at you right through the walls like ghosts

Solved it! Turns out you need to have UseAccelerationForPaths set in Nav Movement even if you aren't using the navigation system

[celest python]

this perception system thing is a dumpster fire

[harsh storm]

Show me on the doll where perception hurt you

[celest python]

i cant debug why perceptioncomponent can not perceive my stimulicomponent on sight sense without deep diving into source code

[harsh storm]

Did you set the collision channel to like ignore or something?

By default the collision channel for the sight trace is like Camera I believe

You can see it in the project settings

[slow bobcat]

Does this sentence mean you added a stimuli component to the un-detected character? If that's the case, remove it. Pawns are automatically registered as stimuli.

[harsh storm]

I actually turn that feature off. I think it is annoying personally.

[celest python]

Didnt change anything unfortunately. this project is something I work on after my day job ends so I was avoiding putting energy to debug AIModule but I guess I have no other option

[slow bobcat]

it's probably something super stupid. My guess is the same as Durox: something in your physics configuration, but you will have to check
Maybe it would be a good idea to watch a youtube video about how to setup senses, see if you are missing a step somewhere or something is off

[celest python]

thanks 🤞

[elfin smelt]

So I tested in build my yggdrasil creation and it is not 8 ms but "only" 3.2 ms with a Load time of 2.6 ms

is there something that I could do beside trying to decrease my number of nodes ?

[slow bobcat]

Even with the number of nodes... It feels like you are evaluating the entire tree. As if you didn't have important branching points that would discard big branches of it.

Also, os that per single tree (you only have one AI on the level test) or for multiple AI's? Key question

[elfin smelt]

the 3.2 ms, is not the Behavior tree execution
It's only it's creation i think

the main BT has 4 big sub trees for different roles in a squad. and in sub trees there is a famous "Reachable/Unreachable" selector that split each sub tree.

I could pass more task in C++ removing there instantiation (I started that) but I don't know If It could give me enough optimization 😅

[slow bobcat]

Removing instancing will reduce the creation time heavily

Moving things to c++, really on memory (which implies no Instancing)

• the befits how how much faster c++ is vs BP

[elfin smelt]

okkk so it is the right way ?
i'm just thinking that removing one small event won't do much on the 3.2 ms 😅

[slow bobcat]

Yeah it's the right way to avoid instancing and rely on c++ when possible. It's a PITA but worth the time

[hasty turret]

check size map of your behavior tree.

if you cast to bp all the character stuf will be loaded with the tree

[slow bobcat]

Yep. That was my next paragraph:

Check your hard references. Chances are you are pointing to characters, animations and what not hard referenced. You should use soft class/object pointers and load the assets asynchronously when needed.
Also, when casting, try to cast where possible to c++ base clases. When you cast to a BP, you load the BP in memory. If the BP hard references anims, meshes, textures.... It will load them.

[elfin smelt]

It should be pretty clean on that side 🤔
The big BT has 4.1 MiB. I don't have valid examples of what is good or bad on the size tho

[final prism]

Question: Is there a way to have an EQS query that scores in a binary format? I.e. instead of filtering I want to give it a low or 0 score, there doesnt seem to be a way to do that?

An example would be something like, from distance 0 to 500, give this item a score of 1, over 500, give it a score of 0 instead of filtering

[slow bobcat]

And what's the plan for the 0 scores?

[final prism]

To still use them, for example it might be possible there are no or only very few points between distance 0 and 500, so I would still have points outside of that range, albeit scored very lowly in comparison

[slow bobcat]

That's what scoring does.

If you just score the items (no filtering, no score and filter, just score), that's what you get

Scores are normalized [0..1] in the end

[final prism]

yes Im trying to achieve a scoring function that does that for me. I am not trying to score linearly or anything

[graceful wind]

looks like you want to group the result

[final prism]

I mean sort of? I dont think eqs has that kind of functionality anyway

[graceful wind]

you can combine multiple test score i guess, but not sure

[slow bobcat]

I think I'm not followign the issue but... isn't that how EQS works for scoring only already?
A test scores and then you average that score. Then the next one applies and averages etc?
In the end, you get the escore obtained after running all the tests.

don't you use these two to achieve what we are discussing?
Sorry if I'm wrong with this, it's been a while since I have to deal with this

[final prism]

Yes but there is no scoring equation that does what I want. In the case I described I do not want to give it a linear score based on distance, it would be a constant score of 1 if within the range and a constant score of 0 if not.
Sort of like a binary step function

Unreal provides a constant function, but that only gets me half of the way there

I would have thought this would be a common use-case, since you usually want fallback points?

I could run a second query with and without the test if the first one fails, but that means running all the other tests again (which include pathfinding tests too..)

[slow bobcat]

And what are you getting with the different normalisation types if not 0/1?

[pastel moth]

is it possible to set tick rates of global tasks?

[upper mural]

Hello, I'm using the Twinstick template and for some reason the enemies seem to wander to a certain part of the map and only then follow the player. They work fine on the default twinstick map but not on custom ones. I can't seem to figure out why they dont follow as they spawn

[graceful wind]

I am going to sleep but this question is bothering me. how do you usually handle firing an event to a State Tree A when the AI is interacting with a Smart Object that has its own State Tree B? I need both State Trees to be aware of the event so they can make decisions accordingly.

Would it be better to directly fire the event to both State Trees, or should I instead store the event in a shared subsystem or global variable that both State Trees periodically check to decide their actions?

or maybe another pattern

[vast relic]

What's the equivelent to a service from BTs, a global task?

Like for example, I want to transition from passive to aggro when an enemy is near

and I'd like to do it all from the ST if possible

or do I give Passive the LookoutForEnemy task and have it complete if enemy is found, transitioning to Aggro?

[harsh storm]

Just running a task on tick really.

(To answer the equivalent for a service)

Having a global task - in your case, is fine for what you want as well

All a service really was, was a task that would tick that you could attach many to a node (in BT). It was sort of a way to achieve running multiple tasks at once.

In ST terms - you just do the same thing. You can put the task on w/e state you need for the logic to run.

I have a task that binds perception component stuff that handles the typical transitions and what not.

[fringe cloud]

Hello, I need help with StateTreeAIComponent please.
I'm trying to use it for my actor (in server side), but nothing happens.

I tried with a test actor and a StateTreeComponent, it works except that as soon as I use the AI version (StateTreeAIComponent) nothing happens anymore, my actors are owned by an AIController and the debugger is empty (no actors with state tree detected)

UE5.5.4

[warped tiger]

You can use USmartObjectSubsystem::SendSlotEvent() to send events to a USmartObjectComponent and USmartObjectComponent::OnSmartObjectEventNative to susbscribe to the events. If the AI is interacting with a smart object it probably has access to the smart object slot handle, which you can use as a parameter to call USmartObjectSubsystem::SendSlotEvent(). Once you receive an event in the smart object component you can try forwarding it to the state tree

[simple crescent]

Is your state tree ai component on the ai controller or the actor? It needs to be on the controller.

[slow bobcat]

Put the AI state tree component in the AI Controller and call Start Logic (or was it Init Logic?) OnPossessed

That should work

[slow bobcat]

Ah this, just saw it.

[fringe cloud]

mmmh ok, yes actually is it on the controlled actor not the controller

i try that now, thanks

[final prism]

It just normalises the values, the function is the same. So a linear function for the values 0-500 will be normalised to be between 0-1, but 250 will still be 0.5 and a value of 600 will give me a value of 1 instead of the desired 0

[slow bobcat]

Aaaaaah I get you now

Ok sorry took me that long

So... You can assign scores to items in your tests. I'm not on my pc but I can show how tomorrow

[celest python]

Looks like

    if (!FAISystem::CheckIsTargetInSightCone(Listener.CachedLocation, Listener.CachedDirection, PropDigest.PeripheralVisionAngleCos, PropDigest.PointOfViewBackwardOffset, PropDigest.NearClippingRadiusSq, SightRadiusSq, TargetLocation))

this if statement returns false

thats why perception doesnt see

I am right in front of the AI though 🤔

[graceful wind]

looks like something worth taking a look, thanks a lot man

[graceful wind]

I think this is the designed way, I forgot that this smart object slots state tree is a feature specifically for new GameplayInteraction framework, state tree is one if the option on how AI behave when using the smart object. this state tree behaviour definition looks very modular that I am interested in. I think I will keep track this slot handle on my interaction manager (subsystem).

In case the AI have gameplay tags "Interacting" I will fire the event using slot event instead of firing the event directly to the AI state tree.

[harsh storm]

@celest python Can you take a screenshot of your complete setup with the perception stuff? I just set up a fresh project and it worked perfectly fine. 5.6

[celest python]

    UFUNCTION(BlueprintOverride)
    void BeginPlay()
    {
        UAIPerceptionComponent perceptionComp = GetAIPerceptionComponent();
        if (perceptionComp == nullptr)
        {
            Error(f"{GetControlledPawn()}::{GetName()} - AIPerceptionComponent is null in the controller!");
            return;
        }

        SetTeamID(1); // NPCs are 1 by default

        perceptionComp.Activate();
        perceptionComp.OnPerceptionUpdated.AddUFunction(this, n"OnPerceptionUpdated");
        perceptionComp.OnTargetPerceptionForgotten.AddUFunction(this, n"OnTargetPerceptionForgotten");
        perceptionComp.OnTargetPerceptionUpdated.AddUFunction(this, n"OnTargetPerceptionUpdated");
    }

(SetTeamID doesnt change the outcome)

[harsh storm]

And the stimuli source?

[celest python]

its not set

    // appearently pawns are already registered as stimuli sources for sight sense automatically
    // UPROPERTY(DefaultComponent)
    // UAIPerceptionStimuliSourceComponent PerceptionStimuliSourceComponent;
    // default PerceptionStimuliSourceComponent.AutoActivate = true;
    // default PerceptionStimuliSourceComponent.RegisterAsSourceForSenses.Add(UAISense_Sight::StaticClass());

i commented it out after bruno warned me about its automatically registered

[harsh storm]

[/Script/AIModule.AISense_Sight]
bAutoRegisterAllPawnsAsSources=false

Throw that in the DefaultGame.ini and then uncomment your stuff.

And then restart the editor as well

[slow bobcat]

have you tried to activate it after registering to the delegates? you might be missing the callback if it spawns in range

[celest python]

gameplay debugger also verifies its not updating

[harsh storm]

I literally have never used perception without that in my DefaultGame.ini and haven't encountered issues

[celest python]

it only updates once

and its only if AI is not moving

restarting the editor rn

[slow bobcat]

that's super weird

[harsh storm]

It is. That's why I'm doing this hail mary play of just mirroring my setup.

[celest python]

I copied body of this function to angelscript and printed the result:

so it also returns true in angelscript

while not in cpp - tho i might be missing parameters maybe

[harsh storm]

Also - have you tried doing the setup in BP?

Maybe some funkiness with AS's bindings.

I know Hazelight doesn't use most of UE's tools.

[celest python]

actually, that makes sense (and adding the new config made my pawn not register as source with my pawn)

same

this is something I have to debug thoroughly i am afraid

[harsh storm]

Did you set it up with the thing I had for the .ini?

[celest python]

yep

[harsh storm]

Oh dang

[harsh storm]

Can't say I've ever seen this

[celest python]

ComputeVisibility call in UAISense_Sight::Update() only returns true on first call, then never returns Visible
thats something I'll have to debug but there are a lot of variables and its the end of the day for me.. will have to continue tomorrow

thanks for walking through it with me

[harsh storm]

@slow bobcat Do you recall if Siggi mentioned if the state tree delegates that came in 5.6 are going to get payload support?

[slow bobcat]

Ummm I recall the payload conversation but not the "when" part I'm afraid.
I will be in 5.6 by next week (upgrade in progress). If you can wait, I can check for you

[harsh storm]

Oh, I'm already in 5.6. I get to be a lot more nimble than you in that regard 😅

I know that the new delegate stuff doesn't have payload support though

[slow bobcat]

@final prism you can set custom scores in your tests doing something like this

{
     flost score = distance < 500 || distance > 500 ? 0 : 1;
    QueryInstance.Items[ItemIndex].Score = score;
}

you have examples in the engine, check how they use FEnvQueryItem

[static raven]

Claiming resources is still a modern approach to solve tasks overlaps? I'm having a few issues with an ability where the AI must override movement and logic behaviors and I solved claiming AI_Logic and Movement resources, but I'm not sure if this is ok or there is a more modern approach for these complex problems

[slow bobcat]

I don't recon this thing you mention. How do you claim them?

[static raven]

there is an in-build GameplayTask and a few functions in AIController aswell

[slow bobcat]

And what are they doing internally?

[static raven]

its kind of complicate, TaskManagerComponent (AbilitySysComp if using GAS) prevent task activation based on available and claimed resources, resolving conflicts based on prioprity, so, for instance, if a task request the "Logic" resource with higher priority than default, BT (BrainComponent) execution will be paused

I still have to read the details, because claiming Movement resource doesn't seems to prevent PathFollowingComponent or CMC to keep with the current movement request

but again, I'm not sure if it is intended to be used or if it is a modern replacement, to be fair, all AIController code looks old and deprecated if you judge too harsh

[slow bobcat]

Aaah I see. Thanks a lot for the explanation. I will check tomorrow if we use it. I don't recall reading (or writiting) code using this feature.

[harsh storm]

That's because it is old 😅

Pretty sure AIController and a decent chunk of the architecture of the AIModule happened when Mieszko was the only AI dev and possibly a junior-ish dev. (I don't recall the last detail)

[static raven]

yeah, like you said, a decent chunk need a little bit of modern love

there are many features that are not correctly integrated, like resources and PerceptionSys. you can't "pause" PersSysComp the way you can pause BrainComponent and just to try to implement it yourself you need to override several classes and copy-paste several functions that are not virtual

[celest python]

welp

i gave up on fixing perception system

gonna write my own again

too much bs just for a simple math function doesnt work

who knows what causing it

[errant mist]

Turning this on in the details panel of the movetotask makes this automatcally true.

[slow bobcat]

Super weird. And have you de bugged the code? Or are you in a BP project or using vanilla engine (you don't compile it)?

[celest python]

I'm on source, thoroughly debugged the code

Having exact same C++ function running on my on character class provides the expected result but doesnt in AIModule's classes

probably I'm missing something very small somewhere but still, too much work and energy to fix a simple cone check

[slow bobcat]

But if you debug the update function in Sight, what fails? Is it the physics ray check or anything else? It's a small function, should be easy to run both examples and compare

[celest python]

UAISense_Sight::ComputeVisibility fails after its first ever call. The dot comparision at the end of the function returns false even though doing same math returns true in my own code

I could continue debugging and find out why somehow but would require a lot more time

it became a chore instead of engineering 😄

[slow bobcat]

haha forgot about the amazing comment here

[slow bobcat]

sounds like you are almost there though. It's a matter of checking the values passed to the function no? I would imagine they are different

[graceful wind]

I hope Eren didnt trap forever within that rabbit hole 💀

[fringe cloud]

Hello guys,
I do not understand why my ai go to the back of the target and do not use shortest path (I tried with a offset but no more result)
Can you help me to solve this please

[graceful wind]

maybe the actor's pivot location is there at the end of the red line

[rich marsh]

I don't see it doing that in the source code and testing did not show that working

[slow bobcat]

I wonder if you are suffering this exact same issue
#gameplay-ai message

You can check by drawing the polygons (option in the recast object) and showing the path using visual logger Navigation category

[fringe cloud]

probably yes

[fringe cloud]

mmh, i don't know how to draw path with Visual logger and i can't find "Draw Paths" and draw polygons bools make nothing (pict)

[slow bobcat]

Record with visual logger, repro the issue and then, stop recording. Right click in Navigation category - hide all but this. Then check frame by frame. One of the marks will show the path and the nav corridor it relies upon

In recast, you need to do this

[fringe cloud]

ok i succeed to display a path with visual logger, thanks!
But this confirms the problem, all AIs want to go behind the target and take the longest path..

I saw on the message you quoted that is probably a engine bug? I use UE5.5.4 for this multiplayer project
And i have a question, why navmesh generate green area IN my tower?? it's good behavior? (pict 2)

I think i have only one solution, get a shortest point between AI-Tower and nearest from the tower (in navmesh) and go to this point

[fringe cloud]

oh sorry, I forgot to answer you, I had already looked at that and no, it is indeed in the center of the cube

[slow bobcat]

the problem is that you are passing the center of the cube as a goal location and that is not on nav (cube affects nav). The engine is projecting that location onto the nav and picking the closest point to it on nav, which is the point selected.

do a test and set to false bCanAffectNavigation in the cube and build nav. Then path. Observe how the path will go straight to the cube

if you want it to force the path to the cube side facing the start point, that something you will need to calculate rather than leaving it to whatever the projection thinks it's the closest point

And i have a question, why navmesh generate green area IN my tower??  it's good behavior? (pict 2)```
meshes will genewrate nav inside or on top unless specified otherwise. This is common. You can add a nav modifier component to your tower and set it to null for example.

[static raven]

Hey guys, I'm having a little bit of a trouble trying to implement IVisualLoggerDebugSnapshotInterface in a custom actor.
I'm overriding GrabDebugSnapshot, I add a new category and just set the snapshot location, but when using the Vislogger my actor doesn't show up

I place a breaking point inside the function and its never getting called

[static raven]

nevermind, I was using it wrong, GrabDebugSnapshot only gets called if you have put any log data for that frame using any of the UE_VLOG macros

[tight oxide]

Is anyone here aware of a way to spawn new nav mesh bounds on runtime? I've got the navigation set to be fully dynamic with navigation invokers and want the nav mesh bounds to be spawned in on runtime

[fringe cloud]

yes i will do it 🙂

[fringe cloud]

Ok i didn't know, thanks you for you help!

[humble spruce]

Good afternoon, hoping somebody can provide me a quick understanding of what I am lacking here....lol

I need to add the sensed actors to my array (not the one from OnPerceptionUpdated) and then from there, I need to use my EventDispatcher from my HealthComponent to remove the actor from the array

Is this the right place to add the actors? I assume its done on FOR EACH, with an add to my array

the goal would be, when the AI has no more actors in the array, it would set state to passive and move on with its life

Right now, it works great... but when it kills an actor it just sits there and watches the other 50 actors in the sensed zone

[slow bobcat]

What's the use case? Usually when people needs this is because they are doing procedural levels and things need to be handled in a different way.
I think you might be able to spawn volumes from code, but let's see if we can help you with a workaround.

[slow bobcat]

you mean that the AI becomes passive even though it has 50 entries in the array?

[humble spruce]

Yeah, I think I figured it out by forcing it to get the first index, setting it to the attack target, and moving on from there

when no valid array index, it sets to passive

[tight oxide]

Actually figured out, so you are totally able to spawn the volumes through the code, the caviat is that you can't use the default brush component. As long as you attach a new BoxCollission component, it works just fine. You do need to set the collission to be fully dynamic, but that was already the case in a project

[slow bobcat]

Why is the brush component an issue?

[tight oxide]

Because you can't actually properly edit it in a packaged build, it's more of an editor concept

So just slapping a box component is simpler and achieves the same effect

[slow bobcat]

But... I'm not following sorry. You are attaching a box component to who?

[tight oxide]

To the newly spawned Nav Mesh Bounds volume

The engine behind the scenes doesn't actually check if it's a brush component or a box component, or whatever else, all it really checks is the component bounds, which can be changed by attaching a box component

The more you know

[slow bobcat]

Aaaaah OK ok. Yeah makes sense

[slow bobcat]

But quesiton still stands: how come you need to spawn nav bounds at runtime? Is it procedural without modules?

[tight oxide]

The players can make stuff in the world basically

Create structures mazes stuff like that

[slow bobcat]

I see. Nice

[strange plank]

Im working with smart objects for the first time, trying to get the position of it using the default state tree tasks, but for some reason the move to task always fails? The "FindSlotEntranceLocation" task seems to be finding a slot, and even outputting a transform, but move to always fails to move to it, any ideas or help would be appreciated

[slow bobcat]

Guessing here but.... Are both tasks inside the same state or in sibling states?

[strange plank]

both tasks are in seperate states, it seems like two weird things are happening, the entry point slot destination doesnt seem to be anywhere near the slot itself, and then move to is failing even though the point is perfectly navigable

[slow bobcat]

Try setting the move to state as a child of the one getting the location
Get location
Move to

And about point being perfectly navigable... Try Project Point to Navigation after getting the location and pass the result to the move to

[strange plank]

Setting the state to a child isnt helping, the states just infinitely check eachother and fail as soon as they enter, unsure why. Thank you for the advice, if you have any other tips i would take them

[warped robin]

I have debugged this several different ways and verified this isn't user error - I have created a decorator, set it to abort lower priority (and self, for the record). It is returning TRUE for a couple of seconds before it activates. it seems like the only reason it does activate is just because the tree evaluates it normally - as in, it's NOT aborting lower priority.

I am overwriting PerformConditionCheckAI. Has anyone seen this? I'm sure I'm just doing something wrong

This is a simple BP test to make sure I'm not going mad. this evals to TRUE but does not abort.

[warped robin]

as far as I can see, none of the flow abort control flags that are set in BTDecorator.h are overridden by any children

[slow bobcat]

If you have a custom decorator and you want it to abort, you need to implement the abort call. Only you know how your custom decorator should abort. What I mean: the bbk decorator aborts upon the bbk changing. A distance decorator could abort when the distance is shorter or larger.

At least in code is how it works