gameplay-ai

[misty wharf]

I think you would probably want to keep the calculations etc. in the controller, with the behavior tree doing the higher level choices like choosing the target towards which the car is driving or things like that

[rich wolf]

yeah i suspected that

[rich wolf]

so basically i have to move many of my AI variables to blackboard

[oblique basin]

Anyone writing custom C++ BT tasks, I'm finding it tricky to use delegates to control tasks as it seems that both Node Memory and Owner Comps are a) only directly accessible when using task virtual functions and b) changeable such that memory addresses for the above objects are subject to change before task completion, so saving pointers during execute is inconsistent. For example, I have a callback to when my character changes state, but binding a task function disables being able to pass node memory around. How do people handle this?

[glad snow]

So here's my Event graph. I'm trying to get A.I. walk from point A, to B, to C, and when it gets to D it stops how do it fix so it walks around the whole area in a loop

[dapper crest]

So I'm getting the impression services in the path to the current task don't run during the execution of a long task? But decorators in the path to that task are still being evaluated etc.?

[wooden stag]

Anyone here work with smart objects? I can't get any of the nodes related to tags to work. I am providing a valid slot handle/smart object handle, but it seems like these nodes just do nothing currently.

"Get Slot Tags" and "Get Instance Tags" always return nothing, even if I input the same slot handle I use to get the slot location, which does work, and required filtering via tag, so I know the slot handle is valid and has the tag, but using these nodes to print the actual tag(s) returns nothing

[uneven cloud]

Use AI tasks for that.

[oblique basin]

What's an AI task? Google refers me to BT tasks only.

[wooden stag]

From forums

AI tasks are an extension of gameplay tasks and are mean as main communication layer between all sorts of logic sources (BT, FSM, level scripting, extermal-induced reactions, etc) and the AI entity itself. In 4.14 I’m going to flip the switch and BT tasks by default are going to spawn AI tasks instead of directly messing with AI agents’ internals. Both Paragon and Fortnite have been successfully using AI tasks for some time now, and Gameplay Tasks are also a super set of Ability Tasks used by Gameplay Abilities System

https://dev.epicgames.com/documentation/en-us/unreal-engine/API/Runtime/AIModule/Tasks/UAITask?application_version=5.3

[oblique basin]

Ok, but I don't see how this solves the problem. The issue is that if you're managing behaviour with a BT, you can only end the task with a ref to the owning controller's UBehaviorTreeComponent, which generally can't be captured in bound delegates. So if you call an AI task from a BT task, you still have problems ending the initial task, as far as I can tell.

[oblique basin]

UBehaviorTreeComponent* OwnerComp = Cast<UBehaviorTreeComponent>(GetOuter());

This is the answer.

[fierce carbon]

What type of optimization methods should I consider with mass AI?

[fiery epoch]

Guys, AI pawn sense doesn't sees me through the wall on the first time. But when I get sight than ai always follows me. Only way to stop it is getting behind pawn sense. So just after chase starts the walls can't help, why?

I wathed few tutorials and this problem only happens with me. I just don't understand why that happens

If getting behind the ai wouldn't work. Than I would understand the problem, but the fact that walls won't help after first sight cracks me

[vague hollow]

Hey guys, silly question about State Trees.

I have an STT that outputs a vector (variable is set as output).
I want to grab as an input that variable in another STT that is not a child of the previous one. How can I do that ?

I'm always able to get the variable from one STT to another if the latter one is a child of the first STT but not if they are siblings

[misty wharf]

You can't. You can only bind values between tasks that are in child states, or if the tasks are within the same same state

[vague hollow]

So what are my alternatives ?

[misty wharf]

Global tasks or evaluators mostly I guess

Possibly extend the schema to support additional parameter values you can store

[light void]

I have made a base AI I am satisfied with. How can I create offshoot of that AI? Like, make other AI that have the same base functions but with the option to override a function or event if I think it's necessary for the other AI? What is the clean way to do so?

[fiery field]

Is there a way to debug the the sight sense for an AI ? Like, see the sight cone basically
Because when I tick show > senses in the viewport, it doesn't show anything

[misty wharf]

Gameplay Debugger should show some info for perception

[vague hollow]

I've used my STT to set a vector variable in my actor, this way i can always keep track of it. Not really satisfied with this option but the prototype is working fine

[fiery field]

I didn't know this functionality. However I activated it, activated the perception debugging, and there's still nothing shown. I might have messed up my setup

[keen crow]

if you are trying to handle a result of some logic (be that component or gameplay ability) in BTTask, i'd suggest you taking a look at brain messages features, i.e. in your UBTTask_MyTask::ExecuteTask you can call WaitForMessage and specify the tag of the message and later call it from whereever by using UBehaviorTreeComponent::HandleMessage(const FAIMessage& Message)

[oblique basin]

Interesting. I may circle back to this. For now, I just subscribe to delegate calls in the AI controller, which itself triggers a delegate that broadcasts a reference to its BTComp. From that you can get the AI controller through GetOwner() and get a ref to its Node Memory and so do pretty much anything you want. Cheers for the idea 🙂

[uneven cloud]

Look at the BT task move to and how it uses the AI task move to.

[uneven cloud]

Don't use systems that were depreciated in 2016.

[uneven cloud]

No one, including Epic uses that old functionality anymore. They used to, but switched to using AI tasks.

[fierce carbon]

I am trying to find ways to optimize my AI (heaps of zombies). If I was to unposses the AI controller, will that disable character movement/ticks? I am trying to disable far away AI and only enable the ones near the player.

[steel moat]

Can someone explain whats happening here? the player is unable to lose sight of the AI and still it goes to the roaming, i still dont understand the issue

This is the behaviour tree of that AI, at first the player is not in sight so it is doing the task on the right most of the behaviour tree when it comes in sight it switches to the task in middle to follow the player. Now when player jumps even tho it didnt lost the sight AI goes back to roaming

the first video and the 2nd video are not same

[dense owl]

I think it’s better to use level streaming for these things

[fierce carbon]

What do you mean? Can you go into more detail/laymen's terms please?

[dense owl]

https://dev.epicgames.com/documentation/en-us/unreal-engine/level-streaming-in-unreal-engine

[fierce carbon]

You mean world partition? I already use that.

[uneven cloud]

No it's not. Turning off AI when the player is not close enough is incredibly common.

[steel moat]

anyone? i need help in this please

[dense owl]

Turning off how? I thought you’d said level streaming does that but alas memory is not what it once was 😅

[uneven cloud]

Never once said that. You turn off whatever you need.

[dense owl]

Ok

Guess I thought if the object is unloaded by WPO/streaming there’s no way the AI that possesses it would be running, but every day is a lesson

[rare ruin]

You need to point at the AI you want to debug, if you're too far, it sometimes don't auto target the AI

[stoic cove]

Hello. I'm facing an issue with UE 5.4 I migrated my project and suddenly I get a crash / error saying that my nav mesh bounds are too large. I have a very large open world and want AI to work anywhere. So what I did in my old project is just place a massively huge navmesh bounds volume because that seemed to be the only way to have invokers be created around AI pawns. This never gave me any issues, until now in 5.4. So what is the alternative / better way to make AI work in a massively huge open world?

What I don't really understand is why the tooltip for "generate navigation only around navigation invokers" says "this will make the entire world navigable" yet it doesn't, because it only does when you still place a nav mesh bounds volume that you have to scale extremely large, which then results in the above error. So how does one truly enable open world navigation? The tooltip seems to be incorrect?

[static crystal]

Ever find a solution to this? project wont compile because of the missing AIMODULE_API

[azure pulsar]

Was wondering, can the default cost of NavAreas or the override cost of NavFilters be changed at runtime?

[stone walrus]

Basically it's a problem with Ur perception angle of view idk what's setting called it allows ai not to lose perception when players jump across ai

Idk what's setting called

[regal drift]

I solved Async Pathfinding and got it to work so i will post my current solution. Is there any way which i can batch several pathfinding request?

[tidal mantle]

it is not losing perception but the stimulus is getting expired

[stuck oar]

I have 3 doors. The player clicks on the door and has an ai walk the camera to the door. When I click on the left door, it walks straight. I am peprplexed

[severe kayak]

Hey, does ue5 support sharing of a behavior tree and blackboard across multiple agents? I am wondering how BTs can handle large number of entities all at once

[dense owl]

Yes, ofc

You can also have subtrees if you want only certain branches to be shared (must have same blackboard tho)

[severe kayak]

But what i am missing (and remember that I have 1k actors), is that there is a BT assigned to an AIController whoch is assigned to my actor. And if I remember correctly, there is a ahed cap of 20/30 ai conteollers at a time.
So I sont understand how BTs can scale to those levels.
And it is totally okay of they are not, I kust want to make sure I am not developing something that is already supported 😀

[dense owl]

When you’re talking thousands #mass might be a better option regardless

[severe kayak]

yes of course it would be my next attempt 🙂

[obtuse igloo]

Anyone able to figure out this out? #gameplay-ai message

I might have to use a different method if not, I’m stuck LOL

[misty wharf]

You can do this via C++, it's a bit fiddly though but possible

[olive prism]

I have a problem, when my AI makes attacks, due to root motion, it often stays above my character and loses the vision sensor, therefore it enters the passive state

How can I prevent me from losing the character that is in focus when stop seeing him?

[keen crow]

So, what's with the BT nodes API issue in UE5.4? I've updated my project from 5.3.2 and all of a sudden I get compilation error pointing to engine's nodes (I assume because I inherited them in some of my own BT nodes). I remember facing this issue when 5.4 only came out in preview and I rolled back to 5.3.2, but back then somewhat of a solution was to manually override all non-exposed methods (OnBecomeReleant. InitializeMemory, etc) just for the sake of it. Is there any better solution now?

[keen crow]

so, uh, do I get it right that epics messed up BT nodes code, dragged it from 5.4 preview to 5.4 release, and 2 months in the problem isn't fixed in EGS engine version and nobody in epic isn't doing anything about it? 🙊

[boreal tundra]

lol why

[keen crow]

maybe it's too dark inside. jokes aside, maybe you have incoherent settings in CMC nav agent capsule shape settings, actual character capsule shape and the one in project settings in navmesh section which is used for generating navmesh?

[boreal tundra]

You mentioned a lot of terms that I've never heard. lol

I'm sure nav mesh could be better since mine does its job but it's super bad looking

[keen crow]

have you changed the NPCs capsule shape? perhaps made it bigger? also I'd recommend checking collisionon the doorframe and checking navmesh overall if NPC is even supposed to pass there

your navmesh is fine, NPC should be able to pass there

but incoherent nav agent (any AI that can walk on navmesh) can still be a problem, because navmesh generation takes it into account

[boreal tundra]

Collisions seem sweet

[boreal tundra]

The fact is that my AI can easily pass around the house without problems but sometimes he starts jittering stuck to doors or during patrol decides to walk against the wall

Like here the sphere shows the area where he should look aroun which is small but in-game he was walking towards the corner on the left

@keen crow nav mesh goes over objects sometimes like bed and chests.. AI still doesn't walk over them so it's fine but maybe it triggers some problems?

[keen crow]

well, again, there can be several reasons why can AI's movement on navmesh look ugly, I can only suggest you check on some of the possible reasons, but giving you a 100% correct answer is just a matter of trial and errors in your project. so I would

1. check that navmesh generation nav agent size, actual character capsule size and CMC size are coherent. If you didn't adjust any of those, you can ignore it, as default should work
2. check that all relevant meshes have Can Ever Affect navigation set to true if they actually should
3. Check that in your BT/ST it doesn't happen that AI is constantly switching destination or something
4. Maybe you've enabled RVO in your NPC's CMC settings and set some big avoidance radius and hence your NPC is constantly trying to avoid player while it's path lies through the player

[boreal tundra]

Character Capsule

One is its normal capsule, another is to trigger stuff when touched

What's CMC?

What is RVO?

How do I check that nevmesh generation nav agent size? Didn't you say that the nav mesh is fine?

[keen crow]

character movement component

[keen crow]

Reciprocal Velocity something something avoidance, it's a mechanism built in CMC (considered outdated) that allows NPCs to avoid other characters (or I guess any actor having velocity) when moving. It's known to produce errors in navigation because it doesn't take navmesh into considerations when it moves the nav agent a bit to the left or to the right to avoid obstacle, also in my experience it can introduce some jitter for the NPCs. CMC has a set of parameters for RVO, but AFAIR it's disabled by default

[keen crow]

The navmesh looks fine but you must understand that navmesh is generated for a nav agent of some radius with some max step height and max step angle properties. You can have multiple nav agent types (small NPCs, big NPCs) and they could have different nav mesh. So it is possible that you have a default navmesh generation settings but for gameplay reasons you've made your NPC capsule bigger for whatever reason. Judging by your answers it's not the case but I'm just sharing my considerations on what I would have checked had I your problems

[boreal tundra]

There's not RVO in my CMC.. Maybe because i'm in ue4?

[boreal tundra]

I'm tweaking cell size and height and it is more good looking now even tho the ai can't climb the stairs anymore now. requires more tweaking.. so far it doesn't seem to be a huge improvement

[keen crow]

it should be, just type in "avoid" in CMC details search bar

[boreal tundra]

Okay I have it on false

[keen crow]

well, then you can try either

1. for testing purposes shrink the NPC's capsule component radius to half the size just to see if it starts to pass smoothly through narrow gaps after that. If so, maybe either tweak the navmesh generation properties or perhaps change the level design to widen those narrow parts
2. if that doesn't help and if it happens not only in narrow paths but everywhere, then try to see if the AI is actually constantly changes final destination point, maybe you have multiple EQS services running at the same time and updating the same blackboard property or something

[boreal tundra]

Boh.. I tweaked a bit.. I hope he won't get stuck when game will be released

[frigid vine]

I'm not at my computer but so figured I'd shoot message in here just for rough idea/direction.

Currently my fps grapple game uses a lot of NAV links to handle enemies jumping to the different platform.

Say I had two equal size platforms side by side, one raised higher. If I wanted to enemy to jump down/up when following player is there any easier/better way to handle this dynamically then just placing a bunch of NAV links around the platform?

[misty wharf]

Easier? Doubt it. Better? Maybe, but get the feeling it would be kinda complicated to set up in a way that works reliably the way you want. You could for example attempt to do some kind of custom test to determine whether the unreachable position where the player is was possible to reach via jumping or something. I think there was also some code in one of the published UT codebases that automatically placed navlinks so you don't need to manually do it

[harsh storm]

Yes, it is jacked up in 5.4 and it is fixed in 5.5 currently. It couldn't make the hotfix patches because it would mess up binary signatures (according to an ex-Epic employee)

[misty wharf]

5.4 seems to be the most broken release in a while...

[keen crow]

fascinating

[dense owl]

There’s a 5.5 on the branch? Neat

[scarlet kayak]

Anyone else experiencing behavior tree problems with 5.4? My enemies work fine in PIE but don't move around at all in the packaged game

[harsh storm]

When do you start the behavior tree?

[scarlet kayak]

I have 4 monsters placed in the level, not spawned. Single player. Is that the problem?

[harsh storm]

That didn't answer the question

When do you call "Run Behavior Tree"

[scarlet kayak]

event begin play > run BT

[harsh storm]

Don't do it in BeginPlay

Do it in OnPossess

[scarlet kayak]

ok thank you!

[harsh storm]

Things may not be valid when BP runs in this case.

That's why

[scarlet kayak]

simple as this?

[harsh storm]

yes

[scarlet kayak]

thank you! I greatly appreciate it 🙂

interesting that it wasn't an issue when I used 5.3

[harsh storm]

It could have been an issue

You were just lucky to not encounter it

There is no guarantee that you'll have a Pawn when the controller's BeginPlay is ran.

[scarlet kayak]

I see... so the ai controller was loading before the pawn was

[harsh storm]

Not necessarily

It is a matter of BeginPlay's order not being guaranteed

[scarlet kayak]

hmm I did another package build and same issue. Enemies doing idle animation but not walking around.

[limpid shore]

| might have solved this kinda
~~Hi, I'm trying to make a function that detects what's in a image (a cat, a dog, a car etc)
I've come through this thing called "neural network inference" the newer version is called "neural network engine", and some stuff have changed, the tutorials from 2 years ago are all outdated,
only tutorial I can find is this, https://dev.epicgames.com/community/learning/tutorials/34q9/unreal-engine-nne-quick-start-guide-5-3
but I've having trouble understand it, should I put the script in an actor or an object?
the old tutorial: https://www.youtube.com/watch?v=LX1w_etaftY
said I need to put the code in a "neuralNetwork" class, which doesn't exist in ue5.3.
why are there red lines under stuff like "INNERuntime" am I missing a dependency? am I missing an include, if so, what is this include.
hoping for directions.~~c

[harsh storm]

Delete the navmesh stuff and recreate it. That has been an issue I've seen in the past.

[sonic yoke]

I'm having a bit of trouble with my nav mesh agent radius. I've changed it in my project settings, but it has no effect since I have a RecastNavMesh-default actor (which spawned by itself?) which changes the value. It doesn't seem to matter if I change the value on it since every time I restart UE it resets to a default value. How do I change this so the change sticks?

[dense owl]

You need to create supported agents

[sonic yoke]

Thank you

[feral sedge]

I have this problem. I asked about it before, and was also told to look at the VisLogger. Can't be sure since I don't know much about it, but I didn't seem to find any answers in there. IIRC, the log is too big to post here.
I mean, nobody seems to know anything about it anywhere on the interwebs. It happens randomly to only some of my AI, usually after going across my map in World Partition. Like you, it seems that the Navmesh is there, but some AI stop moving towards my Player and only attack when you get close (as if no Navmesh). On reload in the general area of the problem (from savegame actor), the same AI enemies or areas then work perfectly.
I've been struggling with this for months. My demo will be in NextFest this week, and I have to accept that this unwanted behavior can't be avoided at the moment, which is a bummer.

[scarlet kayak]

Thank you again- it was a Nav mesh problem!

[scarlet kayak]

I had this problem yesterday. I deleted and recreated my nav meshes, migrated only the levels I'm using to a new project, and then packaging worked

[uneven cloud]

All navigation errors are logged to the visual logger.

[pine steeple]

Can also turn on log lognavigation verbose and log logpathfollowing verbose

For more detailed logs

[spring scaffold]

Hi guys
I have couple questions about navigation's weird behavior. First of all why does ai doesn't choose the shortest path but hugs walls instead? Every time i use unreal navigation i get results as on first picture where car rides on the red path instead of blue. Is there easy way to change it? I really don't want to write my own local navigation because i already had to write my own global navigation using Dijksta.
Second thing is that agent radius parameter in recastnavmesh is always set to 35 in runtime even though i set it to another value in editor. It doesn't matter if i'm using regular navmesh but it does when i use "Generate navigation only around navigation invokers". When i use this option the radius around obstacles is always set to default value. Can i change it someway?

[misty wharf]

I don't know why it wouldn't follow the shortest path, never seen that happen, but check this for the agent radius issue https://zomgmoz.tv/unreal/Navigation-System/Configuring-navmesh-settings

[spring scaffold]

Do you mean it should follow the blue path by default? I've seen many people asking about ai going too close to obstacles but haven't ever seen a solution.

[misty wharf]

It definitely should be following it at least on the left side

[spring scaffold]

I've been already trying to set agent radius in project settings and it never worked.

[misty wharf]

Did you configure a supported agent?

It seems sometimes to be a bit inconsistent about which setting it follows and may require one

(or it might be I never paid enough attention to how I have configured it in my projects where I needed to do that lol)

[spring scaffold]

I'm not sure what do you mean by configuring the agent. I did change it's radius in nav movement.

[misty wharf]

there is a setting called supported agents

[spring scaffold]

Do you have any video where agent takes the shortest path instead of huging the walls? At this point i'm almost sure they do this every time

[misty wharf]

try adding one into that list if you don't have yet, and configure the radius there

I don't have videos of it but I've not noticed this behavior in any of my projects. I'm not using nav invokers or such though, so I don't know if it's some weird issue with them

[spring scaffold]

Oh I've tried this already. The problem I have with this one is that when I use Supported agents my recastnavmesh disappears during runtime and navigation isn't calculated at all.

[misty wharf]

I've heard invokers are problematic but can't really say for sure since I've not used them myself 🤔

[spring scaffold]

Tbh my issue with ai not taking the shortest path appears even when I'm not using nav invokers.

Thanks fpr your help anyway 🙂

[untold quiver]

Hello, I'm trying to write a move to task at the end of my "EnterState" Function in a C++ State tree task. I need to bind to the result of that task to return the correct value of EStateTreeRunStatus. However when I bind a lambda to the Path Following component's OnRequestFinished delegate I get compile errors. Is there a way to do this or am I stuck using the blueprint state tree tasks?

Okay nvm I found a good example of it in the "PlayContextualAnim state tree task

[spring scaffold]

Here is video if you would like to know how it looks

It's just goes to the green ray, not the red ones

[swift bolt]

When it comes to the heuristic weight/scale, what effect does that have on navigating a path from point A to point B? What should the moving objects behavior be if the heuristic weight is zero versus one versus two and in between

[ruby shard]

how to cutomize FNavigationQueryFilter ? why the hell is there wrapper over wrapper ? i just want to override some functions for dtQueryFilter

🤔

[wooden coral]

Hey, I'm trying to use RegisterNavigationInvoker() for my player cause it's really suitable for my game design, but it doesn't seem to be working cause I don't see any Nav geometry rendering after pressing P in the editor, and my NPCs just stuck unable to do any pathfinding
Using a NavMeshBoundsVolume worked fine for me but I really want to switch to invokers

I call RegisterNavigationInvoker() in player's BeginPlay(), and UnregisterNavigationInvoker() in the EndPlay()
I also set bGenerateNavigationOnlyAroundNavigationInvokers to true
Don't know if that's all for the setup

[uneven cloud]

You still need to have a bounds volume when using invokers. There's also an nav invoker component that should be used instead of doing it manually.

[sullen plume]

I'm recreating the AIMoveTo Blueprint Node in C++ here's the C++ code, however, the character is stationary, and when I look at the behavior tree which uses the C++ task, it is instantly calling and ending it repeatedly (specifically im tryna have the player wander so pick a random location within some radius and move there)

And here's the blueprint which I'm trying to recode (the blueprint works fine btw):

Me a couple hours later...
I was recommended to just derive from the BTTask_MoveTo so I did that, and I'm more lost than ever,

NodeMemory barely makes sense, apparently I'm supposed to make custom structs to store values?

I can make my own struct sure but when I look at the example MoveTo Code there's just way too much going on for me to understand.

Like one second the code uses NodeMemory and the next its like fk it we using the BlackboardCompoennt now

when i'm referencing the code ^ I mean like Unreal's "Tasks/AITask_MoveTo.h"

[uneven cloud]

I don't understand why you aren't just separating this functionality into 2 different BT tasks. One to find a location and then just using the built-in move to. You're doing a lot of extra work to not even get the base functionality of the built-in task.

[median socket]

Id clash in non Game and non PIE world. This should be incredibly rare!

Does anyone have experience with hitting this ensure when using NavLinkProxy components in Level Instances?

And does it matter?

It started happening in 5.4 I believe.

For context, we have a door blueprint with a smart navlink component. This door is being used in level instances for houses.

// During PIE or game we just generate a new Id, this is most likely to be from a runtime (non editor placed) prefab like a level instance but could be from 
// a legitimate but extremely unlikely Id clash after loading.
// If this occurs in EWorldType::Editor world it's a legitimate ID clash, currently we do not handle this edge case here as it should be incredibly unlikely to occur
// and we do not save changes when cooking or building paths running a commandlet etc.

This comment is above the ensure.

[ancient wedge]

any luck finding the answer? My AI is also doing this weird behaviour from time to time.

[uncut rune]

Create 1 task that will set Reachable Point, and then with 2nd task MoveAi To that point.

[final thicket]

somebody can help me with my AI Enemy? i took a project where i'm not the first dev so im lost with a dumb ai that dont even know whats happening

i can show the project tho

[spring scaffold]

I have no idea. I don't have this problem in fresh project.

[steel moat]

is it possible to attach the AI controller to a skeletal mesh component i have attached with the pawn instead of directly attaching the controller to pawn? Whenever the character is comes in sight of the AI the whole body turns to face it, I only want the skeletal mesh component(Head) to face the character

[misty wharf]

You can try using the Focus feature, iirc set focus or such. I recall with the appropriate settings you set focus to something and it only turns the head towards the focus target by default

There's nothing automatic that turns anything "when character comes in sight of the AI" so if you have logic there that turns the whole actor, then that's what's gonna happen

[steel moat]

i have controller attached to pawn, and also passing the value of player's location in the set focal point function

[misty wharf]

try using one of the other focus priorities if it turns the whole actor

I'm fairly sure I used this in my project and it turned just the head but I don't recall the specifics

[hallow compass]

is there somewhere a BT and ST pros & cons ?
i find it very hard to find what would fit the best and whats the most "optimized" and "modular"

[dim lynx]

hi, did you solved it?

can you help me with this? I also try to have the AI ​​flank the player 😅

[misty wharf]

Unless you're comfortable with C++ go with BT's. StateTrees are nice but they seem to be kinda heavily work in progress and have a few oddities here and there that you may have to work around or figure out from the code

[hallow compass]

i guess i'll go with BTs
but, what do you suggests if i want to stay in c++ zone ? i was planning to dev the tasks in c++ and use them in the BTs

[misty wharf]

You can write BT tasks just fine in C++

Although personally I find I often write them in BP's since there's often at least some latent actions occurring in them, and those are a bit less of a hassle to work with in BP's at least in my opinion

[hallow compass]

okay

[sullen plume]

regarding both of your suggestions

[sullen plume]

^

even if i do that, my issue is that I'm unsure how to make the ai move. Like the code seems to work but the actual actor ain't moving at all.

I've tried the deriving from the BTTask_MoveTo and doing it myself, the actor stays in place

[harsh storm]

Honestly - it is wild that your solution is to rewrite that instead of investigating why it isn't moving

Considering the thousands of games that have released with this exact setup (one task to find a point and another to move to it using the MoveTo task) - something is wrong with your setup

[sullen plume]

okay if i'm given an actor location

• target location

how do i move to that

like even if i do the setup ur recommending i still dunno how to get to that point

[harsh storm]

Store it as a BB key, then use the move to task and set the target to that key

Make sure you have a valid navmesh as well

[sullen plume]

when you say use the move to task

which func do i call?

[harsh storm]

None

[sullen plume]

and do i just pass in the default ownercomp and nodememory

[harsh storm]

It is a built in task with the engine

[sullen plume]

oh that one

okay just to confirm

are these accurate? if (moveResult == EPathFollowingRequestResult::AlreadyAtGoal) { return EBTNodeResult::Succeeded; } else if (moveResult == EPathFollowingRequestResult::RequestSuccessful) { return EBTNodeResult::InProgress; } else { return EBTNodeResult::Failed; }

[uncut rune]

Can you show your behavior tree? Did you try to debug it if it's even firing? You can look into Behavior Tree to see the flow.

[uncut rune]

and Wander is with that AiMoveTo node?

[harsh storm]

Do the setup that you've been told at least 3 times now.

[sullen plume]

ikik thats my old tree

[sullen plume]

can you please answer this

[uncut rune]

lol then show updated? xd

[sullen plume]

I'm doing ur strat

dw

[harsh storm]

The snippet of code in a vacuum means nothing really.

[uncut rune]

This behavior tree Wander task is written in C++?

[sullen plume]

hmm

    EBTNodeResult::Type result = Super::ExecuteTask(OwnerComp, NodeMemory);
    if (moveResult == EPathFollowingRequestResult::AlreadyAtGoal) {
        return EBTNodeResult::Succeeded;
    }
    else if (moveResult == EPathFollowingRequestResult::RequestSuccessful) {
        return EBTNodeResult::InProgress;
    }
    else {
        return EBTNodeResult::Failed;
    }```

better?

im in a class that dervies from teh moveto

task

[sullen plume]

yeah i prefer C++ cuz i can interact pretty easily with other classes

[harsh storm]

You are doing absolutely useless work

[sullen plume]

ouch

[harsh storm]

Just do the setup you have been told

[uncut rune]

It's better to first create prototype that is working, and then translate it into C++ for example. Right now you have no idea what is wrong, if its the behavior tree setup, code or maybe you dont have even ai controlleron your pawn

[sullen plume]

Ok how do ik if i've reached the targetLocation so that I can choose a new random location to wander to?

[harsh storm]

Use 👏 the 👏 built in 👏 move 👏 task

It literally does all of this for you

[sullen plume]

EBTNodeResult::Type UBTTask_Wander::ExecuteTask(UBehaviorTreeComponent& OwnerComp, uint8* NodeMemory)
{
    EBTNodeResult::Type result = Super::ExecuteTask(OwnerComp, NodeMemory);

    if (moveResult == EPathFollowingRequestResult::AlreadyAtGoal) {
        findNewTargetLocation();
        return EBTNodeResult::Succeeded;
    }
    else if (moveResult == EPathFollowingRequestResult::RequestSuccessful) {
        return EBTNodeResult::InProgress;
    }
    else {
        return EBTNodeResult::Failed;
    }

}```

[harsh storm]

All you have to do is provide a location or actor

[sullen plume]

i know

but let me at least try to do it in c++

[harsh storm]

No, you don't seem to know

Make the task to get a random navigable point in C++

If you want to use C++

[sullen plume]

Alright I did all that and my ai still aint moving
Do I need to setup player input component by creating the move, look, and jump bindings? Does the AI use those when moving around the level or does it use something else

and i have tried with both moveto and movedirectlyto

[harsh storm]

Use Move To

Do you have a nav mesh?

[sullen plume]

yeah ofc

uk how for player to move wasd u gotta setup player input?

does ai need me to setup wasd?

or does it use some other form of moving around

[misty wharf]

AI doesn't care about keyboard inputs or anything like that

It should work for navigation if it's a Character. If it isn't a Character, it needs to have a compatible movement component added to it

[sullen plume]

yeah it has a char movement component

weird i msut be missing smth

[misty wharf]

look in visual logger

movement failures are usually shown in there

[uncut rune]

Why won't you put that Move To node task that is already built in and just check if it's the code problem or not first? xD It's 5 seconds

maybe your code is right and you didnt set something else

did you tell your npc to run behavior tree?

[sullen plume]

sir

thats

what i did!

yeah i did tell him to run behavior tree

i first trieed moveto

then movedirectlyto

both same result

[uncut rune]

show us code inside these nodes

[sullen plume]

the moveto is just default whatever

[uncut rune]

and how are you setting black board key inside find new?

[sullen plume]

the blackboard key works fine like i see the value changing to a valid location

but here it is anyways

EBTNodeResult::Type UBTTask_FindNewWanderLocation::ExecuteTask(UBehaviorTreeComponent& OwnerComp, uint8* NodeMemory)
{
    if (auto pawn = OwnerComp.GetAIOwner()->GetPawn()) {
        if (AAIController* AIController = Cast<AAIController>(pawn->GetController())) {
            FVector actorLocation = pawn->GetActorLocation();
            auto navSystem = UNavigationSystemV1::GetNavigationSystem(GetWorld());
            FNavLocation targetNavLocation;
            navSystem->GetRandomPointInNavigableRadius(actorLocation, 10000, targetNavLocation);
            OwnerComp.GetBlackboardComponent()->SetValueAsVector(GetSelectedBlackboardKey(), targetNavLocation.Location);
        }
    }
    return EBTNodeResult::Succeeded;
}

[uncut rune]

2 things:
show me in Ai Controller on Begin Play your setup
run game, click P and do screenshot of your nav mesh, is it green where it should be?

[sullen plume]

yup everything green:

the guy getting scaled is the ai

and yes aibehavior is set in the blueprint

[uncut rune]

Try to lower Radius of points, maybe it's choosing some crazy number and cannot go there

[sullen plume]

maybe

[uncut rune]

and inside Character did you set Ai Controller Class?

in the Pawn category

[sullen plume]

yeah

maybe he's just lazy

[sullen plume]

you meanunreal's outputlog?

or like the visual studio output i get while building?

[misty wharf]

Tools->Debug->Visual Logger in UE

[sullen plume]

ainavigation is red?

what am i supposed to be looking for?

nth else is red

[misty wharf]

Press record, then play in editor so that the BT stuff runs. It should show a timeline of sorts about different events that occurred during the PIE session

If you then click on the events or use the time marker to move around, it should display more info

[harsh storm]

Did you set it so that the controller possess on spawn and on placement?

Also, don't run the BT on BeginPlay, do it on possession

[sullen plume]

yeah

[misty wharf]

That suggests it's attempting to move at least

[sullen plume]

yeah

[sullen plume]

I made a test blueprint function that uses Controlled Pawn i'll provide img of the workflow
When I play i get the below error

PIE: Error: Blueprint Runtime Error: "Accessed None trying to read property K2Node_Event_ControlledPawn". Node: AI MoveTo Graph: EventGraph Function: Execute Ubergraph Test BTTask Wander Blueprint: BTTask_Wander

[signal island]

we recieve Payload in statetree only in c++?

[sullen plume]

i've done all you said, i think, but uh the onpossessed and beginplay never run

[harsh storm]

Well that points to a different problem

[sullen plume]

yet somehow in the world I do see my controller just chilling in the center of the actor

oh wait

i dont nvm

it aint using my aicontroller

just the default one

[dense owl]

You need to set it to use the correct controller

[sullen plume]

does this look right? tghis is my bp_aicontroller class defaults

cuz rn like everything seems to be working excep the behavior tree which never runs

[harsh storm]

That's because you're not running it

What is the controller for the pawn

[mellow zenith]

Can anyone help me out with some issues im having with my AI

would be easier if it was in call to show properly what the issue is

[dense owl]

Would be easier if you just described the issue and provided relevant screenshots if need be

[mellow zenith]

I can describe it but showing screenshots will be hard as I need to show its movement in action

Effectively though while the AI moves it is like spam sensing and forgetting the player even when infront of them

and its causing the AI to freak out between patrolling and attacking in the behaviour tree

Have 0 clue what is causing this as ive done this type of AI tons in the past

[misty wharf]

Typical causes for this are something blocking the LOS, or the AI's movement logic making it turn enough to lose the player from its sight cone

If you've f.ex attached guns or other meshes into either character, they can also block the LOS trace

[mellow zenith]

@misty wharf So does AIPerception only percieve the player and not attatched components?

[misty wharf]

Perception attempts to do a trace to the target actor, if it hits something before the target actor, it considers the target hidden. It doesn't check if the hit actor is attached to the desired target

[mellow zenith]

Hmmm

Lemme screenshot some stuff in an attempt for you to assist lmao

cus im not really sure what is causing it to bug out

oh wait

hang on

On a controlled AI, where does the AI percieve from

if its a character

the AI Pawn is the top

and the bottom one is the player

Looking at it now I think it might be the static mesh on the character

as we used that instead of a skeletal mesh so we could avoid re-coding its basic movement stuff

[misty wharf]

It's possible. Try setting all the meshes to No Collide

Also, the collision channel for LOS checks is Visibility by default I think... it's configurable from project settings

[mellow zenith]

ah okay

I didnt realise I could change that

thanks

Ill try them out

Yeah it was the mesh

such a stupid bug lmao

[sullen plume]

figred it out thanks

[uneven cloud]

Look at is animation based which is what you should be using for just moving the head. Only pawns can be controlled by ai controllers and they are not "attached".

[uneven cloud]

Why aren't you checking to see if you actually got a valid point?

[uneven cloud]

The behavior tree should not be started in begin play. It should be in on possess.

[fluid ingot]

Can someone explain to me like I'm 5 what "Services" are for and some examples of how or when to use them?
I can't understand them no matter how much I have read already.

[sullen plume]

by valid point I assume you mean the value is set.
So to answer that yeah I was troubleshooting then and hadn't implemented fukll functionality so no decorators.
Having finished troubleshooting, I have added a decorater that checks if the blackboard value is set

And the second point you are also correct I fixed that too

[uneven cloud]

By valid point, I mean that the navigation system returns a valid random location. That function returns a bool. You should use it and have the task succeed or fail based on that.

[sullen plume]

yeah i didn't do that ill fix it rn

            bool bSuccess = navSystem->GetRandomPointInNavigableRadius(actorLocation, searchRadius, targetNavLocation);
            if (bSuccess) {
                OwnerComp.GetBlackboardComponent()->SetValueAsVector(GetSelectedBlackboardKey(), targetNavLocation.Location);
                return EBTNodeResult::Succeeded;
            }
        }
    }
    return EBTNodeResult::Failed;

perfect

[uneven cloud]

Services are executed when the underlying branch is running. They have 2 main uses:

1. runs a tick on a set interval usually to update data. Such as updating targets.

2. Doing something when entering and/or existing that brach. Such as setting the movement speed to be used for that branch of the tree, like sprinting when chasing.

[sullen plume]

i know i've asked this before and i've been told its way too complicated but does any1
know how to, in a C++ BTTask, make the AI Move to some FVector location?

[uneven cloud]

What do you mean?

[sullen plume]

bcuz i've been told just set the blackboard value in C++

and then in the behavior tree just use the built-in MoveTo and pass in the blackboard key location

what I want to do is do that in one step, in c++

this is one of my attempts:

EBTNodeResult::Type UBTTask_LessOldWander::ExecuteTask(UBehaviorTreeComponent& OwnerComp, uint8* NodeMemory)
{
    if (targetNavLocation.Location == FVector::ZeroVector) {
        findNewTargetLocation(OwnerComp);
    }
    EBTNodeResult::Type moveResult = Super::ExecuteTask(OwnerComp, NodeMemory);

    if (moveResult == EPathFollowingRequestResult::AlreadyAtGoal) {
        //targetNavLocation.Location = FVector::ZeroVector;
        return EBTNodeResult::Succeeded;
    }
    else if (moveResult == EPathFollowingRequestResult::RequestSuccessful) {
        return EBTNodeResult::InProgress;
    }
    else {
        //targetNavLocation.Location = FVector::ZeroVector;
        return EBTNodeResult::Failed;
    }

}

the guy doesn't move at all

oh u probably gonna need findNewTargetLocation too

void UBTTask_LessOldWander::findNewTargetLocation(UBehaviorTreeComponent& OwnerComp)
{
    if (auto pawn = OwnerComp.GetAIOwner()->GetPawn()) {
        if (AAIController* AIController = Cast<AAIController>(pawn->GetController())) {
            FVector actorLocation = pawn->GetActorLocation();
            auto navSystem = UNavigationSystemV1::GetNavigationSystem(GetWorld());
            FNavLocation newTargetNavLocation;
            navSystem->GetRandomPointInNavigableRadius(actorLocation, 10000, newTargetNavLocation);
            targetNavLocation = newTargetNavLocation;
            OwnerComp.GetBlackboardComponent()->SetValueAsVector(GetSelectedBlackboardKey(), targetNavLocation.Location);
        }
    }
}

I thought that by setting the blackboard value and then calling super the super would look at that blackboard val and go there, but it wont 😭

other option was using like NodeMemory but since I'm deriving from MoveTo and MoveTo has its own struct i dunno if its safe to overwrite it without ruining Super::ExecuteTask

[uneven cloud]

If you don't set the BB key selector in the BT it will default to self. So that's what the base move to will use.

So it doesn't move because it's trying to move to itself

[sullen plume]

i dont think thats the problem

ignore the right side of the selector

that dont matter

I'm guessing its on the C++ end

don't see any1 succeeding to do this on the web either

[uneven cloud]

There's absolutely no reason to do what you are trying to do

[sullen plume]

😭 thats what the other guy said

alr alr

ill stick with the basics ig

thanks for trying

[steel moat]

calling the Look at function is only moving the head, I need my sight perception to follow the player too, but its only rotating when i use setfocus function but it also rotates the body

[uneven cloud]

Then you need to override the pawn function that gets the view location and rotation. By default it uses the actor's rotation. I like to use a socket attached to the head and use that for the location and rotation.

[steel moat]

by pawn function u mean the functions of AI perception??

[uneven cloud]

No. It's a function on the pawn

[steel moat]

which function are u talking about...?

this is what i'm doing currently

SetControlRotation(FMath::Lerp(GetControlRotation(),rotation , time / 0.1));
TurretPawn->Attachment->SetWorldRotation(FMath::Lerp(TurretPawn->Attachment->GetComponentRotation(), rotation, time / 0.1));

the second line is setting the head rotation, and this is in the controller and im calling it on the tick

When i set the world rotation of the head it works as i want but the cone of the AI perception doesnt rotate until i use the set control rotation

[uneven cloud]

The function that I specifically said

[steel moat]

;-;

[steel moat]

the look at function?

[uneven cloud]

No. Did you read my response at all?

[steel moat]

u said to override the pawn function that gets the view location and rotation

[uneven cloud]

What part of that don't you understand? I told you the function. That it's on the pawn.

[steel moat]

i understood half of ur response, idk about the function on the pawn that gets the view location and rotation so i didnt get that part

[uneven cloud]

Did you look at the pawn functions that you can override?

[steel moat]

 */
virtual FRotator GetViewRotation() const;

/** Returns    Pawn's eye location */
virtual FVector GetPawnViewLocation() const;

this u mean?

i was looking in the blueprints....my bad

[uneven cloud]

Yes

[steel moat]

so i have to override those functions and return the head's socket rotation and location?

[uneven cloud]

Yes

[steel moat]

Thanks it worked!

[steel moat]

i want to check if a given position is navigable or not by the turret, i searched the internet and forums and found that there are two functions that does something similar to what i want, TestPathSync and ProjectPointToNavigation

but i couldnt figure out which one is the optimal? there are mixed opinions on the internet]

[uneven cloud]

Project point only checks if the location is on the nav mesh. Test path checks that there is a path. If you use the hierarchical mode, it's a lot cheaper.

[steel moat]

if (UNavigationSystemV1* NavSys = UNavigationSystemV1::GetCurrent(GetWorld()))
{
    FPathFindingQuery Query;
    Query.StartLocation = GetActorLocation();
    Query.EndLocation = RadialSearchCentre;
    if (NavSys->TestPathSync(Query, EPathFindingMode::Hierarchical))
    {
        if (GEngine) GEngine->AddOnScreenDebugMessage(-1, 5.f, FColor::Emerald, FString::Printf(TEXT("Centre valid")));
    }
    else
    {
        if (GEngine) GEngine->AddOnScreenDebugMessage(-1, 5.f, FColor::Emerald, FString::Printf(TEXT("Centre Invalid")));
        RadialSearchCentre = GetActorLocation();
    }

}

I tried this but it still prints centre valid even if theres no path

[fluid ingot]

Thank you, that was super useful!

[steel moat]

I dont want my AI to go to the location that has been blocked off and instead stay still, so far ive tried Test Path and FindPathSync but they all work the same as shown in the video. Is there anything else i need to setup like navmesh or that modified volume i used to block the path for AI?

[fluid ingot]

How do I avoid the AI getting stuck with each other?
Like, I would like it to detect "hey, there is another actor in front of me, I should get another path".
Is there a way to mark a moveable actor as an obstacle in realtime?

[fluid ingot]

I figured it out, I implemented Reciprocal Velocity Obstacles Avoidance, the results are kind of iffy, but it will do for now.

[dense owl]

There’s also DetourCrowd AI as an alternative

[fluid ingot]

Yeah, I saw it, but I thought that It could be a bit too much for what I need it for.
If I need the improvement, I'll implement it.
Thank you for the recommendation either way, I appreciate it!

[granite robin]

I just want to simply FREEZE my AI -

this doesn't have the effect I want

[sullen escarp]

You could grab the brain component off the AI Controller and call Stop Logic on it.

[sullen plume]

I am using unreal engine's MoveTo in my behavior tree, however, I want it so that even if the player is unreachable the ai will try to get as close as possible to the player. Right now, when the player goes to an unreachable place (still in nav mesh), the AI just stops.

[harsh storm]

Going back to this - I integrated GAS into my project yesterday. Took all of like, one day to integrate 😅. I spent last week learning it though. Was easy to integrate because my game's architecture was already like what GAS wanted. The main difference between my stuff and GAS is that I pretty much just had Abilities and Effects combined into one object.

[sullen escarp]

I'd run an Environment Query with scoring test on Distance to weigh points closest to the Player and filter test out any points the AI can't pathfind to.
https://dev.epicgames.com/documentation/en-us/unreal-engine/eqs-node-reference-tests-in-unreal-engine

[uneven cloud]

You should debug that. Probably has to do with your query not completely set up or allowing partial paths.

[tidal mantle]

how to setup the query?

it always returns true idk why, i even tried the blueprint node AI move to, its also returning success even when AI is blocked off

[vestal olive]

When using a UNavLinkComponent, are there any event callbacks for when a unit enters the link from either side? I know there's some sort of delegate called from an ANavLinkProxy. (when using smart links at least..)

I'm to the point of just spawning Link Proxies in begin play but this feels like functionality that should be inherent to the component version?

[uneven cloud]

Do you know how to debug code? How to find the variables in a struct? How to search the code of examples of a function being used?

[uneven cloud]

Only for smart links. Point links you can only access in C++ and need to add functionality for it in the pathfinding component

[vestal olive]

I see so smart link and bind to the OnSmartLinkReached delegate with an ANavLinkProxy is probably my best bet.

Ahh I see, smart links are designed exactly for performing extra logic anyway, can I not set up a smart link via a component though?

Seems to only be a feature on the actor proxy variant

[uneven cloud]

You can, but I think you need to use C++.

You should however be able to make a nav link proxy to be your base class.

[sullen plume]

is there a simple way to check if some point is reachable by my AI? In C++ preferabbly

[narrow mason]

Is there a way I can tune the accuracy of pathfinding and pathfollowing?

I'm using the standard FindPathSync but with a custom pathfollowing algorithm, but the agents don't seem to follow the best path. They get to the desired location but take a longer route than I'd expect

[sullen escarp]

bool UNavigationSystemV1::TestPathSync(FPathFindingQuery Query, EPathFindingMode::Type Mode, int32* NumVisitedNodes) const

[sullen plume]

for the numvisitednodes parameter what do i set that as? do i hardcode it or sdo something like this:

        //while (!bSuccess) {
        //    val *= 100;
        //    bSuccess = TestPathSync(someparams..., val);
        //    // makingg sure no infinite loop
        //    if (val >= 999999.0f) {
        //        UE_LOG(LogTemp, Display, TEXT("inf loop we out FIXASAP"));
        //    }
        //}

With hardcoding my game works fine, but I assume for larger distances to travel the AI will just stop moving? (correct me if i'm wrong. )
Using the loop, however, just makes my game freezes, I'm guessing TestPathSync takes hella cpu resources so looping that causes problems. Any suggestions on how to proceed?

[open swift]

Hey guys, I saw an interesting post somewhere about getting the tests for a specific EQS query in c++ and running them manually against an item from a previous query to determine if the query needs to run again or if the previous result is still valid, but I'm having trouble figuring out what's actually going on here.

I've got this far, but no idea how to actually check whether the test passed or whether passing in an FVector (CurrentMoveLocation) is going to work for the AddItemData call. Anyone done this before? I can't find the post now that was doing this.

[misty wharf]

It works fine but I think it still has some bugs in particular with the property binding system, but I think there are reasonably easy workarounds

The biggest "problem" with it is that some of its behaviors are not entirely self explanatory so it may take a bit of learning and digging to figure out why it does what it does

[uneven cloud]

Given that you wrote your own pathing algorithm, only you can fix it. Are you not using A*?

[uneven cloud]

Why aren't you just running the EQS with a generator on the left side that checks the validity?

[open swift]

because then i would have to add that generator to every different query, seemed cleaner to do a catch-all that just checks all the tests for whatever query you want to use

[narrow mason]

Let me clarify. I'm using custom pathfollowing, not the usual AI/PathFollowingComponent combo.

I'm taking the raw path generated from FindPathSync and moving the agent along it from point to point.

The problem is that sometimes the agent takes a long route around some obstacles instead of going through a gap that I made wide enough for it.

I don't know if there's any step I'm missing to find shortest path after generating the path to target. Or if its the path generated that isn't just optimal

[uneven cloud]

Except usually you don't want to use every test to check if it's valid. You also have a lot more control over when to use it or not. And you're not having to do something weird to get if it passes.

[open swift]

well im only checking tests that filter (or filter and score), im not sure what scenario i would want to skip a test that does filtering

[uneven cloud]

We use the EQS to find targets. The tests that are run first to check if a target is valid is different than the ones that are for picking a target. Because otherwise, they don't change targets in appropriate situations. For our location quarries, we don't test the path again, because the nav mesh isn't dynamic.

[open swift]

ah

you make a good point about path testing

i was only using this for specifically locations

[uneven cloud]

Doing it manually just gives you so much more control

[open swift]

i guess so

[sullen plume]

FVector UBTTask_MyMoveTo::getClosestNavigablePoint(FVector targetLocation)
{
    UNavigationSystemV1* navSystem = UNavigationSystemV1::GetCurrent(GetWorld());
    if (!navSystem) return FVector::ZeroVector;

    FNavLocation navLoc;
    navSystem->ProjectPointToNavigation(targetLocation, navLoc);
    return navLoc.Location;
}

I want to the ai to get as close as possible to the player bcuz rn if the player enters a closed like box or something in the level, the AI just freezes instead of reaching the outer edges of the box and stopping. Unfortunately, this function doesn't work

[stray widget]

hey devs!! I am having some issues with the AI... so they are supposed to do the following things:

start at a position,
follow the spline
on reaching the last point, find a random point near the target
and Attack

in order to find a random point near the target, I am making use of EQS... but the problem is that they are trying to take the same spot despite trying to maintain a certain distance from each other... currently I am running the EQS only once... do I need to run it in loop until they find a suitable spot??

[uneven cloud]

You aren't testing if the point is reachable. It's better to use the EQS for what you are trying to do anyway.

[sullen plume]

I'm not well-versed in EQS and from the videos i've seen its Blueprint-Based

thus trying to integrate between blueprint and c++ will be another hurdle id rataher not deal with

[uneven cloud]

It's a data asset, not BP

[sullen plume]

any good tuts on it?

[uneven cloud]

Yes. The intro to AI course in the learning library has a section on it.

[sullen plume]

is the learning library something on dev.epicgames.com?

[uneven cloud]

You need better tests in your EQS. Use some dot product tests for smarter location picking. Also a distance test to the ally's destination.

[uneven cloud]

The link is pinned in this channel

[stray widget]

the third test is getting all the enemies and gets a location which is atleast 300 units away from them😐😐... but I will give it a shot with the dot product

[uneven cloud]

That's the current location, right?

[stray widget]

yes... I am aware that they always change after that run... so do I need to do the distance test at every iteration

[uneven cloud]

I said you need to do a distance check on their destination.

[stray widget]

how do I get their destination

as of yet I have only done tests from the querier... but didnt know we could do for others as well

[uneven cloud]

Your putting the result of the EQS in the blackboard for the move to to use, right?

[stray widget]

yes

[uneven cloud]

That's how you get it

Use a context to get them

[stray widget]

so is it like asking... is any of these enemies are closest to that chosen destination?

if there is any.. than choose a different spot

[uneven cloud]

I usually do it as "is this item within x distance of an ally" if yes, filter out

[sullen plume]

this the one?

[uneven cloud]

No

https://dev.epicgames.com/community/learning/courses/67R/unreal-engine-introduction-to-ai-with-blueprints/mX27/introduction-to-ai-with-blueprints

[dense owl]

Using only cpp is just as bad as using only bp

[hoary vortex]

Hello, how do I make a floating AI that follows the exact path the player was in a second a go?

like an actual precise copy of the locations

[obtuse dome]

Has anyone run into issues with AI Move To getting verticality wrong before? I have an NPC set to walk to the pink sphere on the first floor using AI Move To (its acceptance radius is 10, same as the debug sphere's radius), but a fraction of the time the NPC will walk to that location on the 2nd floor instead of the 1st floor and say movement is complete (success)

[dense owl]

Is the pink debug sphere your projected goal location?

[obtuse dome]

Yes, it is the Destination param for AI MoveTo

[dire jungle]

Can anyone point me in the right direction? I have some movable actor obstacles in a level, and i would like them to obstruct the nav mesh as they are moved, so that the AIs dont try to walk across them. What's the right way to go about this? the first thing that comes to mind is to enable runtime generation on the navmesh, but maybe theres a cheaper way to do it?

[tidal mantle]

yea

[uneven cloud]

This is false.

[tidal mantle]

yes

[obtuse dome]

Here's a video. Normally the NPC runs straight through the door to go to the first floor (pink sphere target), but occasionally it stops right out front of the 2nd floor door (immediately above the pink sphere) rather than going to the first floor. It's rather concerning that the behavior isn't consistent

[uneven cloud]

Are they things the AI can walk on? If not, I'd use modifiers only instead as it's significantly cheaper.

[dire jungle]

theyre basically these big cubes that can get pushed around the nav mesh, and the ai shouldnt try to go through them. by modifiers only, you mean project settings -> navmesh->dynamic modifiers only + dynamic obstacle checked on the actor

[tidal mantle]

this is always returning true for me even when theres no path

[stray widget]

am not sure if I am doing this right:

[steel moat]

"allow partial path" is it true by default?

i searched about it and it seems like i need to set it false

[uneven cloud]

Why are you doing a dot product with the ally current locations?

[uneven cloud]

You can look at the struct and answer it for yourself

[steel moat]

it is set as true and i set it to false now i need to call the TestPathSync function, idk whats the last parameter is in

bool UNavigationSystemV1::TestPathSync(FPathFindingQuery Query, EPathFindingMode::Type Mode, int32* NumVisitedNodes) const

FPathFindingQuery Query;
Query.StartLocation = Pawn->GetActorLocation();
Query.EndLocation = Origin;
Query.bAllowPartialPaths = false;

bool bIsNavigable = NavSys->TestPathSync(Query, EPathFindingMode::Hierarchical); 

This is how i am doing

[sullen plume]

I do also use blueprint but only for animations / basic logic, its just I've seen videos of the crazy difference in performance between blueprint and c++ and for something like AI whose pathfinding is resource intensive, I think C++ would be the wiser choice

[steel moat]

Did u fixed ur AI bug?

[steel moat]

.

can u check this out?

[sullen plume]

not yet, im just gonna learn the EBQ thing and use that

[steel moat]

EQS

u mean?

[sullen plume]

YEAH

*yeah

[steel moat]

can u pls ping me if u able to fix that issue? im facing similar issue its just i dont want my AI to move if the path doesnt exist and call a different function when that happens

but my AI keeps going there trying to reach that point

[sullen plume]

my ai just gives up

and stops moving

are you using like directlymoveto?

hvnt tested that

movedirectlyto

[steel moat]

i want my AI to do that....

[steel moat]

nah its simple move to

[dense owl]

Yeah that’s fair.

[sullen plume]

weird

[steel moat]

i fixed it, had to change some more variables in the FPathFindingQuery Struct and then do TestPathSync

[sullen plume]

hol up

r u using the normal moveto

[steel moat]

yes

[sullen plume]

or did u create a bttask that derives from moveto

oh

[sullen plume]

I tried this

[steel moat]

using normal moveto

[sullen plume]

but found it easier just to call super::executeTask and see if that returned failed

[steel moat]

oh

[sullen plume]

if it did, I do my own thing

if it didn't ill just end it right there

[steel moat]

im testing path in another task

not in the move to

                    FPathFindingQuery Query; 
                    Query.StartLocation = Pawn->GetActorLocation();
                    Query.EndLocation = Origin;  
                    Query.bAllowPartialPaths = false;
                    Query.SetAllowPartialPaths(false);
                    Query.NavAgentProperties.AgentHeight = 80; 
                    Query.NavAgentProperties.AgentRadius = 80; 
                    Query.Owner = AIC; 
                    const ANavigationData* NavData = AIC ? NavSys->GetNavDataForProps(AIC->GetNavAgentPropertiesRef(), AIC->GetNavAgentLocation()) : NULL;
                    Query.NavData = NavData; 


                    DrawDebugSphere(GetWorld(), Query.StartLocation, 100, 12, FColor::Green, false, 10, 0, 3);
                    DrawDebugSphere(GetWorld(), Query.EndLocation, 100, 12, FColor::Red, false, 10, 0, 3);

                    bool bIsNavigable = NavSys->TestPathSync(Query, EPathFindingMode::Regular);
                    if (bIsNavigable)
                    {
                        if (GEngine) GEngine->AddOnScreenDebugMessage(-1, 5.f, FColor::Emerald, FString::Printf(TEXT("Centre valid")));
                    }
                    else
                    {
                        if (GEngine) GEngine->AddOnScreenDebugMessage(-1, 5.f, FColor::Emerald, FString::Printf(TEXT("Centre Invalid")));
                        Pawn->RadialSearchCentre = Pawn->GetActorLocation();
                        Origin = Pawn->GetActorLocation();
                    }
                    OwnerComp.GetBlackboardComponent()->SetValueAsVector(GetSelectedBlackboardKey(), Origin);
                    Pawn->bReachedCentre = true;

[sullen plume]

my goal is if fail, make the ai move to the closest navigable location in respect to target

ic

[steel moat]

u mean like this?

earlied the test would be come true even when there was no path

[sullen plume]

yeah

exactly like that

my ai will follow me, except if I jump onto the blue box (img), because the blue box doesn't have a like nav mesh. So in that scenario I want to go to the closest green area

[steel moat]

are u using custom move to?

[uneven cloud]

You should use the agent properties instead of hard coding them for the query.

[steel moat]

you mean in the editor?

[uneven cloud]

No. In your code. You are hard coding the nav agent properties instead of just getting them from the AI.

[steel moat]

how can i set nav agent properties?

if i dont set it it will get the default value

[uneven cloud]

You are hard coding the height and radius for the nav agent properties. Then a couple of lines later you are getting the nav agent properties ref to get the nav data.

Instead you should get the nav agent properties first and set the Query.NavAgentProperties to that.

Also, in UE never use NULL. Use nullptr

[steel moat]

i do that, but i copied that NavData line from the BT_DecoratorDoesPathExist class....i will pay attention to what im copying next time

[coral mesa]

im trying to do custom boids

anyone got any experience with it?

[stray widget]

I initially thought of letting them avoid the direction if there were any enemy in that direction but then I realised that wont be a viable choice as they will start avoiding the suitable spots behind them... so I removed it

[real agate]

When I use the Simple Move To BP node I get a missing component warning, anyone here know what would cause that?

[misty wharf]

What component does it say is missing?

[real agate]

It doesnt say.

[misty wharf]

Show your graph with the simple move to

[real agate]

Yeah, 1 sec

[misty wharf]

This is in a BT task right? What kind of pawn does the controller have? Does the pawn have a movement component?

[real agate]

Yeah it is a BT task, the pawn has a BPC_AI_GMC_Base_Grounded component added to it.

[misty wharf]

Does that inherit from a movement component of some type or is it just a plain actor component?

[real agate]

Yeah, it inherits from BPC AI GMC Base

Which I assume is a movement component, another team member sat this up so I cant be 100% sure.

[misty wharf]

Check what it inherits from then - For the MoveTo's to work, there needs to be something that is a movement component, this is typically something like Character Movement Component, Floating Pawn Movement Component, or other such builtin component type

Also can you paste the exact warning you got

[real agate]

Sure

[real agate]

If I use the custom task I get SimpleMove failed for AIC_Marine_C_0: missing components

If I use the built in MoveTo node I get: Unhandled Exception: EXCEPTION_ACCESS_VIOLATION reading address 0x000000000000024e

/// Movement component intended for animate actors such as humans and animals. Typically organic movement is characterized by quick acceleration to a maximum
/// speed (where most of the movement happens) and quick deceleration to a stop again. Physical forces like air resistance are being ignored with this type of
/// movement due to their negligible influence. This implementation is not limited to bipeds or any particular collision shape.
UCLASS(ClassGroup = "Movement", BlueprintType, Blueprintable, meta = (BlueprintSpawnableComponent))
class GMCCORE_API UGMC_OrganicMovementCmp : public UGMC_MovementUtilityCmp, public IRVOAvoidanceInterface
```It does seem to inherit from a movement component.

[misty wharf]

Hmm based on the error message the problem is that your AI Controller is missing a path following component

I was wondering what causes it and looking at the source code, it outputs that exact error if it can't find a path following component from the controller

[real agate]

Weird since it has one

[misty wharf]

Give this a try: In your AI controller's BeginPlay, get the path following component and check if it's valid

This kind of smells like BP corruption

[real agate]

Alright, will do.

[real agate]

AIController.PathFollowingComponent is not blueprint visible (BlueprintReadOnly or BlueprintReadWrite). Please fix mark up or cease accessing as this will be made an error in a future release. Get PathFollowingComponent This is probably the issue yeah?

[misty wharf]

Nah, that just means the path following comp isn't something you'd normally be able to read in a BP graph

it's accessible via C++ just fine though

[real agate]

Hmmm... Well I cant check if is valid because of that.

[misty wharf]

Try using get component by class and set the class to path following component

[real agate]

Ok

[real agate]

It is valid alright.

[misty wharf]

That's peculiar 🤔

Can you show the graph you used to check the validity?

It's probably fine but just to double check because I'm not quite sure why it would be complaining about this

[real agate]

Yeah sure

[misty wharf]

Yeah that's fine all right..

In your BT task, maybe do the same check there

also I think there should be a node called get path following component you can use

[steel moat]

I m facing this weird bug in my AI sight Perception, it doesnt happen every time but sometimes when the Target detected by AI goes out of its lose sight radius, none of these functions gets called. Usually they do but only sometimes they dont give the call when target goes out of sight. I was unable to figure out the what causing that bug

[real agate]

Did it already, all were valid

[misty wharf]

Well that makes even less sense then lol

Do any of the other movetos work?

[real agate]

Hmm, actually using GetPathFollowingComponent returns not valid in the task.

[misty wharf]

Oh, I wonder why does the get component by class return a valid result then 🤔

But in this case this does sound like blueprint corruption. I would try recreating the controller from scratch and see if that helps

[real agate]

Alright will do! Thanks!

[misty wharf]

Haven't encountered this, are you using any of the settings to control how quickly perception data is forgotten or such?

[steel moat]

i set the max age as 0.1

i only know that setting related to this

[misty wharf]

Maybe try setting it to 0 which is the default and see if it affects it at all

[steel moat]

0 means never forgotten

the tooltip says

[misty wharf]

Yes that's what it says but that is the default

I've never had this problem and I've never changed it from the default, so it's worth a try :)

[steel moat]

lemme try

[real agate]

Yeah this fixed it.

[misty wharf]

👍

[steel moat]

This is with max age 0 its the same when i keep max age as 1

this is with 0.1

[misty wharf]

have you tried adding prints into the perception updated to check that it isn't running or how did you determine it doesn't work?

[steel moat]

This is the bug, as you can see when player goes inside the sight of AI, the head looks in that direction and then when it goes outside its still looking at the poistion where AI last seen the target and also theres a boolean variable in BlackBoard which i am changing when player is detected and player is lost "Player Detected" when AI goes out it should be false and true when its in sight, but here its staying true even when Target is outside and when i go inside and come out again it becomes false. Im changing the value on TargetPerceptionUpdated function and also on Perception updated function

[steel moat]

Im making my AI look at player and follow Target whenever its detected but when i change the max age settings it doesnt work

[fluid ingot]

Anyone has any idea as to why the nav mesh is generated like this in this ramp?

[solid wedge]

well it's time to convert 3 year old AI blueprint into a State Machine

[fluid ingot]

I do not envy that task

[solid wedge]

I'll basically paste these different commented groups into different Task Nodes

with my conditionals as decorators on branches

[plucky hearth]

Will it work the new “500 Animations” Motion Matching sample with ai?

[tidal mantle]

maybe its related to collision or try rebuilding paths?

[misty wharf]

AI and animation are two separate things

[fluid ingot]

They rebuild automatically whenever something changes, and it keeps generating that way

[tidal mantle]

no i mean delete the recast thing in the outliner and then go to the build menu and click on build paths

im not sure if it it works

did u try it before?

[fluid ingot]

Oh, no, I haven't, I will see if it works

[fluid ingot]

It did ended up building the same way again

Oh, I made the slope smaller (by increasing the ramp size in one axis) and it generated the mesh correctly.

[tidal mantle]

oh

maybe it was something to do with nav agent properties

[quaint wedge]

Hey guys, not an ai specific issue but I'd like to know if anyone ever encountered issues with Level Sequences and Level Streaming.
Let's say I have LevelSequenceActors (that does some cosmetic transform changes on objects) in some streamed levels, and sometimes after getting streamed/unstreamed/streamed, the sequence is frozen, nothing moves anymore, no matter if I call Play manually or set is as Autoplay.

[hollow tapir]

I could use some help, In UE4, the AI Perception fully STOPS sensing players its already sensed, if it detects a new player.
Ex: Player 1 gets sensed, AI moves to Player 1. If it spots player 2 it will NEVER be updated/sense Player 1 again, at all. It wont even trigger the Ai Perception

[uneven cloud]

I've never seen that behavior. Are you sure they aren't sensed? As in checked the gameplay debugger? Or is there a bug with how you are selecting targets?

[hollow tapir]

I can clip it if you want. I literally have a print string hooked up every time it'll update, and it straight up will stop firing

[uneven cloud]

How are you selecting targets?

[hollow tapir]

Wym selecting targets? Like detecting by affiliation? -- Or do you mean which sense? Itd be all affiliations, and by sight

[uneven cloud]

How do they pick which target to attack?

[hollow tapir]

Right now it's set up so whenever the perception is updated, and its a player, it'll go to that player. So based on what actor was last updated in its sight

But thats not necessarily the issue, the going to the player works fine. But I dont understand why the perception itself stops working

[uneven cloud]

That's your problem. The updated event doesn't work like you are expecting. The perception doesn't stop working. I recommend using the EQS to select the target on an interval.

[hollow tapir]

Ohhhh dang really?

[harsh storm]

Can't say I've ever ran into this issue either to be honest.

[hollow tapir]

I didnt know that, thank you

[harsh storm]

I use On Target Perception Updated and run some logic based off that for my stuff currently.

Luthage's advice is still the better approach overall - but I'd still be interested in seeing your code

[hollow tapir]

That's what I was doing, and it seemed like it works for singleplayer, but it gets funky when multiplayer is involved

[harsh storm]

My game is multiplayer as well

[hollow tapir]

Interesting, could I maybe see how yours is set up? The On Target Perception, that is

[harsh storm]

No. Because I don't have the project opened. Show yours

[hollow tapir]

Lemme fix it up a little, I had been gutting it a bit to try and figure it out

[uneven cloud]

You don't get the updated event every time a target is sensed. It's only for new sensed targets. And losing sight of one doesn't trigger it. Just better to get all the sensed targets on an interval when you have multiple targets.

No.

[uneven cloud]

Well the silver character is behind it. What do you mean the vision half angle isn't doing anything?

Can you show your sight sense settings

With that setting, it's not going to see anything behind it.

[harsh storm]

I may have a setting set where it forgets actors. I remember messing around with that.

[hollow tapir]

Bro I swear this Unreal Engine is some wizardy shit. I cleared out all code in the actor, and rewrote it all. And now its functioning corretly? I didn't change or alter anything lmfao

[uneven cloud]

Because it limits itself to 40 degrees on either side. So anything more it won't see.

[harsh storm]

That's just papa Tim smiling on you

[uneven cloud]

I missed that you added that. Not sure what to tell you, I have never used that setting.

[olive prism]

Has this ever happened to anyone?

After about 30 minutes, many of my artificial intelligences want to go to the same point, this causes them to get stuck and not try to look for a new point.

How can i fix this?

[sullen plume]

Any benefit to setting up UAIPerceptionComponent in AIController or AICharacter class?
Here is how ti would look in the AICharacter Class:

    perception = CreateDefaultSubobject<UAIPerceptionComponent>(TEXT("PerceptionComp"));
    UAISenseConfig_Sight* config = CreateDefaultSubobject<UAISenseConfig_Sight>(TEXT("SightSenseConfig"));
    
    perception->ConfigureSense(*config);
    perception->OnTargetPerceptionUpdated.AddDynamic(this, &AEnemyAICharacter::onSeeActor);

And this is how it would look in the AIController class:

    if (UAISenseConfig_Sight* sightConfig = CreateDefaultSubobject<UAISenseConfig_Sight>(TEXT("Sight Config"))) {
        SetPerceptionComponent(*CreateDefaultSubobject<UAIPerceptionComponent>(TEXT("Perception Component")));
        
        auto perceptionComp = GetPerceptionComponent();
        GetPerceptionComponent()->SetDominantSense(*sightConfig->GetSenseImplementation());
        GetPerceptionComponent()->OnTargetPerceptionUpdated.AddDynamic(this, &AEnemyAIController::onActorDeteected);
    }

[tidal mantle]

backward offset and near clipping just increases the vision of AI by extending the vision cone slightly behind ur AI

the peripheral angle to 90

[tidal mantle]

i use that too but im facing this bug

[tidal mantle]

This one

[tidal mantle]

generally AI perception should be added to controller only

cuz AI is the one controlling pawn, so the sight and all should be in controller only

and that way u wont need to add perception component in every AI u make

[steel moat]

how can i activate my AI debugger tool right from the start?

throught code, is that possible?

[plucky hearth]

So this is a yes, it will work? I set up previously an Ai, but motion matching require some stuff like a trajectory component on the BP_Char

[misty wharf]

If by AI you mean behavior trees or other such logic to control it, that will still work just fine. You may need to add the new component into your character of course but that's really an animation concern, not an AI concern specifically

[plucky hearth]

thanks! I meaned that

[uncut rune]

Is it possible to get EQS Context be set as value from key from blackboard? RIght now I'm casting to get that info from AIC, but I would prefer to take it from blackboard.

[misty wharf]

You can get the blackboard from the AI controller

[uncut rune]

Found that, should I now cast to the Object and get actor or is there other way?

[uneven cloud]

Yes you should cast the object to an actor

[uncut rune]

Thanks!

[wide robin]

Does anyone know if this has changed?
Otherwise I'm doing something wrong. I have my main EnemyTree that is like this. The BB of the ClosestHostile tree inherits from the AIBase BB but still results as incompatible

[misty wharf]

Afaik it has not changed. I think the problem you're seeing is that you're using inheritance, when it specifically is unsupported for subtrees

[wide robin]

so i need to have the same BB for each subtree i want to use?

[misty wharf]

The BBs used in subtrees must be 1 to 1 compatible with the parent BB, that is, they can't contain any keys that don't exist in the parent and vice versa

Yes

[wide robin]

okay, that's a bit unconvenient

that means my BB will end up having a lot of keys

if i want to make a lot of subtrees

[misty wharf]

Keep in mind that you don't need to use the blackboard for everything imaginable

Your decorators or tasks can also access values directly in the pawn or controller if it's more convenient that way

[wide robin]

yes, that's a good point, i'll keep that in mind. Probably will try to use the BB as a main source of information and see where this leads

thanks for the answer!

[sullen plume]

This problem is ai/cpp so if u dunno cpp just ignore:
At 2:10 in this video the narrator says we cannot define enemies neutrals or friendlies via blueprint. How would I do so in cpp? https://dev.epicgames.com/community/learning/courses/67R/unreal-engine-introduction-to-ai-with-blueprints/D0e/setting-up-ai-perception

[misty wharf]

See here https://zomgmoz.tv/unreal/AI-Perception/IGenericTeamAgentInterface

[wooden trellis]

question - I made a waypoint system for my enemies so they can patrol areas, but for some reason they keep stopping at the waypoint for a few seconds before moving to the next one - when I want them to just immediately move to the next point. Any idea why this is happening? Not sure if there is some default affecting this or not

[sullen plume]

I sort of got the logic behind that, but when I went to override / implement the GetTeamAttitudeTowards(const AActor& other) i am lost here is my bsed implementation bcuz the guides I found are creating like IDs and whatnot and like cuustom enums in the gamemode class:

ETeamAttitude::Type AEnemyAIController::GetTeamAttitudeTowards(const AActor& other) const
{
    if (other.ActorHasTag("Player")) {
        return ETeamAttitude::Hostile;
    }
    else if (other.ActorHasTag("AI")) {
        return ETeamAttitude::Friendly;
    }
    else {
        return ETeamAttitude::Neutral;
    }
}

[uncut rune]

Show behavior tree

[sullen plume]

I have setup two callback functions, the first is called properly when the ai sees my player but the second, the forgotten, aint ever getting called. What am I missing?

        perception->OnTargetPerceptionUpdated.AddDynamic(this, &AEnemyAIController::onActorDetected);
        perception->OnTargetPerceptionForgotten.AddDynamic(this, &AEnemyAIController::onActorForgotten);

why is the ai never forgetting my player?

[sullen plume]

Here's my BP_AIController:

[uneven cloud]

Enum team IDs are far better than using fnames.

[uneven cloud]

There is no default affecting this. What is happening entirely depends on how you implemented this.

[uneven cloud]

You need to turn on forget actors in the project settings.

[lyric flint]

I have navmesh / volumes in a bunch of my levels. They are set to Dynamic Modifiers Only. When I PIE, they work fine and the AI move around. But when I run through the start screen which calls OpenLevel(...), the navmesh isn't loaded in. Is there a way to have "static" (read: dynamic modifiers only) navmesh while still loading different levels?

Actually, caveat to the above... some levels work fine. So I must be messing something up in my settings for certain navmeshes...?

[crisp oar]

#gameplay-ai message

I started following along with this to try and learn behavior trees because lord knows it's hard to get a good video tutorial, but it seems like this is very outdated.

will that cause any serious issues if I follow along with it?

[sullen plume]

where would u suggest I create the enum, GamemodeBase ?

[uneven cloud]

I would not recommend putting it in the game mode. You don't need to put it in an existing file.

[steel moat]

I created the enum in AI Controller class

[uneven cloud]

It is the best introduction to the AI systems.

[uneven cloud]

That means you need to include the AI controller in more places. Not recommended either.

[steel moat]

@uneven cloud can u pls take a look at this?

[uneven cloud]

The perception updated events don't fire when most people expect them to. Such as losing sight. You need to debug it to see what is happening. We can't debug it for you based on a video.

[steel moat]

usually they do fire when losing sight, is there any function in AIPerception OnActorForgotten?

[uneven cloud]

You have to turn on actors being forgotten in the project settings for them to be forgotten.

[steel moat]

i didnt do that

i will try this

[steel moat]

I turned that on, but i cant see any function ```OnTargetPerceptionForgotten

nvm this function was added in 5.3, im using 5.0

[dense owl]

I feel like that’s prly the worst minor version to use

[harsh storm]

Meh - people say that every version

[dense owl]

I didn’t say current version is the worst to use

https://tenor.com/view/im-not-one-of-those-people-ace-trainer-liam-im-not-one-of-them-im-different-im-not-like-them-gif-19328136

[steel moat]

oh well..

[haughty coral]

Looks like state tree will be soon usable also for utility AI.

https://github.com/EpicGames/UnrealEngine/commit/0fa5fd8e366558ee5972e9808bcaae5c12ea557f

[crisp oar]

speaking of state trees, I know different tools work better for different jobs but what are people's opinions on state trees vs. behavior trees? I started learning behavior trees and the blackboard can get a bit clunky and I've heard state trees are better in that regard but I don't know what the tradeoffs are.

[misty wharf]

It seems like a decent system to me but not sure if it's inherently any better

There's some things with it like I'm not entirely sure how you're intended to do things like perform a latent action which results in some data and then use that data to choose what to do next or such

it's certainly possible to do it but some of it just feels a bit clunky so maybe I'm missing something

[harsh storm]

I haven't been sold on it yet personally

I still stick to BT or custom state machine

[crisp oar]

I'm still learning BT, quick question, what would be a simple approach to having an AI search for you after it loses sight of you for a short bit (like go to 3 different locations near where it last saw you or something?) I followed along with this to get started: https://dev.epicgames.com/community/learning/courses/67R/unreal-engine-introduction-to-ai-with-blueprints/D0e/setting-up-ai-perception

[harsh storm]

I'd use EQS to generate those spots and then iterate from there

So pick 3 random navigable points in X radius. Then just refine it from there

[crisp oar]

I haven't learned EQS yet, I was getting hung up on how to make the state tree trigger the searching for player after losing sight of them. If EQS will help I'll look into that section in the above documentation tutorial thing

I mean behavior tree

[harsh storm]

It would be a behavior in of itself

So it would be its own sequence nodes

[crisp oar]

so like a separate task?

[harsh storm]

The most basic would be, far left side of the sequence (not the entire tree!) would be the Run EQS Query, then store the result in a BB key. Then use the Move To task to move to that position.

So like 2 nodes to get the most basic implementation working

[crisp oar]

so is it a completely new behavior tree?

[harsh storm]

But from there, you just iterate and add additional rules and all that

[harsh storm]

Doesn't have to be, but can be.

Then you can plug it in as a subtree in other BTs

For simplicity, I wouldn't bother though

[crisp oar]

I didn't know you could do more than one behavior tree, out of curiosity if you run a behavior tree inside a behavior tree will it temporarily stop the parent behavior tree?

and yeah just asking this part for curiosity

[harsh storm]

Doesn't stop it

It just becomes additional nodes for the parent tree to run

[crisp oar]

So right now I have it set up where on the left it chases player and on the right it patrols randomly, but the left chases part has a decorator that won't run it unless a condition of whether it sees the player is true, and when it runs it aborts the random patrol and what not. I'm not quite sure how to add something inbetween that process (conceptually inbetween) and still ahve the previous system work.

[signal island]

Where do i put this code to Team perception to work UAIPerceptionSystem::OnEvent<FAITeamStimulusEvent, FAITeamStimulusEvent::FSenseClass>(GetWorld(), FAITeamStimulusEvent(GetPawn(), InEnemy, InEnemy->GetActorLocation(), TeamEventRange));?

[harsh storm]

Put this "investigate" behavior in between the chasing sequence and the wandering sequence

BT's get evaluated left -> right

It is really a priority selector

[crisp oar]

alright

[harsh storm]

So the top most priority is chasing the player.

[crisp oar]

I think I get it, just need to set something up to track the player location while player is seen in a variable I suppose

[harsh storm]

You just need the last known location

[crisp oar]

oh you can just call on last known location?

is that EQS?

[harsh storm]

Yeah, it's just a vector

When the player leaves sight, store the location where they left sight at

Then feed that into the EQS

That's the position you want to search around

[crisp oar]

alright, I'll watch the tutorial on EQS then

[harsh storm]

Once you have the basics working, it is easier to iterate and expand upon

[crisp oar]

thanks for all the help

[distant citrus]

Hey all!

I'm doing some tile/grid-based 8 way pathfinding on a custom Recast Nav Mesh in C++

Pathfinding works fine enough for 4-way, but when I allow 8-way directional movement, it's almost like the A* pathfinding prefers the diagonal neighbors, making the path result a little... unwieldy lol

I'm finding the path using the following:
EGraphAStarResult AStarResult = PathFinder.FindPath(StartIndex, EndIndex, FGridPathFilter(*GraphAStarNavMesh), PathIndices);

Any ideas how I might be able to solve this or how I can have neighboiurs that are not diagonal preferred?

Thanks in advance!

[simple verge]

Is there a bug or trick to get AI Sight Perception working in a world partition level?

[vivid fern]

ime from writing a* algorithms by hand many times, whats probably happening is that it seems to prefer diagonals just because of the order of evaluation of the neighbor grid tiles, the diagonals in your test are probably more likely to be evaluated first, and since it seems your diagonal cost is the same as your cardinal cost, it gets picked and doesnt change since nothing has a better score

i would make diagonals cost more than 1

since in real physical space diagonals are further than cardinals anyway

[distant citrus]

This is true, they all have the same cost... and that makes perfect sense.

[wooden trellis]

sorry for the late reply - I just have a Move to Location or Actor node on a newly made enemy actor. There is no behavior tree or anything - The only other function firing off is how it finds the next waypoint to move to (which is a for-loop going through an array). For whatever reason, it takes about 10 seconds for the enemy to realize it's within the acceptance radius

[distant citrus]

By using Unreal Engine's struct FGridPathFilter how can I assign a cost to neighbours?

[distant citrus]

I'm wondering if this is the right area...

struct FGridNavMeshPath : public FNavMeshPath
{
    FORCEINLINE    virtual double GetCostFromIndex(int32 PathPointIndex) const override
    {
        return CurrentPathCost;
    }

    FORCEINLINE    virtual double GetLengthFromPosition(FVector SegmentStart, uint32 NextPathPointIndex) const override
    {
        // We exclude the starting point so -1
        return PathPoints.Num() - 1;
    }

    double CurrentPathCost{ 0 };
};

[vivid fern]

AH i have never used it haha 😄 as i said from my own experience writing this sort of stuff from scratch, its what i've done in tactics games i've worked on (lamplighters league, battletech)

[distant citrus]

Nice...! That's exactly the type of game I'm working on but the approach so far has been using this method...

[vivid fern]

i'd start with maybe the path cost stuff, if you can find the place where the cost from one node to another is computed, you can add the additional diagonal cost there and it should straighten out your paths

i dont how recast handles corners over like open space but you also probably want to disallow diagonals in those cases too

in path solvers i've written i often do things like check for those literal corner cases over empty tiles so i can invalidate the diagonals haha

[uneven cloud]

There is no bug or trick. It works the same way regardless of type of world.

[uneven cloud]

You need to figure out what is happening by debugging instead of being OK with "for whatever reason".

[wooden trellis]

I think I fixed it? the issue was I was targeting the waypoint actor, but when I target the waypoint's location... it's perfectly fine and there is no delay... I don't know why there is a delay for when you target an actor but whatever 🤷 it seems to be baked into Unreal as the same thing happens in a completely new project

also, the initial reason for me posting here was if someone else knew why this was happening - now I know I can bypass it by targeting the location, but I still have no idea why there is a 10 second delay when you target an actor and I have no idea how to adjust that time for whenever I need to target an actor (granted, I can likely just use some other pathfinding method, but it still would be nice to know if and how I can change that delay - but whatever 🤷 ) and yeah, i found several ways around the delay issue, but it feels like they are all "overly complicated" when there likely is just a single variable somewhere hidden in some Unreal submenu or code that would likely adjust it instead - why it's not included in the event node is beyond me

[uneven cloud]

I've never seen a delay like that, so it's not inherent to UE.

[strange bay]

So how could you have a small enough LLM to run in game, and where could you get one that isn't trained on copyrighted works etc.?

[wooden trellis]

when having an AI chase a moving actor, like the player or another AI - it's pretty unnoticeable. I only found all this out when having my AI patrol waypoints - granted the tutorial I followed had them target the destination (they didn't say why) and I figured it'd be quicker to target the actor instead (plus possible future-proofing in case I want the WPs to move or something). The delay is on the Success, when the AI enters the Acceptance Radius, the delay starts ticking - so I assume it would allow for a target actor to escape the radius, but if the target actor is stationary, you will have an awkward moment of the AI just standing perfectly still for a few seconds before it realizes it succeeded. This is really the only code running (aside from finding the WPs but again, that's just a for-loop)

this is also the same blueprint used in my completely new scene, with nothing else being touched. So if there is something else at play, I genuinely have no idea what it could possibly be or how I could find it.

*also, possibly clarifying a potential misunderstanding - the delay is with the success node, the BP continues perfectly fine when linking something to the Exec node, it's just that I only need to know when the AI reached it's destination

[uneven cloud]

Your assumption that it allows the target actor to escape is false. There is nothing in the code that is doing a delay.

[wooden trellis]

my assumption is more on the part of the UE devs (as oppose to anything in-code) - other than that, I have no idea why there is a delay or why we can't adjust it

[simple verge]

I ask because I'm seeing an issue where AI spawned in a world partition level do not have their sight perception working properly, even though sound is working fine. 🤔

Thanks for the sanity check

[vivid fern]

maybe the perception system isnt being activated at the proper time (on posses?)

[uneven cloud]

There is nothing doing a delay. I'm not sure why you can't understand that. The code to go to an actor vs a location is identical after the initial set up and tracking the target. Neither of which adds a delay. The only reasons for what you are seeing is if the actor is blocking navigation and causing a perf issue, or collision is causing it to be blocked and can't get there.

[vivid fern]

fwiw i was wandering around the other day and saw that move to supports locations that move.. you can have it repath and move toward a location as it changes

[uneven cloud]

The BT move to task does that

[vivid fern]

yeah

[uneven cloud]

The tracking is different, because that is binding to changes to the blackboard.

[vivid fern]

nodnod

[uneven cloud]

The first thing I would look into is if something invisible is blocking visibility collision channel, which you can turn on via the show menu.

[simple verge]

How do you mean? Turning on collision or something else?

[uneven cloud]

In the viewport there is a Show drop down menu. You can turn on show visibility collision to see what is blocking that collision channel.

Most of the sight perception issues is due to collision. The AI can't see something when it's visibility is blocked.

[harsh storm]

Looks neat. But not enough info in workflow and all that.

[uneven cloud]

They don't look very smart

I always find it a bit dubious when people start talking about smart AI when it's about architecture. Done well, a FSM can make smart AI

I'm well aware of the differences between the architectures.

[simple verge]

tyvm, it was indeed something like this.

[vivid fern]

yeah... "smart" ai in games is mostly about the design and player feel, the system that backs it is only in service to those things and the performance constraints

basically ime game ai is very subjective and its mostly a "soft" target as to how it feels to plyaers. Game ai is more or less about fulfilling a very different need from "being actually intelligent"

the goal is to be fun, rewarding, and make the player feel like they are awesome while still presenting a challange or puzzle

like look at the enemies in mario games, not very complex, except maybe boss behaviors, no one would argue that these games arent great and fun for many people, and i rarely hear players complain about "oh the ai is bad".. maybe simple yes, but for those games its exactly what they need

ime too players can react badly to actually tactically clever AI, they'll say they're cheating or the level spawned them in behind the player, etc etc etc.. as ai programmers in games, you fill a very weird kind of engineering role that works less on logic and reason and more on feel and subjectivity x_x

[silent fox]

Figured it out, had to switch target base class to actor from object in blackboard. Kinda stupid I cant make AI look at an object with it

[uncut rune]

Can something like this be done in Tasks or should I rather create function in AI Controller for that?

You need to expose it and set public inside the task

[dense owl]

Well… you’re doing it in a task so yes ? 🙂. Mind you, it might be better to keep the goal location function separate and use the built-in MoveTo task

[uncut rune]

Hmm, even if I want to move to couple of target locations? So I would have to create sequence with decorator that checks if path is set and iterates unless x is <, then it goes to the next BT node?

I let player set locations to which the unit will go. Something like patrolling.

[dense owl]

I’ve done patrolling between 2 points before but can’t recall what I used, off the top of my head

If it’s more than patrolling between A and B, BT might not necessarily be the way to go, since you’re manually directing the AI rather than letting it follow a behaviour tree and make decisions based on what happens

[uncut rune]

The key here is that AI is already moving between 2 points from 1 to 2 and then back. And on top of that I want to let player add additional Path Points in between. This node going back looks funny, I’m not sure if that is proper way to hack “do while” , but it works 😄

[dense owl]

There is a While node in bp

But not super great

[uncut rune]

Was just wondering if usually things like that are as functions inside BP of pawn or just as tasks.

[dense owl]

IMO, if it’s predetermined behaviour, task is fine. But if you’re giving direct control to the user such as manually placing a number of path points, that may need to stay outside of it, even if it ultimately changes a BB key, from which the AI can path

But this is just what I think, this supposition could be entirely false, as Luthage likes to remind me 😅

[harsh storm]

Interesting: https://github.com/EpicGames/UnrealEngine/commit/40347062007f8fce492dcd754dc2d7bd4b5c1615

[uneven cloud]

https://tenor.com/view/movies-the-heat-melissa-mccarthy-book-throw-the-book-gif-14851681

[uneven cloud]

If you put this functionality in the AI controller, then you need to add a bunch of things such as a delegate when it's ended.

You do have problems with your current implementation, such as not being able to abort it or handling failure.

[potent iris]

Hey, would recommend the #generative-ai channel as there are people who are interested in LLMs and generative AI
Generally you won't find an LLM not trained on possibly copyrighted work because LLMs are pretrained on massive datasets which are impossible to properly filter
Unless you train your own (which takes an enormous amount of compute)

I think the best bet is using something like phi3 3B with 8 bit quantization (3GB of VRAM), should be able to run on a potato
Or llama3 8B Q4 (4GB of vram) for something beefier which still runs on a potato

Using GGUF converted models and embedding a GGML based inference backend like llama.cpp is I think your best bet

[uncut rune]

I basically Add Path Point and Remove Path Points (by Player Controller), from the Actor Component that is on the Unit. The task runs only when the Unit is generally set to ''Transport''. When there are no points, but unit is set to transport the task will just get finished without ''following path'', if there is point it will follow it first between already going from first point to second (like main objectives from building to building). I guess keeping in on task is fine?

[fiery field]

Hey guys. With the "senses", (sight hearing etc.), is it possible to get all the actors in sight at any given moment (if so, how ?), or do I have to build list of actors myself ?
Also, I noticed that when starting the game, the actors in sight at the start of the game don't trigger the perception

[misty wharf]

get perceived actors or something along those lines should return a list of them

also I'm fairly sure that things immediately in sight should trigger perception 🤔

[crisp oar]

I'm currently trying to set it so that if the AI loses sight of the player it searches for them from by moving to three separate random locations in a reachable radius around the player. Two questions:

1. Does this structure work for that (as in will it stop chasing and start the selector that will be searching for the player when the target actor is unset and then abort that and resume chasing if it sees them?)

2. I can't quite figure out how to have EQS return more than one result, and have the result be a random selection from the list of possible locations. Suggestions?

but now that I look at it the parent one aborts when the target actor is unset so I probably need to make it not in the chase branch, however I also don't want it doing the searching for player constantly while it doesn't see them, so I'd need to find out how to make it only do the search both when the actor isn't set and only after it loses sight of the player and only once after that happens and allowing it again when it chases again.

I found the single random item from best% thing for run mode on the EQS but since it only returns one thing I imagine I'd need it to do a conditional loop or something but not sure how to make it only loop 3 times. My current thought process is I'd need it to loop until a blackboard boolean or other variable is unset, which maybe I could do a task that makes a value that starts at 3 go down by 1 each time it's run in the loop but also have the value reset when the loop finsihes...

if I run a task that has a variable that has a default value that changes during the task, will that variable's starting value reset to the default value when the task runs again?

sorry one more thing, but for some reason this new task I created is looking for target actor location and self actor even though the task doesn't even have those two blackboard keys in it. Really confused.

[uneven cloud]

None of that changes what I said.

[uneven cloud]

Why 3 random locations instead of the last seen location? The perception component has the last seen location. Getting a location near where the player actually is will look like the NPC is cheating.

[uneven cloud]

The behavior when sight is lost should be a lower priority than chasing. Not higher.

[uneven cloud]

There's already a loop decorator.

[harsh storm]

Hey @uneven cloud - would you mind if I DM you a question? Not specifically about UE per se.

[uneven cloud]

Go for it

[crisp oar]

I wanted the AI to search 3 locations before giving up, I'm still struggling to figure out how to set it up, I can share a screenshot a bit later today

[celest python]

https://github.com/LindyHopperGT/AIFlowGraph

[vivid fern]

the general way most games seem to do this is:

• Move to last seen, play some look anim
• Use an EQS or something to find points near by that are "suspicious" aka maybe around a corner out of sight etc, Move to that location, play the "look anim"
• Repeat 2-3 (or however many times) or alternately wait for a time out, say when the perception age expires for example
• Drop searching and resume a lower priority behavior (patrol? random wander? etc)

pretty much how i've approached that problem for years 😄

[tidal mantle]

isnt EQS still experimental?

is it safe to use?

[uneven cloud]

EQS has been experimental since at least 2016 and hasn't really changed since

[tidal mantle]

my goal was to do n number of patrol in r radius, the count for patrol and radius could be defined in BP, i used a conditional loop and BT Task to achieve this. I tried using EQS but with eqs i couldnt find how to make it dynamic

i setup my eqs to generate the patrol locations that out of Ai's sight

locations that r being blocked by sight

so if there wasnt any obstacle in its sight it wouldnt move

also i wanted to loop the eqs n times, the number of loop should be defined in BP

but i couldnt find a way to do that, i tried using custom bt decorator loop but didnt worked maybe i didnt properly set up but yea

@vivid fern did u encounter any issues like this in eqs and how did u solve it...
thanks and sorry for ping

[vivid fern]

i just threw EQS out there as a UE specific example, but you could easily have some other method to gather potential destinations and store them in the blackboard (or somewhere that you can access from your BT). Keep in mind im approaching this from a point where you might write your own tasks, or some c++

in my own little prototype, i just use the EQS query to generate a singular destination randomly from the top 25% results, then i just rerun the query every loop to find the next search location, based on the last sight location, so they will end up sticking around the last seen location, but moving around positions near by until some exit condition happens

(run out of search loops, or run out of time)

[tidal mantle]

thats what i did, tho im just generating random reachable points

and assigning it to move to task

in a conditional loop

[vivid fern]

i just use a loop with a repeat count

[bitter wyvern]

Is there a way to place nav link proxy in a Blueprint?
or can you change the navlinkproxy position at runtime in the level?
I have not found any functions that let me change the point link positions at all

[steel moat]

is there any way to make that repeat count to take the value from blueprint variable?

[vivid fern]

honestly didnt look, however if not could probably write a decorator that could

[steel moat]

I tried creating a child class of the loop decorator...but i couldnt figure out

[vivid fern]

truthfully i'd probably derive just decorator but i've not looked at it

[steel moat]

well...

[vivid fern]

i havent looked at the loop decorator's code so couldnt say one way or the other

looks like UBTDecorator_ConditionalLoop might be a good one to use, override the CalculateRawConditionValue function maybe

like i mentioned i havent tried but seems a good starting place ¯_(ツ)_/¯

seems like it already tracks blackboard keys tho

[hasty turret]

in unreal's environment query system, any one knows why there is "bTestQueriesUsingBreadth" option provided? What are the pros and cons of running query tests seperately before it all finishes?

[steel moat]

i think i tried this too

[steel moat]

yea rn im using the inbuilt Conditional Loop, it tracks a bool value

[spring scaffold]

I think I've found solution to this problem. Parameter which changes this path is called "Heuristic Scale" and is located in RecastNavMesh. The lower you set it's value the more rounded path AI will choose. First picture shows path for 0.1 and second picture shows path for 99.
Path which will be chosen is shown by these shorter red lines on the bottom.

[ancient wedge]

ah, interesting finding, I wonder why it's not happening to anyone else though. Gut feeling tells me it's the invokers

[spring scaffold]

I think it's not invokers because I had the same problem with regular navigation. I've seen couple questions about the same topic but never found answer. I don't guarantee this is correct answer but tried this couple times and always worked.

[ancient wedge]

okay, thx for sharing. I'll try it out in our project 🙂

[spring scaffold]

I would really appreciate your feedback when you check if it works in your project

[dreamy surge]

Is there anything I can use to put chatgpt inside UE5 where it can have a mesh and answer voice commands?

[misty wharf]

Probably but that seems like quite a fair amount of work

[distant citrus]

Hey thanks for this advice - worked on this today and had a lot of success... I owe you one 🙂

[vivid fern]

sweet glad it worked out

[sullen plume]

Why is it that my btservice (Update Plyaer Location) isn't calling OnCeaseRelevant when its turn is finished?

(Previously was on the right side and the service was ticking properly then, after switching to left side ticking stops but onceaserelevant never called)

Here's code if it helps:

{
    NodeName = TEXT("Update Player Location");
    bNotifyBecomeRelevant = true;
    bNotifyCeaseRelevant = true;
}

void UBTService_UpdatePlayerLocation::OnBecomeRelevant(UBehaviorTreeComponent& OwnerComp, uint8* NodeMemory)
{
    Super::OnBecomeRelevant(OwnerComp, NodeMemory);
    auto blackBoardComp = OwnerComp.GetBlackboardComponent();
    if (APawn* targetPawn = Cast<APawn>(blackBoardComp->GetValueAsObject(TEXT("targetPlayer")))) {
        if (AHorrorGameCharacter* targetCharacter = Cast<AHorrorGameCharacter>(targetPawn)) {
            UE_LOG(LogTemp, Display, TEXT("HasBecomeRelevant"));
            targetCharacter->shakeCamera(60*60); // huge val (1hr)
        }
    }
}

void UBTService_UpdatePlayerLocation::OnCeaseRelevant(UBehaviorTreeComponent& OwnerComp, uint8* NodeMemory)
{
    Super::OnCeaseRelevant(OwnerComp, NodeMemory);
    auto blackBoardComp = OwnerComp.GetBlackboardComponent();
    if (APawn* targetPawn = Cast<APawn>(blackBoardComp->GetValueAsObject(TEXT("targetPlayer")))) {
        if (AHorrorGameCharacter* targetCharacter = Cast<AHorrorGameCharacter>(targetPawn)) {
            UE_LOG(LogTemp, Display, TEXT("HasStoppedBeingRelevant"));
            targetCharacter->shakeCamera(-1); // disabling camera shake
        }
    }
}

void UBTService_UpdatePlayerLocation::TickNode(UBehaviorTreeComponent& OwnerComp, uint8* NodeMemory, float DeltaSeconds)
{
    Super::TickNode(OwnerComp, NodeMemory, DeltaSeconds);
    UE_LOG(LogTemp, Display, TEXT("Tick"));
    auto blackBoardComp = OwnerComp.GetBlackboardComponent();
    if (APawn* targetPawn = Cast<APawn>(blackBoardComp->GetValueAsObject(TEXT("targetPlayer")))) {
        blackBoardComp->SetValueAsVector(TEXT("targetPlayerLocation"), targetPawn->GetActorLocation());
    }
}```

[simple river]

Hi! Does anyone have any ideas as to why my AI's path following state randomly goes to idle while its pathing? Basically it stops moving abruptly and I can't pinpoint why

What I've looked at so far

• Nav mesh covers whole area, its inconsistent regardless so it can sometimes path without issues
• Behavior tree is correct and still on the follow task
• Blackboard target is correct

it just seems like it randomly fails navigation and doesn't retry

yeah the path seemingly gets aborted and never tries to repath. This gets hit

[harsh storm]

What does the visual logger say? It typically has pretty good info about this stuff.

[vivid fern]

also is the player in a space where the ai cant build a path to?

like you go up a ramp, is it wide enough for the ai nav radius

[simple river]

it is, yes

but shouldnt it try and repath after going back into an area the AI can path to regardless?

[vivid fern]

hard to tell whats happening in your video tbh

[simple river]

It basically seems like it isnt trying to repath after failing once

you actually are right, it seems like the ramp isnt navigatable for the AI. But then after I leave the ramp back into navigatable area it doesnt repath

bWaitingForRepath gets set to false for some reason, meaning it doesn't try and repath

[vivid fern]

yeah it probably hits a fail state and gives up

you could probably do some kind of fall back: fist try to path, if it fails, try to find a nearby point to path to, else enter some kind of wait state until destination is reachable

[steel moat]

you are using selector, and in behaviour tree its always left to rigth and top to bottom order of processing, so here in ur case the tree wont call the tick until the condition on left is not false.

[terse panther]

try do a check of path, like if the path exist or not, if not abort the task, as the reason for the AI not recalculating the path as its still trying to go to an invalid path....
there might be one more issue here(not sure about that), that AI can't go on the ramp even if there is navmesh becasue of the capsule radius of the AI, sometimes you see navmesh in a certain AI, but depending upon the AI capsule size, that area might not be valid for the AI to travel....

[fiery field]

I have a character with a sight sense. It properly detects other characters, but not a simple actor with a mesh (an apple). I ticked all the affiliation (neutral etc.). Why doesn't it sense my basic actor ?

[uncut rune]

Do you have stimuli on that apple?

[fiery field]

Yes, I put an AIPerceptionStimuliSource. I check it as "auto register as source".

Woops, I didn't register it for the specific sense I wanted. Sorry, it's working now

Thanks for the hint

[uncut rune]

If I switch Behavior Trees on actor, does the previous blackboards values of that behavior tree that actor was running gets reseted or will they be ''remembered''?

basically each BT has it's own BB, if they get reseted on Behavior Tree switch then I probably need to use the same BB if I don't want the values to reset?

[vague hollow]

Hey guys, do you have any idea why my perception system might not work when spawning an actor through a custom subsystem ?

If I manually place the actor on my map, it will work fine but It won't work if the actors are spawned throughout the game

[uncut rune]

check this in class defaults

[vague hollow]

😦

It's kinda weird, I'm trying to compare both the placed in actor and spawned actor (comparing their AI_Controller and actor BP) and they both look indentical

[misty wharf]

Show the code you use to spawn it

[vague hollow]

fixed, thank you

this is some dumb shizz i had dw

[misty wharf]

heh :)

[azure pulsar]

Bit of a noob question:
I've noticed that Decorator functions are pretty much a superset of Service's
Since Decorators can have a OnTick override and can change blackboard keys, when should a Service be used instead of a Decorator?

[misty wharf]

Good question, not really sure if there's a big distinction

I guess it's mostly a question of splitting it for clarity

[uneven cloud]

The BB is reset.

[dreamy surge]

How can I import chat gpt into my game so it can be inside my AI and respond to voice commands and see what I see in game?

[vivid fern]

maybe ask in #generative-ai ? otherwise, through a lot of work i imagine... game ai isnt the same thing as chat gpt fwiw

[severe solar]

that's for posting ai generations

[severe solar]

my guess is, using unreal engine source code and GPT API

anyways, is it just me who finds using blackboard AI with C++ useless? Ain't it easier to code it yourself since, unlike blueprint, it won't look like a mess

[dense owl]

Incorrect

See channel description, Allora’s direction was the right one

[severe solar]

i mean shoot your shot but after hours category is for stuff mostly unrelated to ue

[dense owl]

Ig they had to put it somewhere. This channel is mainly for UE’s AI system, rather than what people call “AI” currently (GPTs)

[vivid fern]

what do you mean "using blackboard ai with c++ useless"

[severe solar]

Yeah ik but im also saying that channel isn't for unreal engine stuff

[severe solar]

what makes it more beneficial than just coding it myself

[vivid fern]

i guess the fact that its already there and integrated into the engine is a benefit, if you want a node graph for your nodes, its also already got that too

of course imo its pretty easy to write behavior trees from scratch too so i guess its whatever you feel like

[uneven cloud]

It's not at all useless. Hard coding things isn't good practice.

[severe solar]

?

elaborate

[uneven cloud]

The BB is how you data drive the BT.

[misty wharf]

Tbh I'm not entirely sure what does "blackboard AI with C++" even refer to specifically

If it's about behavior trees specifically, you certainly could write BT's in C++ also (although UE doesn't provide much of an API for it out of the box), but I don't really think it would be any less messy... it would just be harder to change

[severe solar]

yeah mb i must've mixed them up

[misty wharf]

Every time I come back to look at GameplayBehavior_BehaviorTree it makes me wonder how exactly was this even intended to be used... It seems to be missing like 95% of the actual stuff that you would need for it to make any sense... lol

The BT it runs can't even get access to the smart object

And funnily enough if you google it, you find some stuff that relates to cheating in fortnite lol

Oh yeah and it's actually not even calling Super::Trigger which makes it look even more like some sloppy implementation that was never used for anything...

[vivid fern]

yeah this is more what my point was, what does that statement even mean haha

[coarse hornet]

So I'm having a weird problem, for some reason when I remove the repetitive eqs query and add the one time one, it works, but like this it fails. Does anyone have any ideas? (Do ping me if you have a response please). Oh also if more context/screenshots are needed, I can 100% provide some

[vivid fern]

What does the eqs service do? Get a new location at some interval?

What do you mean "it fails" as well?

[coarse hornet]

Yes, and I mean it doesn't give me a "movelocation"

Also I won't be able to respond for a bit, thanks for the help!

[vivid fern]

Whats the eqs centered on? Is it possible its reference location isnt set

[harsh storm]

New tutorial for State Tree https://dev.epicgames.com/community/learning/tutorials/lwnR/unreal-engine-your-first-60-minutes-with-statetree From Epic

[uneven cloud]

The EQS doesn't immediately return, so you don't have a valid move to location. I recommend keeping the task and moving the service to the move to task, so that it only updates while the AI is moving.

[marsh fog]

hi, does someone have any idea on how to make the AI climb the wall? the "move to" doesnt work....

[elfin hinge]

Hi, what could be the reason for Rotate to face BB entry to not work?

• Animation is not set as** root motion**.
• CMC Used controlled desired rotation is set to true.
• It get's stuck at Rotate to face BB entry until AI faces at 0 angle player, anything but 0 angle will make the task stuck.
• Target is found, looking at visual logger it's set as focus.

I'm not sure what else could cause this.

Ok, found issue. AI Controller Tick was disabled.

[spring jetty]

Hey I'm trying to build an RTS prototype and I have a weird issue with forming a formation with units. the units sometimes move to the correct position and sometimes not exactly and they stop their movement in an offset. (usually the first time fails and the subsequent attempts work just fine on a button click). I'm stuck at this for like a week..

TArray<FVector> ARTS_PlayerController::CalculateRectangularFormation(int numUnits, int unitsPerRow, float unitSpacing, const FVector& Origin)
{
    ensure(unitsPerRow != 0);
    TArray<FVector> Positions;
    Positions.Reserve(numUnits);
    int row;
    int col;
    for (int i = 0; i < numUnits; ++i)
    {
        row = i / unitsPerRow;
        col = i % unitsPerRow;
        //
        float unitX = col * unitSpacing;
        float unitY = row * unitSpacing;

        Positions.Emplace(FVector(unitX, unitY, 0.0f) + Origin);
    }
    return Positions;
}``` this is how I'm calculating the points, but it doesn't seem to be problem. I think it has something to do with AIController but I dunno what exactly..

these points are always forming a rectangle I have checked them

void ASoldierBase::MoveToLocation(const FVector& TargetLocation)
{
    if (AAIController* AIController = Cast<AAIController>(GetController()))
    {

        FAIMoveRequest MoveRequest;
        MoveRequest.SetUsePathfinding(true);
        MoveRequest.SetGoalLocation(TargetLocation);
        MoveRequest.SetAcceptanceRadius(1.f); // Adjust as needed

        FNavPathSharedPtr NavPath;
        AIController->MoveTo(MoveRequest, &NavPath);
    }
}``` and this is how I'm moving each unit.

UAIBlueprintHelperLibrary::SimpleMoveToLocation(SpawnedSoldier->GetController(), GoalLocation);``` tried this one first and then used the one above. both behave the same way..

Here is a short video showing the problem, the first button click doesn't arrange the units properly.

the behavior is the same with other formations as well, Triangular, linear..

[coarse hornet]

Gotcha, thanks!

[misty wharf]

Well that's a bit of a gotcha... Apparently calling PerceptionSystem->RegisterSource does... absolutely nothing unless you've previously registered sources with specific senses you want to register :P

[olive prism]

Does anyone know why the rest of my blackboard variables don't appear in the MoveTo node?

[uneven cloud]

Climbing a wall is not functionality that is in the engine by default and you need to add it. You need to break it down into 2 separate problems:

1. The AI needs to know that they can get from point A to B via climbing. The easiest way to do this is to use nav links.

2. The NPC character needs to do the actual climbing. This typically includes animation + implementing the movement. If using nav links, you can get notified when entering and exiting a climbing link. How that needs to be set up depends on if you are using C++ or BP

[uneven cloud]

Move to only allows vectors or actor key types. You need to set your target key to an actor.

[ruby granite]

my ai does not seem to want to target my player

it can see me

[vivid fern]

what do you mean "doesnt want to target the player"

wont attack? wont follow? are you storing the target?

[ruby granite]

im new to the ai stuff so i followed the tutorial in the documentation, any issues i had i found solutions for but this

it can see me, it updates when i enter its los, but it will not change the bool in the blackboard

im sure its a very outdated tutorial, i just cant find out whats going wrong here

[ruby granite]

before hand i hooked up a print string between each set value as bool which is this

i do have it set to neutral friendly and hostile

if i connect the print to tell me the bool coming out of the AND its false, which i assume is because of the player tag not working right

yes it is an actor tag not a component tag

even when i disable it tho, and it returns true, it still doesnt update any of the variables or start chasing

[vivid fern]

have you set a break point there to see what the value is?

[spring jetty]

anyone any idea about this ? I'm still stuck..

[misty wharf]

Ugh, every time I poke around the smart objects system it's like ?????

"Hmm I want to get a list of slots from this smart object component, I wonder now that 5.4 has a lot of updates to it if I can now do it"