gameplay-ai

[tawny raven]

I thought so too but it both has ups and downsides

The navmesh modifier is a way less intrusive method in my view - the "tool" that would generate the area mesh for this modifier actor would just be a separate step, and the modifier itself would work like any other modifier does after that - making it a clean separation and also not requiring you to adjust anything in the ways you generate your navmesh (maybe you also already have a navmesh generated and dont wanna switch it out)

Whereas a custom generator means you need to adjust things in the world and project settings, throw your old navmesh and nav boundaries out, set the settings again and generate the whole thing just to see if it does what you want it to do

As a user i would much prefer a tool that just "spits out" an area geometry for my modifier - this means i can also manually edit it afterwards

But as a developer i right now am just trying to see if ANYTHING works 😛

What keeps me away from the navmesh modifiers is that you end up having two generators (navmesh itself + this area mesh generator) in the end and both will need parameters etc :/

More parameters just mean more headache and conflicts

I will check some algos for turning texture to mesh 😮 then i think about the problem again

[wary ivy]

you'd essentially do some sort of triangulation and then convert that to the volume

[tawny raven]

Yes, thats what i m looking for, basically a binary cutoff at some user-defined threshold that decides where the edge between inside and outside should be based on pixel values, then need to place these vertices at the right height and extrude up and down with some predefined value to ensure it doesnt miss the navmesh

But of course the polygon would have to be smoothed etc

Might be a quite costy algorithm after all

Heh i just realized i did these algorithms already at uni 8 years ago, marching cubes etc

[wary ivy]

yea the hardest part is translating that into whatever UE does 😄

[tawny raven]

I did some research now and marching cubes/squares seems to be the standard still :/

I m a bit scared to run into a timesink here with this, this was just supposed to be a nice-to-have for my prototype haha

@wary ivy in which case would you want to feed a custom texture into a nav mesh modifier? From what I have seen - and from what I personally want - the most common demand is to use the painted landscape layers which are also just textures

[wary ivy]

I just prefer decoupled systems

that way you aren't marrying the end user to use landscape layers

[tawny raven]

i think once the landscape layers work, adding a texture as input is not a big deal

[wary ivy]

instead they could just generate the navmesh modifier volumes with whatever texture they want and use whatever kind of texturing technique for landscape

[tawny raven]

yea, true

[wary ivy]

because some people use single layer "auto materials" and what not

[tawny raven]

Good points, thanks

[tranquil oar]

Hi guys how is possible I did not publish it in here
😱
https://youtu.be/69coRSn80u0

[ocean wren]

Hmm, two unreal engine bundles on Humble this week

[uneven cloud]

If the nav mesh is navigable and it's just the drawing that is clipping, you can just adjust the drawing.

[tawny raven]

yea with the offset, i know, but i showed the clipping intentionally to show how much it differs

[plush cargo]

is there a way i could set up StateTree so i only need to cast to the actor once

i feel when i do a cast for every task that wants to change on actor, i am doing something wrong

[harsh storm]

Just don't cast?

ST's get bound to a specific type of class anyway

Then in your tasks, you'd just have a variable of that type

And you just bind it in the details panel for the task

[plush cargo]

right

[harsh storm]

Now - obviously, this ties the task to THAT specific type of enemy.

[plush cargo]

my variable types was set to actor

[harsh storm]

So it reduces the reusability of the task itself.

[plush cargo]

its ok player and enemy is the same

and every humanoid

[harsh storm]

If you want it to be more reusable, you are just going to have to cast.

[plush cargo]

just a misunderstanding you sorted it out

variable type was not set

[plush cargo]

whats the advantage of using BT or ST over simply writing a state machine in blueprint?

[keen crow]

c++?

[plush cargo]

https://tenor.com/view/what-gif-25726276

its a serious question btw

everything i imagine i want to do with an AI i can visualise in BP

[tawny raven]

I am not the best person to answer this maybe but I was waiting for StateTrees to be out of Beta and want to start using them soon and I like using BPs a lot for many things and my expectation towards StateTrees is that they give a more organized structure and flow for an AI's decision-making than a BP could give

[uneven cloud]

State Machines have the problem of transition logic being in the state itself. It's fine if you have a simple AI, but the amount of work to add new states grows exponentially the more states that you have.

BTs separate the action from the transition logic and makes it more modular.

State trees appear to be a hybrid, but I haven't dug into them much as they are still currently in a very rough state.

[tawny raven]

I would not present BTs as "only good", there is no silver bullet for AI, BTs also have disadvantages and can get messy and costy

[plush cargo]

hmmm ok

[tawny raven]

In any case I would rather use a ST or BT for AI than a pure BP because the BP in itself has no structure for this sort of flow so you most probably end up modelling something like a state machine or behavior tree inside it to tackle your AI problem

[plush cargo]

yeah its mostly how new ST is that is difficult for me atm. dont want to waste my time in BT.

[tawny raven]

Do you mean it is difficult because it is not fully finished or lacking tutorials or what do you mean?

[uneven cloud]

There is no "correct" architecture to use. It's entirely dependent on what your game and team needs are. I use a Utility/BT hybrid, because that's what the game needs. I wouldn't recommend it for a more scripted experience, because you will be fighting the system the entire time.

[plush cargo]

lacking tutorials or examples but im also new to UE in general

[harsh storm]

I think the docs do a pretty solid job at getting you up to speed personally.

[plush cargo]

so in a way im also new to AI, atleast the scale of it. Which leads me to chosing StateTree because its newest, so i can grow into it.

[harsh storm]

I think your issue is more that you're new to UE in general.

[plush cargo]

im stupid

[plush cargo]

yup

[harsh storm]

My advice would be to use what is stable and learn the engine before venturing off to the new shiny stuff.

[plush cargo]

thanks

there is so much in my design that will be tuned against AI so i want it early, atleast a version i can expand on as i learn more.

[tawny raven]

How is StateTree not stable? it is officially out of Beta.

The better question is, to follow up on what @uneven cloud said with which i fully agree: is the ST the best tool for the job?

That is less a question of which tool is easier to get into than the which tool fulfills the requirements of the job better

[harsh storm]

You are interpreting it as the wrong type of stable than what I intended. BT's in UE are battle-tested and have a plethora of resources. ST's, not so much. No actual games shipped with it as of yet. It is stable in as to say that Epic changed it to stable.

[keen crow]

can (should) i store TOptional or FVariant in NodeMemory? I have a decorator to set a value (single) to blackboard fields of different types (I choose types by a custom enum) and I want to reset previous BB value in OnCeaseRelevant but I don't want to declare a struct like

struct FSetBlackboardValueMemory
{
  EMyBlackboardKeyType MyBlackboardKeyType;
  bool bWasSet;
  float InitialFloatValue;
  int InitialIntValue;
  FName InitialNameValue;
  FString InitialStringValue;
  FVector InitialVectorValue;
}

instead I'd like it to be simply

struct FSetBlackboardValueMemory
{
  FVariant InitialValue;
  bool bWasSet;
}

so is using FVariants/FOptionals good idea here?

[crystal hatch]

Only PODs are really safe to be used in BT node memory. BT instance memory is one big pool for all nodes, allocated at once, initialized with zeros, and that's it, no constructors are being called. The memory for individual nodes are just reinterpretations of memory offsets.

With this in mind you can work around it to a degree by having a flag in your memory that would (due to how the memory is initialized) be initially '0' and when detected you could manually initialize the things that need that. I say "to a degree" because there's no way I can see to "fake" destruction - the whole memory pool is released at once, no destructors are being called for individual memory "cells" (because BT instance memory doesn't even have that concept), so you risk memory leaks.

[plain wasp]

I made a AI Sense Stimuli in my player.

I added AI Perception component to my AI.

I set the configuration to hearing and did all settings everyone online said to do.

I have all Sense update nodes going and a noise being made every second and not a single update fires on the AI despite the player certifiably making it through the "Report Noise Made" function

Like AI perception seemingly looks like it was given up on by the UE devs considering unsupported/broken it is. Soght worms. Somehow hearing had an entirely different system. Like, what tf was dynamic about that to the ue devs that did AI perception? Like pawn sensing was better.....

Like pawn sense was as straight as hello world after trying to figure out a simple task in AI Perception.

like, i am confused what is wrong with hearing.

i mainly want to use hearing since using sight wouldn't make sense mentally for a mob with no eyes

[keen crow]

thanks, btw can TArray<float> be considered as POD? is it safe to store TArrays in NodeMemory in general?

[keen crow]

what's under the Senses Config > AS Hearing Config?

[plain wasp]

I found it after you commented that

[plain wasp]

They gotta start that expanded..... lord Jesus

[crystal hatch]

no, but a float[] can if you really need an array.
Of course you can always make your note "instanced" and there store whatever you want as node's properties.

[keen crow]

hmm, okay. what about TWeakObjectPtrs to UObjects/Actors?

[crystal hatch]

TWeakObjectPtr is fine since its construction boils down to zeroing its internals, and it doesn't require any destruction.

[dawn schooner]

I'm doing some research about pathfinding and navmesh prior to writing the code for my Mass Entities to do pathfinding (the paths are requested per-battalion, not per unit, so it should be relatively cheap). I've been looking at the AIController, CMC and Pathfinding Component source code to get an idea of how Unreal implements all of this on top of Recast/Detour. Is there anything else I should study/read, any resources, to get a better understanding on the topic?
I'd be also interested in async pathfinding requests (so that the current frame is not overloaded when multiple requests are made simultaneously) and multithreaded requests (if supported at all).

[forest tundra]

I want to spawn a character with a behiover tree in C++, what should I use?

[crystal hatch]

See how UAIBlueprintHelperLibrary::SpawnAIFromClass does it.

[celest python]

pathfinding is already async if you are moving to an actor

it works like this:
A* from nav geometry
string pulling (post processing the path)
move request
path following follows the path segment

instead of calling move to, if you do async path finding and call requestmove on pathfollowing comp you can move FVectors with async pathfinding too

[dawn schooner]

I see. Given my setup (consisting of Mass entities) the movement is always frim vector to vector. I was looking into lower level pathfinding (not using the pathfollowing component nor CMC).

Probably after some research, I'll be using: https://docs.unrealengine.com/4.26/en-US/API/Runtime/NavigationSystem/UNavigationSystemV1/FindPathAsync/

[plush cargo]

My prototype is First Person Melee
Humanoid enemies will range from zombies to overpowered knight types
To mention a few specifics i have in mind:
Unarmed and look for weapon or run away
Block on incoming attacks
Separate aggressive and opportunist action

is this best done in ST or BT? Or can it just as well be made with BP?
I am new to UE so either way it will be a challenge either way

thankful for any speculation and i can elaborate if needed

[celest python]

pathfinding gives you an array of FVectors

pathfollowingcomp just sends direction to movement component

towards next FVector index in the array

[young lintel]

I think the general idea is that behavior tree is far more efficient than doing similar logic with blueprints

[plush cargo]

yeah but

at what point does a particlar solution become relevant

[young lintel]

When it works for your game

[plush cargo]

ok

that doesnt help me

[young lintel]

The issue being if you make the whole system in blueprints to discover that it dumps your CPU performance

[plush cargo]

im asking from the perspective of someone new to UE

and i gave a small summary on some particulars

[harsh storm]

As Luthage pretty much said before - it depends™️

It is really dependent on you.

That's just how a lot of gamedev is

What works for one project, might not work for another.

In your description - either tool will work fine.

[plush cargo]

ok then

i'll keep grinding on statetree because the system im thinking about sounds driven by states and substates

[young lintel]

State tree is probably the best of both worlds, it's just an experimental engine feature so it could change

[plush cargo]

i figured it was almost ready

[young lintel]

¯_(ツ)_/¯

[dawn schooner]

yeah, that's exactly what I needed. I'm close to getting it to work, not difficult to be honest once you reach the correct functions. Thanks a lot for the help. BTW, what's the differemce between normal and Hierarchical pathfinding?

[celest python]

optimization

and afaik compatibility of recast navmesh

[dawn schooner]

mmm have no idea what would be better for an RTS

will test both, profile, and see what works better in context

thanks a lot!!! 🙂

[young lintel]

What is the feature called in games where the geometry between the camera and player gets hidden or removed? Project Zomboid or The Sims come to mind as examples

I'm trying to figure out how I can implement this but I have no idea what to search for

[harsh storm]

Wall occlusion is what I typically call it.

[young lintel]

Well, right, but searching for occlusion doesn't produce much in this context

[harsh storm]

So add "wall" in front of it

Occlusion masking is another way I've heard it described

[young lintel]

Oh, sorry

[harsh storm]

"How to make walls hide like in Diablo" - works too

[young lintel]

Thought I was in general

[harsh storm]

@young lintel There is also this https://www.unrealengine.com/marketplace/en-US/product/mask-occluding-objects-from-camera

If you're the marketplace buying type

¯_(ツ)_/¯

[young lintel]

I mean, I'm not unwilling to consider it... that does look pretty good

Thank you, but I'm headed back to #ue5-general lol

[harsh storm]

https://tenor.com/uC4l.gif

[plush cargo]

i seem to have hit something

my new AI controller with AI perception doesnt register stimulus

sight perception and stimuli doesnt generate any updates

i can see the debug info about the perception but adding moving stimulis into the area of sense does not do anything

[misty gale]

Saved be a tiny bit ! About 2 ms if memory serves 😅not mass tho but still

[plush cargo]

again i had missed to check friendly/neutral in the ai perception

wasted time on that

[autumn sluice]

Hey, quick question: I have AI Characters that are using MoveToLocation. One AI was fine, but 16 AI causes the CharacterMovementComponents to take (combined) 50ms ish to complete each frame. Does anyone know how I should be moving these AI so they're not destroying frames? I'm going to need more than 16 of them in the long run.

[celest python]

It didnt save me much..

0.5ms only, though I only have 50 pawns

[plush cargo]

with statetree, how could i design a pulse/clock/timer i can use to get periodical updates?

[autumn sluice]

I have an actor component that does my timer logic and sends an event to the StateTree

I don't know if it's "right" but it works

[plush cargo]

ok! can i ask how you distribute the event?

[autumn sluice]

I just get the StateTree component and call Send State Tree Event with a custom gameplay tag for the event

[plush cargo]

oh

i havent looked at it like that yet

thanks

[autumn sluice]

I've had what appear to be a lot of bugs with StateTree tasks completing properly so I've had to use Events for several things

[plush cargo]

yea i noticed that states arent progressing without some debug overrides if empty

and such

[autumn sluice]

Yea I had the same issues. I have a StateTree helper component I made that has a function that sends the StateTree an event. It has solved most of my issues, and I've found having a helper component suuuuper useful for async tasks that need input/output between state siblings

[plush cargo]

yeah i havent been able to make intricate statetrees yet

still dipping my toes into it

for now i will just use an integer increment for clock

[autumn sluice]

I absolutely hated using BehaviorTree, so I'll do just about anything to get off of it. While there's still basically 0 documentation for it, I much prefer how StateTree is laid out and I find it a lot more intuitive to use than BT

[plush cargo]

im almost in the same situation

i didnt think the little i learned about BT made me able to approach AI in plain english as much as i want

[misty gale]

Yeah i got a few more 😅 that and paralellfor on the tracing did wonders for the gamethread

[dawn schooner]

Have you profiled to see exactly what is the bottleneck? Maybe with blueprints it's slower, but we could have a few dozens / hundreds of characters with AI and pathfinding without many issues. I'd profile first to make sure what's slowing down everything

[autumn sluice]

I don't fully understand this yet (insights is kinda new to me) but it's what lead me to the CharacterMovementComponent

I assume this is very high

[dawn schooner]

I'm seeing a capsule and sphere component underneath, what are they doing?

[autumn sluice]

Capsule is their collision, the sphere just triggers some distance items when the player enters it

[dawn schooner]

if you are sphere casting every frame, and the radius is big, that will probably be the cause

[autumn sluice]

Hmm...Let me look at the code for that, I generally don't cast but maybe I messed something up

[dawn schooner]

sorry wrong word

I mean Spheretrace

[autumn sluice]

Man... That sphere is just doing OnOverlap events with a pretty simple check but I guess that's enough to cause the issue. I'll have to figure out another way to handle the logic in that overlap

[dawn schooner]

A big sphere cast on tick is very expensive. I'd look just in case if you can set the physics to only do overlaps (might be cheaper) but I'd find another way. For example, you might cast the sphere overlaps on a timer (not every frame) and distribute them over multiple frames.

Async traces are also a bit faster and multithreaded iirc so that could bring an improvement as well

Have you tried to deactivate the sphere completely and see what performance looks like? Just to be 100% sure that's the cause

[autumn sluice]

Literally the only thing this sphere does (it's massive) is check if the player is in range and activate the nameplate/healthbar/statusbar. I disabled the sphere entirely and performance only dropped 10ish frames instead of 30 frames this time. I haven't looked at it thought insights yet but it it was definitely the bulk of the issue. I THOUGHT the OnOverlap was cheap, but I'll just have to adjust it

[dawn schooner]

Regardless of your usage, the sphere is actually colliding and querying spatial structures to get actors in radius. The bigger the radius, the worse it will get

It's not cheap at all

Also if those enemies are not many, you can just check the distance to the player manually (via vector maths) and you won't need the spheres at all

[autumn sluice]

Is there a better way to determine when the player is in range? I'm going to stop using this immediately. Maybe I'll just move it to a timer

[dawn schooner]

Querying spatial structures makes sense to reduce complexity of checks for many agents

If you are just checking the range from enemies (AIs ?) to a single player, just do your own distance check based on actor positions. Even if you do it every frame it will be infinitely cheaper

[autumn sluice]

I can do that instead I think. I only have four players so it's not the end of the world I don't think

[dawn schooner]

That's gonna be a nanoseconds operation

[autumn sluice]

It's an easy fix, too. I don't even need it every frame, just every few

Yea just popped open insights, now the largest cost are the two skeletal systems that I'm using (long story, but they're going to be replaced soon). Thanks for the help!

[dawn schooner]

np glad that made a difference!

[forest tundra]

Is it more logical to write with ai c++ or bp?

[celest python]

same rules applies to #gameplay-ai like anything else

use whichever makes sense

[plush cargo]

any tips on how to approach a melee AI that challenges a first person perspective

design wise

im already getting a conflict with a rotate towards location and move to location

the latter snaps

if i am unclear let me rephrase

what UE nodes are useful for making an actor move to and around a specific object or location

[dawn schooner]

Is it possible to query a single point in a navmesh to see if it is navigable.or not?

Also, do navmesh queries take height into account?

[celest python]

project point to navmesh with given box extent

box extent is your radius

if it returns false it is not navigable

though it doesnt account partial paths etc etc

just checks if there is navmesh there

[dawn schooner]

great, it might be exactly what I need. Although I fear it might be expensive and limit my multithreading possibilities if it is only in the game thread

[celest python]

you must access a uobject to do it

so its not 100% thread safe

but also 1k of navmesh projections are only expensive as calling a single BP function

(based on vblanco...)

in theory its just a fancy trace

[dawn schooner]

ohhh lol. Nice then. I had the idea to use them to solve a ver specific problem related to boid formations, pathfinding and obstacles

I only need to do a few per frame

[misty gale]

Isnt this kinda what sphere does?

[misty gale]

Either helper actor, or rotate vector around axis

[dawn schooner]

no, sphere trace does physics queries to get actors in the radius. Checking the distance between two known actors (two vectors) is infinitely cheaper

[misty gale]

Hm. Im sure someone has told sphere traces are incredibly cheap due to basically just checking vs radius

[celest python]

DistSquared is one of the most cheap things you can do

[misty gale]

PointsNear is even cheaper

By a magnitude

Not precise tho

[celest python]

hmm, i dont think so

[misty gale]

My tests says so

4.27

[celest python]

cant beat that argument, though cant get convinced without testing my own too

[misty gale]

Fair enough

I should provide my data, but i dont store it anywhere logical 😅

Id have to retest to provide anything

[dawn schooner]

It's just not true, look at the profiling graph shared some minutes ago.

And I also had a huge performance issue with sphere casts

[dawn schooner]

It is not checking vs radius. It is doing a 3D sweep with a sphere shape

[misty gale]

Not gonna start a big argument, but im not sure theres a big diff between a 3d sweep and checking vs radius

Anyhow ill take your word for it

[celest python]

i'm suspicious of its performance benefit over distsquared is, it does three ifs and one >= meanwhile distsquared is only a mathematical operation

any branch should be more expensive than math operations

[misty gale]

I can go back to dist squared and verify, i got it set up ready ti rumble

Gimme a minute

[dawn schooner]

But why are you saying that Sphere casts only checks radius? It's just not how traces work. Traces do physic queries

It's like a linetrace, but bigger due to the sphere shape

[celest python]

They are pretty close to each other on gcc but distsquared slightly has less instructions

[dawn schooner]

Yeah but that is not what a sphere trace does 😅

[misty gale]

not that big diff

~10%

[celest python]

Use stopwatch and profile in shipping btw

[celest python]

I was talking out of the context

[dawn schooner]

Ye I feel like the conversation deviated 🤣

[celest python]

oh god.. i just switched to MSVC on godbolt and instruction count went to triple

[dawn schooner]

My original point was that Sphere traces are much more expensive than anything that just compares two points

[misty gale]

It was a simplification of what it entailed, but it doesn't matter. It's pretty clear that its quite a lot more expensive than a vector comparison

[naive peak]

Is this a good way to handle the different senses that are detected?
It works well for me, but wondering if its the proper way

[sturdy iron]

Hello, what are the solutions for

UAIPerceptionComponent + UAISenseConfig_Sight

not recognizing a player moving in a vertical direction for example running through a set of stairs
It just loses sight and my system kind of allows the AI to only move to a target only if it sees it

[celest python]

Just had chance to check StateTree..

god damn that data binding system is awesome... 😌

[plush cargo]

what is a helper actor in this context?

[misty gale]

Could be an actor that just attach itself to target and applies a steady world rotation

[plush cargo]

so something like an empty object used for relative positioning?

i did a lot of that in 2D for camera related positioning

[misty gale]

Yeah exactly

I guess it would be a 2 component actor, a scene root and a scene targetlocation which is offset from the root

Perhaps it doesnt even attach, it just follows with some movement speed

Was the first thing that came to mind atleast

[plush cargo]

not a bad idea

but at this point with my little knowledge i have nothing to compare it with

im still a bit confused about how to move without changing look direction

for an AI character

[misty gale]

It was to provide two different directions basically , one look direction, and a separate move direction

Which ofc could just be two direction vectors

[plush cargo]

aye i get that

i just ment the toolbox of UE functions

what im confused about i mean

[forest tundra]

do you know a video that shows how to spam pawn ai for c++

[plush cargo]

so having dabbled in StateTree i have a question

a couple of questions

is it common practice to treat state tasks as aynchonous events ?

should i write most of my actions on my AI actor and use state tasks to execute these actions ?

if so, how should i make callbacks to the state tree?

[ocean wren]

I modified where it did the raycast for the sight sense so it did a different raycast to different parts of the body each update.. so basically choose a point on the skeleton and cast to that.

[plush cargo]

Anyone?

[harsh storm]

No one really knows. Just guesses. ST hasn't been out long enough for people to formulate real-world practices. Only theories.

[harsh storm]

Tasks should be done in the task itself. Use the character as the data holder. So ST becomes the brain more or less and the character is just how the ST gets information from the world. (At least one way)

[plush cargo]

i wasnt able to run AI Move in a task, which is where my confusion starts

[harsh storm]

Couldn't say why without digging into it. But even in BT/BP, people run into issues with using that node. So it could very well just be a setup issue. And I don't have the time to dig into it.

[plush cargo]

i see

Ok

this somewhat points me into a direction i can use

[plush cargo]

im truly looking forward to StateTree knowledge

[solid snow]

hey guys I am quite new to BT. and I am running into this issue.
I set two decorators,
IF Player can't see the enemy -> go to random location
IF Player can see the enemy-> find a place where to hide
The 'PlayerCanSeeTheEnemy' bool works fine but when returning TRUE nothing happens. The Player should look for a place where to hide but he doesn't and it stays stuck, no movement. In the BT the Root is highlighted yellow.
Weird fact is that if I remove the condition both task and EQS query works fine. Can someone help me with this? Thank you guys!

[uneven cloud]

One thing to keep in mind is that the BT evaluates left to right. So the one on the left has higher priority, so if you flip them you won't need the Can't See Player.

I would use the visual logger to debug. Likely one of your tasks is failing and the vis log will tell you which one.

[solid snow]

I have another decorator on top of this, I managed by 'merging' the two conditions for each sequence and now seems to work fine 😄 thank dude

[autumn sluice]

Anyone have any idea why this option would be greyed out? The documentation doesn't seem to help.
https://docs.unrealengine.com/5.1/en-US/world-partitioned-navigation-mesh/

I was using Navigation Invokers, but it turns out that AI cannot move to locations outside their invoker range...which becomes pretty useless when I need AI to move decent distances

[versed maple]

I haven't been able to get my hands on it yet, but the first thing I'll try is to put the actions in the control/character, like I do in BT, where I send the actions with an Interface.
What happens if you control an enemy that is not a character? Would you have two practically identical ST just for that?
Your brain tells you where to move but it's your legs that make you move.

[plush cargo]

i did a quick test, i can access the ai controller and move to location in a task

but it also worked just casting to an event on the actor

i guess its very versatile

[versed maple]

Yes, I'm from the anti-interface club but if you have a bipedal soldier, a creature, a drone and a tank. It's hard to come up with a hierarchy of AIController and characters that works for everyone.

[harsh storm]

Yeah - it is honestly. But also at the same time, I don't know how I feel about it being tied to a specific class. Actor context that is.

[celest python]

i still couldnt understand how we are suppose to access owner actor from tasks/conditions

i guess its strictly meant to be used with the data bindings

I was thinking about a UCLASS generator with yaml until i realize this

but for example for a move to task, it feels weird to not have an "owner" to move

[plush cargo]

this worked fine

i have an evaluator on tick that gets various values every nth times

like perception data and so on

[young lintel]

Alright, I admit I have no understanding of how this is supposed to work, but does anyone have any hints about why this isn't doing what I want it to?
(If "IsPlayerControlled" returns true I basically want the behavior tree not to run at all)

[plush cargo]

its funny

my attack method is flipping a variable from checked to unchecked

i have a task that flip to checked in a task. no task tht flips it back

yet the AI attacks

[young lintel]

Yeah.. if I delete this connector between Root and Selector, my AI still runs the whole behavior tree

But without any of the handy debugging information

They're becoming self aware

Your AI just wants to kill, mine just wants to wander

[plush cargo]

😄

its just...

as if the variable on the character is addressed via a task, it somehow gets flipped back because reasons.

but its not consistent either

[young lintel]

I guess need to just tell the AI Controller to stop or start thinking whenever I toggle this

I was hoping for a more intelligent solution allowing it to make its own choice, so it would still run certain things even when under player control, like responding to attack or starvation problems

[harsh storm]

Create a variable in the task/condition and put it in the "Context" category.

Obviously it should be a child of actor

[celest python]

Does it automatically detect and bind the context actor in the schema?

[harsh storm]

You still have to manually bind it.

But it actually shows up if it is in the Context category

[celest python]

Ah, alright

[harsh storm]

#gameplay-ai message

"Enemy Actor" is the variable inside of the STT_Destroy task.

[uneven cloud]

How are you switching to player controlled? If you are changing controllers (which is recommended) then you should be stopping all logic on the AI controller.

[young lintel]

No, the player is independent of the characters in the game, they can merely choose to give orders to them

The AIController should, ideally, still be responsible for what the character does after you tell it to do something

That isn't how it currently works, but that is how I would like it to work

Currently I have blueprints that control moving to things and interacting with them when the player gives a command

But the autonomous AI will need to be capable of all of the same interactions so it feels like unnecessary work to have to create the behaviors in two places

Think of like... The Sims, I guess

For a very loose comparison

I was hoping I could have the player tell the character to, for example, mine a rock, just by adjusting some values in the blackboard, while still giving the AI autonomy over how it accomplishes that task

However I was also hoping to be able to block certain sections of the behavior tree if the player is actively assigning tasks

So like if the character is starving to death it would override the order to mine a rock in order to solve the starvation problem before continuing to mine

But it wouldn't just start wandering around like a non-controlled character

[verbal shore]

Sounds like you could make use of utility scoring system. Afaik there is no built-in functionality for that but you can build your own solution or look at Github or marketplace

[young lintel]

Well, yeah... but in the meantime it would be good if I could just lock out sections of the behavior tree, lol

I guess I could run a task in every single sequence that fails if the player is controlling

But it seems like decorators should be capable of this if I was using them correctly

[verbal shore]

Your execution was wrong in the picture above

You can use decorators for that

Let me check one of my behavior trees

[young lintel]

Oh I have to move it down lower

[verbal shore]

You have to have selector above

[young lintel]

I guess that makes sense

[verbal shore]

Like in the picture. Also if you don't want to do anything if the player is controlling, put wait or some task there even though there is nothing going to happen

[young lintel]

Well, ideally, I'm going to add some behaviors that will be the sorts of behaviors a player could assign

[verbal shore]

I had some troubles before so that's how I do it.

Yeah, then there would be no problem at all

[young lintel]

Cool, thanks

[celest python]

I feel good because StateTree exists and it looks like almost a perfect tool for general purpose usage
I feel bad because StateTree exists and it abstracts away around 5k lines of code I wrote to handle what StateTree could do easily in editor

[harsh storm]

Funny thing about ST - it can be modeled in Godot almost exactly 😅

[celest python]

thats surprising

[harsh storm]

Not really. The scene tree easily leads to this kind of design 😅

As a matter of fact - that's the most common way people create state machines in Godot-land

Not a 100% replica of ST mind you, but the rough idea.

[celest python]

hmm

it really feels alien to think outside of the UE scope

[harsh storm]

Is this your first engine?

[celest python]

yeah

[harsh storm]

Ahh - makes sense.

[celest python]

one year ago when I first learned about ECS it was also another shock of my life

[harsh storm]

Fancy for loops

[shadow furnace]

a fancy for loop + a vaguely database like query

there ya go fps go up

[ocean wren]

I don't suppose anyone here has a lot of experience with NLP do they?

[keen crow]

How would one make an AI to approach a player/location maneuvering (i.e. strafing left and right while still approaching target) ? So far I only have an idea of running eqs query in a service on a move to to reevaluate next point with providing players LoS and locations in EQS contexts for EQS tests but idk... intuitively I feel like this is wrong

[ocean wren]

EQS is definitely the right way to do it

the squeenix guys used that and just did a series of concentric circles as target points for their "waiting in a queue to attack" agents

so they use EQS to generate points around the target.. then the strafing comes from simply spinning those EQS locations by animating the heading for the generator

They had some sidestepping logic so that they faced the target and then stepped left/right if the EQS point was far enough away for a sidestep

[plush cargo]

is this a talk or tutorial or something?

how can i treat a state as enter dosomething and exit on condition

as opposed to enter dosomething and exit

[random ridge]

hi, my behaviour tree moveto is failing, but AI Move to in the BP is working

[plush cargo]

check that the target you specify is being communicated in the moveto

usually my problem, target actor var doesnt contain anything

[plush cargo]

use debug

[random ridge]

it's set and valid, it needs to pass the BT check

which debug?

[plush cargo]

print debug

from the moveto task

[boreal tundra]

Hello, do you know why since I've changed the model for my enemy he became dumb af? He can't come down the stairs, if you go behind a sofa standing he can't see you

[random ridge]

not sure how to debug in behaviour tree? seems like I cant add comments etc

[boreal tundra]

with the previous model he wasn't like that

[plush cargo]

compare them

[plush cargo]

find a way to see if that key is valid

because i dont think moveto throws any error if the target is none

[random ridge]

well, it has to be, the previous node Blackboard: TargetPlayer is set checks it

when the AI cant see the player, it runs the left branch, when they see them, it runs right right branch, the TargetPlayer var becomes set & valid, and then I move to it

if I print out the value in BP it returns the player character

the moveto function just fails, but it works in bp

ok weird, its working now after refreshing a BP didnt change anything ;s

[plush cargo]

anyone know if parent transition is the first above

or back to root ?

[faint fox]

any idea how to figure out which node aborts a task in bht?

[autumn sluice]

From my experience it will go to IdleState

[plush cargo]

ty

[ocean wren]

What do you guys think.. functional programming for all now in UE? https://simon.peytonjones.org/assets/pdfs/verse-conf.pdf

[autumn sluice]

Anyone know the intended way to have AI move over large distances in World Partition? I was using Navigation Invokers but when I give the AI a location outside of their invoker range they just won't execute it. I was trying to get the navmesh to use Tile Pool Sizes but my IsWorldPartitionedNavMesh is grayed out and I cannot seem to get it to activate. Not sure what setting is missing as I have all the documentation settings correct.

[celest python]

Publication date: January 2023
I wonder how did you access it 😅

[harsh storm]

@celest python You still messin' with ST's?

[celest python]

yeah currently trying to inherit from FStateTreeConditionCommonBase

[harsh storm]

One thing that I am not sure about the usefulness of is the evalutators. I know what they're for but like...you can just get the same information in the given task.

Any thoughts?

[celest python]

Was just thinking about it ten minutes ago, but I dont have any certain ideas too yet

I guess its suppose to update the parameters of the schema directly

rather than local task parameters

[harsh storm]

Hmm - does the entire state tree rerun after a "Tree Succeeded" call?

So after a state transitions to Tree Succeeded - does it reevaluate all that stuff again?

[celest python]

afaik yes

[harsh storm]

I guess an evaluator can be looked at like some kind of service from BT. But for the entire state tree.

[celest python]

yeah, gotta check city sample to completely understand - but its ~100gb

[vale pawn]

Is there a way to call the same sequence from multiple places?

in a blackboard

or group a bunch of tasks together to make one task that I can use in multiple places around my blackboard?

[uneven cloud]

Make it a sub tree

[harsh storm]

And now - seemingly randomly, the "TreeStart" event isn't being called on the evaluator 🤔

[wind surge]

I made a basic ai like this, but when a horde of enemies follow me, they always make a line and it looks silly, what can I do to make it more horde-like and not a line?

[keen crow]

Is sight perception supposed to refresh visual targets after expiration? I'm experiencing something weird - I'm standing right in front of an enemy and after sight sense lifetime perception component doesn't return any valid stimuli source, but inside of it perceptual data there is still a visual stimuli of the player character, but is is expired with like 250ms overdue

[ocean wren]

It was on twitter feed.

[celest python]

yeah just saw it, we are talking about it at #lounge right now

if you wanna join

[ocean wren]

I don't think its actually going to be the next scripting language.. at least I hope not 🙂

Anyone tried doing any compute shaders for AI with Unreal?

[keen crow]

make a BT with a MoveTo task to a location which is set by a Run EQS service that finds a random spot for a horde to move towards your players

or try just deriving your AI controller from ACrowdFollowingAIController

[uneven cloud]

Instead of going directly towards the player, vary the locations. The EQS is really good for this.

[ocean wren]

+1 for EQS

[wind surge]

Thanks for replys, Ill check the EQS

[keen crow]

can I manually refresh expired sense stimuli? For sight in particular

[ocean wren]

why would you refresh it if its expired?

[keen crow]

@ocean wren because of this

it seems that registered sight stimulis are not expected to be refreshed because of the sense's notify type

[ocean wren]

Hmm, it should update the sense stimuli automatically while you've got it in vision.. are you sure something isn't blocking it?

[keen crow]

100%

[ocean wren]

Been a while since I played with it, but I never had any issues with it needing refreshing or anything to my memory.. I guess you'll have to single step it and figure out whats happening

if i remember right, it has a timer since last sensed and just resets that every time it was successfully sensed

well, not a timer, an elapsed value

sense timeout or some such

[desert moat]

Anyone every had an AI skip a task only once trying to get my AI to walk to a location then return but the return task keeps getting skipped on the first loop so the location becomes offset

You can see it here, the first loop it'll play but the Move To Player Check Return plays too fast compared to the other loops where it plays normally and functions normally

So I think my AI Move To is failing on the first loop, after setting up a print to the failed node I can see it printing only on the first loop anyone know what could cause this?

[ocean wren]

What is it you're trying to do? that looks like a very weird sequence

The reason youre node is succeeding is because its going to targetlocation, but you're already at target location

I assume you meant to go back to guard location?

but either way, you should refactor that.. so that its just a lower priority to return to guard post and then allow higher priority if there's a threat or a patrol to do that takes over the priority.. that way you'll always go back to guard post if nothing more important happens

so use a selector instead of a sequence

[vale pawn]

How would I re-create the "Wait" task that's given to you in the Behavior tree?

[verbal shore]

Use Task Wait Delay node in your task

[vale pawn]

Hmmm

I'm trying to figure out how I can prevent a custom task from moving on until it's done

[verbal shore]

I will send an example in a sec

[vale pawn]

Cool, thanks

[vale pawn]

Is there a way I can do it when I don't know how long it will take?

Like the Move To task?

where it moves on once the AI reaches its destination

[verbal shore]

You can finish task whenever you want by using finish execute node

You can do something like this for example:

[vale pawn]

I see

thank you

[verbal shore]

you're welcome 🙂

[vale pawn]

I'm having another issue

with 2 Move To tasks in a sequence

and it does one but not the other

[verbal shore]

Can you send a screenshot?

[vale pawn]

sure

So I get the two locations in services

then move to the first one

then the second one

[vale pawn]

and it should just repeat that

[verbal shore]

You're using selector, you need to use sequence instead

[vale pawn]

lmao dang it

Thank you so much

[verbal shore]

haha no problem!

[desert moat]

Anyone know why my AI Move To keeps failing when my patrol points are too far apart?

[uneven cloud]

Using the visual logger will help you to debug that.

[neat marsh]

How do you get pawns to block navigation for other pawns but not themselveS?

I have setup dynamic navmesh generation etc. but having the pawns affect the navmesh underneath themselves leads to them twitching as they can't figure out the start point

Is there something to make them stick to a path for longer before abandoning it?

[uneven cloud]

You really don't want to change the navigation under them. There is a crowd following system for obstacle avoidance.

[neat marsh]

Thanks, yeah i've abandoned that line of thought.
The crowd following system is a bit awkward for my use case I think. I dont want enemy soldiers stepping out of the way of eachother like they are working together

[lyric flint]

https://www.youtube.com/watch?v=zIDqJ9qDm9w&ab_channel=ReidsChannel
hello i tried to follow this tutorial and everything works, i wanted to add a behavior where npc will roam randomly
so i added this script but it doesn't seem to work. is there a better way for it?

[neat marsh]

I'm now using crowd avoidance which is working great, I realised it respects some of the RVO avoidance settings like mask group which makes it great for my needs!

But now I have an issue where setting the group to avoid at runtime is ignored, I assume it gets registered or something. Anyone know how to change it?

Using all 3 nodes to triple check even though I should be alright with the ignore node:

[crystal hatch]

How do you trigger the Random Roam event?

[plush cargo]

any recommended AI assets ?

im thinking AI might bee too big to make for me

[lyric flint]

i tried it directly to event tick and next to wait event but both didn't work

[crystal hatch]

the way you have it implemented calls for only a single call, and once done it will loop indefinitely. If you call it repeatedly you'll interfere with previous calls resulting in AI not moving at all.

[lyric flint]

i see, do you happen to know how it should be implemented or a tutorial explaining some parts?
sadly i just started on trying out how behavior tree works but i still don't get how some of it works.

[forest tundra]

how do i create behivor tree class in c++

[crystal hatch]

#gameplay-ai message

[crystal hatch]

I do, but it's been described so many times that I suggest you looked it up. In short I'd suggest using BTs and combining EQS task with a MoveTo task. Alternatively you can implement your own BT task that would contain the BP code you've posted.

[forest tundra]

UPROPERTY(EditDefaultsOnly)
TSubclassOf<UBehaviorTree> BehaviorTree;

in .h

and

UAIBlueprintHelperLibrary::SpawnAIFromClass(GetWorld(),SpawnClass.Get(), BehaviorTree ,Location,Rotation,false,nullptr);

in .cpp

BehaviorTree tree gives error

what is the problem

[crystal hatch]

then you say you get an error it's always a good idea to paste the specific error you get from the compiler. But in this specific case it's easy to guess. Check UAIBlueprintHelperLibrary::SpawnAIFromClass's signature and it will become obvious. But I also believe that' pretty much what the compiler tells you.

in short: TSubclassOf<UBehaviorTree> is not the same as UBehaviorTree*

[forest tundra]

oh thank

it worked

I wasted my hours

[plush cargo]

does focus only work in a BT task ?

[desert moat]

Yeah it did, was the nav invoker radius

[crystal hatch]

always, and I can't stress it enough, ALWAYS read the compiler errors, most of the time it's a silly mistake you can address on your own. It's programmer's fault 99.999% of the time 😄

[crystal hatch]

I know BTs in and out but for the life of me I don't know what you mean 😄

[plush cargo]

i couldnt get this to work in a statetree task

but judging from your answer i didnt try hard enough

and from my very limited experience the only other time i used it, was in a BT

bad question i kno i kno

[crystal hatch]

yeah, this functionality is BT-independent. You need to make sure the flags in your pawn or movement component are set. Look for bUseControllerRotationYaw

[plush cargo]

oh ok

i hadnt thought of that

@crystal hatch while i have you here i need to ask, how can i enter a statetree state and not exit until i want it to exit?

[crystal hatch]

Sorry, don't know, not my cup of tea (yet). You'll have to wait for @lyric flint to drop by.

[plush cargo]

gotcha 🙂

[forest tundra]

I want to spawn ai

Behavior tree not working when spawn as actor

how can i make it work

[echo lark]

@crystal hatch Why is SmartObjectSubsystem only creted on authority ? (not clients) ?

[crystal hatch]

Pretty much the same reason AIControllers only exist on servers. If we had one on clients we'd need to keep them in sync complicating the whole communication. We'll make it work for Players, don't worry.

[echo lark]

okay thanks, was mostly wondering about Players and predicting interactions ;

[verbal shore]

Set Auto Posses AI to Placed In World Or Spawned in your character

[forest tundra]

thankk🗿

[terse panther]

hey there, can someone tell me how can i change the rotation here?, i tried to put a custom query of both type, points and actor, But the result is save for every type of actor/vector value . For the default one (EvnQueryContext_Querier) its taking the forward vector of the pawn. but i want to generate behind the pawn i.e. forward vector * -1. Any idea how can i do this?

[marsh pasture]

is there a blackboard present per AI spawned

or would 10 AI use "Spawned" in Blackboards

[ocean wren]

You can create your own Environment Query Context and provide values for yourself..

See those dropdowns? those relate to query context objects.. so make your own

[keen escarp]

Hey, i would like to check if enemy is facing the player or is atleast almost facing the player.. I think i need to get enemy's forward vector and characters location and use DOT-product..Buuut I'm having problem implementing this to work on blueprint, could someone guide me how would u do this kind of functionality

[terse panther]

I tried to make my own, but they are not working don't know why.

[ocean wren]

Yes, you want the dot product.. you should get the vector between your player and the enemy.. (enemy position - player position).normalized and your forward vector, then do a dot product on that to get the angle

You might want to make it only work on the horizontal plane...

There's actually a blueprint node that has this functionality if I recall?

[celest python]

Creating new FStateTreeConditionCommonBases was more wonky than I imagined

I dont want to think about the boilerplate written behind the scenes to make it work

[harsh storm]

That's a shame. It looked pretty straightforward to create a new task at least. I only glanced at it.

[celest python]

there must be some very hell of a reflection system magic happening

to optimize it to moon

for example you manually put

    UPROPERTY(EditAnywhere, Category = Parameter)
    bool bInvert = false;

to everywhere

base class doesnt have it

and you must have this constructor

    explicit FStateTreeCondition_AngleFromLocation(const EStateTreeCompare InInverts)
        : bInvert(InInverts == EStateTreeCompare::Invert)
    {
    }

and declare another struct to put your data

etc etc

[harsh storm]

Sounds like another use case for BP 🤣

[celest python]

this one probably cant be done in BP

this is to expose things to data binding system

[harsh storm]

Oh - I thought you were just doing custom enter conditions

[celest python]

I'm creating that Compare X thingies

yeah

finally managed to make it appear

i guess that "bind" dropdown menus dont appear magically in BP

if it is, fml

[celest python]

gonna write a yaml to FStateTreeConditionCommonBase 😂

[harsh storm]

Definitely can do in BP

[celest python]

https://tenor.com/view/anya-spy-x-family-shock-anya-kazumaozk-gif-25861816

[harsh storm]

So uhhh - look at what the BPBase does and copy that 😅

[celest python]

it doesnt seem to do anything, but my R# doesnt work at StateTreeModule

[harsh storm]

I think there is a setting you have to flip for it to work on the engine module/plugin stuff. Rider doesn't index that stuff for me either.

[celest python]

oh yeah i didnt enable that on r#

thanks

oh...

things just make sense

i was wondering why StateTree has a linker

to handle that BP data binding

and custom c++ boilerplate stuff i mentioned

so its not possible to copy what BP base class is doing

because BP classes are handled internally

and probably it boils down into state tree compilation process

oof

here comes yaml parser then

[keen crow]

remind me does a BT which is ran in Run Behavior task has to have the same blackboard asset as the invoking BT?

[ocean wren]

Hmm, what class does that Bind expose? that a delegate?

I need to have a bind-able entry in a list of matches for an NLP system

wondering how to make the UI bindable... basically a list of token strings and types with a matcher method and some method to build if a match is made

guess I'll give statetree a look

[keen crow]

no you can't

ok another question. is it possible to set FBlackboardKeySelector anywhere outside of BTs? For example in some data asset?

[celest python]

you mean this? i dont understand yet but its somehow handled at StateTreeLinker

anim graphs also has the same functionality but you can also bind function calls if signature matches

[uneven cloud]

Yes. You do need to be able to initialize from the correct blackboard asset. Your class also needs to use IBlackboardAssetProvider and you need to implement the function to get the blackboard to show the drop-down for the keys.

[keen crow]

thanks

[main hollow]

in ue4, is there a way to have nested dynamic behavior subtrees? for example, the parent BT has a Set Behavior Dynamic node where you add a behavior tree which also has a SetBehavior Dynamic node with a different tag?

I tried setting a behavior corresponding to the tag within and it fails

it doesnt look beyond the existing nodes on the parent BT for that tag as far as I can tell

[keen crow]

there is, but it's weird. the thing is when you call SetDynamicBehavior, the UBehaviorTreeComponent (or something inside of it) searches for BTTask_RunBehaviorDynamic only in loaded trees in tree stack. So what I ended up with is create my custom Run behavior dynamic derivative that re-injects dynamic behaviors after starting the next dynamic subtree. To optimize it a little bit I also created my own UBehaviorTreeComponent that allows to set dynamic behavior for trees starting from some index in tree stack

[main hollow]

ohh.. I will have to look into overriding some of these then 😅 At first, I thought it is a matter of delay or something coz I was trying to set the internal BT as soon as I added the external one.. so I tried adding with 5 secs just to test.. but no.. it wasnt that 😆

[main hollow]

thanks for this.. I will give this a shot!

[haughty coral]

A logic should run when BT branch is entered and left. What is better choice? Service or Decorator?

[keen crow]

any, but doing it with a service is easier (although makes more sence making it with a decorator). just override OnBecomeRelevant and OnCeaseRelevant and set bNotifyBecome/CeaseRelevant to true in node's constructor

[haughty coral]

Good, thanks. I was wondering is there is any standard for it but there is probably none.

[celest python]

Question for StateTree - if mieszko or mikko ever see this:
Are there any plans for integrating binding "getter functions" to data bindings on ST? like how anim graph does it

[celest python]

UE::StateTree::PropertyBinding and UStateTree::Link() btw
FStateTreePropertyBindingCompiler

[celest python]

oh no.. you also have to write that linker stuff for tasks too

but good news is you have something like TStateTreeExternalDataHandle

[celest python]

        InstanceData.AbilityEndDelegateHandle = ASC->OnAbilityEnded.AddWeakLambda(Actor, [this, Actor, &Context] // @todo this is extremely dangerous.... 
        {
            InstanceDataType& InstanceData = Context.GetInstanceData<InstanceDataType>(*this);
            InstanceData.bAbilityIsRunning = false;
            UAbilitySystemComponent* ASC = GetAbilitySystemComponent(Actor);
            if (ensure(IsValid(ASC)))
            {
                ASC->OnAbilityEnded.Remove(InstanceData.AbilityEndDelegateHandle);
            }
        });

I made something like this in EnterState to hack the const correctness in statetree

cant wait to deal with ambigious bugs this will produce because of &Context

[celest python]

i guess at some extent instance data type also gets to manage things like this to avoid const limit on core functions

[terse panther]

hey there, any idea how can i rotate the arc direction of the EQS, i tried which multiple vector values for the direction context by making custom querycontext , but none of them are working. I want to generate the donut in semi circle behind the AI, like in the image its generating in the front, but i want it at the back.

[celest python]

@harsh storm just fyi - i was looking at source code.. which seems to be not updated properly for 5.1

we dont need that boilerplate I've shown

its easy as creating two structs and declaring data properties on one and override relevant functions on another to create tasks or conditions

[harsh storm]

Yeah - I saw that siliex set you straight 😅

(Why is he so darn good 😭)

[celest python]

hehe, he also spent at least a few hours 😄

there was a small phase he was asking "wtf" questions about ST at #cpp and twitter

[harsh storm]

Are you actually trying to use STs for your zombies?

[celest python]

yep

like, for anything

game director, zombies, boss fights, boss fight managers

[harsh storm]

Alright - cool. I like BTs, A LOT - but screw it. We're making zombie games together, but separately. I'll switch to using it just 'cause 😅

Are you doing #multiplayer? @celest python

[celest python]

yes

[harsh storm]

🤩

[celest python]

[harsh storm]

Though - I think where we differ is, you're doing #gameplay-ability-system if I recall, where as, I'm not.

[celest python]

yeah, though i use gas in a weird way.. mostly to benefit from it's replication features lazily

the way how UE evolved in a year caused me a lot of trouble.. i went through ~7 heavy refactorings because of my learning progress and new features

like for now state tree literally makes #gameplay-ability-system redundant for me

[lethal helm]

how come the left task isn't executing?

I've set the decorator to abort lower priority and the decorator is passing. I'm watching the breakpoints on the task and it isn't even trying to execute so it's not like it's just instantly failing

[harsh storm]

@celest python Can you link me what you read for your ticketing system? It sounds kind of interesting for knowledge bank. Might use it for my game as well, though I also might not even need it.

[median socket]

I think the PatrolTarget condition is what needs to abort in order for execution to pull back

Hence the "abort" -- it only aborts what is relevant and running, it doesn't monitor all conditions on idle branches; just pulls back when a relevant decorator is no longer valid

I could be wrong though

[lethal helm]

that doesn't sound right

that would make behavior trees worthless

[median socket]

Hm, maybe you're right

[lethal helm]

the left decorator is set to abort self and abort lower priority and it even highlights the right node as a lower priority abort target

[median socket]

Yeah, that makes sense, sorry

Then I have no clue

What composite are they connected to?

Actually, I guess that doesn't matter if you've monitored it

[median socket]

Are there more nodes to the left with lower priority abort decorators?

[lethal helm]

yes but they aren't firing

[median socket]

I found this, not sure if it's relevant to you

If there are multiple nodes that are all set to abort “Lower Priority” (or “Both”), they will all have to be satisfied before the lower priority tree will be aborted. That way, you won’t have to worry about separate value changes repeatedly aborting your lower priority tasks when the higher priority task can’t actually execute.

[lethal helm]

it's not. that just means if I had three decorators it wouldn't abort the lower priority task until all three pass

[plush cargo]

anyone figured out how to stick to a statetree state yet?

[median socket]

The word "node" rather than "decorator" in that sentence made it unclear to me, but that makes sense

Well, I doubt I can help you; hope you figure it out

[lethal helm]

my debugger appears to be lying

task_followspline is being executed but it's not being executed

end me. it was being pissy because I was using a timer in my task

switching that logic to a one second tick on the task itself fixed the problem

probably some horrible async deamon leaking up the abstraction layers

[slow bobcat]

I have a question about a problem with nav modifiers I haven't noticed before (I'm impressed that I didn't tbh).
Turns out that, if you have a mesh, let say a box, and you add a nav modifier component to it, the collision calculated for it is by doing some short of parent-wrapping-collision. The problem is that, when rotating the box, this happens (image).
Is there a way to make the nav modifier tight to the box so it only cuts from the nav the section intersecting the nav-mesh and the collision?
Maybe something we can override in the engine?
We happen to have lots of assets that are re-used all over the place and this is becoming tedious to fix by level designers.
Tagging @crystal hatch since he probably knows the most about this.

[crystal hatch]

NavModifierComponent is using parent actor's bounding box (I'm simplifying a bit). If you marked your StaticMeshes as nav obstacles it should do what you expect.

[slow bobcat]

Hey! Thanks for taking the time to answer.
Do you mean like this?

our designers where looking for something more like this

[crystal hatch]

you can use "null" area to mimic the component approach.
Yeah, I figured as much. Can you mess around with collision settings, like using ComplexAsSimple and such?
What your designers want is impossible. Think about AI agents wanting to go underneath the thing.

[slow bobcat]

Can you mess around with collision settings, like using ComplexAsSimple and such?
We can definitely try and see what happens.

What your designers want is impossible. Think about AI agents wanting to go underneath the thing.
This is a good point. I was already wondering myself how to solve it and, well... it's solved the way the system works already

Thanks a lot. This gives me enough to give them a reasonable "no can do".
Cheers

I guess I could do some dark magic to calculate the position, rotation and size of a nav modifier volume to get something like this

[crystal hatch]

that's definitely the way to go if you need something like this, but you mentioned thousands of assets/instances, so I strongly suggest figuring out an automated process for this 🙂

[slow bobcat]

yeah, not all cases are problematic but... enough to put time in it.
We get cases like this too

which are currently solved by hand

[crystal hatch]

Note that AI will have issues navigating around that static mesh

[slow bobcat]

oh yeah, that's another story...
Can't wait for ML to solve this shit for me XD

thanks again. Much appreciated.

[indigo rain]

If you want to kill an ai (i.e. shoot the pawn), is it best practice to cast to the pawn and do something at that level or then get it's controller and do something at that level?

Or do you go a further level and do something to the blackboard specifically?

I was just thinking something like play animation, spawn object dead at the location, destroy original actor might be less taxing, but could have issues?
So maybe going into I am dead on the blackboard and make a dead state where it plays I am dead then stays on the ground post animation..... It's been a long minute since I did anything with AI and I'm helping a friend on his jam project.

[crystal hatch]

if it's for a jam I'd say:

1. Stop AI logic (via the controller)
2. unposses the pawn
3. play "die" animation on the pawn.

[indigo rain]

I figured something like that. I'm pretty rusty, but those seem like the right keywords lol.
Is there a better 'best practice' you would use for longer projects? I'd love to know if there's a bog standard way of doing things.

[celest python]

You mean the token system?
Its just the DOOM and Spider-Man GDCs

https://www.youtube.com/watch?v=2KQNpQD8Ayo
https://youtu.be/LxWq65CZBU8?t=1353

Also mentioned in Division 2: https://www.massive.se/blog/inside-massive/gamedev/myths-in-video-game-ai/

[harsh storm]

Schweet. Thanks. Also - make sure you ping me when replying 😅.

[celest python]

Alright 😄

[crystal hatch]

the details depend on specifics of the project. The overall structure is like I said, the details regard what to do with the pawn and/or its skel mesh (if it needs to stay on the ground, for how long, if you want to pool pawns, etc).

[young lintel]

When using navigation invokers, are there any commonly accepted solutions for being able to provide them a target location on the other side of a very large landscape and allowing them to eventually navigate there?

I have some pretty severe limitations with a pre-baked navmesh and I'd love to use invokers but my AI needs to be able to travel long distances between points, I thought about making road splines they can follow in between pathfinding I was just curious if there was some already accepted solution to this type of problem

[uneven cloud]

What are the limitations that navigation streaming doesn't solve for? Invokers are incredibly expensive (far more than dynamic regen) and cause issues like you are seeing.

Typically I don't have them navigate when the player(s) aren't around and simulate instead. That usually includes a grid or spline based simple navigation.

[young lintel]

Navigating across a large landscape, from one side to the other. Assigning an AI character a target location on the other side of the landscape even with a fully generated navmesh often results in them making lots of wrong turns, backtracking, and eventually getting stuck in a place that doesn't connect

I have a design goal of simulating small groups of traders walking from one side of the map to the other, which I could probably fake with some effort but I also need to be able to allow a player to follow them if they choose to

Which doesn't really work if the whole trade caravan navigates into a ravine and gets stuck trying to walk up a cliff

I'm not against using splines for roads that AI could follow for longer distance journeys like that but I wanted to do my due diligence researching that type of solution before I start implementing it

As it is, my goals of a large terrain have been reduced down to 2x2km because of limitations in navmesh complexity

I don't know, I'm new to Unreal so I don't really know what my expectations should be

I'm sure there's probably things I can do to improve the way I'm using the navmesh I'm just not sure what those are

I'm not currently using invokers at all

I was looking at Kythera AI to maybe try to solve thee problems but I'd hate to buy it and discover it doesn't solve these problems, lol

[celest python]

its.. damn expensive probably

I'd say ensure hiring somebody/somebodies for your specific goal wont be cheaper than Kythera AI if you are going to buy it

[young lintel]

Yeah I've asked for some pricing information from them but haven't heard back yet

Hiring people is never really cheaper, if its a contract you might spend less initially but any problems will have you paying them again

It could give you a better solution, perhaps

At least better tailored to your specific application

[sleek hull]

How much ai do you have in your game? is it multiplayer or just singleplayer

[young lintel]

It's singleplayer but every character in the world is AI

[sleek hull]

oh interesting. How many characters are there?

[young lintel]

It's not densely populated, but eventually there will probably be something like a couple of hundred

Probably on the high end

Most of them aren't wandering the earth, they'll be hanging around towns, gathering resources, and fighting

[sleek hull]

ok. what do you mean by limitations though? is it computing limitations and low frame rate or something or you mean the degrees of freedom of the ai?

[young lintel]

The prototype will probably only be a couple of dozen unless I can get the world size larger without breaking everything

Well, the navmesh doesn't even like to generate with good precision on anything larger than a few square KM

And when it does, if I actually set an AI controller to navigate from one side to the other, the pathfinding gets severely confused

[sleek hull]

I'm a unity dev so maybe it's just me but I thought that you can directly control precision. I have an idea though ...

[young lintel]

You can change the tile sizes and cell sizes of the navmesh, but it's a trade off

You can make larger cells at the expense of precision

But then you have AI that can't get closer than like two meters to an obstacle because the navmesh isn't precise enough

[sleek hull]

Why don't you just create an invisible lower poly mesh that has the same topology as the mesh you are trying to create the navmesh for and use that instead?

Like you can take your terrain or something and make it a quarter of the verticies or something

[young lintel]

I mean my terrain is already effectively 2 meter resolution just to try and stretch everything enough

[sleek hull]

oh yh it may not work then. Do you have a picture of the terrain or something?

[young lintel]

I mean, it's pretty basic

Like I said it's just prototype stuff, imagine a 2km x 2km square landscape

With some hills and valleys

[sleek hull]

ok

[young lintel]

village A on one side, village B on the other, if you have AI at village A, and tell them to Move To a location at Village B, they walk in the general direction for a while, until they find an obstacle, then they might turn around when they realize they've followed a bad path, or they might just stop

Or they might walk back and forth until the AI just gives up

[sleek hull]

Is this the default unreal engine navmesh/ai system you are using?

[young lintel]

yeah

[sleek hull]

wow that sucks. I don't think issues like that happen in unity.

can you adjust your geometry to make it easier for them?

or is that too much

[young lintel]

I mean, maybe but the landscape is already more open than I want it to be

[sleek hull]

no i mean you can make a set of specific paths

[young lintel]

I would be fine with defining ideal paths between major points of interest if necessary

[sleek hull]

or you can code out your own ai system

[young lintel]

But they need to be convoluted

It's never just a straight line

I mean, yeah I'm not really talented enough to be writing my own AI solution

[sleek hull]

for example you have groups of empty transforms pointing in the paths you want your ai to take

and your ai can pick them at random and add some noise to it's path

[young lintel]

I could possibly fudge something like that

Hopefully Kythera's licensing isn't insane

I would love to find a more robust solution that would allow for a larger world size, anyway

The long journey is a lot less intimidating if it's only 2 km

[sleek hull]

it looks like it is free for noncommercial use

https://www.kythera.ai/licensing

[young lintel]

Yeah, this will be commercial though, heh

[sleek hull]

but they take royalties for commercial

[young lintel]

I asked them for a trial, anyway

[sleek hull]

that's insane imo

[young lintel]

I mean you're paying Unreal royalties, too

[sleek hull]

royalty seems like a lot for just an ai engine

[young lintel]

And Steam, for that matter

or Epic, if you go that way

[sleek hull]

oh yh i guess but that seems like a lot imo

do you prefer royalty or flat fee

[young lintel]

I prefer being able to make a game rather than not

I mean I could make a little arena shooter without buying middleware but it wouldn't be very interesting

If they can provide a solution that enables me to make the game I have in my head, it might be worth paying them a few percent

We'll see what they ask for

I'd prefer if it was all just open source

But, shit in one hand, and prefer in the other, and see which one fills up faster, as they say

[sleek hull]

yh

can you code at all or do you just use unreal blueprints.

[young lintel]

I mean, I'm capable of coding, I'm just using blueprints for now until I have a better understanding of the engine

I'm a full stack python web developer at my day job, so I'm fine with writing code if I need to

C++ isn't my favorite but I can muddle through

[sleek hull]

oh ok i understand. yh I was just asking because I think the AI you are trying to make doesn't seem that difficult to develop yourself as opposed to some of the other game AI applications I've been researching.

[young lintel]

No, it's probably not

I mean I could probably come up with a solution just in blueprints with some work

I was hoping there was best practices, or common solutions, for these sorts of problems

You'd think with all the hype about the huge terrains you can make in Unreal that the AI would be capable of using them

It's unfortunate

[sleek hull]

yh

How much of a performance hit is there with your AI enabled if any?

[young lintel]

I mean its too early on to know

And I can obviously disable characters if they're far enough from the player, but if the navigation system isn't even capable of generating a usable path across the map, that is kind of a deal breaker

I don't know I think I probably just need to fiddle with navmesh settings, this seems like too serious a limitation to be intended behavior and not just my own bad implementation

[sleek hull]

yh another problem though is that it isn't reliable. even if it were to work there are chances your ai can get stuck once in a while

[young lintel]

I can handle once in a while

I can send monsters after stuck AI

[sleek hull]

ok

[young lintel]

But if none of the AI are capable of not getting stuck it's an issue

I'm sure I could run splines across the map between towns, tag the nodes appropriately, and have AI just find the nearest spline to their destination and follow it

If I put roads in the landscape along those paths, it works out okay, but, I would prefer more flexibility

[sleek hull]

makes sense. ai is probably a big part of your game

[young lintel]

Yeah

If it comes down to it I can reduce the scope a lot by just making the AI not travel across the map, leaving the player characters as the only agents capable of doing it, but, that does detract from the experience I'm going for

trade caravans, roving bandit gangs, migrating monsters

Looks like Mercuna could be another option, they have an indie license for $2k

[sleek hull]

2k a year?

[young lintel]

It looks like a flat $2k

For development budgets under $500k, we offer licences with limited support for £4,000 for Mercuna AI Navigation, or £2,000 for either 3D or Ground Navigation.

Since I'm just looking for ground navigation

[dawn schooner]

Is it possible to get the navigation's mesh "width" from a given nav point?

I know that the navmesh can seek different paths by radius, so I guess that it has some way to query the width of a path.

My use case would be the following: I do pathfinding per battalion, and there is a small number (5-15) of units per battalion, which move in formation. The problem is that the formation positions are unaware of their surroundings, causing units to try to walk through walls if the position overlaps, let's say, a wall or corridor. So being able to tweak the formation per PathPoint would be great.

[young lintel]

Might be better to make each unit do its own pathfinding while attempting to stay at a given offset from the leader

It would look a lot more natural too

[sleek hull]

I think waypoint solution can work as well and you just make sure all the waypoints are a distance aways from wall.

Can you manually edit the navmesh or no?

[young lintel]

You could certainly do some line traces or sphere traces from the path point to make sure there's room on each side at any given point

I'm not sure the navigation system exposes enough of its guts to be able to determine any sort of "width" of a selected path

Well, you could get to it, deep in the bowels of C++

But it's probably non-trivial

[dawn schooner]

It is a simulation for many agents, so individual pathfinding is going to be very costly probably

[dawn schooner]

Yeah, that's not a problem, the game is quite complicated and I expected to dive deep into the nav system

[young lintel]

This guy dug pretty deep into it, https://maladius.com/posts/manual_detour_navmeshes_1/

I don't think you can use his little code snippets but this might give you some hints

Honestly use creative traces and working from that sounds like a more time efficient solution, but I'm not an expert, haha

[dawn schooner]

Oh great, thank you! I've researched quite a lot but didn't find this

[young lintel]

Yeah I stumbled over it earlier today, it's not really a solution to anything but it does give some pretty good insight into how it's all put together

[dawn schooner]

I can query nav points which is very close to what I need, but knowing the exact width would be perfect

bcs then I can adapt the formation in advance, instead of querying frequently the outer formation spots to see if there is navmesh there or not

[young lintel]

Yeah

[dawn schooner]

although I'm thinking that I can do it somewhat

[young lintel]

Wonder if it's as simple as just exposing a variable

[dawn schooner]

yeah, I'll tell you if I find out :D

[young lintel]

😐

I got the evaluation version of Mercuna, but their evaluation plugin package doesn't have Linux binaries

smh

[young lintel]

Welp, Mercuna gets better results than stock unreal navigation but it still can't find a path beyond a certain length

Like... Less than a kilometer

Not really worth paying a couple thousand dollars for if I still have to hack together my own pathfinding solution

[sleek hull]

what do they mean by kilometer? can't you just scale up/down your meshes to deal with that problem. or adjust their own unit system?

[young lintel]

No

They don't say anything about the limitation but I was just testing it on my prototype map and it works great but only up to a certain distance, if you try to find a path beyond a certain length it just doesn't

My characters are already at 0.5 scale in my effort to get around the limitations of unreal nav mesh

I guess I wait for Kythera to send me an evaluation package to see if they do any better

Well now...

[lethal helm]

I'm confused. Couldn't you just reduce the nav precision to be compatible with your distances?

don't exactly need a nav mesh precise to the foot when your dealing with a km of distance

[young lintel]

The AI still needs to be capable of walking through doorways

[lethal helm]

ooooh thats a nasty problem

you want high detail level and long distance?

[young lintel]

Yeah

[lethal helm]

how about two navmeshes?

[young lintel]

Perhaps

That actually might work but I'm not really sure how to switch a character back and forth between them at will

[lethal helm]

c++ fun

[young lintel]

Merkuna might be capable of that, actually

[lethal helm]

ultimately it's just querying a path from the correct nav mesh then following it

the tricky part is figuring out which function you need to override to change that

[young lintel]

lol, yeah

[lethal helm]

the navigation system has alot of abstraction so actually finding shit sucks

[young lintel]

The other parts are tricky too, I'm like, two months into Unreal at this point

The documentation, what little there is, is not good

[lethal helm]

has anyone tried to implement target dependent modifiers for the ai sight sense?

[young lintel]

As an alternative solution to my pathfinding problem I'm thinking of something like, each of my villages has a zone already, what if I add spline components to each one and hand build a spline that describes the road route from each town zone to each other town zone, when an AI is in a town and wants to go to another one, they can find the spline component they need, find the node nearest to their location, then just MoveToLocation each node in the spline...

I guess I could use actual road splines too, two birds with one stone

[lethal helm]

yep thats pretty easy

[young lintel]

It only addresses the need for AI in towns, though, doesn't help any that are outside of towns but that might be alright

[lethal helm]

the ones outside towns should navigate to the nearest road if they want to go to a town

heres a simple spline following task

[young lintel]

That looks great but it's compressed too much to read, haha

[lethal helm]

https://blueprintue.com/blueprint/r12b-g9b/

[young lintel]

I was thinking one-way splines but yeah I guess I could give each spline component a "beginning" and "end" town zone, so the AI can decide whether to ascend or descend the index keys

[lethal helm]

in your case you would want to use "move directly to" instead of moveto

[young lintel]

Well, no I think it will work, the nodes in the splines will be close enough that the normal AI pathfinding should work fine

[lethal helm]

true but the spline in your case already guarantees a successful path

so might as well save some cpu juice

[young lintel]

Yeah... I still want them to be able to be interruptable but still be capable of returning to the same route

[lethal helm]

kick them off the spline when they are interrupted

[young lintel]

I will probably need to build this in a behavior tree

[lethal helm]

well yeah

thats a behavior tree task up there

if (splinevalid) do -> task_followspline

as something lower priority then combat

[young lintel]

Yeah, cool this shouldn't be unpossible

Thanks

[harsh storm]

@celest python Question with ST. How do you get it to mingle with perception? To me, it seems like ST is designed to be on the actor itself, instead of controller. And the perception is ideally on AIController.

[celest python]

i never worked with default perception system before, but i'd go with the way of having perception related things on tasks actually...

because it lets me enable/disable perception utils per task

[harsh storm]

🤔

[celest python]

but its not feasible to refactor

so other way around would be having an evaluator update the perception and evaluator can update schema parameters

[harsh storm]

The perception stuff is a component that goes on controller or actor. So how would you get it working on a task?

Traditionally - the perception stuff would just update the variables in the BB

[celest python]

without knowing the internals of the perception api: making perception component a parameter in schema is actually a good idea.. you can take perception comp as "in" parameter from tasks and evaluate there. though you would probably replace BB with a custom data object

maybe something like UPerceptionMemory

this is also lets you run multiple perception comps btw.. some advanced AIs require multiple ones based on the design -- out of context

[harsh storm]

It's wild to me that you've spent a year on AI and haven't even touched the perception stuff 😅

It's like...one of the coolest parts of the AI toolset imo.

[celest python]

my game is an arcade twin stick shooter

every AI is on the screen

though I wrote a custom perception system.. a lightweight one for a client

i was trying to mock what Alien Isolation did exactly

and how perception interops with BB wasnt my taste (for that specific AI), I wanted a custom structure between systems

[harsh storm]

@celest python Figured out another way. Using gameplay events. When the perception is updated, can send the state tree a gameplay event with certain gameplay tags. Perception.TargetAcquired or w/e. And then use that as transition conditions.

I didn't really understand how it would work by making the perception component as an in param to the schema. 'Nor could I actually figure out how to pass it in BP land.

[harsh storm]

Well - that at least lets me trigger the state change. Not sure how to actually do anything with the payload data though 🤔

Just need to figure out a way to pass the actor that triggered the perception event. I could put it as a bound variable in the context actor, but I'd prefer not.

[celest python]

nice. I was more of thinking an idea like embedding perception into states rather than using it as a top-down system, thats why I came up with an extraordinary idea

[harsh storm]

My smol brain can't wrap my head around that idea. Because it is a component that lives on an actor. Then when one of the senses gets triggered, it broadcasts a delegate, which of course you can subscribe to. So, trying to get that into the task itself - not really sure how.

Can't seem to pass the component reference through the schema either.

I just...need to figure out a way to pass that actor that triggered the perception delegate. I don't like my first idea of binding to an actor variable on the context actor.

[harsh storm]

@crystal hatch I know you said to tag you on the PR itself, but it didn't seem to be working 😅
https://github.com/EpicGames/UnrealEngine/pull/9910

And of course - now it did. Shame.

Sorry for the double pings 😅

[keen crow]

I have a situation when I need to run some sequence under a selector indefinitely while blackboard condition is sattisfied. Which is better in terms of performance: create a simple parallel and put "finish with result in progress" as the main task and run my sequence in background or create a sequence with a loop decorator? The latter would require me to have 2 sequences: one with the blackboard condition and the other with the loop decorator and it looks a bit uglier compared to the first approach

[manic karma]

KISS

[keen crow]

maybe lets start with a coffee?

[celest python]

just like Keep It Simple Stupid method coffee is also proven to increase the productivity

[manic karma]

Keep It Simple Stupid.

Does your game need optimizing right now?

[keen crow]

does KISS imply that I shouldn't ever bother about performance on early stages of development? I was just wondering which way is better if any of them is

[harsh storm]

Do what is simple and works

And definitely don't pre optimize

It's a massive waste of time

[celest python]

BTs execution flow is extremely optimized for you anyway

you only get to care about context of your BT nodes

[harsh storm]

I still need to figure out how to have the State Tree actually do something with the payload you send through an event.

[nimble perch]

Hey guys, there is there a way to make specific actors block specific AI navigation path?

so AI 1 can move through Actor 1, but AI 2 will have to pathfind around it

[boreal tundra]

Hello guys, do you know why my navmesh disappeared from stairs randomly and I didn't touch nothing !?

[plush cargo]

what is the stuff called for whatever makes AI avoid other AI ?

[haughty coral]

@crystal hatch @lyric flint

I wrote down also a short feedback on ST. You probably know about all of these points but I wanted to write it down anyway in case it is at least a little bit helpful. I hope you don't mind.

- Live view of ST execution when debugging. This is not really missing unless something goes wrong and you need to understand what is going wrong.

- Being able to understand ST just by looking at its tree. (?) conditions are a mystery unless you select every single state, expand menus and understand the task and the condition. BTs are very good in this because the flow is defined by the BT architecture, so there is no need to understand transitions. Additionally, BTs display all the info like node names, descriptions, keys, values on the nodes, so you most of the time can understand whole behavior just by looking at the graph. Maybe ST could have some expanded view which is less condensed and displays more details.

- If I understand correctly there is no out of box solution for having event-based execution just like BTs had observer stuff, so transitions checking variables need to execute in some interval. For some stuff it is of course possible to use events.

- It still suffers from exponential growth. If you have a multiple states at the same level, you need to define transitions for all of them and maybe add a transition to the new state from every other state. This is probably too crazy of an idea but I wonder if it would make sense for ST to support BT like Abort behavior for entry conditions of states. So when some entry condition succeeds, it can abort lower priority states. I believe that would help with the exponential growth.

- Setting EQS context is a bit more difficult from ST as there is no single place from where both thing could get data easily. In BTs you usually use Context which gives you something from the blackboard. But this is more shortcoming of EQS than ST.

- It can be unclear which task supports Succeed/Fail and which tasks are running indefinitely. You need to understand it in order to build and understand ST properly.

- BlueprintBase of BT uses events. BlueprintBase of ST uses functions. Because of that it is complicated to use latent nodes or event-graph only stuff. I created my own blueprint base derived from the default one but the task picker is not displaying it. I assume there is a hardcoded list of parent classes in the ST module's details customization.

- When having a bindings access to multiple instances of the same task, it is unclear from which task the binding is. I would like to have just multiple instances of the task named MoveTo but I can't because otherwise I am not able to understand which is which as both show up in bindings just as MoveTo.VariableName. Maybe it would be better as StateName.MoveTo.VariableName? 

---------
On the other side, it feels like it is really well done. The property binding is great, it feels robust and the API is nice. Perfect base for the future development. For now I will probably use BT for the core of my AI and ST maybe for smaller subparts or other stuff which requires general state machine. Missing debugger is probably most critical point together with the readability. I learned it after I got back to it after 2 weeks pause :D

This is the BT I converted and tested it with

[boreal tundra]

Can someone help me with my ai ? It became stupid af all of a sudden

lol

[boreal tundra]

Do you know how to make my AI go around obstacles?

[young lintel]

Your questions are not specific enough for anyone to be able to answer them.

[boreal tundra]

Okay sorry I was waiting for someone to have interest and then explain better.. Anyway, I need my AI to chase me and it does.. now I need him to avoid obstacles and go around them in case there's one since when I go down the stairs faster than him he just stands there menancingly and watches me

[uneven cloud]

Is it a non moving actor? Then use nav areas and filters.

[uneven cloud]

AIs can't go places the navigation isn't built at.

[boreal tundra]

He can.. he's just not supposed to go in a straight line but to find a path to reach me

[uneven cloud]

Is there a path? Do you have "partial path" set to true in the move to node?

[boreal tundra]

he gets stuck like a chicken

[uneven cloud]

The visual logger will help to determine if it's a pathing problem or a character movement problem.

[boreal tundra]

The visual logger spams "can see" " cannot see" continuosly

[uneven cloud]

Turn off all categories that aren't navigation related.

[boreal tundra]

No

actually

he can see me

he just remain in chase

pursuin

[boreal tundra]

so he just get stuck because he doesn't circumnavigates obstacles

it didn't used to before

[uneven cloud]

Navigation is how it navigates around obstacles. There are a number of reasons why it wouldn't be working as you expect, so you first need to figure out exactly what it's doing. The best way to do that is to use the visual logger.

[nimble perch]

is a nav obstacle modifier not really an obstacle?

[boreal tundra]

You mean the apostrophe?

[nimble perch]

Meaning, if the target is there, it can go to it

[uneven cloud]

No that's the gameplay debugger. The visual logger is a different tool: https://docs.unrealengine.com/4.27/en-US/TestingAndOptimization/VisualLogger/

[boreal tundra]

now ? lol

[uneven cloud]

A nav modifier is used to modify the nav mesh, usually by setting the area class. If an AI can navigate through it depends on the area class. If I remember correctly, obstacle area class means they will avoid it, but will navigate into it if they need to.

[uneven cloud]

If you uncheck the categories that don't say navigation, you can walk through the timeline to see how it navigates. It will also draw the paths in the viewport.

[boreal tundra]

ok so I leave only AI Nav or Nav ?

i guess AI Navigation

[uneven cloud]

For navigation, I keep both of them on. I always forget which one does what.

[nimble perch]

i am using a filter and modifier to make an ai avoid a specific area class that my modifiers use

but, this will still go to it if no other option exists

[uneven cloud]

If you don't want it to go into that nav area, you can set the filter to exclude it.

[nimble perch]

oh I see!

[boreal tundra]

Ok

The visual logger says that the killer was in chase

and could see me

as the gameplay debugger said already

[nimble perch]

weird

that still makes the same behavior

[boreal tundra]

which problem did I fix with this thing? That the killer doesn't go around obstacles during chase as I already knew ?

[uneven cloud]

Like I said earlier, you need to look at what the navigation is doing.

[boreal tundra]

Bingo

We are close to the culprit Detective

[uneven cloud]

Looks like the pillar isn't affecting navigation.

[uneven cloud]

Are you setting the filter on the AI controller or the MoveTo?

[boreal tundra]

I've tick "Can affect navigation" to yes but the chicken still stands there

[boreal tundra]

what should i do ?

[uneven cloud]

If you look at the actor in the scene, does that have Can Ever Affect Navigation set to true? What are the collision settings?

[boreal tundra]

You mean the stairs? It's a blueprint and its core element is set to true yes

[uneven cloud]

I mean the pillar.

[boreal tundra]

Yes

Collision Block All

I tried to set Can Character Step to No

but still doesn't work

[boreal tundra]

😩

[ocean wren]

You have to debug it... look at what the log is saying, what the visual logger is saying.. it will likely tell you whats wrong

[boreal tundra]

...

[unborn jungle]

Is it possible to extend the range at which a nav link proxy reaches?

It seems after a fairly short distance, the secondary darker green arrow stops getting created