gameplay-ai

[dawn schooner]

I don't, those actors are rendered through Niagara anyway

but I don't remember if the position updates were multithreaded, lemme check

[celest python]

here's what I did:

• MoveComponentImpl does the trace, returns it result immediately, saves the actual FTransform update request to array
• On last tick group, a singleton loops them in ParalellFor
• After parallel for finished, another GT loop happens to call render dirty

Didn't test, no idea if works. So I only implemented on code level not on practical level

[dawn schooner]

excelent idea, did you observe a significant performance improvement?

ahh well maybe you didn't profile then

xD

was very curious

[celest python]

I didnt even run the project after lol

I was just so bored and had to fill my mind with something

[dawn schooner]

LOL

Well I checked, and my transform updates are NOT multithreaded - mainly because the setup on Mass and the Niagara sync requires a single thread setup

so I didn't bother, performance was very good anyway

but it was significantly faster than the built in transform update

[harsh storm]

Imagine, just for a moment, Epic fixed the transform update 🙏

[harsh storm]

World peace

[dawn schooner]

at least

I hope Epic product holders are reading this

[harsh storm]

Nah, they want to have a critical part of a game be slower than Godot

I've actually heard its more that so many things rely on the transform system to be done the way it currently is done.

I just don't think I've ever heard a good thing about the transform system of UE 😅

[dawn schooner]

it's only fair

[dawn schooner]

can imagine xD

[dawn schooner]

well - it works - that IS a good thing

[celest python]

this is what keeps Epic from being innovative on gameplay framework

when someone finally compete them on their precious ✨ metaverse ✨ fortnite and make their engine publicly available idk what they are going to do

[dawn schooner]

xd

[harsh storm]

I don't know for sure, but I'd be willing to bet that it is cheaper to move objects in Roblox than Unreal. And if so, then this is already done.

[wise sluice]

Thanks!

[maiden cargo]

If this make a 20x increase in speed for moving objects, what's the catch and what does it breaks for it to work?

[misty gale]

Did you share what you did or is that hush hush? 😅

[dawn schooner]

no, I just applied what vblanco says

potentially I mostly copied and pasted vblanco's or megafunk's code snippets

[celest python]

It skips tons of things that a random scene component attached to a AActor doesnt need

Including collisions

[misty gale]

I see

[dawn schooner]

just in case it's useful:

void ARTSUnit::SetLocationAndRotationFromMass_Cheap(const FVector2D& InLocation2D, float InHeight, float InRotation)
{
    SCOPE_CYCLE_COUNTER(STAT_CheapTransform);
    UnitRootComponent->SetComponentToWorld(FTransform(
        FRotator(0, InRotation, 0),
        FVector(InLocation2D, InHeight),
        FVector::OneVector
        ));
    
    UnitRootComponent->UpdateBounds();
}

this one only updates root component, megafunk had one recursive function that updated all components from an actor iirc. I'd probably not make it recursive since I'm not sure if it hurts performance

[maiden cargo]

If the scene component doesn't need it, it's not a catch is it? :p

[harsh storm]

I think MF said the recursive one was like a 4x improvement.

[misty gale]

Guess this is less worth it for me as i dont have components at all

[celest python]

thats the thing

[misty gale]

Sounds like thats the biggest benefit

[celest python]

it does run same expensive logic even when a scene component doesnt need it

[harsh storm]

Everything has components.

[misty gale]

Uobjects?

[harsh storm]

So you have absolutely nothing in the world?

No characters?

[misty gale]

Singular

[harsh storm]

No meshes?

[celest python]

FSceneProxy go brr

[misty gale]

A single ismc

That doesnt move

[harsh storm]

So your game is just a single ismc?

[misty gale]

Well, for moving "characters" , yes

[harsh storm]

What are you even making?

[misty gale]

I might have to add a few more, depending on how VAT resolves

Casual city builder

[dawn schooner]

yeah for sure, I just would use the leanest way unless you need to loop through an unknown number of nested components

[harsh storm]

So, in your game, everything that moves is going to be an ismc?

[misty gale]

that's the plan, yes

i've tried swapping for hism, but it's far slower at the updates...

[rare ruin]

Mass is handling most of the agents as one ISM, what's the issue @harsh storm ?

Also you can turn agent that are close to you to actors

[maiden cargo]

Please spoonfeed me, I don't think I understand then.
if this is such an upgrade for no cost, why isn't it in engine already?

[harsh storm]

I didn't say there was an issue

[rare ruin]

Oh ok my bad, thought you was concerned about it

[harsh storm]

And I know this. And once this happens, you enter into the land of having more components. Because Actors are just a collection of components.

[misty gale]

probably just an uncommon approach

[harsh storm]

Not at all

[dawn schooner]

Niagara ftw

[harsh storm]

Not at all either. I've seen many people go this approach

[misty gale]

Yeah i'd prob benefit from that

[dawn schooner]

for things that move yes

for meshes than don't move ISM/HISM is fine

[misty gale]

wouldnt mass benefit from niagara ?

[dawn schooner]

they are independent systems

[misty gale]

Im at an acceptable limit currently tho, so im not gonna convert my system anytime soon

[misty gale]

yes, but for visual representation, instead of ism

[dawn schooner]

your game will run faster if moving static meshes are rendered through Niagara

but there are caveats

mainly lack of physics

we have to implement a new click-selection system because linetraces won't work on Niagara meshes

[misty gale]

how many actors are you aiming for ?

[dawn schooner]

also with Niagara LODs had to be implemented manually, although they'll change that soon (if not in 5.3, haven't checked)

[dawn schooner]

2k

[misty gale]

wouldn't think you'd need niagara for that,

but it saves cpu time so

[dawn schooner]

goes more brrrrrr

[rare ruin]

I enabled collision on an ISM recently
I got 1000 agents moving with physics on each agent. It took me 2 ms to update all the physics objects

[harsh storm]

The only reason I was pressing more was curiosity more than anything. It's not something I haven't seen plenty of times.

[rare ruin]

That's insanely high but i don't need that many 👀

[dawn schooner]

meanwhile our whole Niagara setup (which also draws minimap, fog of war, etc) takes 0.200ms

[harsh storm]

Goin' for that 2 FPS?

[dawn schooner]

sorry

maths

[harsh storm]

I was like, "I know GB said Niagara is slow, but jeebus" @dawn schooner

[dawn schooner]

LOL imagine that

Fx NO STONKS

[harsh storm]

But like everything is slow for what GB wants 😅

[misty gale]

< 1 mill is not acceptable

[harsh storm]

GB is just on a whole different planet

[misty gale]

I was thrilled when 20k moved lol

I expect the number to drop significantly once vat enters the room tho

[dawn schooner]

in my tests VAT vs no VAT barely made a difference

what did a difference was the quality of the meshes (vertex counts basically)

we really need LODing

[misty gale]

Thats promising

Hism brings lods but

[dawn schooner]

with aggressive LODing I've seen 100k meshes in Niagara

[misty gale]

The update time was like 4x

[celest python]

Imagine you have 10 components attached to each other

First one is the base capsule of ACharacter

other 9 is just visual only components, like niagara VFX, that doesnt require collision

if you move ACharacter, it will call MoveComponentImpl of root component, which is the capsule of ACharacter

it will run a very compherensive, accurate but also expensive logic to do factor collisions and sliding towards sides if you are continously moving to forward while getting blocked by things

then call MoveComponentImpl of its childs recursively

then other 9 visual only components also do the same

and waste CPU

vblanco trick solves the issue for other 9

[misty gale]

Sounds like it should be a bool in scenecomps

[celest python]

while you keep the main logic for only capsule

[maiden cargo]

Yes, I got that part but I wanted to know why it wasn't in the Engine already or with a PR?

[celest python]

There are a lot of things could be in the engine but doesnt, this is just one of them

[dense owl]

The same could be asked about many things, that’s not a valid question around here

[misty gale]

Fortnite didnt need it :p

[celest python]

UE is being engineered towards Fortnite

also the reason why AI sucks in UE

they're so careless that they fund* paid solutions to get better so people can use them instead of AIModule

[wise sluice]

So StateTree and Mass are not made by Epîc 🤔 ?!

[celest python]

Mass is not an AI framework

StateTree is not replacement for AIModule too

[wise sluice]

MassAI sorry 👀

[celest python]

I meant Kythera there https://kythera.ai/

[celest python]

MassAI is just a module hooks StateTree API to Mass

[wise sluice]

Not only

It also include MassNavigation and MassBehavior

To handle SmartObjects and ZoneGraph and any system that you could build around

[harsh storm]

And yet it is still the best out of the other real engine choices.

[celest python]

*publicly available and real

[harsh storm]

That should go without saying

Obviously Frostbite or RE engine don't exist for us

The only engines that matter to us are ones that we can actually use

But make no mistake - I'd drop UE in a heartbeat if I was ever able to use RE engine

[celest python]

Its others being incompetent not that AIModule is arguably good enough tbh.. It just works™️ until it doesnt

but who am I shitpost

[rare ruin]

What parts of the AIModule are bad?

[harsh storm]

Well, Godot will never have an AIModule equivalent. Reduz is firmly against it (as are others)

"AI is so unique to games, we can't offer a generalized solution out of the box"

Not an exact quote mind you - but that's pretty much the sentiment

[celest python]

Majority of it

[dawn schooner]

@celest python that reminds me I should swap the StateTree actor components by another way to setup state trees with our units, that is quicker to tick, for now I just reduced tick frequency

btw I've been working with State Trees for a while and it's quite fine, probably I only use the basic stuff but no bugs so far

also probs I'm doing it wrong but it works so

[celest python]

Fun fact recently one of the ex-epic people at benui's server got puzzled by one of the AIModule's implementations too lol

[harsh storm]

The navigation stuff specifically

[dawn schooner]

lol

[harsh storm]

People like to rag on the AIModule - but I actually have very little issues with it.

Other than maybe how much indirection there seems to be.

[celest python]

did you even ever interact with AIModule though

[harsh storm]

I also don't do that complex of AI to be fair.

[harsh storm]

Define "interact"

[celest python]

Extending

something different than inheriting from BTTask, building BT etc

[harsh storm]

Not really. Haven't had much of a need.

I haven't had to make a custom brain component for example

[celest python]

if this part was also problematic "it at least works" argument would be invalid lol

people rag because they go a bit more deeper through API than that

I almost never read/interact with low level stuff

I am still not comfortable

[harsh storm]

I do think the code itself is a bit hard to follow though. So that part is annoying.

But at the same time, I cut them some slack at least. Because it was one person doing all of the AI for a AAA company

And maybe some of those decisions made sense at the time

Not to mention them admiting that they were much younger and more foolish back then.

[celest python]

I think there was also a junior

Mieszko wasnt alone

zoombapup was saying blackboard made by a junior

[harsh storm]

Either way - its low amount of resources for a high expected output.

[celest python]

I never blame individuals anyway

I saw with my eyes that when individuals do 100x better than how they were when they are at Epic

ofc this depends on team or project but whatever

[harsh storm]

Part of the issue, I believe, is a lot of this stuff was built for a specific project and then tried to be retrofitted to be more generic but just never got done.

On top of...well Epic seemingly never actually just removing an API. Instead just indirection after indirection after indirection, etc...

UE's tech debt is 100% going to be their downfall imo.

[wise sluice]

Also retrocompatibility + Unreal is used by ton of ppl

Is that not easier to do something individually without care about other projects that could rely on it?

[harsh storm]

Oh, 100% they've already said they have to do some of this stuff because of contractual obligations.

[celest python]

when fortnite needs it desperately they're very comfortable with removing whole API though

[wise sluice]

Which one?

[harsh storm]

Because at the end of the day, it is still Epic's engine. And the needs of Epic will always come first.

[celest python]

from what I can see - could be true or not - majority of the things are up to initiative of the teams

[misty gale]

Is it to slow or what?

[celest python]

there isnt a gigabrain council deciding on things and keeping people off from doing things better

[misty gale]

I wanted to swap my ai setup for statetree for visuals and debugging etc but ..

[celest python]

once you modify something you're responsible of maintaining

if something gets broken Epic directly contacts you (this part is true)

so when the initiative doesnt come from fancy people teams dont refactor things for the better

especially since they are busy

and as we all already know fancy people at epic busy with fortnite

[celest python]

Mesh drawing pipeline

4.22

idk what it covers completely, but I know a few big projects affected heavily

they even told at some talk "we change things frequently, be in touch with us if you're a client"

[wise sluice]

I guess you have to do that at one point when you realize you made a misstake

It has been completely refactored not removed right?

[celest python]

I dont know, affected parts are not my area of interest so didnt care to check

but I assume like at least %80 of the API got wiped out and rewritten?

lol actually

they refactored one part of it again

and its awesome this time

FSceneProxy stuff

I was ranting about that thing a lot

its now both faster and easier to extend

[rare ruin]

I mean if that's good now, it's worth the pain no?

[celest python]

of course

this is what we want

[misty wharf]

I hope they make a sprite based fortnite

because my other project needs paper2d to be better

lol

[harsh storm]

Epic has never cared about 2D in the entire history of the company since UE1.

I know - I was there

Pretty sure I've played like every one of Epic's games, lol

[celest python]

I probably wasnt even a sperm when UE1 released

[harsh storm]

I was really into Starcraft when it was released.

[rare ruin]

Starcraft editor?

[harsh storm]

Nah - custom game modes.

[uneven cloud]

Hard disagree. That the UE AI "sucks".

[harsh storm]

https://tenor.com/baZvE.gif

[celest python]

Agree to disagree then 😄

[harsh storm]

I agree with Luthage personally. But I've already stated that.

[misty wharf]

UE's AI stuff has worked pretty well for me tbh

It's obviously not perfect but I really don't want to build my own systems for this from scratch

I have more important things to do :P

[harsh storm]

I recall there was a point where people were complaining about BT all the time and I felt like zomg and I were the only ones who actually enjoyed it, lol

[misty wharf]

lol

[misty gale]

I avoid BT at all costs 😅

[misty wharf]

Yeah I've been pretty happy with BT's. I'm moving from a lot-of-small-nodes approach to a self-contained bigger nodes approach with it for my other project

For certain bits the state management and action selection etc. works better that way

[harsh storm]

Perception and EQS have also helped with doin' stuff easier as well.

[celest python]

There is no framework that all programmers agree its good or bad I think, not to mention usability of things relative based on your scope and habits

[misty wharf]

I'm messing around with StateTree in a smaller project now just to see... it's pretty good but it has a lot of weird little gotchas

And for supposedly being "general purpose" it seems rather difficult to run without being in a state tree component in an actor

[celest python]

One could argue my problems with UE AI is skill issue but then I saw better programmers than me also not enjoying it 😄

[harsh storm]

Maybe another skill issue? 😈

[misty wharf]

I suspect if you know exactly what you want/need then you might be better off writing your own systems

I just want something that I can use as the basis for building stuff on top of so I don't have to deal with the low level stuff because the low level coding inevitably ends up taking so much time that I'd rather use building my game logic

[uneven cloud]

It's complete hubris to say that it sucks. Lots of games are made with it just fine. And the user experience is significantly better than most custom built solutions, including Kythera.

[harsh storm]

And user experience is super important.

I recall you talking about the experience of actually using Kythera - sounded like a big nope for me 😅

[celest python]

Kythera loses by big margin since its web browser based

[misty gale]

Biggest issue with custom stuff

Atleast as solo/indie dev

[celest python]

It also crashes often when it's trying to update things from browser to UE assets

[celest python]

Lots of games are made with it just fine.
There are tons of terrible in-house engines that released just fine games too

Amount of workarounds and lack of UX when developing something is what matters for me

[harsh storm]

Not to mention UE has some pretty cool debugging tools for AI especially!

[celest python]

Yeah visuallogger is pretty good

[wise sluice]

Mass Debugging tools are really awesome

[uneven cloud]

Skill based and the "not built here" attitude that a lot of programmers have. As someone who has worked on proprietary engines and on UE projects with programmers who think they can do it better, they are always wrong.

[celest python]

So no one can do better than UE?

[uneven cloud]

I have yet to see a toolset that focuses on UX to the level of UE.

[celest python]

I respect you disagree with me and I acknowledge your experience since I also refer you to people I discuss when they're stuck on something related with AI on UE but its bold to assume no one did better than UE or cant do better than UE tbh. Just because UE is best around its rivals doesn't mean UE is the best among all

[unborn sphinx]

Cost vs benefit vs time. UE ai systems win.

The time it would take to research another option alone, then negotiate licensing, or write from scratch. Unless you have a pro dev team, doesn’t compute.

[uneven cloud]

I said I haven't seen it. Every single project I've worked on that some engineer replaced a system in UE, because he thought he could do it better ended up with it being worse. Every. Single. Time.

I've worked with a proprietary engine used by a well known studio where there was one guy who could expose data to design. You had to request it and wait 1-2 months.

[celest python]

I remember you mentioned that earlier. And it would be funny if a single person could replace UE AI completely, no idea why anyone thought that'd be good idea and allowed it 😄

[uneven cloud]

I don't like Mass, because the user experience of it is terrible. I had to rewrite smart objects, because the UX is also bad.

UE isn't perfect. Far from it. But if you try to fight it you'll just be putting yourself through a lot of pain.

[dense owl]

MieszkoZ would die 😀

[celest python]

I dont enjoy developing my games with Unreal AI, but I do nevertheless, its not like I have any option. But it sucks for me because I often find myself working around things instead of it could be supported at the first place. Like any other module UE's AI system has a solid base to extend upon, but from my experience compared to others in the engine AIModule is extremely unpleasant to work with. I also generally dislike the design principles of UE's BT, I dont find it encouraging me to be more modular on the front-end side, decouple my logic better etc. - not to mention I dislike reliance to AIController with passion. Turns out Mieszko also doesnt like this idea anymore

You might be finding those more pleasant to work with unlike me but I'm arguably a fresh programmer for AI world with no prior experience on other AI systems. so while being aware of the fact UE is pretty good compared to what we have around the industry, I am not into the idea of cheering upon how AIModule is better than others because obviously its also way more worse than others we dont see

[uneven cloud]

I remember him admitting they didn't work with design on Mass and improving the UX is something they want to do.

[wise sluice]

Would love to know what you did to improve SO 😮

[unborn sphinx]

what are you trying to do that the base systems can't accomplish? why do you find yourself fighting it?

[harsh storm]

"Hey design team. So uhhh, I did a thing. Help?"

[uneven cloud]

I had to fully rewrite it, primarily because it doesn't support multiplayer. A player on the client can't tell if an AI is using an object on the server.

[wise sluice]

Ah right, i was not aware of that, i am focusing on singleplayer game right now

Good to know

Did you see they are extending the logic to include player ? They will probably take that in account

[uneven cloud]

We spent more time trying to find a non intrusive way to make SO work for our needs, than the 2 weeks it took for me to rebuild it.

[misty wharf]

There's a couple of issues I had with SO's as well for my project

I was able to so far do without rewriting it and just hacking stuff on top - it would be easier if it was so locked down

For example if all SO slots are claimed, the smart object effectively no longer exists as far as querying for them is concerned

I couldn't think of a good way of allowing actors to queue for using an SO as a result of that

[harsh storm]

I haven't even looked at the SO system to be honest.

[uneven cloud]

We also needed more filtering for picking a SO and an interaction.

[unborn sphinx]

I was going to guess that that was why you redid it. Do you think just giving GAS abilities to inanimate objects would work for multiplayr smart-like objects?

[misty wharf]

That probably depends on what features you need for it

You could probably emulate the gameplay behavior functionality quite well via using a gameplay ability on it instead

say the ability gets granted to the actor using it, and then activated

[unborn sphinx]

Yeah I am thinking typical use: workstations that can be "worked" at or benches to be sat on etc

[unborn sphinx]

right yeah. Maybe the smart-ish object doesn't need an ASC itself then?

Luthage wat did u do?

[misty wharf]

Unless you need to let the SO execute abilities or have effects applied upon it, I don't see why it would need an ASC

[unborn sphinx]

They will be critical for the simulated economy, they will produce economic outputs

[wise sluice]

Gameplay Tags is not enough?

[unborn sphinx]

so I figured I would need to track attributes, etc

[misty wharf]

Ah, yeah it can make sense if you find the attributes helpful

[harsh storm]

At the basic level, a SO is an object that tells the interacting actor what animation to play really. So at the most basic level, you just store the animation/montage on your SO and then when you interact with it, get that anim to play. Obviously this is a super basic implementation approach.

[misty wharf]

The other minor annoyance with the SO system is that it doesn't seem to support multiple behaviors in a nice way

[unborn sphinx]

Hmm I guess SO are not as complex as I thought they were and I could probably make somethign else work

[misty wharf]

So you basically need to have two smart object definitions if you need two different kinds of behaviors on the same object

[unborn sphinx]

wild, yeah... hmm..

[misty wharf]

In my case, I needed a behavior where an NPC is "browsing" objects on a shelf, and one behavior where the NPC picks up something from the shelf

[unborn sphinx]

that is so brutal that they have to be separate lmao

[spring inlet]

you could use different tags?

[wise sluice]

Ohhhhh i'm happy I fixed an issue with Mass 😄 I'll talk about that in #mass

[misty gale]

So would shelf be one SO, and items another?

[misty wharf]

You could potentially use something for it but then you'd need to make your gameplay behavior somehow determine the action the NPC is trying to take, and I couldn't see any really good way of doing it at the time

I think they added some new functionality into SO's in 5.3 again so maybe there's something to help with it by now

[uneven cloud]

We do give GAS abilities when they use it. However, with the built-in system the status of a SO is not replicated. And can't easily be made to replicate.

[unborn sphinx]

I love the free upgrades every version lmao.

[misty wharf]

In my case the shelf just has two separate smart object definitions on it, one which has the browsing activity tag and the other which has the pick up stuff activity tag

Seems to work ok

[uneven cloud]

Nope. We needed more than just gameplay tags.

[unborn sphinx]

So my guess is ur new version of SO is some actor that replicates its "in-use" flag?

and then grant abilities to users

I have a bit of time to think about this because my first playable will have only combat lmao... but I really really really have to focus on non-combat more because I am falling into that trap of just combat AI and nothing else god damnit

the whole point of this game is to discourage combat (while always providing it as an option so ur choice of non-combat truly means something)

[uneven cloud]

No you don't. But the filtering of which behavior isn't very good.

[misty wharf]

That might have been the issue I was having also :)

I don't have the project open atm but I vaguely recall that it wouldn't let you put two GameplayBehaviorSmartObjectWhatever things into it or something like that

Or some other problem in if you had two of them in there

[uneven cloud]

It's a lot more robust than that. I've built a SO system in several projects now, so it didn't really take me long.

[unborn sphinx]

lmao shouldn't have told me that cause in a few months I'm gonna be in here trying to get this workin and I hope I do not annoy u

[uneven cloud]

A lot of what we did was improve the UX of it. That way designers can use it. So we have things like previewing an animation in the definition.

[unborn sphinx]

nice

[dawn schooner]

ye basically, ticking in actors/components without aggregating it is very expensive

[misty gale]

What a shame

[unborn sphinx]

@royal mist is doing this. is he bein silly? what does that navigation grid point node thingy do?

[royal mist]

ya i still can't figure it out lol, it might be a bad idea. im not honestly sure what the solution is but idk unreal well enough

[unborn sphinx]

what does that node do tho? like

[dense owl]

Did you try navigating to native to see?

[unborn sphinx]

No that is not my screenshot and I am busy working on static mesh assets lmao just asking for him as revenge for throwing me under the bus

[strange jackal]

is it ok idea to make AI logic inside NPC event graph itsrlf just using enums ? 😀
(event tick->switch on enum (ai current state))

something like this, will this work for ai states?
or its a bad idea and i must use BehaviorTree with Blackboard ?

[misty wharf]

Yes that will work

BT's and such are optional. If you don't need them, don't use them

[modest token]

how do i make unreal engine handle a lot of ai characters like vermintide/darktide did?

[daring locust]

I want an enemy to return to it's roaming behaviour if the player dies. Can I completly reset the behavior tree somehow, so it starts from the very beggining?

[misty wharf]

This is a fairly complex topic as there are many different factors that go into it, including character movement and AI logic performance. I don't know how many vermintide and such have in general though

[misty wharf]

Isn't there like a restart function on it or something? I might remember wrong.... in any case you should be able to just set up some kind of decorator into your BT that would abort the tree if the player is dead and move into another branch

[modest token]

there are usually hundreds of enemies like 100-200 maybe more

[misty wharf]

You can probably do that with UE's normal stuff with some optimizations on character movement and such, but there is also #mass although I'm not really familiar with how it works

[minor geyser]

in smart object definition slots, where exactly is it looking for the "user tag"? is it looking on the pawn somewhere?

[misty wharf]

iirc on the pawn yes, through IGameplayTagAssetInterface, or possibly through IAbilitySystemInterface since the ASC also carries tags

[minor geyser]

for example on this slot, i only want pawns which have "1" as a tag to be able to use it, i am just not sure where i put the 1 for the smart object to find 😛

[misty wharf]

Looks like it's actually passed into the smart object system through the filter parameter, f.ex. on the Claim function

or the FindSots function since the Claim function looks like it's deprecated in 5.3

[minor geyser]

yeah i ran into claim being missing already 😛 i only work with blueprints, hopefully i can access these user tags via that and not only through real code

[misty wharf]

Yeah from the looks of it the tags can be passed in as part of the smart object request as well in some cases

It looks to have a property called Filter, which contains a property UserTags

[tiny elbow]

Does anyone have any advice or resource for how to best approach organizing and managing multiple "Classes" of AI that share similar behavior?

For example, would be it be best if each class was it's own blueprint, or just make one blueprint that lets you select an enemy type and it changes the model shown and pulls its data from a Data Asset similarly to how I've some guns be coded.

[ionic beacon]

Can you make a 3d grid with EQS, like a 50x50x50 of points? Can only seem to find out how to do a generic x*y

Tempted to jsut manually generate a xyz grid of points but if EQS can do it id prefer that.

[stark tangle]

how to tell to a behavior tree in a service node immediately drop current task and go to left branch?
in my AiController i have an OnTargetPerceptionUpdated Event, which is set EnemySighted Key on a Blackboard
i have a left branch Attack logic, with Blacboard Condition Decorator EnemySighted and right branch Move-to-Waypoints logic with a Service node EnemySighted
but left Attack branch is run only when right branch is complete
pawns see enemy correctly, but dont drop move logic, they attack only after move is complete

[misty wharf]

It really depends on how you need to customize the logic. If the logic is shared between different looking AIs, then a shared baseclass or data asset based solution as you described could make sense.

[misty wharf]

I don't remember if there's a builtin EQS generator which does this, but it should be fairly easy to implement a custom generator which does it in 3D

[misty wharf]

Services cannot affect control flow. You need to use a decorator which causes an abort, either on the currently active nodes (using "abort self") or on a higher priority node ("abort lower priority")

[misty wharf]

5.1 removed the ability to have per-state evaluators
I was wondering why the docs keep referencing to state-specific evaluators but that didn't seem to exist anywhere :P

[jagged frost]

Could anyone point me in the right direction of making an AI that uses a physics based movement system? Using Addforce and a direction for movement, think Snooker/Pool or Golf. I have tried googling for good ways to do this but can't find one, I was thinking about making a system where it would launch 50 objects invisible to the player and pick the one closest to the goal but not sure if there is a better way.

[misty wharf]

What exactly is the problem you're having with it?

[jagged frost]

It's not really a problem, just struggling to find a good way to do it and wondering if someone could point me in the right direction or show me any good resources for it.

[misty wharf]

Well I'm not quite sure what the physics based movement has to do with it

So it kind of sounds like the issue here is something to do with the way it would move

(eg. the builtin MoveTo nodes don't support physics based movement and such)

because the AI itself (eg. decision making, choosing what to do, and so on) shouldn't really be affected by how it moves

[keen crow]

do BTs have some known problems with services under primary task in Simple Parallel node? For some reason in vislog I observe how my service gets
LogBehaviorTree (Verbose) Search node update[Remove]: BT_SWAT_Defend_Overhaul::Aim at[110] and it's missing later in the next vislog records

[keen crow]

ok nvm I just moved the service to the simple parallel and everything's working ok now

[rough ginkgo]

Looking for some material to help me with 3d ai using nav grids as patrol points, anyone know where I can find something

[little swan]

@rough ginkgo There is an old plugin for UE4 - maybe it can help you out - https://www.unrealengine.com/marketplace/en-US/product/don-s-3d-pathfinding-flying-ai

[rough ginkgo]

ty

[ember agate]

What are the common causes when AIPerception sight does not work?
I have an ai and a character I play to test, Sight was working fine and the ai was detecting(Seeing) my character, But after adding detection by affiliation the ai stopped detecting(seeing) my character

Also the overridden function "GetTeamAttitudeTowards" does not get executed unless I uncheck "Detect Enemies" in the ai sense but "OnTargetPerceptionInfoUpdated" is still not executed

[ember agate]

In case anyone else runs into this, I was following a guide and added IAISightTargetInterface to my character without overriding CanBeSeenFrom

[sand kettle]

Hello! This is gonna be a bit long so bear with me. Would really appreciate some guidance!
I'm trying to make chasing feel more natural for my game so the enemies don't look like a swarm of ants. I made an EQS Query that tries to find a valid visible path close to the player. I also want the NPCs to stay at a distance from eachother so i'm trying to simulate some sort of influence map. I made a query context that gets all AI target MoveTo's, and I use that context in a distance check that filters the points close to the MoveTo's.
Right now it feels kind of janky. Its kinda decent when the enemies just try to chase you in a straight line on a flat surface because once they pick a target point, they try to stay close to it. I don't want them to look like they are strafing much in the chasing branch, but rather go to your general direction without taking too similar of a path
When they have to recalculate and the point becomes invalid they often move a lot and some try to go towards the same point until the query finally takes the current AI influence into account (see 1:50 in video). And some of them just stay still for way too long until they decide where to move (0:15). I think this is because the queries for chasing are run at relatively the same time in a BTService and for the influence to be taken properly into account, I think they have to finish in order. So an individual AI query has to finish before running the next one in some sort of manager.
Any thoughs? What else can I do to improve chasing and manage an influence system?
Video: https://youtu.be/711f2F1h09E?si=VwJiNsHUekk_WZ7t&t=11 (don't mind the snapping, gonna add directional animations soon)
EQS Query: https://gyazo.com/0b47f00f7d83f28da1c380c657c9abbe
Behavior tree: https://gyazo.com/74e1781c90a988b64c77fcdfd6096f9c
Testing pawn: https://gyazo.com/8cd7dca66f73733894e72ef9046b5a37

[keen crow]

Can anyone explain me what RestoreSubtree and StoreSubtree values are in UBTNode::InitializeMemory and UBTNode::Cleanup memory? To be more specific, how/why can it be that node memory is stored and restored upon leaving/reentering the subtree? I though that whenever subtree is left, all BTNodes that were being executed in that subtree are finished, aborted or whatever and hence node memory should not matter anymore and be purged, but right now I observe how in UBTNode::InitializeMemory when the InitType is RestoreSubtree the node memory still has non-zero values from previous executions 🤔

P.S. while writing this question I've had a thought - could it be because of usage SimpleParallel BT node when one or both of the children (both primary node and background node) have Run Behavior node within them? I don't really know how BehaviorTreeComponent works in this case, maybe with simple parallel and 2 BTs it like begins to have two instance stacks and whatnot? 🤔

[dense owl]

Guessing you didn’t implement generic team interface, hence why using affiliation failed

[unborn sphinx]

Off the top of my head, one idea would be to have a instance-synced bool to indicate the player is already being chased. If the bool is set, the other AI can go into a different fake chase sequence? It might give u a tiny bit more control

[reef wharf]

well that was annoying. Fiddled with the navigation project settings then it broke my maps navigation until i realised the recastnavmesh was broken. and it wouldn't be remade when inserting new nav mesh bounds. thought i was going mentally ill. but duplicating the map seemed to have fixed it

[misty wharf]

Worth trying to just delete the recastnavmesh actor and such, which contains the actual generated navigation data

[reef wharf]

yeah thats what i did

[misty wharf]

Interesting that that didn't fix it 🤔

[reef wharf]

but then inserting a new nav mesh bounds wouldn't recreate it for whatever reason... i didnt know normal behaviour for it was to create both the navmehs bounfs and the recast

i had no clue why it didnt work and only found out when one post on unreal forums was saying it was supposed to

funni engine

[misty wharf]

Yeah having the navmesh bounds in there and then generating paths should recreate the recast actor

Sometimes it doesn't trigger the generation though so you may need to do it from the Build menu

[reef wharf]

where do i manually trigger navmesh build?

where is build menu?

[reef wharf]

i do remeber not seeing a navmesh be generated at all..

even tho auto build was on

ok ty

[harsh storm]

Not 100% sure why this fixed my issue, perhaps someone else could help me out. I'm using sight perception and was having an issue where the sight was snapping to the new direction instead of rotating with my AI. This would lead to some jank because visually, the AI would rotate and should see the player, but because the snap may have caused the player to be outside of the Peripheral Vision Half Radius, it doesn't detect.

So I found someone saying that they just overrode the GetControlRotation method on the AIController, and replaced it with return FRotator(0.0f, GetPawn()->GetActorRotation().Yaw, 0.0f); and it would rotate as expected.

I may have missed it, but I didn't see anywhere in the perception system where it was calling that method. It seems to really rely on the GetActorEyesViewPoint method.

My pawn isn't using control desired rotation stuff either.

[misty wharf]

🤔

Maybe it's using the control rotation for it? In that case the control rotation can probably snap directly into the new direction. Not sure but that'd be my guess

That would explain why changing the get control rotation to use the pawn's rotation would fix it at least

[obsidian igloo]

Is it possible via AI perception to sense the actor as a whole instead of just the root component center? a character can be half occluded by a crate or something and the AI will lose sight of the character when he is apparently visible

[misty wharf]

If you implement the IAISightTargetInterface, this allows you to customize how the perception linetraces are performed against the actor, so you can for example trace against different body parts

This is C++ only mind you, so in blueprints you'd need some kind of different solution

[sand kettle]

How do you listen to a blackboard value change in c++?

[wise sluice]

@sand kettle

    /** register observer for blackboard key */
    FDelegateHandle RegisterObserver(FBlackboard::FKey KeyID, const UObject* NotifyOwner, FOnBlackboardChangeNotification ObserverDelegate);

From the Blackboard component ?

[sand kettle]

yeah thats prob what i need ty

[dapper iron]

good day gentlemen, i come to you with a simple question:

is there any way to have a dynamic obstacle affect only ONE of two (or more) navmeshes that live on top of eachother?

currently this dynamic obstacle is causing both navmeshes to recalculate, when it really only needs to affect a single navmesh,.

[misty wharf]

As far as I know in order to choose what affects what, you would need to create a custom subclass of the navmesh bits

I have set it up in one of my projects so that I have two navmeshes where only certain things affect the other one

[uneven cloud]

Don't assume everyone here are men. Look into area classes and filters.

[dapper iron]

grow up, i was clearly only adressing the gentlemen in here, not everyone.

[dapper iron]

this is a good lead, thanks for the input!

[dense owl]

Check #rules

[misty wharf]

This is what I had in my notes for this particular thing, maybe it'll help:

1. Extend ARecastNavMesh, have its constructor set `bUseVirtualGeometryFilteringAndDirtying = true` and override `CreateGeneratorInstance`
2. Extend FRecastNavMeshGenerator, override `ShouldGenerateGeometryForOctreeElement` 
3. Have your custom ShouldGenerate function do checks on the `FNavigationOctreeElement`, it contains the component which is being considered for the navmesh. Can do anything there like check if it's from a certain actor type or such 
4. Have `CreateGeneratorInstance` return an instance of your custom Nav Mesh Generator class 
5. In project settings under navigation system, set up an agent type which uses your custom recast nav mesh instead of the default one

`bUseVirtualGeometryFilteringAndDirtying` needs to be set to `true` or the function on the nav mesh generator won't be called.

[misty wharf]

Pretty sure area classes and nav filters don't let you change which actors affect which navmesh unless I've missed something 🤔

[unborn sphinx]

Are you using orient to motion in the CMC? I experimented with that setting, which can be neat for more natural turning while walking, but it made it so the npc head was not always pointing where it was seeing. Characters would turn around instantly to shoot as where the controller was pointing was not where the head of the npc seemed to be looking.

Probably not relevant to ur issue, but sounds vaguely similar.

[dapper iron]

no you're exactly right. and this is exactly the reason i came for input here :)

[uneven cloud]

No, but they can be set up to have the same effect without major C++ changes.

[dapper iron]

no, what i'm trying to do is optimise, there's no need to recalculate both grids if i only need one grid recalculated

[misty wharf]

Hmm.. I guess maybe if you set certain actors to be of a certain area class 🤔

Oh yeah now I remember what the problem with that was

[uneven cloud]

Congratulations, you get a block.

[misty wharf]

You can't have multiple area classes in one place

[uneven cloud]

That's what filters are for. Filters change how the area class works.

[misty wharf]

Well it doesn't make it possible for one area to have two separate area classes applied to it at the same time

[uneven cloud]

It makes you not need to have separate area classes

[dapper iron]

this is completely irrelevant to the question at hand luthage

[misty wharf]

If the area class is used to indicate the "type" of an area it doesn't really help

I needed to do pathfinding queries to determine when certain objects are placed correctly

[uneven cloud]

Give an example of a use case?

[misty wharf]

Only certain objects were supposed to block this pathfinding query, but I couldn't really do it with area classes because you couldn't have a different area class plus the area class for not allowing the pathing for this particular query

[dapper iron]

this is almost my exact use-case :)

[uneven cloud]

The area class can be changed at runtime.

[dapper iron]

in my case, i'm running a nav query to make sure a destination can be reached before permitting the construction of another object that's a path-blocker

[misty wharf]

Sure you can change it at runtime, but if you need to apply that the area is of a certain type, and you need to apply another class to tell it that certain areas an impassable, now you can't have both

because the impassable area class will now not count as the area type anymore

[misty wharf]

Yeah sounds pretty similar

[uneven cloud]

Area classes use bit flags. You can use bit flags to signify multiple bits.

[misty wharf]

I couldn't figure out how those were supposed to be used

[uneven cloud]

I've only used them when entering, but you get the area flag and do a bit check on it.

[misty wharf]

Hmm

Maybe I'll look into it at some point again

The recast bits aren't too bad to adjust for it but there's been the occasional oddity with it that I've not been able to figure out despite it being implemented more or less the same as the default

[harsh storm]

Yeah, it does. As for the rotation snapping though, you just need to change the rotation rate. So the character's mesh still visibly turns slowly. Just, seems under the hood it snaps.

[unborn sphinx]

Yeah I have that sorted for sure.

[harsh storm]

That's the thing, I didn't find any calls to the control rotation. So it has to be affected elsewhere by it.

[unborn sphinx]

oh I see what you mean

I don't know if I understood ur problem fully. Sorry.

[harsh storm]

@uneven cloud Sorry for the ping, but it's a burning question I have. I have a solution, but not sure why it works. Would appreciate any insight you might have.

[unborn sphinx]

I also misstated what was happening in my example. THe NPC would look where they were moving and the tracers would come from the barrel of the gun, through the backs of their heads behind them

[harsh storm]

Pretty much this:

The single lines are the vision cone. The black circle is the AI. Player is "Play". When the AI would rotate according to the blue line, visually to the player, the AI should see the player. But the vision cone was snapping instead of rotating along with the character. So, in gameplay it would look directly at you. But the cone snapped to the full rotation instead of moving along with it.

And overriding GetControlRotation the way I did fixes this.

[uneven cloud]

I'm not sure why either. I use a socket for the view location/rotation and it works with look at, so it's definitely not using control rotation. If you want you can try to debug it.

[harsh storm]

Slap a socket on the skele and then just override GetActorEyesViewPoint (or w/e its called) and return the socket info pretty much?

[unborn sphinx]

hmm well that seems like an interesting change

[uneven cloud]

Yeah. We put the socket between the eyes, so when just the head moves it moves

[unborn sphinx]

So either overriding the getcontrollerrotation or putting a socket, the result is that the pawn's rotation is used instead of the elusive controller rotation for many things?

[uneven cloud]

Get control rotation is used for a lot of things, so making changes to it might have other repercussions.

[harsh storm]

Yeah - that's why I don't like this solution.

[unborn sphinx]

yeah but at first it almost seems like it would be better... but then again I'm sure there are times it would be worse

I really wonder if my NPC exhibit this behaviour now

at first I had such a wide FOV for them that this wouldn't really present itself. But with a new narrower FOV this might affect them

Yeah the probably surely would have this. The rotation of the pawn via the CMC is slowed to 360 units/unit (what are those units anyway? I assumed degrees/sec). But, in reality that controller is snapping so yeah it will miss anything during the invisible snap

[uneven cloud]

To socket version also works for using the AI controller function for line of sight check.

[unborn sphinx]

So typically the controller provides the sight location but you can replace it with a socket on the actual skeleton?

I know I am repeating myself but I am learning lmao sorry

[uneven cloud]

The controller doesn't provide the location. It's gotten from the pawn. Get pawn viewpoint

[harsh storm]

For perception, it calls GetActorEyesViewPoint in the Update loop if I recall correctly.

[unborn sphinx]

oh ok so what is the benefit of the socket?

[harsh storm]

More control over where the "eyes" are really.

And it would move along with the bone

[unborn sphinx]

ok yeah because in debug views I see that my pawns sight is like above their heads lmao

but maybe that is because it's just a 2d cone, right?

[uneven cloud]

By default it's pawn location + base eye height

[unborn sphinx]

oh wild I see ok so then the rotation is just the basic pawn rotation

not the head's specific rotation

[uneven cloud]

Yes

[harsh storm]

Pretty much, yeah

[unborn sphinx]

perfect thank you

[harsh storm]

So what she is saying is to put a socket on the head and when the head is moving around, the sight perception's cone moves along with it.

[unborn sphinx]

u could basically do a whole gdc talk that boils down to that lmao those things have so much fluff this one tidbit is huge

[harsh storm]

Being a bit more "natural" so to speak

[unborn sphinx]

honestly there is so much valuable tidbit knowledge here someone's gotta at least write it down in point form

and share it..

[harsh storm]

I saw elsewhere that someone used the socket as well. But at the time of me tackling the problem, I didn't want to do that. I can't remember. It was late and I was thinking about Resident Evil anyway.

[uneven cloud]

Lol might have been me. I recommend it often

[harsh storm]

Honestly - I find the AIModule a bit frustrating to read through to be honest. So many weird things to me. Like creating a whole enum class or struct inside of a method.

Can just be annoying to parse for me. I can't speak on if it is quality C++ or not though.

[unborn sphinx]

editing skeletons is scary but now with 5.3 not so much lmao

actually adding sockets was never scary was it

[harsh storm]

Not at all, lol

[unborn sphinx]

right so that is a great solution

[uneven cloud]

Using a socket allows art or design to edit it easily.

[harsh storm]

That too

[unborn sphinx]

yes amazing

sometimes the simplest distinctions have so much thought put into them and work so well it is easy to overlook due to the simplicity

all the complexity of actually coming to the solution is optimized away in the final result of "use a socket" it seems so simple but it is so powerful

[tiny sinew]

Hello, i have an issue with AI

When they are near each other they stop attacking

and they stometime stop moving too ( the one on top )

but their brain is still working

(But it seems that sometime the behavior tree turns inactive)

[dense owl]

Are you using stop logic anywhere?

[tiny sinew]

is it a node? if its a node then i dont use it

[dense owl]

Yeah, fair

[tiny sinew]

My " theory " is the bug come from this (wait

is this correct

( it switch from the 2 BTTask_Attack )

well its switching suickly between the Simple paralel itself AND the bttAttack alone

[dense owl]

Would prly need to see what’s inside your attack task. On a side note, a 0s Wait is kinda pointless, should be a bit higher to serve its purpose

[tiny sinew]

(as it should)

[uneven cloud]

What does your attack task do?

[tiny sinew]

this is my Attack task

this is my chase player

[uneven cloud]

Why are you not using the built-in move to?

[dense owl]

BTs have their own MoveTo nodes for starters (don’t need to use AIMoveTo)

[tiny sinew]

Also, they stop attacking if they are too close from each other ( it seems )

[uneven cloud]

Why are you using a delay in your attack task? Delays are just timers you can't abort.

[tiny sinew]

I dont know, i just follow this tutorial : https://www.youtube.com/watch?v=LL9hpaSLD5g

[tiny sinew]

Because i want it to be sync with the damage+animation

now i get the damage at the exact moment of the animation touch me

[tiny sinew]

what does it change ?

[harsh storm]

Ah - Gorka Games. I'd definitely stay far away from them

[tiny sinew]

they actully help me a lot for what im doing..

[harsh storm]

They write really awful code. They're whole channel is just pumping out extremely low quality content.

[tiny sinew]

i cant be autonomous code yet and i follow tutorial, if you have better than them i take it lol but they have everything, short and good explained

But i trust you

[harsh storm]

Even Ryan Layley's stuff is better

[uneven cloud]

The Ai with Blueprints course on the learning library is better than a random tutorial. Especially ones that give you bad advice, like that one.

[tiny sinew]

i cant know if one tutorial is good or not since i have no knoledge about

[harsh storm]

And Ryan does messy code as well. But understand that tutorials aren't going to teach you good code. The point of a tutorial is to teach a concept.

If the concept is not taught, it's a bad tutorial.

[tiny sinew]

So what should i do 😢

[dense owl]

Good, then take our word for it

[dense owl]

👆

[dense owl]

One node instead of a task with several nodes

[harsh storm]

Definitely pay attention to what Luthage says. She has years and years of industry experience with multiple shipped games. That doesn't mean she is infallible though!

Here is the course she was referring to.
https://dev.epicgames.com/community/learning/courses/67R/unreal-engine-introduction-to-ai-with-blueprints/mX27/introduction-to-ai-with-blueprints

[uneven cloud]

Yet you are using an observer abort without being able to handle when the task gets aborted.

[dense owl]

lol snuck in that backhand eh? 🙃

[tiny sinew]

Actually i just want something like this :
Ai : Patrol in X Area
If AI See Me = Ai chase me
If ai is close to me = attack

[harsh storm]

Not a backhand. More to make sure they don't read it as, "Her word is gospel. Never question it."

[dense owl]

Eh, I’d wager they should until they’ve actually garnered some real exp

But then again the same could be said about Duro 😛

[harsh storm]

I'm not saying ignore her right meow 🤣

[harsh storm]

Oh - I want people to question me about things that I say.

[uneven cloud]

Programming is about making a lot of decisions. Decisions based on experience. There is NEVER a right answer.

[harsh storm]

Heck - Luthage called me out some time ago about GetAllActorsOfClass. Had she never done that - I would'nt have actually went and looked at what it actually does.

[dense owl]

Yeah, I try not to recommend get all actors, unless absolutely necessary

[uneven cloud]

Get all actors of class is just fine.

[harsh storm]

It's actually fine to use depending on the class you're filtering for.

If you only have 4 players - you can do that crap on Tick with zero issues.

It literally just reaches into a bucket that Unreal already tracks for you and gets you that list.

[dense owl]

Right, it just ends up being one of those heavily misused nodes taught by crappy tutorial makers in all the wrong situations

[harsh storm]

You're not actually going through each actor in the world.

[tiny sinew]

I Checked this :https://dev.epicgames.com/community/learning/courses/67R/unreal-engine-introduction-to-ai-with-blueprints/K1X/chasing-the-player but im sure about where it will help me to solve my issue
im actually looking for debug more than create it from scratch( i guess? )

[dense owl]

Yeah I seem to rmbr that convo

[harsh storm]

So yes - question me!

[harsh storm]

Don't try and solve your immediate problem. Learn the system.

Because that is your actual problem

[uneven cloud]

It teaches you the systems without giving you bad ways to do things. You can't debug without understanding the systems

[dense owl]

You can start by removing delays, changing the MoveTo , increasing the wait node, but ultimately you’ll need to understand the system, and likely have to debug using gameplay debugger, vis log etc, all taught in that course iirc

[tiny sinew]

Why should i increase the wait node?

[dense owl]

Cause 0.0 is pointless

[tiny sinew]

i thought its 0,5

[dense owl]

If you have it for fallback, make it a little higher at least

[tiny sinew]

Oh

you right

i did it to test if its bc of this wait node my issue

Also, is my issue can be related to delay?

because i use this delay on purpose

[dense owl]

Likely yes

[harsh storm]

The way I'd do it in BP, is bind to an event dispatcher that says the montage is finished.

Not use a delay at all.

[dense owl]

My guess is your branch fires over and over again rn and keeps reevaluating before the delay is over

[harsh storm]

But you really should just go through the linked course.

[tiny sinew]

without delay it looks like this

they spam attack

im in the linked course but i now sure what i should look about that i dont already know about the basic system

And this courses doesnt say " dont use delay " or things like this

this is why im here to see if people have the answer from previous experience on kinda same issue

and without delay they are also buggy on the animation

[dense owl]

Like I said, your branch keeps firing over and over, the delay was masking the symptom

There’s stuff like cooldown you can use, or even simple wait nodes, but to make it work right you gotta understand the basics

[tiny sinew]

also they are now punching themself

i mean, not themself but the other enemy

[uneven cloud]

Basics like event dispatchers so you can wait for the montage to end

[tiny sinew]

By montage you talk about the animation?

because i even remove the animation and everyting, add a printstring at the end of the attack to see if they actually attack

they do attack

but if i move so they are near each other, they stop attack

but IF i kill 1 of them AND move, he attacks again

So it means the problem is not about attacking or not but about the Why they stop attacking

And it looks like there is a information saying " if there is more than 1 contact in the sphere = stop attack "

But i have to know its from the Behavior tree OR from there BP eventgraph

and i think i should change something from here.. cant tell xhat

[tiny sinew]

I solved my issue by changine the Sphere by a Multisphere

(it seems i solved it)

[tiny sinew]

So now my AI is OR attacking me 4 times by 4 times , OR not attacking if another enemy is in its sphere trace

My issue is here
they collide between them and so they stop attacking me

[dense owl]

If your code tests (i.e. casts) for a specific actor but it finds the other AI first and ignored the other results, it’s likely your key is not being set. Should be easy to debug that last part using the gameplay (AI) debugger

[misty gale]

Anything changed with pathfinding in 5.X ? Regular Async pathfinding now results in Error ..

Cant get much info out of it either

[misty wharf]

I haven't noticed anything

I think I had one navmesh disappear but build paths fixed it

[fickle vault]

Do the AI that use Behaviour Trees need the NavMeshBoundVolume?

[harsh storm]

If you're trying to navigate around, yes. BT is a separate system from navigation though

[misty gale]

Seems like the only thing (i can see) returning error is that it cant find NavData

[misty gale]

Hm gotta be my code , the bp node find path to locatin works

[misty gale]

Works in 5.2.1, not in 5.3

[fathom sun]

Hello, does anyone know how EAIOptionFlag::Type ProjectGoalOnNavigation works in UAITask_MoveTo?

It seems to be breaking what I want to do. All I'm trying to do is tell the AI to move at 0,0,0, and there's some stable ground, however it seems that the destination location change to 0,0,-140 for no reason. I also have tried to pass Disabled for this parameter, still the move request is modified internally resulting into the very same destination location of 0,0,-140.

Apparently it tries to build the pathfinding query, but it results into this weird location. What might be wrong? All I did is create the navmesh volume, and extend it on the whole area. The ground the AI is walking is mostly flat. I don't see any reason to make it try to navigate beneath it

[tropic sundial]

Hi All,

I'm trying to use Smart Objects with State Trees in Blueprints (for now).
Note that there is a solution/c++ project as I have added a C++ file to it. And I'm currently using Rider to compile/run it if any of that makes a difference.

I've got to the point of 'Use'ing the smart object, I can use the item, but the return value (the Smart Object Behaviour Definition) gives me a valid value,
but I don't seem to be able to use it in any way.

LogBlueprintUserMessages: [TASK_UseSmartObject_C_2] GameplayBehaviorSmartObjectBehaviorDefinition_1 is what I get back.

I think I should be able to get the GameplayBehaviorConfig from that, which would then allow me to setup the behaviour?
In the code I can see references to the GameplayBehaviorSmartObjectBehaviorDefinition, I don't see to have it available in the blueprint editor
anywhere. No casts, can't create a subclass, doesn't appear to have GameplayBehaviorConfig available under it/etc.

Does this mean I have something misconfigured? Do I need to add anything on the CPP side?
I have plugins turned on: GameplayBehaviourSmartObjects, SmartObjects, GameplayStateTree.

Thanks in advance for any pointers.

[misty wharf]

Doesn't calling Use Smart Object or whatever it is called trigger the action on the SO?

At least I thought that "Use" was the part of it which would trigger whatever behavior is defined for the SO

Yeah at least AITask_UseGameplayBehaviorSmartObject does whatever behaviors are on the SO

and if you're doing mostly BP this is probably the one you would want to be using

[tropic sundial]

I thought that too. I'm certainly no expert (obviously hehe).

But it looks like if you use UAITask_UseGameplayBehaviorSmartObject, that will do all the bits an pieces, so does the full process.

However, I'm using SmartObjectSubSystem->Use.. Which appears to only do a bunch of checks/locks and then returns the behavior

Can you use an AITask from a state tree task?

[misty wharf]

StartInteraction in the AITask goes from a behavior definition into triggering the behavior but I'm not sure if all those functions are BP exposed

[tropic sundial]

I (possibly mistakenly?) had thought they were for Behaviour trees?

[misty wharf]

I think you can since you can just set the controller of your pawn as the AI task owner

[tropic sundial]

Hmm, yeah it pops up. I wouldn't be able to interrupt it at all or anything like that though I don't think?

Might be a decent fallback option.

I'm still a bit worried that I don't seem to be able to access the classes it seems like I should be able to access.

Makes me think I haven't setup something properly

[misty wharf]

Nah the smart object system is just raw lol

It works but it's clearly still work in progress

For example TriggerBehavior is the function used to trigger the actual smart object behavior and that function isn't exposed to BP's now that I looked at it

I built a bunch of stuff around the SO system in one of my projects so it's certainly usable, but it's not super blueprints friendly unless you use exactly what they're giving you to do so

I made some custom extensions to how the SO's function, and made my own AI Task to use them, but even my own AI task is very similar to how the builtin task for it works, so that's probably the way to go with it

[tropic sundial]

Fair enough. I'm eventually going to convert everything to CPP, but find blueprint a good way of just throwing things against the wall so to speak

Thanks 🙂

[celest python]

Anyone saw zoombapup around lately?

[dense owl]

Died with Auther ig

[wise sluice]

Hey ! Does StateTree are Hierarchical State Machine?
It seems to be one but i'm not sure 🤔

Ah crap, read the documentation first ... nevermind

[tiny sinew]

Hello
I had a problem like sometimes the AI STOP moving but the Behavior Tree was still running
And apparently, link "On Fail " to finish execute fix my problem, is it a normal way to fix that?

[fathom sun]

Does anyone know whether it's possible to generate path towards some location/actor and store it in the PathFollowingComponent, but not immediately move using it? The thing is that I need to check whether the destination is directly reachable, do something, and navigate towards it afterwards

[dense owl]

You should just use the built-in MoveTo task if you don’t need to change the acceptance radius dynamically

A task always needs to finish so if it has no way out it’ll get stuck

The other issue is your move is failing so you’ll need to test your goal location, nav mesh etc to find out why

[tiny sinew]

What does it change to use the built-in MoveTo task?

Actually i think i solve all my problem with the AI

It works as i want but i want to learn more so i have to understand why something is better than another one lol

[dense owl]

Because then you don’t have to make your own task and account for every pin?

[tiny sinew]

i keep in mind
Other than that, does it have something to do about the perf?

[dense owl]

Doubt it

[uneven cloud]

It projects the given point onto the nav mesh. So that the AI can move there.

[fathom sun]

The problem is that the project point is below the floor

[uneven cloud]

You can check if the destination is reachable before trying to actually move there.

[fathom sun]

Wouldn't that generate a path as well?

As I understand, it would generate the path to check whether the thing is reachable, and then, when I'll try to move to, it'll generate yet another one

[uneven cloud]

Is there navigation generated on the floor? From your image it doesn't look like the floor geo supports navigation.

[fathom sun]

It's like that. Pretty much flat

[uneven cloud]

No. Has Path does not generate a path and is cheaper, especially if you use hierarchical mode.

[fathom sun]

But what if I've never called move to beforehand? The path wouldn't be generated, and the path always would be invalid. In that case, I need to somehow find out whether the target is reachable. I was using NavigationSystem->FindPathToActorSynchronously(), but it generates a path totally unrelated to any component

I was wondering whether it's possible to do something similar, but on the AI's path following component, and store it there, so that the next MoveTo would use that (obviously if the destination is the same)

[uneven cloud]

Asking if a destination is reachable is TestPath. I do not understand why would want to get a path and sometime later actually move to it, as it'll cause bugs.

[fathom sun]

Oh, I've found out how I can fix that. In my case I had the ground a bit lower than 0 Z. Every time the nav would try to project the destination, it would go to -140 (which is perhaps the thickness of the ground or something). After I've set the ground to be at 0, the projected destination Z is 60

[fathom sun]

Asking if a destination is reachable is TestPath

TestPath? What do you mean?

I do not understand why would want to get a path and sometime later actually move to it, as it'll cause bugs.

I'm using coroutines to call the move to, and before that I would like to do different things depending on the reachability. So anytime I call MoveTo, the code after that doesn't run until MoveTo finishes

[uneven cloud]

Test Path is a function in the navigation system. You can test reachability by using test path.

Or you can rewrite a large chunk of code to make it do what you want. I recommend not fighting the engine.

[fathom sun]

Do you mean UNavigationSystemV1::TestPathSync()?

[dense owl]

What happens if something blocks your path while you’re busy doing whatever ?

[uneven cloud]

Yes

[fathom sun]

AI is already capable of handling bumping. If it bumps into its alike, it'll switch to another state, cancelling the previous move to, and try to walk around the bumped another AI. If it bumps into player, it'll get some aggro. If it gets stuck somewhere, there are some checks on tick that check its speed, so that it tries to jump some direction to get unstuck. If it even that doesn't help, it'll try to change the destination. If nothing helps for long enough (like 30 seconds), it'll suicide 🙂

The game has a lot of AI characters (more than hundreds), so killing those that are stuck doesn't make much difference

[dense owl]

My point was why save the path at all, and not just check if reachable, do your thing, then path to it and go? Is this a perf cosideration?

[uneven cloud]

TestPath is pretty cheap. I don't recommend doing hundreds of them in a frame, but it's significantly cheaper than finding a path especially if you use hierarchical mode. It's not finding the BEST path it's saying there is A path.

[fathom sun]

Oh, if it's cheap enough, I guess I can go with that

[uneven cloud]

It's actually cheaper to test a path before finding a path, because finding a path when no path exists is actually really expensive.

[fathom sun]

Initially I was wondering whether it's possible to check the reachabilty and do the move to without re-creating the same path twice losing perfomance (and since there are a lot of characters, it would be considerable). But, as Luthage has said, I can use TestPath that will do the thing relatively cheap, so I'll try to go with that. Still, I would like to achieve the initial idea. If TestPath will turn out to not take that much performance, I'll just end up using that

Which PathFindingMode should I be using in the TestPath? Among Regular and Hierarchical the first one seems to be cheaper

[uneven cloud]

Hierarchical is cheaper

[fathom sun]

Oh, all right, thanks

[uneven cloud]

You're welcome.

[celest python]

@potent iris pin please?

[tranquil ice]

what can cause a delay when I spawn a third enemy which uses behaviour tree? the glitch is when it's spawned it chases me not right when it was spawned but 1 sec later, at the mean time two first spawned enemies run towards me as they were spawned.

and after those two, each enemy has a delay when is spawned and only right after that it starts to chase

the sight perception causes the delay I don't' know why 😦

[thick gorge]

I made a BT that has characters pick random nodes, walk to them, idle, then walk to another. It works on the characters I have placed in the level, but when I spawn the same character at runtime, he just stands in the spawn spot idling.

Is there some step needed to initialize/activate a BT at spawn?

[harsh storm]

How are you doing it with the ones placed in the world?

At some point you're telling them to run the BT - where is that code?

[thick gorge]

Thanks Duroxxigar, looking for that now.

[harsh storm]

It's most likely the same class, no?

Just how are you running the BT?

Is it on possession?

Make sure the character gets auto possessed on spawn as well. It's a setting in the character.

Near where you select the controller class

[thick gorge]

Got it. I am making progress, I found where the BT is being run.

[harsh storm]

And where is that?

[thick gorge]

Looks like a separate BP. I put this together using a tutorial without understanding how all the parts work.

I'm adding some Print Strings to figure out which parts are executing when. But I do see the Run Behavior Tree node, so I understand that is key to making sure that is happening on placed, and spawned characters. Thanks!

[thick gorge]

Hmmm, I'm seeing a setting that might be the issue Pawn > Auto Possess AI "Placed in World" vs "Spawned"

@harsh storm That's the issue I was having. Pawn > Auto Posses AI was set to "Placed in World." Changing it to "Placed in World or Spawned" fixed it. Wow, such a literal option. UE5 has so many options for built in solutions. 😅 (only took an hour to find)

[harsh storm]

#gameplay-ai message 😅

[thick gorge]

Thanks a bunch Duroxxigar!

[thick gorge]

Here's my test of a crowd of Ninjas fighting uphill to bombs being tossed at them (music is incidental): https://www.youtube.com/watch?v=k2MG3C6n2AU

[fickle vault]

If BT doesn't need NavMeshBoundVolume, then why does AI Move To need it?

[keen crow]

BTs are just a form of ordering AI actions, AI Move To is an implementation of AI moving in the world and this implementation relies on navmesh

[fickle vault]

Getting confused here. So does that mean i would still need the navmessh for BT as well?

[keen crow]

you don't need navmesh for BTs, BT doesn't require it, but navmesh is the only built-in way to make your NPCs actually walk around, so you need navmesh for their navigation and the only built-in way in BTs to command NPCs to go somewhere is MoveTo task which, inside of it, uses AI Move To AI task (not BT task) and that one uses navmesh

is there anything stopping you from just placing a navmesh bounds volume on the map?

[wary ivy]

This video contains content from UMG, who has blocked it from display on this website or application```

[fickle vault]

laziness mostly

So basically, if I use BT to get AI to move, I won't need navmesh. But if I use AI to Move, i'll need navmesh?

[ivory rune]

You need navmesh anyway

If you wanna AI to move to some place by path finding.

[keen crow]

if I use BT to get AI to move,
as long as you don't use the only built-in BT node (MoveTo) to make NPCs move - yeah, you're fine without it. like you could make your own 3d navigation based on SVO, that'd be easier than putting navmesh on the map

[fickle vault]

SVO?

[keen crow]

sparse voxel octree

[fickle vault]

ah

[keen crow]

not goida

[fickle vault]

In the end, navmesh is a hassle

but it is also the easiest option atm

The thing is, i have this minion ability and I want it to be able to move on its own

best if no need navmesh

[young plank]

Hi, I'm wondering what's the best way to implement an event-based decorator. In broad strokes, the algorithm looks something like this:

1. Start listening on the external event on OnBecomeRelevant call.
2. Once external event has fired, store EventPending flag inside NodeMemory
3. Return state of EventPending flag inside CalculateRawConditionValue
   The main issue is there no good place to reset EventPending back to false. We have OnNodeActivation callback, but it fires only if all other conditions for node activation are met (e.g. some other decorators on this node might evaluate to false and OnNodeActivation would not be called). But without this callback it's impossible to know when to reset the flag. Technically it's possible to use CalculateRawCondition to reset the flag, but in case of successful node activation, this method is called multiple times and the return value has to stay true in order for activation to finish succefully.
   Has anyone managed to find a good way to implement something like this? I've studied UBTDecorator_GameplayTagQuery from the engine's source, but it's not exactly what I want.

[misty wharf]

BT's aren't great for events. I have something like this in mine but the value it gets is a bool in the blackboard and it has a separate task which resets it - maybe it's possible to do within a decorator itself since you can do it this way also

fwiw, StateTrees have support for events out of the box, but they do have a bit of a learning curve and gotchas due to nearly zero useful documentation beyond the very basics, and some odd bugs here and there (which you can work around, but in any case)

[young plank]

Well then I'm not sure how to properly structre somewhat complex behaviours. For example, If I want an actor to react to some instantaneous transient events in the world (e.g. grenade just exploded, player started an attack) I need a way for an agent to react to those things. You can't poll everything.
Problem with resetting the event from outside is you have this brief period of time where the decorator will keep returning true even though the event has been already processed. It can lead to some nasty bugs.

[lyric flint]

What are good game genres to use behaviour trees for? I feel like with most indie game scopes, it's overkill compared to HFSMs. Cause in an indie shooter or a zeldalike you only have like what, 2-3 states per enemy with clear patterns.

I just need to work with BTs more so I can put a game using it in my portfolio.

[twilit nova]

Hey, I have an issue with some of my agents, I think it has something to do with StateTree, sometimes they get "stuck" if you see what I mean.
If anyone knows why, I'd appreciate the help ! Thanks

[keen crow]

What's the best way to implement NPC going to some location X while keeping it's distance from it's target (player)? So far I can think of 2 approaches

1. After I've got X, do a series of MoveTo with small paths. In BT, under some selector, have a looped sequence that runs an EQS that finds a point that is closer to X and at least N meters away from P until some decorator says that NPC is at acceptable distance to X
2. Somehow utilize recast navmesh (perhaps override something like FRecastQueryFilter::getVirtualCost) and/or any other related entities (put navmesh modifier volume on player, consider player in crowd manager, honestly idk)

[wise sluice]

I would suggest to continue this discussion in #mass

[twilit nova]

Yeah, I already ask there but got no answer :/

[wise sluice]

Ah sorry yes! I encountered the same issue during my test but i'm not sure how i solved it

At first it was because, i had an Idle state and the tree was stuck inside it since there was no MassSignal to refresh the StateTree
But here you got an AI that doesn't finish its path

[sudden citrus]

What's a reasonable way of preventing navmesh generation on top of static meshes? Say trees etc.

Or do we just let it happen and check for islands accordingly?

[sudden citrus]

Basically just generate an intermediate location to the side of the player, first go there, then to the target. It will look good to the player and that's all that matters.

I wouldn't worry too much about making it a nice curve because the environment will have a bigger impact on the path than your math

[wise sluice]

I just checked the setup that I made i got the same issue @twilit nova
Some agents are just "stuck" at distance 0
The only way to solve that is to dig into the MassNavigation logic and add some kind of acceptance radius or just debug it?

[twilit nova]

Yes it happens at the end of paths but also (sometimes) at ZoneShape intersections (polygon shape type)

[twilit nova]

Well at least I'm not the only one 😅 Thank you

Do you know what those numbers mean ?

[wise sluice]

Hmmm i think it is the ID of the State
Let me check

That's how much the state tree changed since start

[twilit nova]

Ok, I see, thanks. So they really are stuck but the question is why

[wise sluice]

If you can dig into c++ check the MassNavigationProcessor

[twilit nova]

Sadly I'm not very familiar with c++

[wise sluice]

More exactly
UMassSteerToMoveTargetProcessor

[dense owl]

That’s a #mass thing no? 🙂

[twilit nova]

I asked here also because I was pretty sure that it had something to do with StateTree and not just Mass but yeah, the discussion evolved into Mass 🙂

[dense owl]

Not saying you shouldn’t have, only that they’ll be able to provide more support there if you’re dealing with mass

[sand kettle]

I'm watching this SquareEnix AI presentation https://youtu.be/BV2GTGbSjq8?si=w3nSNJqtGJUgEVwx&t=2556 . At 42:35, how come their agents don't move after they reached their destination? I tried their exact same BT and EQS setup and mine always keep moving around because obv eqs keeps recalculating the destination. Could it be some logic in their custom move to node that prevents this?

[harsh storm]

I wish I could understand Japanese - they give some really neat talks 😅

[sand kettle]

I don't understand it either but it still really helpful since they screenshot/visualise all their queries 😅

This is my result 😦 https://gyazo.com/5b7811b146a4333d8ff81349b95886b8

[sand kettle]

I tried to achieve something similar yesterday before watching this video. Felt really janky. But I kinda got them to stop moving by completely filtering points near other targets, limiting donut angle to 90 and prefering lower distances from querier to item. They dont do that at all in their video so idk how they dont keep moving around after finding a good spot

[uneven cloud]

What you need to do is really understand how BTs work and when they should be used vs other architectures. Since you are talking about a portfolio, which usually means looking for a job.

[misty gale]

What im gonna do with my custom ai task setup is shortcircuit my current task with a really low (good) score, and mostly only let it invalidate itself if the target goes to far away from it .. not sure if saying that gives you any ideas but..

When a spot is first picked it's gonna be superior to other spots unless something drastic happens

[uneven cloud]

You can just use the EQS and as it's moving use a service to update the location. You are overthinking things.

[uneven cloud]

Nav mesh modifiers.

[dense owl]

Realistically if your tree bp has can ever affect nav on, that shouldn’t even happen

[sudden citrus]

The base - sure

but someone asked me what to do if there's navmesh on leaves high up

like imagine you have a cube, and the cube produces a hole, but then has a poly on top.

my experience so far is that generally you should be more clever about querying the nav, because you will inevitably have silly islands

curious if there's a sane way of not generating silly islands where you can help it

[brittle socket]

I'm doing an optimisation pass on my plugin and I've realised we're losing about 3ms a frame to a small number of checks to get the closest point on the navmesh.

I'm not sure exactly what I'm looking for, but is there some knack to getting a nav location or check whether there's a valid path to a point without tanking performance (or is it known to be very expensive)?

I can imagine what calculations I'd need to do to get the same result from say a triangle mesh, and I just can't see why it'd be so expensive, unless it's doing a load of other things under the hood that I don't need it to do.

[harsh storm]

#gameplay-ai message

#gameplay-ai message

Just recently pinned too

[uneven cloud]

We can't help you if you don't say how you are getting the closest point on the nav mesh

[dense owl]

So your leaves are not part of your tree mesh? 😀

[brittle socket]

Project Point to Navigation to get the point

GetPathLength to check whether there's a navigable path

[uneven cloud]

Get path length is your problem. That's even more expensive than just getting a path. And completely unnecessary. To test if there is a path, you should be using Test Path with the hierarchical mode.

[sudden citrus]

not my project 🙂 I assume the rest of the tree just doesn't have upward-facing polys

pls refer to the cube example

[dense owl]

If that poly is its own separate actor, then it will have the same option. Otherwise, just do what Luthage said and use modifiers

[sudden citrus]

you can't attach a modifer to a static mesh, and we're talking about huge levels

Proc gen trees etc. You could say use proc gen to place nav modifiers... This isn't generally something I would be concerned with but I'm making sure my reasoning is solid

So tl;dr is there are tools in the engine for this, and they're not "do stuff to your static mesh to magically not be navigable", right?

[sand kettle]

Yeah...I need to find some good conditions for when the EQS for one agent has to be recalculated. Right now i'm thinking of recalculating it when the player either moved too far from the position where he was when EQS happened or when there is no direct path/los to the player

[sudden citrus]

To add to this: Look at BTDecorator_DoesPathExist and see how it's doing a hierarchical check. There is no hierarchical path search in the engine but there is a hierarchical reachability check

[dense owl]

Seems a pretty dismissive approach for a reasonable suggestion to use OOP to give every single instance of a mesh the “magical” ability to not have navmesh generated on top of it.

[brittle socket]

Ah thanks, I didn't spot TestPath, I'll switch to that.

The ProjectPointToNavigation call is still noticeable despite the search volume only being very small (100x100x200) and spanning at most 4 tiles (containing maybe 15/20 nav triangles). I guess that's the cheapest way to get the closest navigable location to a given world location though?

[misty gale]

Yesh thats sorta the same as im planning

[sudden citrus]

Okay, but I'm asking about that approach, and I'm being told here to do something else...

[dense owl]

Who said to do something else? You were given two different solutions, pick one

Figured I’d give you another way to skin the cat, I would never say don’t do what Luthage said if that’s what you thought

She empirically knows better

[sudden citrus]

1 is to separate the actor, 2 is to use modifiers, neither are that. These are plain SMs placed in the level

My question is basically if anyone has done the "don't let SMs generate nav on top" solution, was it neat enough to implement, do they recommend it

[misty gale]

I remember seeing a trick for a case that sounds like this

It was some hollow cube with an "internal cliff" so to speak where the navmesh would generate in the middle

[sudden citrus]

There's a setting in the engine to fill collision underneath a mesh to prevent that, the problem I was asked about is generating nav on top of something that shouldn't have any

[misty gale]

Perhaps that was just nav modifier really

Plain sm's is also their own actors(sm actors) btw

Generally you have a test if a point is reachable or not, which should exlude all islands (unless nav linked )

[uneven cloud]

Yes it is the cheapest.

[sudden citrus]

Yeah that's my approach. Deal with the data at runtime, don't try to guarantee it's spotless. Just sanity checking the other way here, maybe there is one

making trees BPs is probably it if it doesn't carry a huge overhead over StaticMeshActors.

[misty gale]

It sounds a tad ludacris to me but what do i know

Not that i care about a cost, its probably very very cheap(free?)

Id expect it to only have a tiny cost during nav mesh generation

[dense owl]

Ludicrous is the actual word, Ludacris is the singer/actor that used that name as a play on words 🙂

Luda! 😀

[misty gale]

Sry, my fat phone fingers and spelling is off today(too)

[dense owl]

It didn’t really matter it was just funny hehe

[misty gale]

Sometimes the joke is in the misspelling of the thing 😆 (row row)

Like automagic

[dense owl]

Accurate

[sand kettle]

Ok I think I got it to work pretty well, the query screenshot from the video is wrong. It should be prefer lesser for tests 1 and 2

[misty gale]

Nice

[lyric flint]

That's the thing, every AAA game I looked at so far was using HFSM, from DOOM to Final Fantasy to Monster Hunter. So far BT seems useful for shooters and stealth games where you don't have cool behaviour patterns like in a boss fight and where you instead want fluid realistic human-like behaviour.

[uneven cloud]

Lots of games use BTs. All but 1 AAA/AA/Indie game I worked on used BTs. Many also use hybrids. There's a good GDC talk that is pinned in the channel about how we select architectures. I highly recommend watching it. There's a lot of considerations that aren't just what the behavior is.

[lyric flint]

Tbe arborist one or the flame retardant one?

[deft sedge]

ever run into issues with this? This is the last message I'm getting before engine crash atm

[uneven cloud]

Flame retardant one

[sand kettle]

What happens when you attach multiple generators to the query root? Does it have any use? Seems like nodes on the right never get executed anyway

[harsh storm]

I believe the 2nd will only run if the first doesn't meet the criteria or something like that

I think it explains it in a tooltip or the docs on EQS

I've just never done it before

[ember river]

Quick question, I duplicated some AI controllers and blackboard. The AIController I want to use seems to still be linked with the old blackboard. How does one change that?

[harsh storm]

Blackboards are tied to a BT asset. Unless you have a variable for a blackboard in your AIController that you made.

[ember river]

Indeed. I have a behavior tree that is linked to the wrong blackboard more like

Question really is, how to I set the behavior tree to the right blackboard or vice versa

Is it even possible or should I just create a new one with them properly linked

[harsh storm]

In the BT editor, at the top, you can switch to the BB editor. Then you should be able to switch it there.

[ember river]

Aye, that's where the problem is, when I switch to the blackboard, it's a different blackboard

nvm, figured it out. Had to click on Root and change it

Thanks for answering

[harsh storm]

Yeah, I don't change it often. Forgot exactly where you changed it at. Just knew it was in there somewhere.

[ember river]

Dude, I haven't did anything AI in almost 2 years haha

I feel you. I learn stuff, then, move onto a different process in the project. All that old info is semi-lost

[harsh storm]

Anyone happen to know where the debug drawing stuff is for the perception debug visualizations?

[uneven cloud]

Gameplay Debugger Category Perception

[cosmic gull]

I can't seem to make my AI actor follow me vertically. It follows me around just fine near the ground but when i go up in the air too much it seems to "lose sight" and just stop moving towards me. Both of us are using flyer movers and im using the AI MoveTo function with Target Actor set to my character. Any ideas what might be wrong?

[dense owl]

Your AI can’t nav off the mesh

[cosmic gull]

Oh ok ty!

[bitter wyvern]

Hey so in editor, moving my navmesh works fine at runtime and it updates and everything. when I make a build and play it, the navmesh does not update ever. Anybody know anything about this? for reference the video is what is happening in editor. I change the position of the navmesh and it updates fine. when I play the packaged game build, the navmesh never moves.

[uneven cloud]

You should not be moving the nav mesh. You should move the bounds volume and then call On Bounds Updated in the navigation system.

[bitter wyvern]

Sorry that is what i am doing. It seems that is not working in my build however

[uneven cloud]

And you have it set to dynamic regen?

[bitter wyvern]

Yes

[uneven cloud]

Single player?

[bitter wyvern]

No it is local co-op,

[uneven cloud]

Are you moving it and calling On Bounds Updated on the server?

[bitter wyvern]

I don’t think I have a server, it’s local co-op. I’ll go look where I call it when i get home

[uneven cloud]

It depends on how you set it up. Sometimes it can be set up as one of the players is the server. That was my last idea. I'd put in some log statements and do a packaged development build to debug where it's failing.

You can also look to see if it's working in a standalone play (when it opens another window), because it's easier to debug than a packaged build

[bitter wyvern]

Ok thank you for these suggestions i will try them and lyk how it goes!

[uneven cloud]

You're welcome!

[bitter wyvern]

I am calling this from a BP_Actor that i place in the scene. Also it works fine in Standalone window. I will try a debug build next.

ignore the extra setactorlocation lol

I do assign both the navmesh and recast navmesh in the details panel here, could that have something to do with it? kinda new to unreal idk if thats best practice

[wraith veldt]

Hey all, I've got an issue with a misbehaving navmesh. First a little background info: I have a procedural map generator that works by sticking rooms together. Each room is a Dynamic Level Stream that contains a Nav Mesh Bounds Volume, with generation set to runtime so that the navmeshes join together once their rooms stream in. This part works fine.

Enter the problem (image one), and the likely cause (image two). The problem room is always the start room, which is located at the world origin. The room in image two has a set of brush stairs that line up perfectly with the ghost-ramps, and the ramps only appear in maps that include the stairs room.

Updating the navmesh by opening or closing doors, adding new meshes, etc. doesn't do anything to make the ramps disappear. I'm guessing the issue has something to do with Dynamic Level Streaming leaving some kind of collision artefact but I wanted to see if anyone here's had a similar issue with brushes / streaming or has any idea how to fix it.

[bitter wyvern]

it does not work in the debug build. :/ let me know if u have any other ideas on why it may not be working. I am gonna try to figure out how to log statements and view them from a debug build.

[bitter wyvern]

I have gotten it to break in editor now, by just trying to print out the return value from set actor location on the navmesh. is there a better way to change the location of the navmesh bounds and then update it?

[uneven cloud]

Where are you setting the reference to the nav bounds volume? You also should always do an Is Valid check on any reference.

[bitter wyvern]

in the details panel, ill add an isvalid check

[uneven cloud]

You really can't do that. You need to get it at runtime.

[bitter wyvern]

okay thank you im sure that is my problem. damn unity. is there a node that does this?

[uneven cloud]

You can use get actor of class, since there should only be one.

[bitter wyvern]

Hey ho my problem has been solved. Thank you so much for you help kind human!

[uneven cloud]

You're welcome!

[gleaming star]

yo what's good guys, i am struglling with some behavior tree stuff where my AI seems to get "stuck" on one of the tasks and i have no idea why...

disclaimer: this is multiplayer game

so

• i have an enemy AI that roams, has AI perception, and on sight will attack, and on hearing will investigate

i want to have it so that when i attack the enemy from anywhere, the enemy immediately targets the player , but it seems that the enemy gets "stuck" on the "moveto" node, but it doesn't really make sense, as i am (to my knowledge) setting this variable of the target correctly.

i'd be happy to share any and all of the current logic if anyone may be able to assist.

i am using an AI controller and a BP for the main base enemy and then using a behavior tree and blackboard all together. it is kinda complex and confusing even to me right now but i still do think everything should be working.

the target reference is being set on "event any damage" on the enemy BP and the variable is being sent into the AI controller to change the state of the enemy to "attacking" but it seems that the "target" isn't being taken or it gets "lost" somehow.

the enemy on any damage will sometimes go towards the player that attacked it but most of the time it will randomly move to a spot and then "reset" or get "stuck" doing pretty much doing and i have no idea why, but in the Debug it is stuck on the "move to " task.

would anyone be able to assist me in finding a potential solution to this?

thx in adnvace. im about to pass out, so feel free to dm me, or write me here, i'll be fresh by the morning. ty in advance

[coral mesa]

im using floating pawn component why is it that while following player, they tend to go out of map and cant go back to navmesh?

[harsh storm]

Btw - just yesterday I did a quick measurement of this. Perf should be fine (didn't measure memory complexity though!). That is 300k actors in the world and it still took a fraction of a second. (@uneven cloud Just a ping so you can see this as well, in case you wanted something to point to)

[raven nacelle]

Tell me, is it possible in Unreal to configure the calculation of the navmesh path relative to the size of the object?

Is it possible to support multiple Recasts simultaneously out of the box?

[sand kettle]

some progress https://streamable.com/9o1b6y 🙂

[dense owl]

Use gameplay debugger and the Visual Logger to find out what actually happens

[gleaming star]

The enemy get stuck on “move to”

[dense owl]

You’ve said that. Did you actually do what I said?

[celest python]

How deep was the class hierachy?

I think rather than actor count the hierachy depth matter more

[gleaming star]

Lol yes how else would I know that

[dense owl]

Really? Show a screenshot of the vislog where the move gets “stuck”

[harsh storm]

It does matter more. I was going to test that later today. But often times, people use this on like 3-4 deep.

[harsh storm]

@celest python This is with a 12 class deep hierarchy.

Still, pretty fast. (But also slow at the same time)

@dense owl You might be interested in this info as well.

Do note, as siliex pointed out though, this isn't taking into consideration memory concerns.

[celest python]

which CPU you have?

[harsh storm]

Ryzen 5950x

[dense owl]

Yeah I mean I don’t personally think that’s a bad node to use, I just avoid recommending it to people because it ends up being misused everywhere

[celest python]

in BP its bad enough though 😄

that array copy will cost you

C++ semantics eliminate tons of overhead while BP cant

[dense owl]

Wait, this was not in bp?

[celest python]

hopefully not

in BP you pay for duroxx's benchmark cost + copy of 8 * 300.000 bytes with some branches inside of the reflection system

gets real slow

[dense owl]

Well then what was the point of this @harsh storm , I thought we’re talking about the get all actors of class bp node

[harsh storm]

Because of a question in #cpp

[dense owl]

Oh

[harsh storm]

I'll spin it back up for the BP node version

[dense owl]

Yeah, I mean most things are fast in cpp 😀

[harsh storm]

@celest python @dense owl

void AJRealControlActor::CalculateTiming(TArray<AActor*>& TheList)
{
    StartTime = FPlatformTime::Seconds();
    TArray<AActor*> FoundActors;
    UGameplayStatics::GetAllActorsOfClass(GetWorld(), AJRealControlActor::StaticClass(), TheList);
}

void AJRealControlActor::FinishCalculation()
{
    EndTime = FPlatformTime::Seconds();
    UE_LOGFMT(LogTemp, Log, "Total actor count: {0}", GetWorld()->GetActorCount());
    UE_LOG(LogTemp, Log, TEXT("Time taken in seconds: %.6f"), EndTime - StartTime);
}

So I'm literally pretty much doing exactly what the GetAllNodesOfClass does now.

[celest python]

this isnt 1:1 with node

have a BIE and benchmark its call, and in BP use get all actors of class from that BIE's entry

oh nevermind

[harsh storm]

I'm literally calling the exact same method from gameplay statics

Which is what that node is

[lunar grove]

I didn't end up using the sight sense like that

[uneven cloud]

Yes. There is an array of supported agents in the navigation property settings. I wouldn't make a lot of them, because nav mesh costs a lot of memory. 3 buckets of small, medium and large usually is all you need.

[uneven cloud]

What are you measuring? Is this a packaged test build or in editor?

[harsh storm]

Editor. I know it still isn't optimal.

I only pinged because of our previous talk about GetAllActorsOfClass

I'd expect it to be better in packaged anyway.

.1ms in editor is quite acceptable

I later did one with a class hierarchy of 12 deep as well.

[uneven cloud]

Get all actors works differently in the editor vs a packaged build. In editor it iterates through all the level actors. In a packaged build it uses the world TMap, so it directly gets the actors of the class.

[harsh storm]

Yeah.

I'll do a packaged one later today, after work.

[uneven cloud]

Profiling anything in the editor is frankly pointless.

[celest python]

BP node always go through level actors

If I'm not wrong

[uneven cloud]

No it doesn't

[harsh storm]

It doesn't

[uneven cloud]

The BP node eventually uses TActorIterator. Which works differently if you are in the editor vs in a packaged build.

[harsh storm]

I just threw something quick together. Heck, I used python to create all of my classes 😅

I'll set up a better one tonight and post the results.

[gleaming star]

i feel asleep but im back

would you be interested in hoping on VC to take a look? i mean, it gets stuck and break in numerous ways and AI is super compilcated when considering the character/player, the enemy BP, and the AIController

communicatingbetween all of them is super complicated, espcially in a replicated senario

it almost seems like it works for the "first" enemy but the logic "breaks" when many enemys are in play as if the AIC doesn't know who is who and doesn't know how to properly set and forget targets. like the AIC should have a specific instance for each enemy but for some reason it doesn

seem to work like that

[dense owl]

Sry i don’t really do VC. As for the actual issue, to reiterate our story so far:
I asked you to use the debugging tools to get more details on why your move is failing .
You continued to say it’s stuck.
I asked if you actually did what was suggested.
You doubled down and said yeah you did that already, how else would you know it got stuck?
I then challenged that by asking for a screenshot of the vislog.
You then proceeded to fall asleep, came back and forgot the entire conversation and now asking to hop in VC because “AI is super complicated”

Based on this, there is nothing I can do to help further, but I wish you well

[gleaming star]

lmao alright man, lemme send you a screenshot so you believe me

lol it's so broken i have no idea how to even start to explian

ggs

it does this one time, does that another

im going to start over

[harsh storm]

@celest python @uneven cloud @dense owl Development packaged build. Bottom is 12 class hierarchy deep. Top is 3. Both only have one actor the represents the class in the world.

[unborn sphinx]

Why is EQS still listed as experimental? Is it not in widespread use?

[harsh storm]

Because the editor tooling isn't up to snuff according to Mieszko

[unborn sphinx]

Hmm it seems fine to me. It was definitely confusing at first but now that I know it it seems wonderful. My friend EdWordy is refusing to use EQS because it is experimental. He is struggling to get a grid of points projected onto terrain/navmesh and I feel like EQS could do it for him very easily.

[dense owl]

Do you guys use enhanced input in your game?

[unborn sphinx]

Lyra does so yes I do too lmao

[dense owl]

Well that’s experimental, you can’t use it !

[unborn sphinx]

We are not working on the same game

but good point yes

[dense owl]

Ok well, your friend can’t use it

[unborn sphinx]

he probably wouldn't

I see him struggling with ancient ue4 workflows lmao

not struggling actually

he is quite capable, moreso than me

[dense owl]

Some experimental flags have been there for a long time, only still there because the system is not considered flawless yet, but may never get a chance to get there or not for a long time

The experimental warning is more so there for liability purposes probably

[unborn sphinx]

Yeah like if their entry-level designers are all struggling to learn eqs (it seems super complex, scary, and foreign at firsct glance) Epic can say we never said to use it!

"their" being random gamestudio

[royal mist]

i just dont get how ur supposed to use EQS to navigate a grid

u can use simple grid generators thru EQS but the AI runs straight to the target, the grid is seemingly ignored no matter the density, unless im misunderstanding the use case

[dense owl]

Did you watch 3rd pinned video in full?

[shell mirage]

I am looking at UBTDecorator_BlueprintBase, there's a warning at the top which discourage you to use it as a base for creating native C++ classes.
Base class for blueprint based decorator nodes. Do NOT use it for creating native c++ classes!
I am wondering why that's the case, there does not seem to be anything in there that could cause problem. All there is, is a buch of helpers to make your life in blueprint easier no?

In general why must there be a distinction between blueprint bases?

[royal mist]

which one?

[dense owl]

Sry, it got pushed down a bit, 4th link

[royal mist]

whats the title?

[dense owl]

Unreal Engine AI with Behaviour Trees

[royal mist]

kk

i still think i need a grid object

like eqs is cool but i need selected tiles as well

or atleast a visual representation? idk

[deft sedge]

what did you do instead?

[royal mist]

my main issue is i have a grid, that i use for navigation, that doesn't conform the the landscape

and idk how to make it generate based upon the landscape, i tried some janky stuff to no avail

so i might just do away with the grid

idk maybe i can use eqs

[deft sedge]

@royal mist @mechdawn solved that same issue already

idk, why he's not in this server though

[royal mist]

wym sorry, what did he solve? the grid conforming to the landscape?

[unborn sphinx]

u don't. You use EQS to put points on a grid in navigable areas projected onto the navmesh. Then u get those points that you have and do wat u want with them.

[uneven cloud]

The larger problem seems to be is that they are using a grid instead of the nav mesh.

[gleaming star]

i fixed issue

im legendary

figures

[unborn sphinx]

I think that is a stylistic choice they want to pursue. To have the characters move along the grid in a grid like fashion to and from only other grid points

[uneven cloud]

Sure. But you can't project onto a nav mesh when you aren't using a nav mesh

[unborn sphinx]

Well, that is true... however they are using the AIMoveTo nodes in BP, so afaik they are using navmesh

Basically I believe they have a landscape, with a navmesh, static meshes etc in there. They also have a custom grid system that has grid points that are not yet conforming to their landscape. They can point and click and tell the characters to go to these points, but they just pathfind directly to the point and don't walk along grid lines. They want the character to walk along the grid lines and for the grid to conform to the landscape. I suggested EQS could at least get them points on their landscape and work from there

Since already using navmesh, just project points on grid. You would have to iterate along some path of your own creation to force the pathfinding to be along grid points but that shouldn't be too too hard

just have a goal to move to in real XYZ coords. Then project a grid of your sizing. Find the point in the cardinal direction that gets u closest to where u wanna go and move. repeat lmao

it seems silly though because u have to essentially build another pathfinding system on top of the previous one...

The more I think about this, the more I think this would be a viable solution. The grid the player sees and interacts with is totally separate, but must be conceptually linked to maintain physical consistency, with an EQS grid that you project for movement with world origin as context. somethin like that

if u can assure ur character stays on the "grid" by carefully only moving to other grid-spaced points, it might work to omit the world origin context and have the player move only by grid amounts

character rather

this is just a crude way to get a grid that neatly kinda conforms to the landscape

but actually it doesnt solve his UI issue of not having a grid that conforms to the landscape

lmao

[royal mist]

yeah ive been thinking hard about it and i think i wont use a grid, the choice is causing too many issues.

[royal mist]

but i think i can use eqs to solve certain distance based queries, like getting path length and such

[uneven cloud]

That's all a very roundabout and hacky way to do grid navigation.

[unborn sphinx]

agreed. what is the better approach, in a few sentences?

[lunar grove]

I ended up writing some custom vision tracker. The problem I've encountered with sight sense for players is that it's cone based, while I needed that for players, that have a pyramid based FOV, with sides usually unequal (monitors are usually wide).

Can't really tell you all the details, as there's a lot of project dependent stuff going on, but essentially I registered all the actors that wanted to take part of this vision tracking process. I had the so called "observers" (players who observe) and "targets" (things being observed by players). At a certain interval, for every observer, I iterate through every target, and firstly check whether they're even in vision range, secondly I check the observer rotation relatively to target, and if they're rotated enough, I consider things in the FOV, but it be sure that the target can be seen. Finally, I launch a trace using visibility channel at the centre of the actor, because it might be behind a wall.

The solution is not perfect, as when target is too close it might be considered non-observed, as its centre is not seen (note that I check whether the actor's pelvis can be seen). There are tweaks to do depending on your project, but for my case that works fine.

[dapper iron]

just wanted to come back to this from a few days ago.
I've managed to get around my issue by setting one navmesh to have 'dynamic modifiers only' and one to 'fully dynamic' and then just using a 'non modifier mesh' (a cube) to affect the second navmesh, while being ignored by the first navmesh

[inner zephyr]

I am totally confused. My nav mesh is displayed correctly in edit mode but as soon as I start the PIE session it is placed much further down. It just floats in the air. What could cause that?
Weird. Can't really make out what broke it but I just created a new level, dragged the old one inside and broke it so it's not an instance anymore. Now it works fine 🤷‍♂️. But I'll have to setup the streaming levels again 😭

[gleaming star]

Sometimes my enemy will know I am the target but can’t get to me (or at least thinks it can’t get to me).

How can I make the enemy a bit “smarter” in knowing that it can just go up / down a hill or find a route to get to me? This error still occurs with a nav mesh that covers the entire map ect

[shell mirage]

Oh they're structs, no wonder I couldn't find them via the class creation menu