gameplay-ai

[ocean wren]

you can access the navmesh and connectivity yourself, this is probably one of those use-cases

The other thing to think about, is if you want it pathing to your player, or your AI?

I'd have thought calling in backup would want to path to the AI calling for backup?

[cerulean girder]

its for the ai

yeah theyre heading to the ai

that called back up

[mossy nexus]

can you even access a raw dijkstra from UE4 natively?

[cerulean girder]

i dont think so

ild have to write custom for that

[mossy nexus]

that was my thought as well

[cerulean girder]

which is fine, just didnt want to reinvent the wheel if there was already a way

[ocean wren]

right, so get the navpoly under the AI's current position (assuming its on the navmesh), then do a dykstras from that navpoly, looking through the list of EQS positions you'd generated and stop at the first result

[mossy nexus]

I mean you could copy the UE A* and just throw away the heuristic

[ocean wren]

You don't really want paything as it is

pathing even 🙂

[mossy nexus]

well pathing is a pay thing

so it's not wrong 😉

[mossy nexus]

I got a query

suppose I have something like a rotating camera that can look for the player. it is static, but positioned in such a way that it has jurisdiction over a large area. the area is big enough that the camera doesn't have the cone size to keep watch over it all at once. I'm looking for a way to keep it relevant to the player as they traverse the area without seeming conspicuous or gamey. my initial thought was to force the pitch rotation to change in relation to how close/far the player is to the camera position. but I'm wondering if there are other, and perhaps, better ways?

cmon AI peeps

[midnight scroll]

Is there a particular reason that AI Focus doesn't seem that widely known? This is concerning me, because it's exactly what I want. It looks fine, it works perfect. But I've been listening to a lot of tutorials, reading through a bunch of AI stuff. Yet I haven't seen a single mention of it. I went to start implementing a small system on my own and ran into the SetFocus call. Yet tutorials and guides and everything else basically just tells you to rotate the AI yourself towards a target using tick and interpolation math. It isn't even a basic BehaviorTree node to set a basic target, nevermind to use the different levels of focus, so what the hell is wrong with this system that everyone avoids it so hard?

[mossy nexus]

I don't know. I think this is the second time I hear anyone mention SetFocus

but I've been doing my own system from the get go

I generally stay away from proprietary functions both to not need extra includes or hookins to things I don't need, and because I have greater control over my own code

[midnight scroll]

Which is fair. I won't need anything more than what this offers. I'm just really confused as to why it's so unknown. It would literally replace dozens of "How do I tell my AI to face a target" tutorials. Single BP call, Done, GG, no tick implement.. Oh right, that also means no videos and Youtube views.

[ocean wren]

CE: what do you want the camera to do? keep the player in view? lead the players current motion so that you can see the world in thier forward velocity?

[mossy nexus]

stay relevant to the player. so like, no matter where the player moves in that area, the camera has a chance of spotting the player

[ocean wren]

I've typically done that.. computed a look at position which keeps the player and some future position (based off velocity) in view. Basic trigonometry to calculate FOV and look at really.

Similar to what Mario 64 did 🙂

[mossy nexus]

I don't know what you're referring to

[ocean wren]

you don't know Mario 64?

[mossy nexus]

I haven't actually played Mario 64

[ocean wren]

have a look at a video of it being played.. notice how the camera predicts the future position of the character.. that sort of thing

its basically calculating a dynamic look at position that is ahead of the target agent scaled by the agents velocity

[mossy nexus]

okay but this isn't the player camera

this is just an AI camera

[ocean wren]

well, it doesn't matter, I'd still do it anyway

that way, you never lose track of the character

[mossy nexus]

I don't think it does what I'm looking to do

the point is not to not lose track of the character though

I can handle that once the character is spotted

[ocean wren]

Well, how about this.. find a video of what you want it to do, and I'll explain how to do it 🙂 assuming I know

[mossy nexus]

but this is staying relevant as an obstacle

as in, the camera is an enemy. you are sneaking past it

[ocean wren]

oh right, so you just want it to sweep an arc?

[mossy nexus]

the arc isn't the issue

the camera is overlooking a large area. at some point the player can get close enough to the camera to hide "beneath" it. I want the camera to adjust in some way so that the player can still get spotted closer to the camera

[ocean wren]

So tilt the camera down?

[mossy nexus]

yeah that was my first thought

but it's going to seem a bit conspicuous, being that the player hasn't been spotted yet

[ocean wren]

Why not make a camera thats one of those security ones, they're just basically a dome 🙂

[mossy nexus]

like the camera knowns roughly where the player is

and I'd have to explain that

[simple crest]

That's gonna look hella weird lol. "It knows I'm here but it doesn't know I'm here"

Have the camera do one sweep at long and one sweep at close pitch?

[ocean wren]

why even do it? let the character have the benefit of not being spotted?

[midnight scroll]

It's a super high tech camera system with heat and air displacement sensors?

[mossy nexus]

I'm already accounting for that

[ocean wren]

have overlapping camera arcs?

[simple crest]

Or just use a pie slice detection instead of a cone

[mossy nexus]

I guess this could work, but sweeps are fairly time consuming to complete. perhaps I could bob the pitch dynamically to move it in a wave like fashion?

[simple crest]

Indeed you could

[ocean wren]

I've never seen a security camera do that 🙂

[mossy nexus]

I think I need some elaboration here

[ocean wren]

Why not make it a drone? then you could have quite interesting behaviour

[mossy nexus]

how could that be justified for a single lens potentially?

[mossy nexus]

because it being static is kind of the point of the whole level

[ocean wren]

Or how about this.. have a mounting with TWO cameras? just different arcs?

[simple crest]

Lol well there you just answered your question. Duct tape two cameras together

[ocean wren]

I've seen multi-camera thingies

[mossy nexus]

got a reference?

[ocean wren]

google image search is the best reference 🙂

[mossy nexus]

I have no idea what to search for

[ocean wren]

CCTV cameras

[mossy nexus]

I'm not sure a multi-camera thingie is a technical term

[ocean wren]

literally the first page 🙂

[mossy nexus]

yeah okay

I guess I can look into that, thanks

[ocean wren]

I was thinking something more like this:

[midnight scroll]

Panoramic cameras. 😄 https://ellipsesecurity.com/wp-content/uploads/2018/02/ds-2cd6986f.jpg

[ocean wren]

See that one would naturally have views at a shallower angle too

The ones we have in our computer labs look liks this:

and the technicians can rotate/pan the camera remotely

I think they're called PTZ cameras

pan/tilt/zoom cameras

That last one would be a bit more visible about where its pointing, but definitely looks PTZ.

[misty gale]

Ptz cams are nice

We use them alot

Ptz dome cameras

[ocean wren]

Seems they're popular because you can remote control em 🙂 lazy shits 🙂

[misty gale]

Indeed

We remote control entire supply stations

[ocean wren]

And not ideal for when someone is trying to stealth rob your bank, too easy to catch them

Good for ML stuff though, having the machine learning model do the camera control and tracking and whatnot

[mossy nexus]

I keep thinking markup language when you say ML

instead of machine learning

[ocean wren]

machine learning markup language 🙂

[mossy nexus]

MLML

MLUML

[ocean wren]

confusing?

[mossy nexus]

not as confusing as lisp

[ocean wren]

damn, almost 2022

[mossy nexus]

here's to another year in isolation

[ocean wren]

YES!

alien isolation? 🙂

[pine steeple]

🤔

i hate those multi camera setups

scares me, like big brother is watching me

[midnight scroll]

KAOS. Best AI pathing that avoids getting stuck?

[pine steeple]

stuck where and how

[midnight scroll]

Like, RVO stuff or crowd pathing thing.

[pine steeple]

RVO does not care for navmesh, so the agents can end up outside of navmesh bounds

crowd pathing does care for navmesh (slightly), but its a lot more jankier than RVO

i did try using both (turning on/off rvo/crowd following depending on situation) but could never get it working nice

best one for me was Crowd following

especially with big crowds

just reduced how often avoidance was calculated

every frame is ridiculous

[midnight scroll]

Hmm.

[pine steeple]

http://thegames.dev/snaps/plastic_cBj3h1ETy9.png

[midnight scroll]

I really wish there was a way to affect navmesh, except for self.

[pine steeple]

http://thegames.dev/snaps/plastic_ZSYvehXTSD.png

i implemented this on engine level

there isnt

and that would be another entire headache

sounds good in theory, bad in practice

[keen crow]

Is there a good way to make my character just stay still in BT in an infinite in-progress task so that only a higher priority decorator may abort such idling? At some moments of my gameplay I want my NPCs to do nothing but without reevaluating each decorator every BT-tick, instead I want only blackboard-observed values to abort such behavior in higher priority decorators

[mossy nexus]

I made a custom task for this. It can only be aborted

[keen crow]

and what's the logis inside the task? just return EBTNodeResult::InProgress?

[mossy nexus]

yeah

[keen crow]

then how is different from default task Finish with result In Progress?

[mossy nexus]

I guess it isn't. I had no idea that existed. but I do like not to use proprietary nodes if I can avoid it anyway

[keen crow]

ok got it. thx

[ocean wren]

Just have a wait node

it'll keep going back to your wait task until something more important aborts it

so have a wait task as your rightmost

[mossy nexus]

I guess the wait task is cheaper?

[keen crow]

but in this case after each wait completed BT will evaluate decorators on its way to wait task that I don't want to be evaluated

[ocean wren]

why not?

[mossy nexus]

but is it cheaper?

[ocean wren]

Well, it doesn't do anything, so how expensive can it be?

cheaper than what anyway?

[mossy nexus]

cheaper than having a perpetual progress task

the tree will tick regardless, no?

[ocean wren]

Well, it depends on how long you wait for

theoretically its supposed to be event driven.. not sure if the wait is using a timer or ticking though, haven't looked

[celest python]

iirc it just updates a float with deltatime

Time =+ DeltaTime;
if(Time > WaitTime)
{

finish()

}

[ocean wren]

ok, so it polls the time

I mean you're not going to get much cheaper than that

[torpid pier]

Anyone may know what it means when your AI has a red circle instead of the normal green circles in the AI view? My guess it's something with not spawning in correctly but its still moving but can't see it

[mossy nexus]

wouldn't FinishWithResult set to in progress be cheaper?

[ocean wren]

Wompa: usually if there's an error, its written to the log, so check that

CE: Not sure what you mean?

[mossy nexus]

there's a node called FinishWithResult

that can be set to In Progress

[ocean wren]

Yeah, but what does that have to do with waiting? if you return inprogress it'll just execute the node again right? so yes you could implement a wait by timing it and returning completed instead of inprogress.. but why not just use the wait node?

[mossy nexus]

yeah so it'll execute the node again, but wouldn't that be cheaper than ticking the wait task?

[ocean wren]

it'd basically do the same thing

its always going to evaluate the node, the node always has to check if some time threshold has elapsed.. same thing, different approach, but personally I'd think the wait node made more sense.

[celest python]

Zoombapup, is there any possibility you know the actual difference between three?

I'm having issue with by path planner component, might be related with I'm using the first element in the enum

[ocean wren]

Well, the first one is using what it says, a 2D raycast against the navpolies in the navmesh, I suspect the second is using some kind of heirarchical data structure to speed that up (likely the octree as I've seen that in the nav code), the last is probably using the regular A* pathfinder

Whats the issue you're having?

[celest python]

Path points (offsets to actual target location) was sometimes not updating their location but I realised it was related with something else

Been trying to solve it since two weeks, was getting suspicious about everything in the end. Randomly I found the actual issue and it's fixed now 😄

[ocean wren]

Good stuff

[celest python]

Out of curiosity, is there any accuracy difference between those?

I guess first one just provides "has path" or "does not have path" info

While others can provide actual info about navmesh

[ocean wren]

Well, the first one is just a raycast, so its a line test against the navpoly bounds, gives you a hit when the raycast hits an edge of the navpoly that isn't connected to another one

done in 2D, so its basically a line/line intersection test

I've used it a few times to probe the edges of navmeshes for cover generation, but ended up using physics raycasts instead

[celest python]

noted, thanks

[tranquil robin]

I have a scene in space and I want the actor to fly to certain point, only problem is that from certain angles there's one another actor on the way. My simple timeline changing the actors location towards the target fails as it collides with that actor. Is there some pathfinding node that works in 3D space that would easily just allow it to detour around the other actor?

[ocean wren]

There's a free plugin on the marketplace that might do what you want

but in general, no

[tranquil robin]

Thanks, I know the plugin but it's outdated, not working on the current versions

[ocean wren]

Ah, you're out of luck then

[tranquil robin]

I made sort of vector swing arm type of solution that has center point on the obstructing mesh, but unfortunately on short distances that also fails since my obstructing mesh is spinning and has some extending parts

well, i guess it is in to watching tutorials and learning how to build flying AIs from scratch

[ocean wren]

You're basically in need of some collision avoidance, which you could do reasonably with simple steering behaviours, but it depends on your flight physics model

[tranquil robin]

it might work with some capsules to detect collision that then push the actor to other direction, but since my dumb linear, or optional pivot around, system is just on timelines, feels like it gets super messy if try add another layer or manipulation on top of that timeline. Anyways, thanks for the input, I'll go search for some vids on how to make flying AI

[ocean wren]

Why are you using timelines?

[tranquil robin]

there wasn't suppose to be any AI in the beginning, but then there's this specific case if the player gets lost and needs to be "force" returned to the starting point (without just teleporting)

[ocean wren]

Timelines seem a very weird solution is all.. I'd have thought a sequencer sequence? or is this just an interpolation value?

[tranquil robin]

so in most cases I can just have timeline lenth based on the distance to fly the player back to beginning, but it doesn't work if the player happens to be exactly behind the goal where he suppose to fly to begin with

it's interpolating yes

since it's empty space, except the station, that's the only object player can collide with. problem is someone might fly so far of the course they have no idea where the station even is and needs to be forced back to beginning point before their time runs out

they are suppose to dock with the station and that can only be done from one side

[ocean wren]

Why not teleport them?

[tranquil robin]

it's a simulator, you'll be sitting in actual capsule, looking out of window seeing the space

that would be highly non-immersive if it just jumped location

[ocean wren]

so you need some way to point them towards the station then?

or some kind of auto-docking thing?

[tranquil robin]

there are guides and such, but i just need fail safe automation to fly them back to front. auto docking is already there basically, in case they fail to do it... but it just doesn't work if they close to the station. linear flight works on front, pivot around works from behind the station, except if they are too close to the station

maybe trying to do this with timelines was fool's way to begin with

[ocean wren]

Yeah, that sounded a bit of a weird approach

[tranquil robin]

i just thought AI is too big topic for me to go into

i'm new to UE and no programming experience, just somehow ended up doing this thing

[ocean wren]

Hehehe.. thats how things work 🙂

[viral plover]

Anyone here have experience with the AI sight rotation instantly snapping towards the target location when using the "MoveTo" AI behavior? I can't seem to fix it in a way that doesn't make matters worse. I would like for the sight perception to follow the actor rotation, which doesn't seem to work when the AI behavior is in charge.

It works fine when I work with character rotations outside of the AI behavior tree though.

[midnight scroll]

I had a problem with that. I had to override the Pawn's C++ FaceRotation and not call Super. I use the following settings in the images there.

void ATarrionCharacterBase::FaceRotation(FRotator NewControlRotation, float DeltaTime)
{

}

[viral plover]

Thanks for the tip. How did you override the C++? Can it be done in a blueprint only project?
Enabling use controller rotation yaw makes the pawn also instantly snap directions now.

[mossy nexus]

just make your own BT task where you do a smooth rotation turn before doing a move to

[celest python]

Sadly its not exposed to BPs

Enabling use controller rotation yaw makes the pawn also instantly snap directions now.
Usually people lerp the rotation in FaceRotation and it makes it act like bOrientRotationYaw

[viral plover]

It's a backup solution I've considered, yes.

[midnight scroll]

Yeah, that's far out of BP only range. If you don't need the pawn to look at the control rotation for strafing or whatever, you can just orient rotation to movement.

If you need control rotation looking, you'll have to pretty much override anything in BP by disabling it all and doing it yourself.

[viral plover]

That's what I feared ^^'

[mossy nexus]

all of this just makes me glad that I made my own system

[midnight scroll]

Looking at the function. I'm not sure if I needed to override that actually..

Oh. Well thanks. You just pointed out I can remove that. 😄 Different settings.

Uncheck that, and use my other CMC settings and it works fine.

[viral plover]

Set Focal Point seems to override the the direction snapping, and be compatible with interpolation called upon from the actor blueprint

[midnight scroll]

Nah, you don't need to interpolate. CMC will do it for you.

[viral plover]

My character doesn't rotate at all though when I use the focal point blueprint 😅 (While moving)

[midnight scroll]

Should. I'm using the default focus system as well.

Without the FaceRotation override, which I just took out, I'm fairly sure nothing I've done is not BP accessible.

[viral plover]

Can the MoveTo C++ code be accessed in a BP project though?

[mossy nexus]

the move to task is just a wrapper for AIMoveTo

which exists as a node in BP

[viral plover]

ah

[tranquil robin]

Talking about AIMoveTo, I was just trying this node.. does it only work on flat level where you define the area on which actors can move? Or can it be used in 3D space?

[midnight scroll]

As far as I know AIMoveTo relies on the navmesh.

[tranquil robin]

What about "Find Path to Location"?

[mossy nexus]

anything AI related relies on navmesh. so unless you have some 3d navmesh then no

[midnight scroll]

There were some plugins like a year ago I was glancing at for flying AI. No memory of them though.

[mossy nexus]

I was toying with the concept of having sky banded navmesh, but my world is not high tech enough to require that

[tranquil robin]

One more question, can these AI related nodes be used in other blueprints? For example, keypress spawns something,followed by these nodes to get path... or do these exclusively have to go to the AI blueprint?

[mossy nexus]

they require an AI controller

it doesn't have to be in the AI controller blueprint

but you need an AI controlled pawn to make use of them

[tranquil robin]

So I found AI controller on the create blueprint list before, but I couldn't figure out how to integrate that together with the actor I want to be AI controlled

[mossy nexus]

it needs to be a pawn

[tranquil robin]

And on the blueprint where I spawn this actor, error is that the AI controller doesn't exist when I try to call it... so I figured that much that somehow it needs to be manually created

[mossy nexus]

pawn is the term for anything AI controlled

pawns also have an option to automatically spawn an AI controller of a given type in their default settings

[tranquil robin]

so how do I combine the AIcontroller and Pawn? Because on pawn I can't find the AI controller

As in I was trying to add AIcontroller to the pawn using add component, but I can't find it there

[mossy nexus]

that's not how controllers work

controllers possess their pawns

I'd look up MathewW's videos on AI controller

you're obviously not aware of how the system works yet so no point in me explaining the very basics

[tranquil robin]

Yeah, I also saw that AI Controller Class thingy, but figured since I couldn't get it to work I missed something. Anyways, you already helped me understand few things. I was actually checking his video on AIMoveTo, that's why I asked if it requires that guiding area, NavMesh, like you mentioned to. But I'll look up his video on AI controller too

Thanks

[midnight scroll]

Does anyone happen to know off hand how to stop AI from slowing down when choosing a new moveto target? Like if you tell them to run to a place far in the distance, they start running, you tell them to run to a point just a few units away from the far away point, it's not even a full degree of rotation away from their current target, but they sort of pause for a quarter second and then start running again.

Seems that RequestedMoveUseAcceleration in the AI's CMC was what I was after. Disabling it stops that behavior.

[misty wharf]

Interesting, I would have thought it could be a result of the pathfinding

but good to know that's a thing

[midnight scroll]

Some people have reportedly also created full BT Tasks that do the entire movement. Not sure what that changes, but I guess it also solves it if you never call finish execute on the task until movement is done.

[mossy nexus]

yeah I consolidate movement behavior in sequences

[fleet pebble]

Hi
This is my first week with the engine.
I have a pack of animations, a pack of characters, and an environment.
I want to walk the characters around the environment giving them some level of autonomous control.
Then I'll film cartoons in the environment.
I figured out how to retarget the animations from the animation pack to the scanned archviz character pack cause they use the UE4 mannequin
then I made a character object and added the skeleton and the static mesh and an idle animation
then I added a blueprint to the character object that I made and used a simple move to relocate it to a position on the map
all of this was sort of done automatically for me by the engine.
I have like...a million animations in this pack.
Is there a guide that someone can point me to that will help me do things like switch to the walking animation, walk the character somewhere, when they stop, change to the idle animation, make them stretch periodically, etc?
I've gotten to the point where I have to actually learn something lol and I'm not sure what to look up

can I just control all that from the character's blueprint?

because I watched a video last night that says I can make an AI component and a behavior tree ... can I buy a pack that does that? Any recommendations?

[ocean wren]

You'd probably be better off using sequencer and just playing the animations rather than trying to do AI

Theres a video on the Unreal Engine youtube channel from the Epic guys on how to use sequencer

[fleet pebble]

sweet

thx

[midnight scroll]

Am I missing a setting somewhere? Is there any way to turn EQS queries into the local space of the querying AI?

Meh. Screw it, will just write my own.

[mossy nexus]

do it. join the NIH crowd

https://tenor.com/view/simpsons-homer-bart-lisa-join-us-gif-13066208

[midnight scroll]

The wut?

[mossy nexus]

https://en.wikipedia.org/wiki/Not_invented_here

okay admittedly they don't paint it favorably

but I think it's entirely fair to write your own stuff in lieu of existing elements

[midnight scroll]

Oh. Yay. The custom EQS generator class crashes the engine.

[midnight scroll]

So. I'm curious why this doesn't work? This doesn't create points when used in an EQS testing pawn like other generators do. Hard coded some world points for testing.

Ah. Wasn't enough to change the Generated Item Type in the EQS menu. Had to change it as a class default for the generator.

[celest python]

Just FYI, I dont know what changed but overhead decreased to 0.05ms

I guess it differs from Standalone to Shipping

Now I can do infinite amount of pathfindings 😄

[misty gale]

just serialize them and you can do infinite anyways 😛

well.. capped by amount per frame and distance to travel before new pathrequest but thats not an issue at 75

[dawn loom]

what is the "best practice" way of running a behaviour tree?

is it with a BT component or with a Run BT blueprint call?

and how can I get the BT/BB from blueprint?

[celest python]

With sync path finding it was consuming 2ms on my machine for 50 pawns

On a tick with 0.2s interval

[celest python]

Though I didnt understand this, can you elaborate more? What serializing has to do with pathfinding? 🤔

[dawn loom]

@mossy nexus after thinking about it I guess I should have a service that loops on a .5 second delay that checks if a new task is available when an AI is idle

[mossy nexus]

well

I would have the AI query the director when it's idle

[dawn loom]

it's better than making the manager itself check the BBs of the AI controllers to see if they're not doing anything

[mossy nexus]

and the director could then say "do x"

[dawn loom]

yeah that's what I will implement I think

[mossy nexus]

yeah this is a no no

[mossy nexus]

sure but you can do that without a service. It's a task to query.

you could also have the director assign roles to your AI and do role based behaviors

[dawn loom]

it's supposed to be a really dead simple background AI I don't think I need to setup anything complicated

they're basically NPCs that will move around a little bit to play animations in front of locations

[mossy nexus]

a role would be a struct containing an array of positions, which could themselves be actors or structs

[mossy nexus]

okay, it works all the same though

[tardy talon]

When the other actor gets out of the perception radius, does this then also change the value of the blackboard key back to nothing(or default)?

[dawn loom]

but yeah I will also have to map the locations to an animation to play somehow (probably through an enum for the location type)

I guess you could call it a "role" actually

[mossy nexus]

That's silly. just use an actor or a struct

my setup is

Point actor that holds information such as animation, duration of stay any potential custom logic

[dawn loom]

how do you get the actors from the manager?

[mossy nexus]

you assign them at editor time

[dawn loom]

I'm thinking about having different zones with different managers so I can't simply query all in map

it sounds kinda painful tho no?

[mossy nexus]

what is your alternative?

for me it's two nodes in the level BP

[dawn loom]

I was thinking about what kind of gismos can be seen in the level editor

something like the spline one but with only points would be cool

[mossy nexus]

that sounds like a pain

[dawn loom]

or at worst having actors that are attached to the manager on the level editor

[mossy nexus]

that's what I am doing?

[dawn loom]

ah you meant attach them by assign them

I thought you meant selecting them in the property editor

[grizzled hazel]

"Never do at run time what could be done at compile time" - some smart guy who I don't remember the name of

[mossy nexus]

but the director doesn't know which tasks are supposed to b active until you tell it

[grizzled hazel]

If it's really basic ai then have them all set up in editor and assigned

[dawn loom]

technically it's kinda compile time

at runtime you would just need to get all the attached actors

[mossy nexus]

this is probably one of the few places I would advocate for level markup

[dawn loom]

and yeah, I'm not the level designer

so I have to make it dead simple

[mossy nexus]

actors in world. put them into an array. activate. director goes brrr

[dawn loom]

why put them into an array tho? you could simply attach them

in the world outliner

[mossy nexus]

stop

you're doing silly stuff

[dawn loom]

that's what I'm going to do 😆

[mossy nexus]

do it right

[dawn loom]

I can't stand the idea of having to add items in the array one by one

it also adds room for error like adding the same actor twice

[mossy nexus]

well get used to it

at some point you're handing it over to a designer

[dawn loom]

that's the whole issue

if you tell them "just drag and drop it on the manager in the outliner"

it's much simpler than

[mossy nexus]

that's not how it works

[dawn loom]

"add empty elements in the list then select the proper actor"

it works tho

[mossy nexus]

yeah but it's really bad

[dawn loom]

why?

[mossy nexus]

esoteric thinking

[dawn loom]

how is it bad?

[mossy nexus]

stick to simple things that work

you can optimize later

You're trying to optimize before even doing anything

[dawn loom]

???

hold my glass of water

😳

[mossy nexus]

because there is no good way of just dropping a manager into a level and expecting it to understand everything

it's a fallacy

you can automate behavior and registration

and tbh I would advocate using a subsystem in CPP instead of a manager actor

[mossy nexus]

that's silly

[dawn loom]

why is it silly? :0

[mossy nexus]

what happens if those tasks are done?

[dawn loom]

I meant to put tasklocation

[mossy nexus]

no global overview

[dawn loom]

it doesn't get deleted

tasks are never done

[mossy nexus]

attaching also make ordering a pain in the behind

[grizzled hazel]

At this point just have a spline to move along if multiple objects are hard

[dawn loom]

as I said it will pick a random task

it's attached only on the editor so that when the map is loaded I can know which manager has which tasks

[mossy nexus]

you should have only one manager

not a million

It's like you want to make it hard for yourself

[dawn loom]

I need exactly 2 because NPCs can't move between a certain area and another

[mossy nexus]

doesn't matter for the manager

you're conflating tjings

[grizzled hazel]

This is starting to sound like how darksouls 1 has its ai set up and limited

[dawn loom]

well it does since when a task is assigned the npc will always do it

[mossy nexus]

and that's why your thinking is esoteric IMO

[dawn loom]

I think you guys don't realize the dead simplicity of what I'm doing

[grizzled hazel]

Just use tags if you want specific areas

[mossy nexus]

i think you're trying to oversimplify how complex AI is

[grizzled hazel]

Or a plugin

[mossy nexus]

tags are fine for this. the important thing is having a single robust manager

[grizzled hazel]

NIH Be damned

[misty gale]

Not actual serialization but basicly just doing a que , and have a central place to process the requests

[dawn loom]

I think you're trying to project your complex AI system in what I'm trying to do

[mossy nexus]

no

[grizzled hazel]

All AI is complex

[mossy nexus]

I'm not

[grizzled hazel]

It's the I part

[dawn loom]

the manager could be replaced with a literal array

[mossy nexus]

you're completely misunderstanding the concept of a manager class

[dawn loom]

there is no "task" to be done, only locations to be in

[grizzled hazel]

Then a sequence would suffice

[dawn loom]

no because it's a crowd

random AIs will be in random locations and move around

but that's all

[grizzled hazel]

Then some fluid dynamics using meshes is in order

[mossy nexus]

except those locations are supposed to have animations as well. you need durations for themz because they can't stay in one place indefinitely

[dawn loom]

that will be handled by the AI

it will play the animation a random amount of times

[mossy nexus]

how will the AI know what to do

[misty gale]

On beginplay -> register at manager?

[dawn loom]

exactly what I'm going to do

I'll just get the attached actors and register them to the manager

[grizzled hazel]

Bad practice

[mossy nexus]

lol

[dawn loom]

how is it bad practice wth?

[mossy nexus]

esoteric thinking

[grizzled hazel]

Because it can just be set before runtime

[dawn loom]

having an array where you can place twice the same actor is better?

[grizzled hazel]

The compiler can handle that

[mossy nexus]

stop conflating things and start listening instead of trying to have your own opinion all the time lol

[dawn loom]

what?

what compiler are you even talking about?

[dawn loom]

excuse me? 😂

[misty gale]

Wait, I'm getting this wrong right? Is it bad practice to let the AI auto register, but good practice to manually pick them? This must for sure be me missing some memos

[grizzled hazel]

The engine you're working in, unreal, has a compiler and can at compile time fix stuff like multiple arrays

[mossy nexus]

you obviously don't know architecture very well but you act like you know everything.yet every step you want to take is painting yourself into a corner

[misty gale]

Or misunderstanding whats been said, at the very least :p

[grizzled hazel]

If it's known at compile time

[dawn loom]

wow you know me so well

[dawn loom]

the heck?

do you even know what a compiler is my friend?

[mossy nexus]

Literally just what I see from your behavior here

[grizzled hazel]

...

No need to be rude

[dawn loom]

if I put an array in my manager and put the same task location in it multiple times how can the "compiler" know that?

[mossy nexus]

if you don't want to listen why are you even here?

[dawn loom]

because I'm trying to have a discussion, and the only person here trying to force their opinion on others is you my friend

[mossy nexus]

ok, good luck then!

[dawn loom]

I gave you a valid criticism and thought of an "esoteric" better way for me to handle things

[mossy nexus]

yes because you conflate things

[dawn loom]

but that's wrong because all I say is wrong?

[misty gale]

Im a bit confused about the multiple tasks attatched to the manager tho

[dawn loom]

they're just task locations

I'm not even understanding @mossy nexus's solution because they never explained it fully

[mossy nexus]

maybe because you never took th time to listen?

[misty gale]

Wouldnt tasks also be actors in the world?

[mossy nexus]

you immediately went NO I DON'T LIKE THAT

[misty gale]

With their own locations, and logic?

'Smart objects' or whatever

[mossy nexus]

arrays bad

etc

[dawn loom]

but if I got it right, I should make two BP classes, one for tasks and one for the manager, have only manager even though it handles two distinct areas that are unreachable from one another, put the tasks in the map and then register it through an array manually by putting the tasks on the array

my criticism was, why not attach locations to their manager instead of letting designers add it to the array

because having an array means the same task could be in it twice

which means two AI could be sent to it at the same time

and having one manager instead of two would just mean having two arrays instead of one and switching between them

and again, why not attach them, there is literally no reason to not attach them

[mossy nexus]

what happens once you want to switch over to using something like subsystem instead? every level will have to change its approach because you conflated ownership in level with mapping

[dawn loom]

why would I ever move to a subsystem?

[mossy nexus]

why wouldn't you?

it's the premier way of doing things

[dawn loom]

and how would that literally change anything ?

[mossy nexus]

accessible everywhere without hitches

[dawn loom]

if my BP classes are in a subsystem or not won't change the fact that they can be attached or not

[mossy nexus]

yes it would

a subsystem isn't an actor

[misty gale]

Subsystems are awesome, i've heard

No clue how to make one yet tho 😄

[mossy nexus]

the fact that you still insist on having a manager as an actor I don't think is helping you

[dawn loom]

sorry I think I misunderstood what you meant by subsystem

but still

why would I ever move to a subsystem?

[mossy nexus]

it's just driving you down this path of making decisions that will leave you with huge tech debt

[misty gale]

Also makes sense for thr ai to query the manager

Not the other way around

[mossy nexus]

yes

[misty gale]

At best, manager idles , while all ai do their stuff 😄

[dawn loom]

as I said the "manager" is nothing more than an array of locations

[misty gale]

Instead of constantly checking what every single one is doing

[dawn loom]

it wouldn't

[mossy nexus]

you still have the issue of ordering

[misty gale]

Ending up with a ton of edge cases and scenarios where the manager must handle it

[dawn loom]

it would only check what they are doing when it needs to assign something to one of them

[dawn loom]

ordering what? I told you it's about randomly picking tasks

how can you order a list of location meant to be picked randomly?

what would even be the point?

[mossy nexus]

what if a task requires more than one location?

[dawn loom]

as I said, idk what you have in mind but if you listened to what I'm saying

it's not even a task

it's just standing

and playing an animation

[misty gale]

What is the ai really doing here? Random idle walking?

[dawn loom]

yes

that's literally all

[mossy nexus]

why even have a manager then lol

just assign random points to AI and done

[dawn loom]

I said it's dead simple but I guess I'm just thinking "esoterically"

[misty gale]

Not sure it counts as ai at all tbh 😅

[dawn loom]

because it needs to go to predefined spots

[mossy nexus]

your thinking is esoteric because you conflate things

[misty gale]

Random points, wasnt it?

[mossy nexus]

not because of your "simplicity"

[dawn loom]

I think the only one conflating things here is you my friend

[dawn loom]

a random location from the list of predefined ones

[mossy nexus]

great way to take criticism

just lol

[dawn loom]

what criticism 😅

you're criticizing me instead of criticizing any part of my system

all the while asking me to make it much more complicated

for absolutely no reason

and packing things together when they don't need to

that's the definition of conflating 😅

[mossy nexus]

I'm criticizing your thinking because you're trying to do 3 things at once

instead of focusing on a layered system

if you truly want a dead simple system. don't have a manager

[misty gale]

Could still have a manager

But not for the ai

[dawn loom]

how would an AI know if a spot is empty or not then?

[mossy nexus]

just assign random points directly to the AI

[misty gale]

For the locations

As they are fixed

[mossy nexus]

RVO?

[dawn loom]

it's more of a locations manager than an AI manager actually

[misty gale]

Predefined*

[mossy nexus]

or not even, just collision detection

[misty gale]

Youd want to know before that

[dawn loom]

that sounds completely unnecessary

[misty gale]

Before sending an ai to a spot

That you know is already taken

[mossy nexus]

that's too complex

they wanted dead simple

[misty gale]

SimpleParalell :p

[dawn loom]

😅

[mossy nexus]

if the tasks never change you don't need a manager

[dawn loom]

I would have been done with my system if I was working instead of chatting 😅

[misty gale]

Id be done with my game if i uninstalled discord

[mossy nexus]

if no overview is needed you don't need a manager

[dawn loom]

there is

[mossy nexus]

just assign points on the AI

done

[dawn loom]

no because they could go to the same spot

and leave others open

[mossy nexus]

oh right you hate arrays

I forgot

[dawn loom]

wtf is wrong with you?

jesus

[mossy nexus]

how if the spots are random?

[dawn loom]

I said I think building the array from attached locations is better from a designer UX point of view

[misty gale]

So manager must know if a spot is assigned to an npc

If it is, pick another

[dawn loom]

I said it picks the spots randomly

[misty gale]

Or keep them in sorted array

[dawn loom]

for the 542123th time

telling me I'm not listening

[mossy nexus]

exactly

[dawn loom]

when literally not reading what I'm saying

it picks one of the spots randomly

not a literal random location

I said it when you were talking about order

[mossy nexus]

if two AIs aren't even sharing the same spots, how could they collide?

[dawn loom]

I told you order doesn't matter because it'd pick them randomly

and spoiler, I don't hate arrays

I just don't see why I would make a public array when it could fetch the attached actors and copy that on begin play

I would have literally been done if I wasn't talking

[misty gale]

Copy? What? 😅

[mossy nexus]

ok buddy, good luck

[dawn loom]

ty

[misty gale]

I missed the copy memo :(

[dawn loom]

this is what I meant by copy

this is in the "manager"

[misty gale]

Ohhh

I see

Id just let the actors register to the manager directly but i suppose this does the job

Being registered, they could use dispatchers to communicate with the manager

If needed

Like if their state changes

[dawn loom]

they don't need anything really

initially I was thinking about having literal vectors for the location

I don't even need an actor

I could have a struct with a vector and an enum for the type of animation to play

[mossy nexus]

I think it's pointless to argue here

[dawn loom]

it's pointless to try to force a certain solution on me*

I take any criticism of my system and when my manager was broken (before I even had had a good thought about it) and you gave me valid criticism I listened and changed the way I see things

[misty gale]

Actors are easier to manage, since they can hold info beyond just their location

They could also have custom logic if you wanted expansion down the line

[dawn loom]

but here it's a tradeoff between making it a pain for designers and a potential source of bugs vs ???

I guess the downside would be if I want to change my BP to a subsystem it would be an issue but why would I ever do that

[dawn loom]

to quote our dear friend I think that would be conflating the task actors which only really need to hold the location

[dawn loom]

I really don't

in fact all of this is unnecessary

our game didn't even need any of this

[misty gale]

If thats decided, then its all good

Im often ovetthinking how flexible i want my systems to be

[dawn loom]

I'm only doing it to get used to BTs and BBs only

[misty gale]

Overengineering is a true curse

At times

[dawn loom]

the NPCs were literally skeletal meshes looping an animation

I thought it would be more fun if some could move around locations

but that's all

[mossy nexus]

better to be future proof in the case of system architecture

[misty gale]

For sure, when possible

Its a tradeoff tho

[mossy nexus]

I think having to unwind years of tech debt trumps overthinking easily

but your mileage may vary

[dawn loom]

I am 100% with you on this one @mossy nexus

in fact I'm rewriting everything in our project (not talking about AI) specifically because they used drag and drop BPs and made spaghetti code

but again here it's dead simple

and the risk of having the array not be correctly filled by the LD is just too big

[mossy nexus]

famous last words

[dawn loom]

I will always be 100% against selecting items that are in a level on the details panel of a blueprint (be it an actor or a subsystem)

and I would much rather have them attached or be part of the actor itself

[misty gale]

Takes some experience and knowledge to properly plan ahead at most times^^

[ocean wren]

Hhaha.. you guys really just talked over each other 🙂 read the thread back..

so... how I'd do this.. I'd have a BP that auto-registers itself with an actor (you can find specific actors in a level onBeginPlay). That means you don't need to have your LD's do anything but drag positions for your AI to go to. The actor they register themselves with would be the one that AI uses to get the location to go to. I take the point that they're just locations, but in my experience they never are "just" locations. You have to reserve them otherwise you have multiple entities going to the same spot etc.

the point I think people were trying to make, was that parenting actors to other actors is kind of lame as a setup. Plus UE editor doesn't make that ideal.

[misty gale]

Nice summay ^^

[ocean wren]

I'd suggest it was a better idea to just have the "place you can go to" register itself as a place to some central system (doesn't matter if its a single actor or a system) mainly because your LD's can have a nice gizmo based position to edit and see and group in the outliner and such

[dawn loom]

the issue is that there are two areas that won't be reachable from one another

and I don't need to make the actors have a "reserved" tag because they would simply be removed from the array when they're taken

and then put back in it when they're freed

[ocean wren]

So add a "area ID" to your BP that registers a position so that when an AI queries for positions it gets only those

[misty gale]

Also, two weeks from now some gd says "we need the ai to do more"

"Would be cool if the npc would do x"

Then your system is ready for this with smart actors :p

[ocean wren]

well, ok, having two arrays vs an array with a reserved bool.. whatever 🙂

[misty gale]

Or potentially smart* actors

[dawn loom]

I'm the GD technically (and I already have the AI doing more, it can stare at and follow the player randomly if it's nearby)

[dawn loom]

not even two, the AI can "put it back" itself

[dawn loom]

I guess this might be a better idea 🤔

but what's so wrong about attaching actors?

[misty gale]

But then something outside the manager is modifying what the manager is ment to control?

Containment

[dawn loom]

OOP is a curse

[ocean wren]

So generally, I'd have whatever data needs to be associated with a position within the position object, and have queries on the manager actor/system to filter them

[dawn loom]

let's not go down the rabbit hole of OOP

[ocean wren]

well, the attaching actors thing is just a pain in the arse when you want to move them properly and when you want to version control the level or do multiple editor stuff. But hey, if thats what you like, go for it

I mean if you're the only one working on it, you do what you want.. but my experience has been that the editor doesn't do well with heirarchies of objects for some reason, I guess I might have seen some edge cases but still

[grizzled hazel]

Also wouldn't just a nav modifier do the job of separating areas?

[dawn loom]

I find it harder to have to make people put area tags or fill an array manually from instances in the level

Than to attach even if it's a pain

[ocean wren]

You don't fill the array manually

[dawn loom]

Especially since the parent won't move

[ocean wren]

thats the point

they fill themselves into a central system

[grizzled hazel]

Like there are nav modifiers made specifically for separating areas like you want, no need for multiple systems

[dawn loom]

Yeah I meant if I don't put an area tag (referring to the previous solution the other person was trying to force me to use)

[dawn loom]

I didn't think of that!

[ocean wren]

Well, if you use individual actors, you can set multiple properties on all selected actors at once, so using a tag or id for each area would work there

just add an id/tag/whatever to your position objects

select the ones you want in each seperate area, give them different values

that'd just be part of the query setup right

"give me points near me in area 1", "give me a point in area 2"

[grizzled hazel]

Just gonna throw this here

https://docs.unrealengine.com/4.27/en-US/InteractiveExperiences/ArtificialIntelligence/NavigationSystem/CustomNavigationAreasAndQueryFiltersLanding/CustomNavigationAreasAndQueryFilters/

[dawn loom]

I think it'd be even simpler to figure areas from what is reachable or not

[ocean wren]

he's not using navigation in this particular aspect though

[dawn loom]

no need to have a tag or an ID or anything

[ocean wren]

so nav isn't the issue

[dawn loom]

technically it's the root of my issue

I think this might be a perfect solution

[ocean wren]

so its not JUST selecting a point then

[misty gale]

😅

[dawn loom]

it is

[ocean wren]

its selecting a reachable point

[misty gale]

It almost never is

[ocean wren]

at random

[dawn loom]

yeah

[ocean wren]

that isn't already being used

[dawn loom]

yup

[ocean wren]

THAT ISN'T "JUST" A POINT 🙂

[dawn loom]

it's virtually the same though

because if it's used it won't be selectable

[grizzled hazel]

You see why we've been at this for HOURS now?

[dawn loom]

and all points will be reachable if they're in the array (either attached or from deducing it with the area)

[grizzled hazel]

God I should've remembered nav areas earlier...

[ocean wren]

I dont think there's a way to contrain a thing within the navmesh though

constrain even

although to be fair, it'd only be a simple construction script change to do that

[misty gale]

Wait, howd nav mesh fix this you said?

[dawn loom]

also sorry if what I said to you earlier sounded rude, I didn't mean to, I simply wasn't understanding what you meant with "compile time" as I'm a software dev (and I studied regular computer science), when you start the editor the "compile time" is already done, I guess you meant on construction @grizzled hazel

[worthy python]

Hello. Where can I find an example of this visibility algorithm? I want separate the safe and danger zones.

[dawn loom]

the points are reachable so they can simply get the navmesh they're in to know which manager manages that area

[grizzled hazel]

When you ship for development it compiles it

[misty gale]

Is that easily accessible?

Didnt think it was

Atlast not in bp

[grizzled hazel]

This is why things like fast paths are important for anims and so on

[misty gale]

Worked out some custom crap to handle it instead

[dawn loom]

ah I don't know about that

[dawn loom]

yeah I see what you're talking about

from a recovering performance maniac point of view I understand how doing something on begin play is technically at runtime

but I'd argue this would be so cheap it wouldn't even be too insane to do it multiple times per second

or even on tick

god forbids

[misty gale]

I use tick on a daily basis for ai 😄

[grizzled hazel]

You better not have any bp casts...

[dawn loom]

gasps

[grizzled hazel]

Tick on ai is fine just not in bp

[dawn loom]

btw I'm more of a C++ person than a BP person and if it was only up to me I'd have made my own state machine in C++ but I'm trying to be considerate of others

[ocean wren]

I thought this was a one man thing?

[dawn loom]

no I'm the only dev but there are 2 others on the team

[grizzled hazel]

Do it in cpp

[ocean wren]

DO IT 🙂

[dawn loom]

I won't tick anything

[ocean wren]

fuck the others, they hate you secretly anyway 🙂

screw those guys

[dawn loom]

I was thinking about making a BT with BTT/BTD/BTS in C++ but it looks like an absolute hellish pain to use those

[misty gale]

My tick is mostly for movement , for now

Its how i handle my 2k ai

Other methods go sluggish

[dawn loom]

so right now either I say fuck BTs, BBs, and may hell ensue

and do everything in C++

[ocean wren]

you can use BT's and BB's in C++ just fine

[dawn loom]

or try to be considerate and use BP (but probably make anything that is frequent enough in C++)

[misty gale]

Are they reviewing your code?

[dawn loom]

meh

[misty gale]

Peer review?

[dawn loom]

nah

they could move some stuff around in the BT if I made it simple enough

but they wouldn't understand anything if I make it in C++

[ocean wren]

But they wouldn't see the C++ would they

thats kind of the point

[dawn loom]

speaking of them I have a meeting on discord with them in 5 minutes

[grizzled hazel]

Do they need to?

[ocean wren]

all they see is the BT in the editor, not the C++ code underlying it

[dawn loom]

they might, I don't know if I'll stay in this company for long but I could get some shares so it's in my best interest to leave them something they could modify easily after me

oh you mean about BTTs in C++

[ocean wren]

I mean the BT nodes are all in C++ anyway

[dawn loom]

nah that's just for me

if I go full C++ I'll scrap the BT

BT nodes in C++ look painful to make

[ocean wren]

oh I get you.. not sure why you'd do that

its just deriving from a base class.. I don't see why you'd think that was painful

[dawn loom]

I mean the BTT_MoveTo which sounds simple as hell is like 400 lines in the .cpp

[ocean wren]

Because it isn't simple

navigating a path on a navmesh isn't simple

especially one with string pulling

and potentially collision avoidance

[dawn loom]

we're not talking about the actual AI move tho

it's only the task but well

[misty gale]

Clean code for your buddies (and others who might see it) would be like 80% cpp and 20% bp exposed if you want the best possible rep :p

Showing clear lines between the two 🥲

[dawn loom]

yeah that's what I do mostly but well I'm still kinda learning

[misty gale]

Same

So im more talk than game, if you catch my drift

[uncut python]

So ive been working on improving my level a ton and now my level is more advanced that my AI Director can handle, i want to play test to see what changes I need to make. I want to have hordes spawn procedurally along my level, but only spawning in in front of me, the frequency being determined by numerous variables. How can i define whats forward progress in my level and backwards progress using the navmesh

i currently have a system with EQS but the AI are spawning a bit more randomized than id like, Id like to somehow maybe make my grid so that it paths out along the remaining (linear) level that i havent traversed yet

[mossy nexus]

this entire discussion, summed up

[uncut python]

I was thinking maybe a spline path if i have to but id like it to be reliant on the navmesh itself

[misty gale]

How would one handle npc pickups? Would they register to the npc manager aswell, and any idle npc would be assigned to pick it up and deliver it to its destination?

[mossy nexus]

I'd do this as part of a point

as in

an actor that has animation etc. on it already

[uncut python]

Thats an example of a path that id follow

[mossy nexus]

you could put custom logic into that (or a custom component) that handles the pickup. the pickup could then have a tag or a variable stating type, and the AI could then have an overview over all spaces that would access that pickup type

[misty gale]

Its a resource in my scenario,

[mossy nexus]

right, same difference

[misty gale]

Transporting a resource from a production building to some designated storage area

But yeah its the same

Was planning on tag for the 'item' data, which uses a db to get the mesh and stats for thenitem

[mossy nexus]

if you have an AI director you could query AI director for high level things, such as spaces that accept your resource

anyway more complex than that and you're probably looking at an inventory system of some sort

[tardy talon]

Is this a good use of casting to avoid them sensing other enemies? Enemy class is quite light and should always be loaded when this is hit

[misty gale]

I do indeed. I did start on a second component for the buildings, so i can keep a clear line of responsibility

[tardy talon]

I know you can filter by using C++ but I dont know C++ hehe

[mossy nexus]

ehhhh... I'd have a set of targets

[crimson galleon]

I prefer tag but at least aip doesnt run on tick

[misty gale]

Buildings being any building, including storage areas and whatnot. Also makes it easier for high level player control of them, and collecting/showing stats and whatnot

[tardy talon]

How can you use tags for this?

[mossy nexus]

check tag against tag

[crimson galleon]

You check if the actor has the tag 😋

[mossy nexus]

sort of like faction

[tardy talon]

oh lol I thought there was a setting for filter by tag on the component

this makes more sense hehe

[mossy nexus]

I generally have a set of actors to look out for, and then check faction on those down the line

[tardy talon]

or well it doesnt there should be a blueprint setting for this >: (

[misty gale]

There is? 😶

[crimson galleon]

Put a tag on your enemybase then see if it has that tag

All actors already have a tag feature

[tardy talon]

No I mean there should be a setting on the perception component

like this

[crimson galleon]

Try breaking the stimulus

But i getcha there. Detection by affiliation sucks though - for bp

[tardy talon]

Is the get tag then branch more efficient than a cast? why use 2 nodes when 1 do trick

[mossy nexus]

yeah I mean this is what I'd use instead of tag. I'm not sure if the perception component handles targets as well though. like for instance, what if you have an enemy structure?

[crimson galleon]

I just said what i prefer and that cast is fine too 😁

[tardy talon]

What does that mean?

Haha fair enough thanks!

[crimson galleon]

Its an output value, below actor

[crimson galleon]

Think of tag this way though: maybe not all enemies of the same class should be sensed.... or maybe different classes need to be sensed.... all they need is the tag to be checked 😏

[tardy talon]

True true

[misty gale]

If only tag was gameplaytag

[tardy talon]

for now its fine like this but in a bigger project for sure

[crimson galleon]

You may not need to break the stimulus. I usually do to check if it is a successful result or not because aip fires when they spot and lose sight of anything detectable

[tardy talon]

good points

[celest python]

I dont remember exactly but iirc what UE was doing is pretty same already. Registers pathfinding reqs into an array, loops them, then dispatch to game thread

[misty gale]

I see

My stuff isnt time sensitive so im just doing one request per frame

Gotta add something extra for avoidance soon tho

[celest python]

Just to understand your mindset a little more; I still dont understand why did you use 'serialization' word at the beginning? I didnt study serialization features much so maybe I'm missing something 🤔

[misty gale]

It was a bad word to use, my bad :p

Not a native englisher (obviously), so was the first word that came to me when i wrote it

Calling it a queue was more accurate really

[ocean wren]

detection by affiliation is broken because you can't have an affiliation for the player, which is just bizarre

[mossy nexus]

good to know

once again seems that I made the right choice in making custom perception 🦾

[ocean wren]

Its not broken in general, its just that the component is added to the AIController and not on the controller class where it should be

or rather, on the pawn side of things, instead of the controller

not typing well tonight 🙂

[mossy nexus]

so only fixable incpp?

[ocean wren]

Yeah, this is a case where I'd do it myself in C++

Using an interface

Then you could have a BP version too

its one of the downsides of inheritance trees, if you get it in the wrong part of the tree you get it wrong, which they've done

[uncut python]

why does my ai keep getting stuck like this? when i initiate them to chase me if theyre in a crowd they just rotate towards me and not move

is there a collision settings i need to fix?

[crimson galleon]

you got a navmesh built?

[uncut python]

yes

they work up until i melee them, they fall down and get back up and are supposed to keep running

I can get some more screen shots when i get back from work

[crimson galleon]

so they move initially but not after some event?

[uncut python]

correct, the behavior tree is also showing the correct info, its retriggering the move to after the melee fall event occurs

anbd theyre rotating towards me in what it looks like an attempts to come after me, but they stutter a little and wont move

[crimson galleon]

sounds like the fall event is changing something then....?

could they be ping ponging between two BT states/branches?

[uncut python]

ive checked the bt and it appears to the run the same way before and after

I also notice something similar near door ways they start to bunch up with eachother and just stand there ujable to move after no matter my position

like 1 will go thru the door fine then 2 come and they swarm up after 1 gets caight

[crimson galleon]

so will one Ai alone work without problems entirely?

[uncut python]

Yes

[crimson galleon]

then maybe multiple are messing each other up by affecting the navmesh area

[uncut python]

Possibly, I tried pulling up AI debugger but I couldn’t find anything to show that, is there a way to visualize the navmesh area

[crimson galleon]

console command: show navigation

[uncut python]

Sweet thanks I’ll try it when I get home

Thank you

[uncut python]

I also forgot to mention I’m using Detour Crowd Ai controller does that have any affect on this?

[ember agate]

Hello, i made a simple character and on event play i have "ai move to" but its not working, it was working but i did some changes to map now it stopped working, i also deleted navmesh bounds and re added them but still not moving, move to location or actor works so the character can move

[misty gale]

Navmesh move it

[ember agate]

If you mean navmesh bounds it doesnt work if i move it

[misty gale]

Was worth a shot making it rebuild

Been known to work more than a few times before

Does the ai move to respond with failed?

[ember agate]

nope, it doesnt fire the on success or failed

the simple move to location works too

[misty gale]

Does it execute at all? Got a screenshot?

[ember agate]

Weird, plugging "self" into pawn input worked, but it i tried that before and it didnt work, also i dont remember i had it when it was working the first time

[uncut python]

So i enabled the nav mesh while in PIE and i still cant see anythign wrong, but here is a video of whats happenign

[crimson galleon]

And your BT?

[uncut python]

i found it i beleive

I was disabling collision on the capsule during the animation so i can walk through the zombies at least until a proper stumble back animation

when i turn back on collision once the anim is over they mustve slightlky clipped into eachother and caused an overlap. Seems to work for now but ill keep an eye on it

thank you

[crimson galleon]

trippy

[celest python]

Anyone knows where exactly previous (or current) velocity get zeroed when a new path requested for AI?

When following player, AI stops for a second before moving to another player and I want to prevent this

Nvm already found it (ApplyRequestedMove() on CMC)

[midnight scroll]

@celest python
#gameplay-ai message

Oh. didn't see you found it. 😄

[celest python]

Thanks though 🙂

[midnight scroll]

Couple different things affect that apparently.

[tardy talon]

AI behaviour trees are the most unintuitive thing I've come across so far 😭

[pine steeple]

i like them

easy to use once you learn them

[celest python]

Avoidance Time is way cheaper than I expected

Interesting

Its literally nothing but 0.02ms on 30 CMC

[misty gale]

what does that include

[celest python]

The avoidance?

Its bUseRVOAvoidance in CMC

[misty gale]

yeah "AvoidanceTime"

does that include all things related to avoidance?

guess i gotta dive into it and see what it does

[celest python]

It's UAvoidanceManager's processing time

Anything related to avoidance yes

[misty gale]

Ah Alright cool

still to expensive 😄

[celest python]

It's beyond my programming skills 😄

Pretty sure Epic also got it from somewhere

Does not look like Epic's work

[misty gale]

probably beyond mine too, so it'll be interesting to see what i can come up with that's dirt cheap, and works sufficiently

[celest python]

It kinda breaks the movement

Controller Yaw and Avoidance velocity direction does not seem to be matching

Pawn looks at the target goal but moves to right without rotating

[misty gale]

Yeah i've seen that , almost as if someone just did linetrace forward and if hit pawn add right vector location offset x)

I'll probably use some combination of some crowdmanagement along with better pathfinding

[celest python]

CrowdManager is too expensive though

Barely can handle 30 pawns

[misty gale]

custom implementation of one i guess

i know there's stuff for predicting existing paths and collision time etc

so that'll probably be the main workload

so the cost will be during pathfinding, which is fine in my case

then some simple movement handling for when roads are too crowded etc

[celest python]

It might be nice if it was multithreaded

Could lower the overhead almost for free

[misty gale]

yeah that'd help a ton

[tardy talon]

Just seems so much more effort and you gotta learn what dexorators and service and blackboard Keys and when I want to communicatie with the BT I need to set up a key, then referenxe that key in another blueprint to then implement it in my BT it seems so much to try and connect to eachother

But yeah should get easier with practise

[worthy python]

Hello. Where can I find an example of this visibility algorithm? I want separate the safe and danger zones.

[celest python]

Been messing with this for a while, I just realized even Back 4 Blood using the default avoidance manager. I thought they had a custom flocking system 🤔

[ocean wren]

RVO was an academic paper called Reciprocal Velocity Obstacles for Real-Time Multi-Agent navigation by Jur Van Den Berg: https://gamma.cs.unc.edu/RVO/

and the associated library

from back in 2008, so there's likely a ton of better approaches now

There's a chapter from one of the Game AI pro books by Ben Sunshine-Hill about that and ORCA (another approach): http://www.gameaipro.com/GameAIPro3/GameAIPro3_Chapter19_RVO_and_ORCA_How_They_Really_Work.pdf

I'm not a huge fan of RVO's 🙂

just feels wrong to treat paths as geometric constraints to me

[celest python]

I did not like it either

It's too aggressive and dominant over the movement component

Pawns literally slide

And somehow they manage to slide towards back

[misty gale]

I wonder what AutoStore ™️ uses for their pathing for robots in the warehouse

probably some simplified stuff since its really just a 2d grid

Along with a relatively limited amount of agents

[ocean wren]

I don't think the sliding is the main problem, its more that there's no giving way to other people are any of that social stuff we tend to do unless there's a huge crowd

humans are surprisingly pliable 🙂

RVO's are just too rigid

[mossy nexus]

british people after a night out drinking, 2022, colorized

[celest python]

lmao 😄

[celest python]

I think sliding is a big problem 😅

[ocean wren]

You can cover sliding by having foot placement take care of it..

its only a problem because the sliding isn't natural

people don't slide left/right, but they could step left or right and it'll look better

[celest python]

But shouldnt it accelerate towards to somewhere instead of sliding at the first place? 🤔

Sliding towards right-front or left-back for example is okay but it looks like they directly slide to right and left

Which makes it even more un-natural

[ocean wren]

Well, sliding is just moving away isn't it.. you can cover that in the animation system

but yeah, the main issue is just that people wouldn't do it anyway

they'd avoid any of that in the first place if they could

it looks weird because its got rigid colliders

[celest python]

That's a "maybe" for my project 😦 I have ~100 of pawns running around and most of the optimization comes from reducing anim BP complexity and movement component

[ocean wren]

I'm sure if you switched the rigid bodies off even that RVO would look better

[mossy nexus]

have you asked GB about his instanced skelmesh thing? maybe you could coax him to give it out to you

[celest python]

Last time I checked GB was not willing to share it, though asking it like that wouldnt be nice 😅 I have a friend in the industry helping me create a new instanced SK currently. Its not working great or nowhere near to finished but I'm hoping we will get to some where in couple months

My only concern is we wont be able to guarantee it's going to work on other platforms so if something goes wrong I wont be able to fix it by myself and my friend helping me doesnt have a UE background at all to fix things

[tardy talon]

My prayers have been answered, they just added a free AI plugin to the marketplace collection so maybe thats gonna be easier for me

^what that guy said

[tranquil robin]

So I have AIMoveTo and it seems to work sort of but it doesn't quite arrive at the target actor I set to it move to( Which has no collisions). But depending the direction where it aproachs, the movement always stops slightly before getting there. Think of if the target object goal would be a center of the clock, instead of navigating to the absolute middle, the AI stops at the number circle, it just falls shot. Not sure where to control it so that I would go exactly the middle all the time

[silent ivy]

Join the Discord if you have questions 🙂

[uncut python]

Woah

[silent ivy]

Ahh i feel it, good luck with whatever you're working on!

[mossy nexus]

check collision view if something is blocking it. contrary to popular belief, collision is invisible

[ocean wren]

"Blueprints" and "advanced" don't go together 🙂

[fiery skiff]

How do I get input info from an AI? I'm trying to get my turn in place animation to run by using Get Turn but it doesnt work on an AI because there is no data coming from the source that Get Turn reads from

[pine steeple]

75 blueprints though

maybe someone should re-write those blueprints into fast functioning c++

[cobalt palm]

Hello, I have an AI that moves to a spot in line, however if an AI leaves the line it tells all the AI that are either in line/moving to the line to stop their movement and MoveTo an updated spot in line.

However, what I have found is that the AI will move to the spot they are told to move to, then they move to the old spot they were originally going.

I used the Stop Active Movement function but am realizing that it doesnt make the AI forget about the orignal location but rather just stops the AI for a split second. Is there a way to have the ai stop doing the MoveTo function?

Ive also tried the StopMovement function via the AI controller but that did not work.

[mossy nexus]

you need some sort of manager to tell the AI to move to the next spot when one opens up

[cobalt palm]

That's what I have. The problem is when the "manager" tells the AI to move to a new spot while its already going to a spot, it goes to the new spot, then goes back to the original spot it was going to.

[mossy nexus]

sounds like you need to check your logic then

to me this looks like you're not changing the target

[cobalt palm]

I know the stop movement is working as I can see the AI stop for a split second, however I need it to abort moving to that spot and instead move to the new one. Its almost as if the AI MoveTo functions are stacking in their own queue.

[mossy nexus]

I don't know why you're using stop movement

just change the target moveto

you're trying to omit a bug instead of fixing it

or at least that's how it sounds

[cobalt palm]

It seems as though I used the normal MoveTo function. So I instead am using the AIMoveTo and it seems to have worked. Well see. Thanks though.

[warped robin]

Hello everyone! I'm working with BT's right now and I've put together a composite decorator, which seems really cool. BUT, Is there not a way to set an observer abort on a composite? you can do it internally but that doesn't work because I'm AND'ing conditions together, I only want to abort if both conditions are true (for example)

[mossy nexus]

the issue is not which function you use but how your architecture is setup

[mossy nexus]

unfortunately not, as far as I've seen

[warped robin]

noooooooooooo.gif

thank you though. I guess I'll make a copy and add that

[mossy nexus]

yeah

I was thinking the same

[tranquil robin]

Thanks, there isn't anything else since it is in empty 3D space. The moving actor is just empty, as in there's no mesh or something that's corner could arrive to the end first and then technically register it there. The agent size is still at -1. On BT the move to radius is 5 but it fall sort more like 100-200 units. Maybe it is just some sort of local space world space issue, gonna look into that next

[mossy nexus]

did you check the navmesh?

[tranquil robin]

this area nav mesh has no collision

[mossy nexus]

what do you mean no collision?

[tranquil robin]

maybe i answered wrong question... i have built the data for navigation and this area has no visualized data that would make the AI stop or go around it

[celest python]

Also check visual logger

[viscid oasis]

Anyway to vary up the MoveTo for AI? They're followeing the exact same path and I'd like to add some variety.

[celest python]

Try RVO avoidance and adding offsets to target goal

[viscid oasis]

okay, I'll look into that. Thanks @celest python

[tranquil robin]

I have empty actor as target to fly to, and another empty actor as AI flying there, and still the AI doesn't reach the position of the target actor. Is it possible that the movement stops just because their bounding boxes are toucing? Even though there's nothing inside with bounds preview I still see blue cube and yellow sphere on both actors, and visually it would make about the distance where the stopping happens

[viscid oasis]

Sounds like it. What's the acceptance critiera?

[lyric flint]

I need my AI to be able to pathfind close to walls. There appears to be a hard cap on navmesh generation at like 35-40 units. My current solution is to scale everything up twice as big but even then, there's still a pretty big gap between the walls and navmesh

https://cdn.discordapp.com/attachments/785317301557657611/928142967309492234/unknown.png

I have some smaller entities that I want to pathfind close to walls too, so this is a pretty frustrating limitation in UE4 and 5

Every item in the navmesh generation settings is minimized

problem solved, never seen this in my life but apparently it's automatically added and overwrites some of the settings you can enter in the project settings.

[crimson galleon]

adjust navmesh agent radius

[lyric flint]

@crimson galleon where?

[tranquil robin]

How do I change this? I can't find any setting in any details panel that sounds like something like this.

To be honest I don't even care about it flying to other actor location, that actor is there just to mark the spot since I don't know how on the blackboard keys I could define a world location. If I make a vector key with 0,0,0 the AI pawn just fly to seemingly random direction and never stops

Also I'm not calling AIMoveTo node manually, it just happens on the BT so it seems I don't have access to what the node would have on regular blueprint

[viscid oasis]

I'm referring to the AIMoveTo - but that doesn't seem to be the core issue you're having.

navmesh agent radius can be found via the CMC.

[tranquil robin]

The radius is at -1.. these are all defaults. I had it on bigger values before and originally thought that would be the culprit but no

[viscid oasis]

hmm sorry not sure

[tranquil robin]

Thanks anyways, I'll keep looking

[ocean wren]

Aren't you making a space sim? AIMoveTo doesn't work for agents that aren't on a navmesh

[tranquil robin]

I have 3D navmesh now

It works, circles around objects, or flies through (if not built into the data) but for some reason the movement to arrive at target always fails short from any direction

[ocean wren]

Usually, that's because it has a value on the movement to allow it to stop near something. On the regular AIMoveTo there's a value that is "how close do I need to get to be near it"

can't remember what the value is called in the task, but it should be obvious if you read the docs

[tranquil robin]

So on BT the acceptable radius is 0

[ocean wren]

You really don't want 0 btw.. have some small value but not 0

[tranquil robin]

It feels and looks like what you say is correct, I just can't find where it actually allows this result to happen. Yeah I have it at 5 now which is the default value, but same happens with 0

seems like it always falls short and visually it feels like it is pretty much the distance between as if bounding boxes of those empty actors were touching, could be coincidence

[viscid oasis]

hmm I had something similar ages ago. I'm trying to remember the situation. Basically the actor would detect how far away it was so it'd move closer It was outside the acceptance radius until it was close enough but it wasn't close enough to trigger the overlapping or whatever needed to return the MoveTo successfully, so it'd get caught in this loop.

It was a few years ago. Drove me nuts. I had to manually do the MoveTo and do a HitTrace on it (or something like that) .