gameplay-ai

[jaunty peak]

Ryans tutorials are decent. That's where I started my AI stuff. 🙂

[desert moat]

I assumed the AI would continue to move to the player if it didn't have anything in there. As I haven't yet added the attack animation as I thought I'd need to resolve this first

[jaunty peak]

That's the trouble with that kind of set up, you'd need to continually check while in the collision volume. Once it's reached the player then that line of code is complete and stops.

the location of the zombie isn't being told to change.

[desert moat]

I see so in theory if I did continue using this method I'd need to setup a while loop and a second collision box to detect when the AI is in a close radius to the break the loop and perform the attack animation.

[jaunty peak]

Doing loops like that will likely need to more issues.

[desert moat]

So you suggest I reattempt Ryan's tutorial?

[jaunty peak]

Yup, I can't remember if I ran into any particular issues with his method. Sometimes there are bugs that he doesn't notice but deals with them in the next episode etc.

So I'd stick with try BT's, once you get the hang of them you'll see the potential of them.

[desert moat]

Okay I'll give it another attempt, last time I got the end of the video after wiring up the random move function and my AI just literally done nothing 😂

[jaunty peak]

If you get really stuck with them, pop into this channel - someone should be able to point you in the right direction

[desert moat]

Anyone know how I can update my animation here. Usually I'd connect this up with a get animation instance then connect it to the mesh on the AI BP. But not sure how to get this for my BT task?

[misty wharf]

I guess you could probably cast the pawn to your pawn class, then grab the mesh from it, and then get the animation

[jaunty peak]

If your animation BP is set up, you shouldn't need to tell it how to move manually. You can also change the max walk speed of the AI if you want to control how fast it actually moves.

[desert moat]

I did have my animation BP set up like that but couldn't perform the scream animation before the AI moves. Which is why I changed it to use a Enum to control which animation should be playing

[jaunty peak]

have a boolean that controls the anim to go into moving state. So it won't move if it hasn't screamed. Let's it also reset it so it has to scream again after a period of time for example.

You can also have distance checks and all that fun stuff inside the BTs for that too.

[desert moat]

That's exactly what I'm looking to do

[stiff gale]

So I'm using ReportNoiseEvent to make noise, but my AIPerception is not picking the noise.

[ocean wren]

Is it in range?

If I recall, there's a distance check done in the sound sense implementation to stop them sensing all sounds

[stiff gale]

I'm reporting noise for 5000 radius

in perception i have 5000 as well

https://gyazo.com/c260914f2d8176b6ee2f2e66fb5071eb
https://gyazo.com/ee659b6c6a0ad055cff78005e611dfc1

[uncut rune]

Does the actor that is producing the noise have an AIPerceptionStimuliSource?

[stiff gale]

No what is that?

[uncut rune]

Its a component that registers the actor as a stimuli source in the perception system. This is what I did for my player character and all the senses work fine on my end:

[stiff gale]

Should i put this in OnPossess of controller or else where. Also would this add the hearing sense to my perception comp.?

    {
        AIPerceptionComponent->ConfigureSense(*SensorConfig_Hearing);
        AIPerceptionComponent->OnTargetPerceptionUpdated.AddDynamic(this, &ANPC_AIController::TargetPerceptionUpdated);
    }```

[uncut rune]

yeah, make sure both the perception component and the stimuli source both have the hearing sense. I have the OnTargetPerceptionUpdated in the begin play function in the controller

I dont call the configure sense function

[stiff gale]

Ii got it thanks

https://github.com/orfeasel/UE4-Cpp-Tutorials/blob/master/EQS_Contexts/MyAIController.cpp

[desert moat]

Anyone know how to check for damage in a behavior tree?

[misty wharf]

Kinda depends on what you mean specifically

In one project I have the ai controller listen to damage events on its pawn and set a bool on the BB when damage is taken, which the BT can then react to

it seems to work ok at least for purposes of that one

[pine steeple]

i use damage perception event

bt decorators/eqs can read this information quite easilty

[desert moat]

That is what I had tbf but I wanted to get a damage taken and death animation working, but I don't know how I can get the health over to the BT so the montage can play and not conflict with the AI thinking it should still be moving to the player/attacking

[misty wharf]

Yeah you just need to choose how you want to handle it

You can for example set it to a dead state which you communicate somehow through setting a value on blackboard, or using a decorator which aborts the main logic in the BT

@stiff gale I guess you just figured that delegate bind issue? :) Was just about to say the signature for it uses a const ref TArray

[desert moat]

So I think I've got my BT setup correctly as far as I know, I am new to BTs. Just not sure where I go from here

[misty wharf]

Yeah something like that. You can set the Is Not Dead? one to abort self if it changes

This would allow it to exit the main processing if the IsDead value becomes true

[stiff gale]

That didn't work either.

By the way , do you know why this won't get triggered?

void ANPC_AIController::TargetPerceptionUpdated(AActor* Actor, FAIStimulus Stimulus)
{
    UE_LOG(LogTemp, Warning, TEXT("................. TargetPerceptionUpdated"));
}

[misty wharf]

🤔

Are the senses configured correctly?

iirc the team settings need changing or it won't perceive anything whatsoever by default

[stiff gale]

Yea via blueprint it works

But in c++ the log won't show or breakpoint won't hit

[misty wharf]

No idea tbh, that's the same event which is used from BP's, so if it compiles it seems like it should work

[desert moat]

I've got that all setup, I think I just need to work out how to set them from the AI_BP

[stiff gale]

Yea got it fixed. I had to include UFUNCTION()

lol silly thing

[misty wharf]

Oh, lacking that usually causes it to crash immediately on call :D

[misty wharf]

you can set them from the AI controller pretty easily via Get Blackboard which then has a bunch of nodes for setting them

if you need to communicate the info from the pawn to the controller, the easiest way would probably be to use an event dispatcher which you bind an event to in the controller

[desert moat]

Okay so I gave this a try, and I'm not having any luck I must be doing something wrong. Any ideas? https://youtu.be/-r_GfJYcqkY

[desert moat]

I'm sure this is where my issue is, but I just don't know why

[pine steeple]

your player character does not have blackboard tho

and that is really weird. i would go follow some more UE4 AI tutorials.

use GetController -> Cast AIController -> Get Blackboard

[queen escarp]

https://www.youtube.com/watch?v=7PaN_7UJrA8

[bright rune]

my AI sometimes just walks out of navmesh and gets stuck. is there something to restrict character's movement, so it can't get out of navmesh? or some setting to make it more 'accurate'?

[cosmic stag]

How exactly does ProjectPointToNavigation work? i had thought that QueryExtent meant: this is how far around the point it looks for a navigable point. But even with a 0,0,0 extent, it tends to succeed even for an object up in the air, not on navmesh

[glossy spire]

Hey folks, hopefully there's a BT GrandMaster that can explain something to me. I have a sub BT that looks at first glance like it's running, and is labeled simulating, but is actually somewhat locked up, unable to actually run tasks or evaluate decorators anymore.

B) Isn't actually running the sub tree. It's stuck and not doing anything until Combo Range or Seen Player decorators become true.

The bottom decorator is going from true to false, and then the selector is going from the wait task to my test sub tree

and it is locked there until the first decorator goes to true again

Seen Player Recently

When I put both decorators (In Combo Range + Seen Player) together, at either level, it works fine as expected

This works fine

This only appears to break Run Behavior sub trees, tasks seem to be fine

Is this some kind of bug, or am I using this incorrectly?

BTW my test tree is just waiting and printing, nothing unusual.

[glossy spire]

It seems like Observer aborts: Lower Priority causes issues with Run Behavior

[stone walrus]

hey guys i am making an ai for ghost character and using ai perception and i set up ai stimuli perception in my character and for object that i over lap with and they can fall and can produce sound . i set ai stimuli source for it also but first it works fine i go to the place where object falls and after then that it start getting location even he dont see me he reaches up to my place . my question how can i differ between my location and object after on target perception updated ... please anyone can tell whats the problem ??

[turbid timber]

Excited to announce my startup WildMeta's new library and Unity/UE integrations to help speed up the adoption of machine learning bots in video games! Thought it might interest some folks here (more techy post from my co-founder also on our blog) https://www.linkedin.com/posts/amandineflachs_wildmeta-unveils-its-new-library-compatible-activity-6840979426063216640-GKro

[slim owl]

Small question about AI, once a BT Servrice calculate a position and the MoveTo say to the AI to move to it's position... Does it wait for the AI to reach the position, or it Tick each 0.4s and update the position each time?

[stiff gale]

I think it stays on MoveTo until it reached its location.

[celest python]

Anyone knows what are pawn actions?

[misty wharf]

A mystery of undocumented code

I've tried to figure them out on a couple of occasions but I can't really tell if they exist or are half-finished or what

[celest python]

lol

[balmy silo]

I want this guy to walk towards me when he sees me. I have some code setup and when he sees me currently he plays the walking animation but doesnt actually move towards me

[celest python]

Feels like GameplayTasks but its prototype version

[balmy silo]

this is my code

[misty wharf]

@balmy silo add some print nodes to the success and fail pins and see whether it is failing

I don't know if you've got a navmesh, but the ai moveto stuff needs one

[balmy silo]

i do have one

[balmy silo]

It succesfully sees the player

[misty wharf]

yes but does it fail the move is the question

which is why adding prints from the success and fail pins would help

[balmy silo]

its failing

[misty wharf]

okay, try printing the result value

[balmy silo]

How do I do that?

the movement result

[misty wharf]

just drag from the pin to the print node, it should automatically be able to convert it I think

[balmy silo]

to fail?

[misty wharf]

yeah the print from the fail pin, since it would go into that one

[balmy silo]

its saying aborted

[misty wharf]

hmm, is the player position actually reachable via the navmesh?

[balmy silo]

it covers the whole map

so i guess

wait

wtf

i just did something

[misty wharf]

oh?

[balmy silo]

hmm

i moved the background back

and it started working

but it says failed

but he moved kinda close to me

this is so weird man

[misty wharf]

Have you looked at what the navmesh looks like with the visualization for it?

[balmy silo]

yes

[misty wharf]

Okay, just a bit hard to tell from the screenshot what the terrain is like :)

[balmy silo]

its not the navmesh i think

[misty wharf]

if they don't move to where they should it's usually some issue with the navmesh

or they collide with something which gets in the way

[balmy silo]

i think i fixed it

but hes floating

when he sees me

[misty wharf]

Collider too large?

[balmy silo]

sending video

[misty wharf]

if the circle that's visible at the very end of the video is a collider then it seems that would do it

since it's just about big enough to lift it up the amount that's visible in the video

[balmy silo]

ohhh

that did it

thanks

[misty wharf]

np

[celest python]

lmao why there isnt any line warping for decorators 😂

w i d e

[misty wharf]

lol yeah I've noticed this as well

iirc there's a way to hide those details from them but don't remember exactly what it was

[celest python]

Can you let me know if you ever come up to it?

All of my actions rely on tags and.. thats just too w i d e to work with 😂

[misty wharf]

untick show property details

that hides them

[celest python]

Couldnt find

Where should I search for it?

[misty wharf]

click the node, it should be in the details panel for it

[celest python]

There isnt one on me

[misty wharf]

Hum maybe it's not available on decorators 🤔 It shows up on task nodes

I guess you would have to create a custom decorator in C++ because iirc it's possible to override how it's displayed there

[celest python]

Alright, thanks for the help 🙂

[obsidian lily]

How would I go about checking if an AI actor is next to the Player character so it will trigger a sequence?

I'm still super new to using the behaviour tree and this is what I have so far when it comes to having the players dog companion follow them when they are visible my only main issue is if the player is too close to them or even slightly leaves sight range it will just start roaming.

[orchid echo]

Is there a way to set blackboard keys outside tasks?
I would need to set an blackboard key enum indicating what state the actor is in.
For example, using the default MoveTo -node would set the enum key depending what kind of task the actor is doing

Then again it wouldn't be much of a trouble creating a task for that purpose

[misty wharf]

Sure, you just need to get the blackboard and then call one of the functions that set values

[orchid echo]

It would have been nifty to be able to set it directly in the behaviour tree without extra function calls or tasks

But thanks

[misty wharf]

Ah

Yeah they don't have any builtins for doing it in the BT

[orchid echo]

That would be pretty cool addition to behavior trees

[misty wharf]

I have a custom SetBool task to do that for bools at least :)

[orchid echo]

Instead of a task I used a decorator and it seems to work just fine for this purpose. Neat stuff 🙂

[ocean wren]

Please don't start adding states to behaviour trees. The whole point of BT's was to get away from the limitations of state machines in the first place 🙂

[orchid echo]

I'm using it for debugging to see is the actor doing what it's suppose to do. Not for anything else 😄

And this seemed to be the easiest way to do it

[obsidian lily]

is this a bad way to go about this?

im just trying to figure out how to stop the ai from sticking near the player even when they are not within sight

but also i want it so that the player can be within a certain distance and also out of sight with the dog just sticking near them

[glossy spire]

Anyone know a way to get some kind of ID for a task or decorator? I'm working on serialization and need to know which particular task I'm on so that I can load it correctly.

[ocean wren]

Maccap: You're getting the distance between the AI Controller and the Player? Why not set a key on the blackboard for the player and use that instead of relying on the GetPlayerCharacter, you've hardcoded the player. What if you want the AI to fight other AI?

Johnl: https://docs.unrealengine.com/4.26/en-US/API/Runtime/CoreUObject/UObject/UObjectBase/GetUniqueID/

https://docs.unrealengine.com/4.26/en-US/API/Runtime/CoreUObject/UObject/FUniqueObjectGuid/

[glossy spire]

Thanks! I was looking at GUID, but wasn't really sure what its all about

[ocean wren]

GUID = globally unique identifier, pretty common system to generate unique ID's for things

[glossy spire]

Are those generally safe for serialization or do they reset every time you run the program?

[ocean wren]

Used in windows stuff too, although I doubt they use the same algorithms

Reading that doc, suggests they're only unique for the life of an object and get reused.. so you'd probably generate a unique one at serialization time

Basically what you'd need to do, is store a map of GUID's to serialized entries and fixup the GUIDs on serialize/deserialize

[glossy spire]

Hmm interesting

okay, that seems like just what I was looking for

thanks!

[ocean wren]

https://docs.unrealengine.com/4.27/en-US/API/Runtime/Core/Misc/FGuid/

If I remember, I used FGUID's for my nodes..

It has serialization save/restore and whatnot

And turns to a string, or saves to JSON, which was my use case.

[glacial sapphire]

I don't know if this fits in animation, but I'm doing an AI where it needs to turn towards me to run. The flip animation goes to 180 degrees, but I want it no matter where I'm at the animation always flips to me and not 180 degrees. Can someone help me?

[lapis sigil]

are you all using the ue4 behavior trees?

i mean i dont

[glacial sapphire]

yes

[lapis sigil]

yeah well

i code a new ai for my game

im curious about ppl who do as well

that ai is more complicated compared to btrees

[glossy spire]

@glacial sapphire I have a similar problem. I've tried using a blend space to play the turn animation, reducing the alpha as it gets close to the correct angle, but I've gotten mediocre results. The best I've gotten is to only play the fast 180 turn when I know for sure the target is behind the character, like for example if they've just been shoved away by the player.

If you find any good techniques I'd love to hear about them!

[ocean wren]

There's nothing fundamentally wrong with UE behaviour trees, but they're not the only solution. Honestly doing your own in BP is probably the worst idea though if that's what you're thinking 🙂

[shadow python]

how one would use FGenericTeamId ? I have team settings based on game state with own relationship test functions
I've implemented IGenericTeamAgentInterface::GetTeamAttitudeTowards which does tests between actors but other code still relies on statics

AttitudeSolverImpl is static, so i can't set it to something pointing to game state

[shadow python]

IGenericTeamAgentInterface looks like an ancient legacy tbh

[misty wharf]

@shadow python override GetGenericTeamId to have the actor return some team ID, then use GetTeamAttitudeTowards to determine the attitude of actor towards another actor. It may be simpler to just create a FAttitudeSolverFunction since this is used by default and is team-based rather than actor based, this is done using FGenericTeamId::SetAttitudeSolver

[shadow python]

i do set solver with lambda in game state postinitcomponents and reset on endplay

the problem is the complex setup

[misty wharf]

Oh?

[shadow python]

players = playercontroller / player primary pawn type + number sattelites
bots = aicontroller / ai primary pawn type (almost same as player) + number satellites
satellites = aicontroller / sattelite pawn type

[misty wharf]

well you should be able to set it up on the pawns, have them return an appropriate team ID from GetGenericTeamId

that's all you need on those

unless you have some ability to change their team IDs

[shadow python]

i did implement GetGenericTeamId in my Playercontroller (with field)
for pawns get generic team id does owner chain traverse to lookup for controller and calls it

with owners go in chain like this ai/player controller > ai/player pawn > sattelite pawn

[misty wharf]

so what's the problem there? It doesn't seem that complicated

[shadow python]

yeah, was concerned if it was right way to do it

seems like it is

[misty wharf]

Yeah, as far as I know that's how it works

the attitude solver is used by default and seems the easiest way to just define team attitudes, and the actor-specific overridable GetTeamAttitudeTowards thing can be used if you need actor-specific overrides of that logic

[shadow python]

i noticed it calls getattitude on actors which often are pawns, so pawns have to implement it and take info from controller which knows about them

[misty wharf]

Yeah, it might call them based on what actor is being perceived

haven't looked at it in that much detail, seems to work on pawn... my implementation is extremely simple though, it just returns everything as hostile vs player's team lol

[shadow python]

yeah, mine a bit too complex with two different game modes that manage relations

with relation data stored in game state the AttitudeSolver calls GameState method to get the relation

[misty wharf]

ah

sounds reasonable

[shadow python]

and now my primary pawn class has 11 interfaces on it 😄

[misty wharf]

:D

[shadow python]

secondary 7

[misty wharf]

that's sounds like a lot, how many of them are engine interfaces?

[shadow python]

4 custom rest engine

[misty wharf]

well, sounds like a lot just as numerically, it's probably not ilke a bad thing if it makes sense architecture wise

interesting, the ones I recall implementing are GenericTeamAgent, GameplayTagAsset and PerceptionSomethingOrOther I think

[shadow python]

the only problem that sometimes placed pawns loose the owner and when loading level sattelites have no valid owner which is super confusing to solve

[misty wharf]

I've noticed the order of beginplays and possesses doesn't seem predictable so yeah

[shadow python]

so far hardest thing is from sattelite pawns reach to real owner that knows about teams

[stiff gale]

How can you improve making noise to notify other ai that an npc has a threat? Idk if I should use appear radius to warn other ai?

[ocean wren]

There was some issue I came across with the initialization order of TeamID because they'd not inherited the TeamID interface in the right place (it was on the AIController and not on Actor, so if the player had a team you had to fudge it. Also the teamid value was set after some particular use case I can't remember.

[misty wharf]

Depends entirely on how you want it to work. One of my projects has enemies just do a sphere check for nearby AI's if they spot the player

[ocean wren]

I've used a notification message in a messagequeue to the squad to notify the squadmembers of the new threat. The basic idea is that the squadmember has alerted and shouted to squadmates, so the logic makes sense.

Definitely recommend looking at message queue's for that kind of thing

[stiff gale]

So I have civilian and guards when civilian get attacked it uses make noise to send tags. Guards hear the noise with that tag then defend the civilian it only works if the guards are close by. Increasing loudness and noise radius didn’t help. .

This seems to be the proper way of doing this. Alternatively I could use sphere channel.

[shadow python]

also it is a bit weird. SetGenericTeamId / TeamID is a property of AI controller with no way to set it per-pawn

and i haven't found a single use of it outside of FAITextSpawnInfo

[misty wharf]

I don't recall how exactly the noise sense works, but it sounds like it could be an issue with how you have it configured if the noise itself should have sufficient range

[stiff gale]

Make noise function has radius and loudness. Isn’t that it? Noise sense has hearing range.

[misty wharf]

Probably yeah

[acoustic minnow]

Hey guys, I am having issues binding a delegate for my AIController where it’s not getting called. Does it look like I am missing anything? It’s worth noting that the PerceptionUpdate works on BP so it is detecting the player so it’s my delegate that’s not working

//In constructor
PerceptionComp->OnPerceptionUpdated.AddDynamic(this, &ABasicAIController::OnPawnDetected);

//function for delegate to call
void ABasicAIController::OnPawnDetected(const TArray<AActor*>& DetectedPawns)
{
GEngine->AddOnScreenDebugMessage(-1, 1.0f, FColor::Blue, TEXT("Message"));
}

[simple crest]

Never ever bind delegates in constructor, You'll wreck your blueprint

[acoustic minnow]

I also tried it in BeginPlay it didn’t work there either

But that’s definitely good to know

[simple crest]

Maybe try it in begin play with a brand new blueprint, and you're not hot reloading are you?

[acoustic minnow]

No hot reloading

I can try making a new BP

[misty wharf]

iirc someone else was having problems binding delegates into OnPerceptionUpdated as well... not sure if they solved it or not, but it was the same that it works fine in BP's

[acoustic minnow]

Making a new BP didn’t seem to do anything

Yeah it’s weird I must be missing something just not sure what

[acoustic minnow]

Update it’s working, in case anyone else has this issue you need the function you are calling to be a UFUNCTION 😅

[misty wharf]

Now that you mentioned it...

He had the same problem lol

Which is a bit weird, because in my experience attempting to call a non-ufunction from delegates causes an error, but maybe it's just a warning in Rider, not sure 🤔

[simple crest]

IIRC you will get an ensure in the editor, it's something that should have been obvious to bring up

at least I feel like it's usually quite hard to ignore that big freeze-up and big red message in the editor output

[misty wharf]

yeah

[acoustic minnow]

Strange, I didn’t get any sort of warning. I assume it was the same for that other guy

[misty wharf]

Yeah must've been

If it trips an ensure as HoJo suggested, it would usually freeze the editor quite noticeably for like 10-20 seconds

[formal lichen]

Hello, I was wondering how I would get my AI to only aggro onto NPCs within the same area of the map as it. Even if it doesn't have a line of sight on it. However, I still want the AI to wander into different areas and only aggro onto the NPCs residing in the new area. Thanks in advance.

[hardy sable]

Hey, so I was wondering if anyone could help me with this. I have large open world map, and I have setup a nav mesh to cover all of it with the 'generate navigation only around maviagtion invokers' setting ticked but I am finding that if the character is far above the bottom of the nav mesh bounds volume (but still within the volume) then it doesn't generate any path for the actor to follow. increasing 'vertical deviation from ground' helps, but it doesn't reach high enough.

to clarify, I am talking about mountains vs the ground

also, it works fine if the character is near the bottom

also I have set runtime generation to dynamic

[hardy sable]

update, the nav mesh seems to just be spawning below the object. Not sure why

[hardy sable]

ok by setting cell height to a higher value, it fixed the problem

[vapid zinc]

is it possible to do ai in c++ without using the blackboard and keys method

[lyric flint]

If you are NOT primarily a scientist, the way you formulated your question is forgivable. If you are primarily a scientist the way you formulated your question is NOT forgivable for tautological reasons :) there is only one logically valid way to answer it which is a tautology (as P V ~P = T aka it is possible OR it is NOT possible which is always TRUE)

[stable void]

anyone know why my perception strength readings are always =1?

[misty wharf]

iirc the default implementation might always return 1

at least for sight

[stable void]

I'm starting to remember why I made my own perception system haha

[misty wharf]

if you implement ISightTargetInterface that allows you to do different strengths at least for that one

[stable void]

I assume that C++ territory?

[misty wharf]

Yeah

[stable void]

😦 I'll revert to my senses, I don't need them to be that precise anyway, thanks for the help @misty wharf you've been very helpful!

[misty wharf]

np

[stable void]

Can someone help me understand the tick parent function please? I understand that every tick would go through the parent tick function first, then the children one but what about delta seconds? should it be connected? what would that do?

[misty wharf]

if you inherit from a class which implements its own function, the child implementation runs first

you have to separately call the parent from the child implementation, and you can choose what parameters to pass into it

so yeah, the delta seconds should connect, otherwise the parent implementation could produce incorrect results if it does something with the delta seconds parameter

[stable void]

can I like buy you a coffee or something? you're always helping me haha

[misty wharf]

just buy my game and leave a positive review if I ever get it on steam

[stable void]

link?

[misty wharf]

it's not on there yet and probably won't be for a long time but the website is https://www.kingoftapes.com/ if you wanna check it out

[stable void]

I've subscribed to the mailing list, good luck with it!!!

[misty wharf]

cheers, that reminds me I haven't sent any emails to it lol

[stable void]

I haven't even thought that far ahead haha

[late rune]

Hi, guys! Does anybody face the problem of Filter class not working on the final build? I have Move to Actor with chosen Filter Class perfectly worked in the editor, but never in final build. Regards.

[raven leaf]

Trying to learn behaviour trees....
What would you suggest I put here for a fleeing mechanic? I want to check if there is a 'safety point' actor nearby, if not then the AI should just run in a direction away from the threat.
What would I put in here? A custom decorator?

[vivid drift]

A service on the Flee selector to try and set nearby safety location as a blackboard var

A decorator on each that checks if the nearby safety location is set

[raven leaf]

Thank you @vivid drift

[jolly quiver]

Hello, I have a very basic AI character that runs from one corner to the other. As seen in the sketch, there's a wall that forces the AI to move around it, which is whats happening, but I need it to actually collide with the wall... How would I force the character to stick to a straight line between A <-> B? I'm pretty sure to detect the wall infront, I'd need to do a line trace or something like that, but I first need him to actually just run straight.

[ocean wren]

grandmaMax: It would probably make more sense to make that selector a sequence and do the "look for flee point" and then "go to flee point" in sequence

You don't really need to have the flee point testing as a service after all

[umbral rock]

Is there a way to achieve the same functionality using a timeline in a behavior tree? Trying to find a nice way to interpolate values so I can have a smooth transition.

[umbral rock]

Are you just using blueprints or are you using behavior trees as well

[umbral rock]

Something to look into is the EQS system in ue4. There are some good tutorials on it. If you can't find one I can share what I have found. Simply put the EQS (enviromental Query System) allows you to put conditions on finding things actors/locations etc (to my understanding) and based on a set of conditions it will create a set of weights. The ai can then use those weights to determine where to move. Thus you can weight certain locations more or less e.g. closer to walls or further away from walls etc.

What are you trying to accomplish by doing this?

[jolly quiver]

Honestly, it's nothing fancy, just a silly idea. I have this little actor dude that's supposed to be running around in a pit, randomly. Whenever he hits a wall, I want him to fall back and pick a new random direction to go to.

@umbral rock I can make him run in a straight line, but I can't dynamically detect when he collides with something. This is the current bp that I'm struggling with...

[umbral rock]

Don’t tell it to go to a place then

Just pick a vector direction and go till collision

[jolly quiver]

I'm really new to this, could you elaborate?

Elaborate on how to actually do it, obviously I know what you mean 😄

[umbral rock]

Can you call me

Driving atm

😅

[jolly quiver]

@umbral rock I don't mind waiting, really don't like calls for these kinds of things. Just mention me when you have some time! Would really appreciate any help, since I plan on expanding this quite a lot! Thanks in advance!

[timid cliff]

Get your current vector and then invert it, so he is moving in a direction away from where he collided. Add slight randomisation to the value and you should have something.

[pale plaza]

Hello all, hope you are well. If I have this check, how do I set it up in the BT to kill any logic if it's true?

[pale plaza]

for instance, I was trying this, but its killing everything even before the check? My AI is going to their routes or killing me during the grapple animation, so just trying to prevent that while the grapples are playing

[umbral rock]

just got back home from a mocap session. Let me know and I can explain

[jolly quiver]

Sorry, didn't see the the ping!

[umbral rock]

no worries

So one thing that you can do if you only want him to go forward is to put an arrow component that faces the same direction as the character

then you can get the forward direction -90º and + 90º

so that you only get directions in front of him

then you tell him to just move in that direction

then on a collision event you can play a montage and pop him back so that he falls

[jolly quiver]

Could you tell me the actual nodes that I need to use?

There are always so many different nodes with different purposes, really confusing x.x

I understand what you are saying, but I can't quite connect the dots... sorry!

[umbral rock]

where arrow drop is a arow component

arrow

then you can add things to the x/y portion

does that make sense?

[ocean delta]

anyone experienced with AI user interfaces? i have a question regarding initialization during multiplayer

[celest python]

How can I get the value of FBlackboardKeySelector?

[misty wharf]

You mean the blackboard value it's pointing at or what exactly?

[celest python]

The value its pointing

FVector, UObject, Float etc

Is there any function resolves that returns it?

[ruby owl]

If I understand you correctly, for example if you want to get an Actor from a FBlackboardKeySelector:
AActor* TargetActor = Cast<AActor>(MyBlackBoardComponent->GetValueAsObject(TargetKey.SelectedKeyName));

[celest python]

Ah, alright, I just remembered this was a thing. Thank you 🙂

[ruby owl]

np 🙂

[real arrow]

Does anybody have an answer to this, I know it’s back from 2019 but I’m curious to know

[misty wharf]

afaik it's just that one gets an array of all actors that are being perceived, and the other just gets the specific actor that had a perception update for it

[lyric flint]

Hey I have a problem with ai sense, they only look at the center of the player , but I want to look in the head

[misty wharf]

you need to implement IAISightTargetInterface on the actor whose head you want it to look at (only possible in C++)

[lyric flint]

Okkk ty

[shadow python]

if i have pawns implementing IGenericTeamAgentInterface changes the team, how to invalidate the perception or tell that previously friendly unit is now hostile

[stiff gale]

In AI Perception. do you know how to ignore self made noise?

[misty wharf]

Isn't there an Instigator or such on the noise events? You could check that maybe?

[stiff gale]

void TargetPerceptionUpdated(AActor* Actor, FAIStimulus Stimulus)
if ((Actor == GetInstigator()) return;

That should do it right

[misty wharf]

GetInstigator() probably returns the this Actor's instigator, if any

[stiff gale]

So that function is in the controller

[misty wharf]

So rather, when you report the noise event, set the actor reporting it as the instigator, and then in the perception update you'd check the stimulus' instigator == this

[faint fox]

anybody has an idea how can i change nav mesh agent type for a character?

[winged grail]

Does anyone know if you can bind EQS data to anything besides floats (blueprint preferred but knowing if it's possible in C++ would be helpful as well)?

I was hoping to have the actor class type I was searching for bound to a variable so that I could avoid creating separate EQS blueprints for each type (e.g., I have an EQS that finds all furniture actors within a radius that is bound to a variable, if I could also bind the actor class to a variable then that EQS could be used anytime I was looking for actors within X radius).

[formal lichen]

Hello, I was wondering how I would get my AI to only aggro onto NPCs within the same area of the map as it. Even if it doesn't have a line of sight on it. However, I still want the AI to wander into different areas and only aggro onto the NPCs residing in the new area. How would I go about doing this?

[modern vale]

Hello. Is it possible to have arrays in a blackboard?

[modern vale]

Arrays or lists

[misty wharf]

No

If you need to store an array in a BB value, the typical workaround is to create a uobject to store it and store the uobject in the bb

obviously you'd need custom decorators to do anything with it but in any case

[modern vale]

Ok, thanks

[soft mantle]

Hi! i'm trying to do a simple MoveToLocation with an ai controller and a character

but my character won't move at all

[misty wharf]

1. Do you have a navmesh? 2) have you verified your navmesh is valid (click in viewport, press P) 3) is the goal location actually on the navmesh and not somewhere else?

[soft mantle]

oh i tought i had the navmesh setup, but it's not poping up green when pressing P

the navmesh position seems to be okey and i build the navigation a few times already

done, there's an option tickled on proyect setting preventing the navmesh to be built

[wild trench]

Does anyone have an answer to this???

[winged grail]

I'm not sure I fully understood the question. Specifically, I didn't understand the desired behaviour due to the following two sentences:

"I was wondering how I would get my AI to only aggro onto NPCs within the same area of the map as it."

and

"However, I still want the AI to wander into different areas and only aggro onto the NPCs residing in the new area."

But if you want an NPC to aggro characters regardless of line of sight, there's lots of ways. Two quick and easy examples would be to run an EQS for the desired unit type within a specified radius or to do a sphere trace and return overlaps with the specified unit type.

If you then want to restrict their movement within a map, you could again tackle it with different approaches. The easiest but least versatile would probably just be creating a collision box and upon ending collision the NPC state is set to return to it's original position.

[formal lichen]

Sorry for wording the first question so poorly, basically what i'm looking for is a way to turn my map into sections. The AI will be able to go to the different sections but can only see the player in the same section as it.

[winged grail]

Ah, I think I get it, and it can see the player regardless of line of sight, but only in that area of the map?

[formal lichen]

yea

[winged grail]

@formal lichen I'm still pretty new, and I'm not sure how sophisticated you're trying to make things, so this might be too simple for your needs, but I would probably just create a third actor (MapSection for instance) that housed a collision box and a variable (or array if multiple player units) to hold the player that was within its collision bounds (UnitsInSection for instance).

Using that MapSection actor and its collision box, you could write something like "On component begin overlap, check if other actor = NPC, if true, using the other actor as the target set MapSectionRef to self. If false, check if other actor = Player Unit, if true, add player unit to UnitsInSection, if false do nothing".

Then have an "On component end overlap, remove player unit from UnitsInSection". (the NPC's MapSectionRef shouldn't matter as it will be updated if it moves to a new section).

Then using your MapSectionRef on your NPC (or on your blackboard, if using a behaviour tree) you can get the UnitsInSection variable from that reference and cycle through them (if there are any) and have your NPC do what you want with them.

[formal lichen]

Thanks, i'll try this out.

[lone onyx]

novice question for AI im trying to make an enemy move towards a player. If i walk upstairs the enemy walks into a wall

would a node system be the way to resolve this issue

[shadow python]

has anyone successed creating AISenses in blueprints? There is a AISense_Blueprint that is blueprintable but i have no idea how to set the ListenerDataType. which is supposed to be a user-defined struct TSubclassOf<UUserDefinedStruct> but it is not possible to pick one

[misty wharf]

Interesting, didn't even know that was an option :P

[timber dagger]

Hi all, Anyone knows some good document about this perception?
I am trying to alert near by friends when an AI gets damaged.

[vapid zinc]

where can i find good ai tutorials for c++ not BP

[misty wharf]

C++ tutorials don't really seem to exist much. Best I can find is usually "ue4 C++ <name of thing>"

usually you can make some sense of things if you just look at the engine code for the thing, and see how it's being used elsewhere in the engine

[wraith anvil]

is there any way to specify a relative location for the perception component senses?

like set the location for the "ear" or the "eyes"?

[timber dagger]

as aisense is not complete 😦

[misty wharf]

afaik it uses the head location of a character by default for sight

[wraith anvil]

how does it knows the head location tho?

hardcoded by "head" bone name?

(that would be awful)

[misty wharf]

the character has a head height value

which you can adjust, it just uses that

[queen escarp]

i think it traces to center of mass on default, u could modify it in c++ though

[misty wharf]

Correct, the perception check traces to target actor origin by default, but the start of the trace is from the source actor position + head height

[limber flume]

Hi all
How does the MoveToLocation call from AIConttoller actually move the controlled pawn? I mean what call from Pawn movement component actually make it move?
I suspect AddMovementInput, but not sure

[uncut python]

Is EQS a good option for AI or is that not the best option to use

And I mean in general, it’s been experimental for a long time so I’m wondering if it’s dead

I plan on using it to find my player (as a zombie) and I think it may be a a more efficient alternative, especially for multiplayer with its weighting system.

And does it work well with crowd controller? (Hordes)

[misty wharf]

@uncut python EQS is fine, but for simple things like "where is the player" you can do much simpler things like just get the player pawn and get the location

EQS is really just a convenient way to create sets of locations or actors (using the generators), and filter/sort them based on sets of rules (the tests you can add)

[uncut python]

I think EQS would be a little easier to generate the points, thank you

[stiff gale]

Should I use MakeNoise or sphere collision and then iterate over the AI NPCs to warn them and change their behaviors?

[lyric flint]

Hey guys, I made recently a dialogue tree but I want some words to be in a different color than the normal text. How to do it ?

Here is an example of what I want to do and the current "code" of my behavior tree

[spring inlet]

guess you would have to use richtext box and add the corresponding styles to your text

https://docs.unrealengine.com/4.26/en-US/InteractiveExperiences/UMG/UserGuide/UMGRichTextBlock/

[ocean wren]

There's a plugin for doing that kind of dialogue. Might be better than trying to get BT's to do same.

But if you're up for changing the engine code, you can easily extend the BT editor to add stuff to the BT for whatever you want.

[spring inlet]

you mean this one? https://www.unrealengine.com/marketplace/en-US/product/not-yet-dialogue-system

but beside of that, he would run into the same problem

[ocean wren]

Well, his problem is displaying text right? Which I assume this system is actually designed for, but the BT editor isn't

ouch, 43 quid? ooooh 🙂

[spring inlet]

it doesn't come with any UMG stuff, you have to setup display yourself, too

[ocean wren]

I wasn't thinking of this one I don't think.

[spring inlet]

ok, maybe you find a link, im curious, too 😄

[ocean wren]

if you look for "dialogue" on the marketplace there's a ton of them 🙂

Might be this one?

https://www.unrealengine.com/marketplace/en-US/product/dialogue-plugin

That one was one of the free marketplace items a while back

Honestly I don't pay a lot of attention because I can create my own, but seems like evaluating the functionality of these could be useful

Hmm, I thought of some project ideas for my students 🙂 that lip sync plugin is terrible 🙂

[old adder]

Hello, what is the difference between CrowdManager and flocking, boids technique?
Is there anyway that we can achieve flocking result from the CrowdManager?

[ocean wren]

Its very much a different thing

One is basically a collision avoidance system based on crowds

flocking is a steering system.

Although there's obviously some overlap in aspects. They're not trying to do the same thing

the Crowd manager is interested in getting each agent from A to B while avoiding collisions and blockages.

But flocking is about maintaining spatial relationships between agents based on forces

I mean you can obviously change the crowdmanager to implement steering forces, because it basically uses a direction vector to steer the agent away from collisions. But they're different concerns really.

That said, I did add a "push out of the way" force to the Crowd system because I got tired of the problems of rigid body collisions on AI agents, so certainly feasible.

[old adder]

Thank you. I am trying to make AI moves as a flocking group while also keep the navmesh pathfinding feature.

Does this mean I should override CrowdManager instead of CharacterMovementComponent to add the steering force?

I been overriding RequestDirectMove() in MovementComponent but it does not seems to work as I think.

So my problem now is to know where I can actually make meaningful override.

[ocean wren]

Its a while since I did it, but there is a function in there that calculates a vector which is the collision-free vector along the path the agent is moving (i.e. it moves in the direction of the path but modifies that vector to avoid other agents), so you need to manipulate that vector calculation

[old adder]

Is it in the crowd manager?

[ocean wren]

just having a quick look.. one sec

[old adder]

Or is it the CrowdPathfollowingComp

[ocean wren]

Mine was using the detour crowd system.. haven't got the code on this machine so having a quick look see if I can remember where in the engine source it is

I take it you're OK with C++ and the engine in general? because this stuff can't really be done in BP

[old adder]

Ah yes, I would be insane to do this in BP.

[ocean wren]

Yeah, looks like its in CrowdManager

void UCrowdManager::Tick(float DeltaTime)

You see that function? it basically uses detour to update a bunch of vector values

You probably need to noodle with the value before that "ApplyVelocity" call near the bottom

Basically, that is the final velocity vector its going to use to move the agent

[old adder]

Yup, I see it.

I think I still have to write extra component so that the pawn knows which group it belong to and calculate base of that.

Thanks, I will have a crack at this.

[ocean wren]

If you want to have grouping and such, then you'd probably be best to have a component that takes the agent vector calculated just before passing to Detour and then modifying that vector. But detour does have some seperation etc in. So maybe you want to just think of it in terms of setting the correct goals to navigate to instead? i.e. set each agent in a group to move to a group point (position + offset) and let Recast keep them together?

Basically, recast+detour is a library that does the collision avoidance part. Keeps agents from hitting each other (well, mostly) and navigates them through navmesh space. So what you really want for groups, is to give them distinations that are near each other and have the detour system get them there while avoiding collisions

I had a squad of soldiers and I gave the squad a move order. That basically calculated offsets for where to move to for each soldier.. Gave that move point to detour system and it moved them there while avoiding collisions

I also added a way for the player to push them out of the way, by having a force repulsion inside that detour velocity calculation (just an inverse of the direction vector away from the collision)

But of course your use case might be different

[old adder]

Yes, I do have a group, calculating group centre, and get offset vector for each units and offset the goal base on that.

So they will stay in the starting formation after moving is done.

However during the path following they would still clump into eachother..etc

[ocean wren]

Yeah, that is in part because UE has rigid collision on characters

so the collision capsules get in the way as agents try and navigate

especially if they're not a good fit capsule

I turned them off

Because for the most part, they're not useful

Or rather, I turned off agent->agent collision

the seperation force is meant to keep them apart.. so better to use that and allow some non-rigid force to keep them apart rather than the very obvious rigid one

[old adder]

I am trying to make them keeping distance from eachothers while moving.

[ocean wren]

Yeah, detour does that

but you don't want to use a rigid force for that, if your agents are moving through gaps

or they get stuck on each other

[old adder]

So I just subclass the AIController from ADetourCrowdAIController?

Do I still need 'Super(ObjectInitializer.SetDefaultSubobjectClass<UCrowdFollowingComponent>(TEXT("PathFollowingComponent")))'?

[ocean wren]

Take a look at the recast+detour library demo on github, you'll understand

Yeah, there's a few things you need to use Detour.. you definitely need the Detour Crowd Controller

there's another step you need.. but I forget the detail. Might be worth looking at the official docs

[old adder]

Oh.. I am not sure how to use this in the first place..hmm

[ocean wren]

Might be detour crowdfollowing component too

[old adder]

Unreal already use recast, does it have the whole Recast+detour combo too?

Or I will have to manually include this to the engine?

[ocean wren]

Yeah, it has a few things.. but I've not used it for a few years so forget the details.

Unreal already has it built in

you have to choose to use it though

[old adder]

nice

[ocean wren]

I don't think I made a tutorial on it sadly. I usually do for these things, but I figured as it was C++ only it was probably a bit too complicated for my audience.

[old adder]

This is quite great direction already, thank you!

[ocean wren]

I do recommend looking at Mikko's demo project on github.. (he wrote Recast+Detour library UE uses), he also did some excellent videos on how it all works.

Mikko Mononen is his name

Funnily enough, I just found out this week that Mikko works for Epic now 🙂 back then he was working for Crytek and more recently for Unity 🙂

[old adder]

Thank you.

[granite vault]

Sorry, is it UE's Github? Because my search function nothing in the repo

[ocean wren]

It should be available on the normal github. The actual library itself is already within Unreal Engine

but the demo is in the original library's github, let me find it

https://github.com/recastnavigation/recastnavigation

And Mikko's blog about developing it: http://digestingduck.blogspot.com/

[ocean wren]

I wonder why it seems that people have so many issues with AI? is it because the toolsets are just too complicated? or is it something more basic?

The concepts aren't that complicated are they? BT's aren't that complex. Although I do think the UE toolset is.. not the best usability wise.

Curios if you guys have any thoughts on what a "good" toolset might look like?

[misty wharf]

No idea tbh. I always found it fairly straightforward

Much simpler than having to write it without BT's and other things from UE

I think the challenge is just how to put it all together since there's no really really obvious way to do things because the systems are very flexible

[ocean wren]

I guess having like a bunch of examples would be useful.. Epic really didn't do a good job of that did they

Mind you, I'm not sure I've seen anyone ever really mention any of their samples.. so maybe they would get ignored anyway 🙂

I suppose actually having developer docs would help too

[winged grail]

I'm sure it varies from person to person, but for me, as someone with no prior knowledge and who works in a field completely unrelated to tech, who began teaching themselves sometime around March-May, there are several major barriers that I encountered along the way (I'm sure there are more, but these are the ones that immediately pop into my head):

1. Planning for modularity and scope. - By this, I mean that it's relatively easy to get a character to do a straight forward task (e.g., move to this location, play X animation). It's much more difficult to create a system that seems realistic and is adaptable (e.g., move to the correct location based on what you need to do, then play the correct animation and/or sequence of animations associated with X task). This is further compounded by the speed with which behaviour trees iterate and the disconnect between BT nodes and blueprint nodes in terms of debugging (i.e., break points). I think this complexity is again compounded by the fact that BTs visually appear to function differently than something like the AnimBP state machine and the fact that you need to be very specific with when, where, and how, variables that affect state-based changes are being set/changed.

2. Understanding how something works, when it works, and why it doesn't work in a given situation. A good example of this is the "Simple Parallel" node. It's not immediately (or even latently) obvious that the background tree cannot have a Finish Execute task. It's not in the description, it's not in the documentation (last I checked) and googling doesn't immediately produce results. Another example would be instances in which decorators don't seem to "fire" or cancel a node below them, despite the observer aborts section being correctly entered. Again, it's not immediately evident that the decorating check is not being updated, in what specific instance it does update, and/or how it updates (e.g., does it update if a lower selector continues to fire a task that returns true, what is the order in which a decorator is evaluated, etc.).

3. Experience with the engine. This is honestly the biggest one, I think people are eager to get started on their projects and it's difficult to temper that excitement with developing strong foundational knowledge. The more experience I get, the more I'm going back and revamping systems I thought were super clever at the time and now seem redundant, bloated, or unnecessary. In many cases, this occurs as I learn something new that the engine can do (a non-BT related example being switching State Machines in an AnimBP based on enumerations).

TL/DR - Planning, organization, foresight, documentation, and patience/experience.

[ocean wren]

Yeah, I could see those factors being important. Especially for the last one, because honestly spending any time here on discord is so depressing with so many unanswered newbie questions I just feel bad for everyone.

The second one I think is partly usability of the current BT system, which if I'm honest just isn't as clear as it could be. Partly because they tried to be tricksy and optimize the BT rather than do the typical "run from the root" that every other BT I've used does

And some of the usability was always a bit suspect due to some HCI stuff. EQS is a good example. Just poorly worded functionality etc.

Conceptually they're fine, but they're just not well designed as toolsets.

[rose crystal]

How can I make car racing ai

Is there any good tutorials

[misty wharf]

there's some stuff on racing AI in Game AI Pro which is free at http://www.gameaipro.com/

dunno if there's anything UE-specific though

[olive schooner]

I am new to unreal engine can anyone tell me which will be better to use for Ai bots,Blueprints or C++

[unkempt barn]

Does anybody have any suggestions for how to use EQS to find the best location to push the player off a ledge? I'm new to this system so just looking for any suggestions or advice

[unkempt barn]

I feel like AI is just complicated not from the perspective of how the tools work (i.e. BTrees) but more the actual design of the behaviors. It's not something that's super intuitive and there's not really a clear "correct" answer.

[unkempt barn]

I think maybe a good single toolset is pretty much impossible because AI is so broad. That being said UEs AI tools are pretty damn nice compared to what I've seen elsewhere.

[ocean wren]

Microsoft has a research paper called "Its Unweildy and it takes a lot of time" that pretty much sums up game AI. Too hard. Poor toolsets. Unclear how to specify. Lots of edge cases. So yeah, I've got some sympathy, but surely thats why you need BETTER tools for it, rather than.. lets face it, Epic haven't really put much effort into AI since I started using UE beyond poor Mieszko doing some work in his spare time.

[queen escarp]

@ocean wren do you have some special tool on your mind? i would love to have a descent utility ai framework

[ocean wren]

Yeah, I'm working on some machine learning based cinematography right now, but have considered maybe creating a new AI editor and runtime toolset. But what turns me away from the idea is mostly the support issue

Did you try Tom Looman's utility system?

I've been thinking that most people would probably be better served by having a system that could generically act given some simple commands. "Go here, pick up that thing". Maybe even using a speech or gesture interface. Trying to imagine what a "good" future toolset might look like. Its such a pain dealing with all the edge cases of boolean logic for instance.

[queen escarp]

@ocean wren machine learning based cinematography sounds fancy and pretty time consuming, are u using cuda torch or tensorflow stuff ?
i am not sure about how supportive epic is tho,
i havent, does he has one ? i just read his blog entry
i would love to code my owen but i am just not convinced yet,
since u need to handcraft the weights to find the best score, i read just once about it to combine utility ai with deep learning

[ocean wren]

Yeah, using a combination of pytorch and tensorflow 2 for the cinematography stuff (using an ensemble of ML models)

I didn't use Tom's utility thing, but saw his posts about it. Figured he'd make a decent attempt at it.

But yeah, deep learning is a different issue. That moves the problem around rather than really fixing it. Because then you have to have training examples

My approach is to train behaviour from examples seen in movies. Hoping to apply that same approach to training behaviour in general (i.e. training from samples of human gameplay for instance).

Its still a bit of a fresh area though, so we're in that "not quite good enough" phrase 🙂

[queen escarp]

@ocean wren yeah its a nice attempt for sure,
i would create training examples lol, but i havent read any paper about utility ai + ml yet, only in gems was something mentioned,
but somewhen i guess i'll plug a nn into it and go for it
ok wow that really is more rnd stuff ha ?
is there some paper comming 🙂 ? i think i watched a gdc talk about it, ai watched the player and learned from him, it was even dl reinforcement stuff

[celest python]

What Tag Cooldown does is, its returning false if we have the given tags until cooldown finished?

[ocean wren]

Intax: yeah, basically lock out and return failed for X seconds

spinynormal: Dave Mark's book "Behavioural mathematics" is about as good a reference on Utility you're going to get. But there's a bunch of GDC presentations. Kevin Dill did a few papers on his "Dual Utility Reasoner" that is worth reading. More in the domain of military simulations, but applicable to games (and Kevin worked for Rockstar on Red Dead Redemption 1)

[shadow python]

in RTS game
if i have a MasterAIController that represents a bot player with PlayerState that manages strategic things
if each of my units is controlled by UnitAIController without playerstate
how would be preferred way to make Pawns aware of player state? (passing master controller as owner when spawning pawn or extra spawn parameter or smth else?)

[misty wharf]

Tbh it probably makes very little difference, so whatever method is the easiest to understand what's going on seems like the way to go

[shadow python]

was a bit tricky to manage teams, with team identifier being in player state, and IGenericTeamInterface implemented by units

and some units not related to any of teams with independent unit controllers

currently set controllers to request team from controlled pawns
pawns with playerstate (taken from master controller) take team id from them
pawns without playerstate take team id from local field

does it make sense?

[misty wharf]

seems reasonable

[ocean wren]

I used a second actor to coordinate agents into squads, then another to coordinate the squads.. so basically a heirarchy of actors with message-based comms between them. Much like the military chain of command.

So individual squad members are commanded and respond to command requests from squadactor.. squad actor takes orders from layer above etc.

I really liked the message-based setup though, because you can just insert stuff into the command message queue and debug the receipt of messages etc.

FMessageEndpoint is what I used if you're interested

In theory you could pipe that over a network for debug purposes too

On spawn you can just fix up the chain of command.. i.e. who owns which units etc. Also do it dynamically for stuff where you want to merge squads into a larger group.

[shadow python]

it would be a bit problematic a bit with group actors in my case

[ocean wren]

Why?

[shadow python]

existing project

can't just redo everything

[ocean wren]

Never say never 🙂

Do whatever it takes to make it work 🙂

[shadow python]

somehow sneaking in IGenericTeamAgent instead of tons of blueprint spaghett

[ocean wren]

But yeah, I get you 🙂

I'd avoid any sort of blueprint for this kind of thing. We had a hellish time with blueprints crapping themselves and just failing to execute randomly in certain builds

and without any clear repro case.. so just in certain builds, without any warning, we'd get a BP failure condition

In the end, we had almost all main code in C++ and just simple glue and component data in BP

[shadow python]

in my case i have several people constantly generating blueprint spaghetti and just me writing in c++

[ocean wren]

ouch. I can only imagine the horror

not even nicely formatted and functional BP?

my students who do BP work tend to do that kind of thing.. they're so excited to have something working, they forget that it needs to be maintained etc.

[shadow python]

does 8 ubergraphs and ~500 functions/events in PlayerController counts? 😄

everything packed into one bp

[ocean wren]

That counts as the seventh level of hell 🙂

[idle cobalt]

Is there any good tutorial that teach you how to make vehicle AI like a tank AI ?

[ocean wren]

They did a livestream on doing that a while back.. on the main UE channel

[shadow python]

there are also bunch of youtube videos with vehicles, with basic behavior tree setups

[idle cobalt]

Thanks

[ocean wren]

Tanks are fun to do though.. I did a multi-level system for a tank game 🙂 because that's my bag baby.. commander, gunner, loader, target selector etc. Fun task to consider the flow of info between them. Probably more complicated than needed to be, but it was a joy to use 🙂

[shadow python]

i did try to implement path following for tracked vehicles, got stuck on math gave up

[ocean wren]

Commander->target selector "hey, find me a good target to the front", target selector->loader "hey, target spotted, load HE" etc..

tanks are easy.. just stop them and turn on the spot 🙂

Basically the same as a human but with a hat for a turret 🙂 a hat that can 360 no scope enemies

Most challenging maths wise was AI for a combat flight sim I was working on back in 2005-6

physics based aircraft, with bits hanging off 🙂

in weird 3D air maneuvres

[old adder]

I...kind of get the steering, flocking to work.

Sadly, I was unable to implement this in the CrowdManager. There is really barely any official Unreal doc about this.

I am actually not even sure the CrowdManager recognize which pawn to add to the crowd sim.

[misty wharf]

iirc characters go into it by default but otherwise you need to implement ICrowdAgentInterface and register it with the crowd manager

(or maybe ai controllers, not sure)

[old adder]

For everyone controller?

If I remember right how to do it was using CrowdPathFollowingComp.

But it does not do any changes for me I am not sure.

[misty wharf]

Yeah not sure, my player character implements ICrowdAgentInterface and registers itself with UCrowdManager, which makes my AI's avoid it

[old adder]

Was it in the controller?

Or the Pawn?

[misty wharf]

it's implemented in the pawn in this case, but you can implement it in either

you just need to register the actor where you implemented it with the crowd manager

and if you implement it in the controller, you need to implement the interface so it reads the apporpriate values from the pawn

[celest python]

Whats the proper way of pausing Behavior Tree? Locking AI not working in my case, I'm only able to unpossess AIController

[misty wharf]

have you tried calling Stop Logic on the BrainComponent?

[celest python]

Yep

Not working either

[misty wharf]

Claim Task Resource with AI Logic as resource class?

and Unclaim Task Resource to unpause

[old adder]

Thanks! I try again.

[celest python]

Gonna try, thank you

[fluid oriole]

Hi, what happens when a Generator has no items (Actos or locations)? Does the EQS Query clear the Blackboard Key?

For example, if I use the Generator to get Actors of a specific class within the Querier's radius but there are none.

[misty wharf]

The result for getting a single actor from it should be nullptr, so if you're using it with a BB, that would set the BB value to none

[fluid oriole]

Thanks!

[mystic solar]

Hello! Looking for some body who could teach us in BT, EQS, AI etc. for sure paid, for 1-2 weeks half time

[lyric flint]

@mystic solar those are very basic applications of AI tho. Are you looking for something specific?

[tawdry zephyr]

What does "Project Goal on Navigation" do? In my head it basically means it will modify a dynamic runtime navmesh so that the goal isn't available to other actors?

[ocean wren]

If its what I think it is, it'll find the closest point near the goal on the navmesh

i.e. it'll "project" the goal to the navmesh

[misty wharf]

iirc yeah that's what it does, if goal is outside navmesh, it attempts to project it to closest point on mesh

[tawdry zephyr]

Ah gotcha

[uncut python]

is EQS gone?

[uncut python]

how did you get EQS to work? it says its deprecated

nvm theres a whole other tab dope

[tawdry zephyr]

How do you know what behaviour tree the selected character's AI controller is running?

[uncut python]

press the apostrphe key, then do num lock on ur keyboard and using the number pad press 2

it brings up AI debugging

[tawdry zephyr]

I don't have a number pad

[uncut python]

hmmm

ngl i think u need a number pad

https://answers.unrealengine.com/questions/568734/gameplay-debugger-and-numpad.html

try that maybe

am i wasting my time with EQS? I feel like its too confusing, i thought it would make my zombie AI smarter but its not that amazing

wtf does this mean\

nvm restarted engine it was some weird bug

[tawdry zephyr]

Never had a use for many years, they're obsolete

I'll just change the shortcut

Huh that just lets you control an AI

How do you get the debug data without a numpad

Most people don't use those anymore

OK found it

Hmm nope

Doesn't activate

Whoever came up with this did it wrong

What's the console command for the equivalent

Anyone? console command for AI debugging?

OK so there's no AI debugging in UE4 with commonly used keyboards

No ones got time for this nonsense

[uncut python]

Ik it’s kinda crazy I just had this issue even with a num pad u have to turn in numlock for it to even work

Crazy

[unkempt barn]

I'm also looking into EQS. It seems like it's halfway between a perception system and a decision making system, I may be misunderstanding it

[uncut python]

I believe u are right, it’s a bit confusing to use tho, I’m not sure if it’s something I should put time into or I should just use regular old tasks and behavior trees

[unkempt barn]

I mean i like the concept. raw perception is not really what our brain uses for decision making, there are functions in our brains (and most other animals) that automatically filters and evaluates the information from our senses before using it to make decisions. So i see the merit in making a peripheral system that is responsible for converting sensory information into useful information

that being said, it's not a super intuitive system (from the little i've done with it)

[uncut python]

Ya it’s very weird to use 100% on in the same boat as u

[unkempt barn]

haha well I'll be sure to share any good tutorials I come across on the topic

[simple crest]

it will be a cold day in hell when i give up my numpad

[simple crest]

EQS is not a competitor or alternative to behavior trees or tasks or perception systems, or anything like that. EQS is nothing more than a target selector. don't feel bad, it is confusing to start using

simplified, the only purpose of EQS is: you give it a list of things, and it will rank them somehow and choose one

[celest python]

How do you guys handle melee attacks in AI? I'm thinking about line tracing with a little distance each n seconds and if the hit actor is enemy disabling BT and play melee attack montage & deal damage

There are many ways to do this I assume but I'm wondering if there is any examples for that

[misty wharf]

My fps game's melee enemies just perform them the same as any other attack

BT triggers the attack, which then does whatever - play animation, do trace, apply damage

[celest python]

How do you check if its in the range or not in BT?

Or just deciding its time to attack in BT?

[misty wharf]

It just has logic that tries to get close enough first

once it reaches the player it doesn't need to check anything since it's close enough at that point

then it keeps attacking as long as the player remains within range

[celest python]

Makes sense, thanks

[neon carbon]

is it good to have a bunch of AI ( for example forest animals ) use AI behaviour trees? will it take a lot of toll on the performance?

frame rate drop etxc

[misty wharf]

it depends on the BT, it's very hard to say up front

[shadow python]

when implementing AIControllers is it preferred to have BehaviorTreeComponent and BlackboardComponents as class default objects or let them be created on run behavior tree call? or create BTComponent and let blackboard be optional

[patent hornet]

they are already on the AIControllers

[misty wharf]

Doesn't the default implementation have those?

Yeah

[patent hornet]

BrainComponent is BTC

[shadow python]

they get created when calling RunBehaviorTree

AIController has only pathfollowing/actions in CDO

[patent hornet]

i never felt the need to alter engine's default behavior here

(obviously, from previous answer)

but if they are created when BT is ran, adding it to CDO might force you to do some manual cleanup if changing BTs

[shadow python]

oh, ok. well. will just make bunch of variables in controller and pass them after RunBT call

was asked to add some configuration pass global variables to blackboard that needed to be applied when BT started

[shadow python]

i feel dumb, i didn't notice InitializeBlackboard so could just override it and don't do anything else

is there any way to prevent huge amount of FName constants being passed when setting blackboard keys?
apparently there are quite a lot of typos all over the project after renaming blackboard keys and leaving Set* methods with bad key names and tracking them all is quite a problem.

since all SetValueAs does not check for result of SetValue and SetValue does not even warn about bad keys

spent several hours debugging just to find out that SetValueAsObject with DefenderTurretActor was missing an 'r'

[neon carbon]

as soon as more than 2 ai actors spawn, my game starts dropping frames

[misty wharf]

Really, does it happen if you don't run their BT's?

[simple crest]

The BT would normally be very small compared to other things - such as a character movement component cost

[queen escarp]

@ocean wren http://mrl.snu.ac.kr/research/ProjectMovingCam/MovingCam.html

[ocean wren]

spinynormal: yeah, there's a few papers using that kind of approach

one in particular uses RL to generate physics forces on a powered skeleton to generate motions from YouTube videos of specific motions (usually backflips and the like)

https://xbpeng.github.io/projects/SFV/index.html

I'm using a non-physics based approach, based on graph neural networks and generative scene graphs. Trained on an extended version of the movienet dataset.

Largely based on this approach: https://arxiv.org/pdf/2008.09092.pdf

[normal river]

I stopped using number pads many years ago as well. They're a waste of space lol

[tawdry zephyr]

Yes

I use a 60% and it has full functionality once you adapt, number pads are obsolete

Then again I taught myself to type (~120WPM now) and I never used numpad when learning so I've always used the row of numbers

So I can understand people who are used to using numpads rely on them and think they're necessary (they're not)

I did get pretty fast on a numpad just so I knew how to use one effectively when required, but then got rid of it and never looked back

This is the first time in my entire life I've been affected by not having a number pad and it's for the silliest of reasons

[simple crest]

Anything smaller than an 1800 layout is only fit for hipsters and book writing, no hardcore technical work sorry. Off topic pseudo-trolling aside, what exactly are you seeing when you rebound your gameplay debugger keys? Apostrophe let you choose an AI, but are you saying you couldn't enable/disable the categories along the top bar with rebound category keys? @tawdry zephyr

Trying to think about it I feel like a long time ago I tried to control the categories somehow and found something else hard coded so I guess it won't surprise me if this turns out to be true

[elder ridge]

I spent some time reading up on UE's implementation of Behavior trees and EQS, and the only thing I have a hard time understanding is how it impacts performance and when. For example if you have 100 moving characters/actors (NPC's), is it better they share one larger tree, or do you normally want to split them up, or should I keep the tree pretty basic and lean more on EQS?

Sure it usually comes down to testing and profiling, but is there any available general guidelines on it? To save at least some time

I also saw the newer "swarm" (niagra?) that was talked about in the UE5 videos. Does it use a totally different system then?

[ocean wren]

EQS is actually kind of useful as a general purpose query mechanism, because it does a time-sliced query for you and generally doesn't kill performance unless you make it 🙂

[elder ridge]

Like a town with people that want to eat, sleep, work etc. Depending on skills, time of day, stats and such

maybe not doing the same things every day and so 🙂

Its mostly for aesthetic reasons though, not as much actual gameplay

[ocean wren]

That said, there are some usability issues in EQS that can trip you up. I think it needs a refresh UI wise, but just don't have the time and energy to do it 🙂

[elder ridge]

So do you think skipping EQS mostly is wise?

[ocean wren]

The niagara thing is more about doing "smart particles" so I wouldn't rely on that for AI

No, EQS is sound, if you can understand it

[elder ridge]

Oh, I thought it was environment query in general

not just sound

[ocean wren]

Its like most of the AI tools in UE, basically works, but has issues of documentation and usability

EQS is basically a response to requiring data about the world for an agent, but wanting that data to be gathered over time rather than in one blocking operation

[elder ridge]

So it might be more common to simply do it yourself then? Instead of the Behavior trees etc?

In that, I mean the default implementation

[ocean wren]

Well, put it this way, using BT's and EQS will work. It'll be a pain, but it'll work. Using something like blueprints? it'll MAYBE work eventually. Maybe.

BT's are pretty standard fare in Game AI, so worth learning

[elder ridge]

I was thinking if you have a large amount of actors, I rather just want each actor to query the current state of the world, and choose certain actions based on some variables. I thought if making them query less frequent (maybe every 10-30s), it would be less heavy on performance

[ocean wren]

That said, the UE implementation isn't the most user friendly and has not great docs.

[elder ridge]

like looking at the time of day, current action, how the current stats are and such

might be easier to make something myself then

Kinda looks like it was the norm 🤔

[ocean wren]

Yeah, you can control how often the EQS queries run for sure, but inside EQS it actually controls how much of a query runs per frame

[elder ridge]

Ah I meant as a custom implementation of a simple state machine, instead of EQS etc

[ocean wren]

i.e. if you're gathering data about 100 raycasts, it doesn't do them all in one frame

for stuff that is pretty much global info like time of day, just feed that into a blackboard entry, I wouldn't use EQS for that

[elder ridge]

Since this is less about what they see, and more about what they do and where they are, so it wont require any rays or per-frame updates, it might be easier to skip the behavior tree all-together

and just have a simple state machine

[ocean wren]

EQS is really mainly for spatial querying (hence the Environment part)

[elder ridge]

ah gotcha, then I think I can skip EQS for now, that clears that up at least 🙂

[ocean wren]

I use EQS for cover selection etc..

You can use it to query where things are on the map for instance

[elder ridge]

I guess I will just have to dump a large amount of AI actors in and look at the performance differences. Was just trying to find some resources or guidelines that might make the process a bit shorter or less painfull to debug 🙂

[ocean wren]

Generally, if I need to do like raycasting in a batch, I'd use EQS

[elder ridge]

Ah okay, yeah that makes sense

[ocean wren]

Its a bit difficult to give you rules of thumb because AI is kind of game specific. But there are some gotchas depending on what you're trying to do

For instance, the default agent count in Unreal Engine is 50. Try doing more than that and some of them won't work 🙂

[elder ridge]

In my case, its simply town NPC's living their daily life, so just doing activities based on their own local stats and some game world states and nothing else

so there is pathfinding in it

[ocean wren]

Ok, so to do something like that.. yes you'd have a BT with some blackboard values for time of day, you'd use EQS for "find me activity X nearby" style queries

[elder ridge]

Ah okay, I will give that a try then, see how it works out. If the agent count is normally 50, it would be interesting to see how people do it in a larger scale then

[ocean wren]

Take a look at the presentation for how they did the AI for the beach scene in Bioshock (its on youtube)

[elder ridge]

sure il look it up! Thanks for the pointer 🙂

[ocean wren]

There's a number in the editor settings, you can set it to a higher value, it just defaults to 50

Usually for sort of "world filling" character stuff. You'd have a smart-object system (smart objects coined by the sims AI) where it announces its availability to satisfy some need.. then you have a query system that says "I need X, find me anything that satisfies X near me, sort by distance" (EQS) and some way of locking that resourse so that others know you're using it etc.

EQS is roughly based on some work by Matthew Jack on CryEngine and their environment query system.. might be worth digging up documentation on that to understand the concepts.

[elder ridge]

Sure thing, will do that then, really appreciate the explanations 🙂

[ocean wren]

AI is fun 🙂 or at least should be!

[queen urchin]

hello guys
Can you help me with AI?
My AI ignore other players and just attack player1

[ocean wren]

I can't read that BP, get a higher res screenshot?

[ocean wren]

How does your blueprint select which enemy to attack?

[queen urchin]

maybe on this BP

I did that just following this tutorial https://www.youtube.com/watch?v=TjoE7Pmank0

[ocean wren]

So the AI sense is fed into a blackboard item. But where is that blackboard item extracted to then use in the attack?

[ocean wren]

Ok, that's the blackboard.. but you need to get the value you saved INTO the blackboard and take it back out to then use to attack

[queen urchin]

maybe this?

AI just chase and attack only player1 and ignore other players even when player1 is dead

[ocean wren]

It's probably because you're only ever setting player 1 in the blackboard

You need to go back over your setup and actually understand the parts rather than just do whatever the tutorial does

[hoary nova]

hey guys, any tips how to create an enemy horde optimized (for replication)? my fps drops after spawning 10-20 enemies (simple character pawns with just a sphere mesh and simple move to location function every 1 sec). Using physics instead of navmesh is faster ? Or BTs ?

[ocean wren]

Whats it for?

I ask, because someone at Epic did a presentation at one of Epic's dev days about creating a massive horde style thingy. Using niagara and some tricks

Pretty sure you'll find it on Epic's website somewhere. Can't remember the name of the presentation though.

[hoary nova]

i want to make smth like back4blood type of enemies

so not very massive, but around 50 enemies i think

[ocean wren]

don't know it.. how many enemies are we talking about?

Have you profiled where you are losing performance? Might be a simple setting you can tweak

[hoary nova]

not sure how to profile the ai thing

[ocean wren]

50 doesn't seem like a lot. But default setups can be a bit crap.

Well, time you learnt then 🙂 there's a video or two on the Epic channel about using insight to profile game loops. Its a useful skill to get started with.

[hoary nova]

ok, i'll check it out

[ocean wren]

If you search for Insight profiling on the Epic youtube channel you should find them. I've seen like 3 different presentations

And one on the aidebugger too

[celest python]

50 of enemies can create significant framerate drops with 4 players especially if its listen server

Your problem will not be related with AI framework, it will be related with RPC sending time, CMC's Tick time and rendering them

Back4Blood probably doing some shenanigans that can not be done easily in UE

@hoary nova

[ocean wren]

Only way to be sure is to profile it

[hoary nova]

hm... so at the moment from insights tool i got that characterMovementcomponent_tick for 60 enemies takes 7-20ms and then i have an extra waituntiltaskcomplete that is also 8ms, but probably because i create a timer for each enemy actor and that is quite bad, so now i'm trying to make bt and see if i can optimize this a bit, also thinking to remove CMC and make only physics actors might be better?

[ocean wren]

I've heard that CMC can be quite heavy time-wise.. but it might be useful to drill down further and see if its a single function being heavy handed

[celest python]

CMC's FindFloor function is notorious for its CPU usage

But thats not only one

Too many traces, RPC sending time, OnRep updating time etc

BT only can optimize the MoveTo function and its doing that indirectly by ticking the system with time-slicing

[ocean wren]

You'd think some enterprising soul would go in there and rewrite it wouldn't you?

[misty wharf]

have you looked at CMC

it's like.. you touch it and everything comes crumbling down

:D

and it makes all kinds of assumptions like the direction of gravity and such and it's nigh impossible to change, but some absolute mad lad has done that for the directional gravity plugin on the marketplace

[celest python]

Yeah that one is insane

[misty wharf]

UE makes all sorts of very annoying assumptions about the orientation of the player character as well

You would think that if you look up, it would look up relative to the orientation of the camera

but haha no why would it do that?

so if you turn the character upside down and the camera with it, now you have to do all of that logic yourself and it just gets kinda annoying to do :D

(I was working on a small space ship type thing at one point)

[granite vault]

Is that dev RomanSmirnov?

[misty wharf]

No idea, if he's the author of Directional Gravity then probably :)

[granite vault]

I see. I did quick search. 1 is "Directional Gravity for Physics and Movement".
Another 1 with indirect name "Directional & Planet Gravity".

One paid. Another is free

[misty wharf]

the free one's the one I tried so whoever made that

[granite vault]

Oh. It says Tefel

[misty wharf]

who's Roman anyway?

That name sounds vaguely familiar

Ah, that's because a friend linked a github page for texture 2d arrays which had his name on it :D

[granite vault]

The author of "Directional Gravity for Physics and Movement". This is paid plugin

[misty wharf]

Ahh I see :)

[granite vault]

https://www.unrealengine.com/marketplace/en-US/product/directional-gravity-for-physics-and-movement

[misty wharf]

Yeah looks similar conceptually as the free one

the free one uses a C++ based CMC-derived movement comp which is probably the main difference

[granite vault]

Yeah. I wonder how he does it in 100% BP

[misty wharf]

I'd guess a custom movement mode and just reimplements the movement logic there

I'm not sure how performant doing all that math in BP's is though... Although it's probably not a big problem unless you've got tons of characters based on it moving about

[granite vault]

Yeah. Switching to non-character would help a lot in this case

[hoary nova]

can one ai controller control multiple pawns?

or that does not make sense?

[ocean wren]

its not designed to

So not really 🙂

[hoary nova]

FloatingPawnMovement component seam to work much better than cmc, it also have auto update tick registration that turns off tick when actor is not moving which helps a lot in multiplayer, just that it looks a bit laggy on client side and i have to interp position

[odd trout]

I have some general questions about best practices in setting up AI. Where should I store and update my pawn's stats (ie hunger)? Should everything be stored and modified inside the pawn and the AI just gets these variables from the pawn or should changes to my variables all be done within the AI then sent out to the pawn?

Also, is it best to move variable values between the AI and pawns by casting or with interfaces. Virtually every tutorial I've seen uses hard casting. Is that a bad practice or something that is okay to do in this instance?

[jovial kraken]

Hiho I have 2-3 general questions about all inclusive.

I figured I would build 2 types of AI for general monsters and 2 types for bosses.

General monsters AI:

• Aggro Range
• Aggro music
• Chase on / off
  ... What else should go in here?

Monster specific AI:

• attacks
• Fight and flight behavior
• Loot and drops
  ...

My Question:
What exactly should be in good general AI?

Is this a good way to build a general AI that counts for all monsters + an additional special AI for each monster?

Would a better solution be General AI = parent + Specific AI = Childes?

Is the death of the monster also regulated via the AI ​​or via the character BP?

[odd trout]

The game I am making also has multiple speed settings which makes this more complicated because I'm unsure of if/how I should go about updating the intervals of services and waits in the behavior tree and/or the frequency of AI ticks.

[ocean wren]

DavidS: It makes more sense to store state data like hunger levels on the character itself. That way you can have it do the logic for hunger. The AI just needs to know about it to change behaviour later, but things like making the character die when hunger reaches a threshold should be part of the character imho.

As for moving variables? I wouldn't worry about it initially. Do it whatever way you prefer and then profile it and rework if its a problem.

Raikun: You can run sub-behaviour trees within a main behaviour tree. So you could build a "general" tree and then have specific subtrees for specific characters.

I think you guys need to stop worrying about doing "Good" or "right" AI and just get something working. Refactoring will be required either way. So get your things working and figure out the pain points.

[odd trout]

@ocean wren Thank you!

[jovial kraken]

Some Ideas?^^

Or do you need more specific Infos about my questions?

[misty wharf]

It makes sense to have a base actor or a base blackboard for shared things yes. It's not required but it can be convenient. For death, it can be done either way, depends on what works better for your specific usecase.

[jovial kraken]

ah thanks ^^

[fossil pine]

I have a question does someone know a way how to spawn an actor behind the player and follow them like that. Finally like in "slender the eight pages"

[ocean wren]

I suspect so, yes.

[misty wharf]

You could literally just attach it to the player lol

[ocean wren]

No, nobody.

[misty wharf]

lol

[fossil pine]

Do you know a video about it

[ocean wren]

I do

In fact I know like a dozen or more?

I even did my own a few years ago

[misty wharf]

Is everyone doing slenderman games

[ocean wren]

Is that what all the cool kids are doing now?

Its hard to keep up with all the trends

[misty wharf]

Well I mean considering you said you knew dozen videos about that and had made one yourself too lol

[ocean wren]

Oh, no, I meant a video about having an AI follow the player

[misty wharf]

Ah :P

[queen urchin]

Hello guys
Could you help me please.
My AI keep attack player even when player is dead. I am not using BT.
AI attack BP:

[queen urchin]

Nevermind
Everything works fine now

[lyric flint]

I have an honest question:
Can u only work on AI through Blueprint? Or is it just be it's faster to start with the basic attributes and then make some advanced mods in the ide/code? Maybe it's not needed at all?
My question is not stupid, I am only ignorant.

[simple crest]

? There is nothing you can "only" work on in blueprint... Some things are more convenient in blueprint, and lots of things are only possible with c++

[ocean wren]

I do all my AI in C++, because its reliable and you can debug it.

[river osprey]

Hey guys, I've got a fairly simple question but can't seem to find an answer! Is it possible to 'restart' a behavior tree?

I have a Tower Defence enemy and want to send it back to the spawn point and restart the tree. Wanted to avoid destroying it if possible.

[misty wharf]

have you tried stop logic or whatever the node was called?

destroying the actor should be fine tbh

[river osprey]

I'll take a look for it, I wasn't even able to track that info down.

Yeah I'm just being a bit lazy. Wanted to avoid tracking status like slow/burn and health. But it might be the simplest solution

[pallid mica]

Aloah, does anyone know what Run Behavior Dynamic is supposed to do?
It allows specifying a GameplayTag.

Ohhh

I think it just made click

It's supposed to allow swapping out the whole Sub Tree Asset during runtime

Not realllly having a usecase for this or being able to come up with one, but well

[misty wharf]

@pallid mica I use it in my AI's - they have a list of tasks they need to execute, and each task has its own behavior tree. The main BT runs until the AI reaches the tasks stage, at which point it runs the task-specific BT's as a subtree, until it has executed all of them, and then it runs some finishing stages. Works quite well for this I think :)

[pallid mica]

And you set those when?

[misty wharf]

The tasks?

[pallid mica]

In the 3rd node from the left?

Yeah, since it's meant to be set dynamically during runtime

[misty wharf]

Ah, yeah the BTT_NextCustomerTask sets up the subtree

[pallid mica]

Makes sense

[pallid mica]

What type of AI is yours?

Some RTS worker unit or so?

(something like that at least?)

[misty wharf]

Yeah sorta like that I guess, my game is a video rental store simulator and this is the customer AI

[pallid mica]

Alright, cool, will keep that in mind when designing the trees

[pallid mica]

Do peeps still use the PerceptionSystem?

[misty wharf]

I'm using it at least

[pallid mica]

Iirc it requires some little c++ to make proper use of it. at least I recall having made my own sight sense instead of using the base one, is that still the case?

Also I only used it on a project that had one AI, are there any performance issues with 10+ AIs on the screen?

Cause Epic likes to overload stuff with more and more stuff

[misty wharf]

You need C++ for the teams part of it, and to tweak some sight-related settings such as if you want the sight perception to look at the head of some character instead of the origin, but if you don't need anything that specific it works fine in BP's only

The main performance thing with it is that it appears to do one linetrace per stimulus source that's within range at least for sight, so you get number of perceivers x number of stimulus sources number of linetraces per tick - but if it goes over a configurable time limit, it timeslices it, which can result in delays in actors reacting to perception

[pallid mica]

But you'd need to do that anyway

Even in a custom system

[misty wharf]

the number of traces can be reduced by using the teams function so that anything the actor doesn't need to perceive goes into some other team and you mark it to not perceive say friendlies

[pallid mica]

I kinda have to trace the targets to check if they are visible haha

[misty wharf]

yeah probably

[pallid mica]

Yeah that's actually fine I guess

We will have team changing AIs, but in most cases it#s x AIs vs 1 Player

So they theoretically only need to trace the player

Looking at the Sight Sense code, lots of stuff going on for something that theoretically only has to deal with some line traces :O

[misty wharf]

It seems to have some kind of "stimulus strength" feature as well but I've never really tried how that works, I'm guessing it might just report the strength of the stimulus in the events it sends

not sure what else it does behind the scenes tbh

[pallid mica]

Have you done any senses from scratch?

Just wondering if that is a lot of work

[misty wharf]

nah, I heard you can create custom ones for it but haven't needed to do that

[pallid mica]

Right, guess hearing, seeing, is fine

They also have Damage, Team, Touch, Prediction

[misty wharf]

I think some people just roll their custom systems, I was under the impression perception was the current thing you should use, replacing things like PawnSensing

I don't think EQS would even work as a replacement for it

[pallid mica]

No it won't work

[runic flare]

Hi, I have several supported navmesh agents defined in DefaultEngine.ini -- I just deleted the old agents and made a new set to match some gameplay design changes

when I build paths on my map it's still building some of the old navmesh agents

Ive tried regenerating Binaries and restarting the editor but these old navmesh definitions are still hanging around somewhere

does anyone know where these are cached and how to clear them out so my navmeshes will build only according to the project settings?

[runic flare]

I have now cleared the nav agent section out entirely and my Nav Mesh Bounds Volume is still generating meshes using the old definitions

[pine steeple]

@runic flare did you delete the recast data from the world outliner?

[runic flare]

@pine steeple I did yeah

[pine steeple]

thats odd

i normally just delete the recasts in the level, delete the bounds

then drag new one in, resize, and voila

[pallid mica]

I have some old code where I used Perception stuff and my own Sight and Hearing sense, where all I did was bAutoRegisterAllPawnsAsSources = false; in the constructor.
Guess what, I have no idea why I did that, but I recall this being a good thing.
Anyone some idea what that boolean is about?

Seems like this is basically just registering every single pawn as a source, even if the Pawn is not relevant

So turning it off, but turning on the boolean for register as source in the config of a specific pawn should cut down on redundant sources

[neat heron]

Howdy, when you're making heavy use of Levels for organization (level for light, geo, logic, ect) and streaming, i take it you want to have navigation on it's own seperate layer(level) right?

Or do you want the navmeshes to be like, baked onto the various geo layers so that navigation logic only loads with the areas? each route seems like it has ups and downs.

or should it be on the persistant layer? i actually forgot the persistant layer existed for a moment (woops)

Fwiw i'm mostly just learning the system and am making heavier use of level than i need to in a short fps game just to make sure i understand them.

[spring inlet]

i would put it on the geo level, as it's the only data affecting the navmesh

unless you work with a lot of navareas and such

[pine steeple]

navmesh is a pain on non streaming levels, and must be on a persistent level, iirc

[pallid mica]

Yeah I also recall it having to be on the persistent level

[pallid mica]

Can it be that SimpleParallel and RunBehaviorTask aren't good friends?

Seems like the RunBehavior instantly returns something that finishes teh MainTask of the SimpleParallel

Took me a bit too notice

Found multiple people having issues but no one that could explain why

The Task itself reports InProgress as NodeResults until the "DeactivationNotify" calls back.
Not sure what that is though

[queen escarp]

show us a screen

[haughty edge]

Hi guys! Can anyone tell me the difference between find path sync vs find path async? Currently I'm trying to see if it's possible for me to multithread the path finding part for my AIs. I don't really have much experience with multithreading in Unreal engine, so I don't really know if I'm looking into the right thing :P

[wise iris]

While using a dynamic navmesh with invokers, trying to find out how feasible it is to add holes/obstacles in the navmesh after it's been generated.
Has anyone here modified the (dynamically) generated recast navmesh at runtime?

[pallid mica]

You can't go backwards if that's what you mean

[pallid mica]

I know someone who did, but it was ugly and caused issues. I think we fell back to just using NavAreas that cut the holes.

[simple crest]

I'm guessing Lorash's Q is asking if two paths can each lead into the same next node... I don't think so

[wise iris]

Spawning a navarea would fix it immediately, even if navmesh already existed?

Or how do you mean?

[pallid mica]

There is a NavArea with that NULL Type

Don't know how it's called exactly

Compared to the ones that just increase cost

You should be able to spawn those runtime with NavModifiers

[wise iris]

Interesting, will note that down

[pallid mica]

Try it at least

Been a while

We also once had multiple NavMeshes

But we had to cut one out due to memory concerns

No each node only has one entry iirc

If you connect another one it disconnects the first

[pallid mica]

Your best bet is to check the source code

Given you want to add Multithreadding, you need c++ anyway, so you should be able to read through source

[haughty edge]

Roger that

Is it feasible for path finding to be multi threaded? Or it might require a huge re write

I'll look into find path async more and then come back :))

[pallid mica]

Yes and no, depends on how quick the response must be

If the code is so heavy that you need to offload it onto a different thread, then that means it will take a while until your AI gets the next path

[haughty edge]

I did an a star implementation with the current nav mesh system, I use it to allow jumping when searching for a path

I'm not very experienced so it's a really wacky solution, it's not terribly demanding if there aren't that many AIs, but I do hope to support as many as I can in the future.

Another not really related question, is there a way to place nav links when the nav mesh is generating? Because I only know how to add them after the nav mesh are done generating, and then it will start regenerating again :P

[pallid mica]

Yeah I think the last time we worked on something like that with the company was when we were making a few thousand AIs

But that was such a custom solution, it's not worth talking about

[haughty edge]

I don't think I'll ever have thousands in my game but that's amazing!

I thought AI move to task is multi threaded but I think I'm mistaken there :P

[umbral musk]

I have an NPC which I need to move to a location to perform an action. As a player, I could be standing in the way. So I call the NPC AI "MoveToLocation" function, and I bind an event to move complete. When that even fires, I look at the result of the event. If it's anything but "success", I have the NPC "ask" the player to move out of the way, unbind the event to make sure it won't be called again, wait a few seconds, and repeat the process.

This seems to work EXCEPT even when the player moves out of the way and the NPC attempts to go to the location again, the NPC still thinks it's being blocked. Any ideas? My navmesh runtime generation is set to dynamic and my "Dirty Areas Update Freq" is set to 2

[dull loom]

how do I get the NavMesh from the AI Controller?

[winged grail]

Hey, I'm just wondering if anyone can point me in the general direction/methodology for solving a problem I'm having with EQS:

I've got an EQS Query that generates a set of items/locations around an actor's "Target Actor" (i.e., the querying actor has a variable called "Target Actor", the EQS Query generates items on a circle using this "Target Actor" as the center). The items have an overlap test/filter that removes the appropriate item from the EQS Return if a pawn is overlapping it.

However, I can't seem to make the overlap ignore the querying actor. I thought that checking "Skip Overlap Querier" would have worked but it doesn't seem to effect it (I'm assuming that the test is considering the "Querying Actor" as the circle centre [i.e., the querier's "Target Actor"] rather than the actor that is calling the EQS).

Any thoughts or resources would be super appreciated, thanks!

[ocean wren]

You can add a second filter that takes out any items that are near the querying actor

The query is basically a set of scoring mechanisms.. so just add another one that essentially removes anything near the query actor

Skip overlap querier is there to ignore the actor doing the querying (most of the time you don't want to target yourself for instance)

[alpine junco]

Just to sanity check - do behavior tree nodes, like tasks for example, only run on the server by default? I'm assuming all behavior tree logic runs on the server only, right?

[celest python]

Yes @alpine junco

[winged grail]

Hey, thanks for the response and sorry for the slow reply! I had a meeting starting up right after posting. The problem I'm having is that my querying actor isn't being skipped for some reason (I'm assuming because the circle itself is not being drawn around the querying actor but rather around a "Target Actor" variable that is set in the querying actor's blueprint).

I'm essentially trying to tell the querying actor to draw a set of approach locations around their target actor, then approach that location as long as it is not being overlapped by an actor other than themself.

I'm not sure how legible this is, but it's a visual to demo the problem I'm having.

Picture of the overlap test as well (with skip overlap querier is enabled)

[west lark]

How would i make an AI that attacks my player and allows to be attacked and has different levels, so like a boss and a pawn type, and also has its own health bar that shows?

[winged grail]

That's kind of multiple separate things and it wouldn't all be encompassed by AI. You would likely make an AI that can attack the player, which would be separate from your "boss" character's blueprint, which could in turn be separate from your "pawn" character's blueprint (depending on the route you went), you would then likely code the health function within your character blueprints or within a function library or component, again depending on the route you took, finally, you would create the health bar as a widget and add it to your character blueprint(s) as a widget component. If you search for combat AI in google you'll find some nice tutorials that should serve as a starting point.

I found some of Ryan Laley's stuff helpful when I was first starting out, but you'll have to extend it on your own as it's usually pretty base level in terms of functionality. There are also some longer tutorial series for various genres depending on what you're making.

[ocean wren]

Butter: Gotcha, so in the EQS system, each filter has a query context. You can use different contexts to get different results. So your query context for the first filter might be the target so that you can use it for evaluating positions around. The second filter can have a different context, that actually provides the player to avoid overlap with. The point is that the filters flow top to bottom down the stack, so you can exclude the results of one with the next. Does that make sense?

You can use the EQS testing pawn to visualize what the results of each does too btw.

So you can check if its using the right objects in your EQS queries (can be useful, sometimes I've managed to mess up and give it the wrong agent in my contexts)

So you might want to check that you are passing the right objects via your query contexts using the EQS testing pawn.