gameplay-ai

[celest python]

Nevermind, it was not 'dynamic' at all, I misremembered. It just letting you know when you hit a specific notify in an animation and can be usable in BPs

But maybe you can check your own conditions when notify hit? Like doing this manually, instead of bindings things dynamically

[mossy nexus]

I mean the binding needs to be dynamic

basically check for the animation to either use the logic on a) anim notify, or b) at the end of animation, if no anim notify exists

[mossy nexus]

I would have to know the exact time per animation, for each animation

perhaps it would be something like splitting the animation up into a startup, active, and backswing setup

and start the logic at the same time as the active part of the animation

but I'm not sure that would gel with the current animation pipeline

[glacial sapphire]

Hello everyone! I'm trying to pass a variable from my AI to a task. what i put in the object?

[mossy nexus]

presumably the controlled pawn?

[glacial sapphire]

no

the AI

how do I take a variable from my AI and set it to my blackboard?

do I use cast or is there another way to do this?

[acoustic minnow]

You will want to cast to the class that has your variable which would be your AI controller or the AI itself then create a blackboard key and set it as an int or whatever type of variable it is

make sure your blackboard key is public. When you add your task to the behavior tree you will now be able to set your blackboard variable to that key in the detail panels of the task in the behaviour tree

[calm kite]

How can I prevent creating this navigation part on the actor I spawn in runtime. I want the actor to affect the navigation, but I don't want to have the navigation on it.

[past lodge]

How do i get vector into a float value for move up/down and turn right/left. I want to use the same movement function for the AI with the player function that bindsaxis for the playerinputcomponent

[mossy nexus]

so for composition's sake (and due to UE4's inherent architectural limits) I'm trying to make a component that I can just slap on anything that needs to be associated with my AI director subsytem. does anyone have experience with this kind of setup? I assume I still have to write a bunch of class specific logic that may be repetitive?

[misty wharf]

I guess it depends on what the component does

[mossy nexus]

... right, so how much logic can I cram into the component to automate the approach as much as possible?

[misty wharf]

All of it? :D

[mossy nexus]

I mean I can't readily put in e.g. an Attack() function because that would be different per AI

so already there it gets class specific

[misty wharf]

Right, so if you wanted it to automatically handle it, you could do an interface on the AI pawns and have the component talk to its owner actor through it

This would probably require the least amount of glue logic to be duplicated across the actors

[mossy nexus]

yeah but that's just a weaker way of doing composition

with the component, I can do the same AND add generic logic that is true for all classes

I don't think glue logic is necessary in either case, it's rather a case of not repeating logic and avoiding too much class specific logic

[misty wharf]

You could have a base class for the component and then have different components for different types of actors as well

[mossy nexus]

the ideal would be having class specific data but have the logic be dictated entirely by component

[mossy nexus]

this feels like just moving the functionality rather than genericizing it

[misty wharf]

Well it sounds like some of it is very actor specific

so the only way to really make that generic is to put it behind an interface or such as mentioned

[mossy nexus]

I don't think I agree with that. putting down an interface means all the logic will be class specific

it's not generic logic that works for all classes then, it's just that each class implements its own version of the logic

[misty wharf]

But so far what you've said is that you have logic which is specific to each actor type

how do you plan to make that generic?

[mossy nexus]

like I said, the ideal situation is having each actor have specific data that is used in a generic way

[misty wharf]

Yeah I guess you'd just need to have the component query for it

[mossy nexus]

now that I think about it perhaps it would be possible to do it like that. have a bunch of items and let the item attribute dictate which function should be used on the component... perhaps that is the way

[mossy nexus]

that could be solved by having a universal struct containing the necessary data I suppose?

[misty wharf]

Yeah

[mossy nexus]

makes sense. thanks for the bounces

[misty wharf]

👍

[restive turret]

AI Behavior Tree flickers when it wants to chase the player

[misty wharf]

The task is probably failing which then restarts the tree

[ocean wren]

Evening fellow AI

[restive turret]

how could it fail ?

[misty wharf]

If it's a move node it can fail for various reasons, such as missing navmesh, or invalid move location

[restive turret]

alright thanks i will look into it

[misty wharf]

Visual Logger usually has some more info on why moves fail so you might wanna check with that :)

[restive turret]

never heard of a visual logger how do i access it?

wait u meant the log?

if yes then i must be a dumbass

[misty wharf]

Window->Developer Tools->Visual Logger

[keen crow]

how do I stimulate one sense because of another? I added a handler for UPerceptionComponent::OnTargetPerceptionUpdated and want to stimulate Sight Sense from Damage Sense when damaging actor is outside of Sight Sense green are but still within its red area

[misty wharf]

just move your handling logic for senses into their own functions and call the appropriate functions from your handlers

this way you can just call that function from your damage handling code instead of having to do some really roundabout way of sending sight stimulus

[mossy nexus]

what would be the absolute cheapest way of getting a route between two points on navmesh?

the route is only a needed for a check

not traversal

[vapid zinc]

i am doing an ai without BTs or BBs all in c++ should i put the ai logic in the ai controller class or the character class?

[misty wharf]

I'd imagine you just have to do NavSystem->FindPathSync or such.. this is what I'm doing for reachability checks between buildable things and NPC's so they don't attempt to move to things they can't reach

[misty wharf]

How you split the logic is up to you. Personally I find it makes sense to create "actions" the character can perform and put those into the character - eg. things like throw, pick up, look at something, etc. - and those can be then called from your actual AI logic or other logic

Basically the character is like a "dummy" which something else orders around

This makes it easy to swap out the AI logic if needed, or just call the functions directly with no controller for testing purposes

[mossy nexus]

that's my setup, plus a director that dictates everything to keep it more in line with ECS

[hearty niche]

how do I bind a function to success or fail of Move To Location in bttasknode?

[misty wharf]

You need the node to be instanced first off, then you just bind a function to it the same way as anywhere else

[hearty niche]

instanced?

I did override the execute task function and wrote my logic

what dispactcher would I be binding to?

[misty wharf]

Oh - if you're just doing it in BP's, just call the regular move to node which has the on finished pin

[hearty niche]

its cpp I am just converting my bp to cpp for performance

I think the right word was delegate

[misty wharf]

ah

yeah you need to mark your C++ task as instanced, it has a function or a flag on it somewhere that you just make true

I don't recall off the top of my head where the delegate for the move is but it should be in the move request or the pathfinding comp I think.... just look at the code for the BTTask_MoveTo or whatever it was called, since it does all this already

[hearty niche]

wait how do I mark it as instanced?

[hearty niche]

gonna try that again as I could not really get how it's doing it.

[misty wharf]

Interestingly it looks like it's actually not instanced

so they did it in a different way for that one

instead they seem to just use the tick feature to check when the movement is done

[hearty niche]

I found a delegate called on delegate finished in the path following component that could be what I am looking for I guess

[celest lodge]

For some reason if my player is at the x the enemies get stuck at the circle, I have a nav link proxy that is left to right only set to have them jump down, and if I am elsewhere not on that platform they are able to traverse the stairs just fine, I can't figure out why they clump if I stand there though

[celest lodge]

I set it under smart link, not simple link was my issue, I changed it there and now everything works

[hoary peak]

I'm using the MoveToActor node to have two AI's walk towards each other and for the most part it's working but sometimes they overshoot and pass through each other. Is there a way to make it update more frequently?

[runic flare]

Try this?

You might also want to tune the Acceptance Radius based on how close you want them to get so they recognise they've "met"

[woeful coyote]

Anyone can help me on this case ? https://youtu.be/skyXVj73gQs

[uncut python]

are there any tutorials on replicating AI? Im having a tough time getting my ai to chase and attack me correctly

[woeful coyote]

same issue here for me...

[celest python]

Anyone tried to use BTs as individual logic trees?

It seems impossible due to design

[celest python]

Like instead of binding to AIController just running it

[uncut python]

im like in a huge mess of bps tryign to do this lol

[celest python]

AI framework is not replicated

Movement is replicated by default if you run move to task in server

Attacking logic is depending on you, you need to replicate it in your character/pawn class, ofc that depends on your way of doing things

[uncut python]

@celest python I see, the biggest issue im having is geting AIPerception to replicate where it will chose the cloest player to go to

so i kind of made a work around but as soon as i switch it to run on server it shits the bed

[woeful coyote]

Do you think you may have issue with the Blackboard variables ?

[uncut python]

Yes, it was wasn’t being set right after this code,. Thank you sm

[pine token]

any idea how to handle navmesh building in an infinite world? i make a voxel game and currently i just move the navmeshbounds volume around and then trigger a navigation rebuild, but that is causing a lag spike although i enabled "lazy" navmesh building and threaded navmesh data collection

[pine steeple]

why are you replicating AI stuff?

@pine token use NavInvokers on large world

[pine token]

i still need a navmeshboundsvolume with invokers

seems like OnNavigationBoundsUpdated updates the bounds

[runic flare]

Has anyone run into a situation where their AIs won't choose the correct navmesh?

I have multiple navmeshes, each with a different agent radius defined

but my agents keep choosing the default mesh regardless of their capsule radius

I've narrowed it down to this method UNavigationSystemV1::GetNavDataForProps -- here NavDataInstance is always NULL so the MainNavData is returned

any ideas?

[stone walrus]

why we use tree meshes in foilage if it already have lod

[mossy nexus]

does anyone know what this decorator actually does?

it seems entirely useless, or very arbitrary

does it discriminate beween blackboardkeyselectors?

[misty wharf]

Huh?

What that does depends on the type of the blackboard value

For objects and such, it does nullptr checks, for numbers it can do greater than, etc.

[mossy nexus]

what does it do for vectors

[misty wharf]

Dunno, try selecting a vector value in its properties and see what options it gives you I guess :)

the options change based on what it can do with the value

[mossy nexus]

it doesn't give options

[misty wharf]

It might not do anything with vectors then I guess 🤔

[mossy nexus]

I don't get notifies either. result change or value change... what's the difference?

what does it mean that a vector value is set or not is not set

it's written so obtusely

[misty wharf]

my understanding is that value change will trigger even if the same value is reassigned, result change triggers only if the actual result changes (eg. false becomes true)

set or not set for vectors doesn't make any sense yeah, for objects that's a nullptr check

[mossy nexus]

so what's going on? looking at the code I can't find anything

there's BlackboardKeyType.h that houses the enums used and some generic functionality, including comparisons

but there's nothing that seems used in that node

[misty wharf]

Not sure what it'd do with vectors, I've only really used that with integers and objects

[mossy nexus]

surely it's means to be used for vector comparison too

[simple crest]

99% sure if you look at the code it's written like, if a vector is 0,0,0 it returns true as "unset" (yeah, it's very lol)

[pine steeple]

isnt InvalidVector MAX_FLT?

[simple crest]

ah yeah. I'm probably thinking of how it treats bool by accident

[keen crow]

I've started learning custom NavLinkProxies and I have a couple of questions

1. Say I have a nav link for mantling on ledges and I have 4 crates in a row and I want my AI to mantle from 1st to 4th. Do I just place my custom nav link proxies one after another on top of each other or is there a better way?
2. Is there any event to subscribe to or a virtual function to override to handle reaching end point of nav link proxy? I have a slide mechanic and I want my AI to start sliding on reaching left point and resume walking when it reaches right point but I just can't find anything to override

[misty wharf]

look into smart nav links, this allows #2 at least

[keen crow]

Could you be more specific what exactly must I do with smart nav links? I'm inheriting my link proxy from ANavLinkProxy and already subscribing to OnSmartLinkReached event to start an action, now I need something like OnSmartLinkLeft event but I can't find one

[misty wharf]

you need to determine it manually based on when the actor reaches the end of the link

[keen crow]

I still don't get you.

when the actor reaches the end of the link
this is the thing that I want to know how to do it

[misty wharf]

You need to handle it in your actor that's moving. You can create a custom smart link component, which has an event which fires when a pawn reaches it, with the destination point for it as well

Then you tell the pawn to move to the destination point, and the pawn can have an event for when this completes, which you listen to

iirc you can also extend the nav link proxy actor to create your custom navlinkproxy which can handle this logic too without needing a component... but if you want to have a smart link as part of an existing actor then you'd need to make a component for it

[nimble vessel]

Im so happy I think I finnaly got it down how I should think when using sequence and selectors

[nimble vessel]

It checks if the vector has been set. Clearing it will make it "unset". Although I have a had problems using blackboard based condition when it comes to vectors a few times. Therefore I wrote my own decorator for checking Vectors.

[mossy nexus]

Set and Unset make no sense as conditions for vector. it could mean that it checks for a FVector::Zero() or it could mean that the value has changed or anything really.

but

I found out yesterday that the Set has to do with whether it's within bounds of the AISystem

so it's completely useless unless you use that

but you won't understand that information from just reading Set. it doesn't implement arithmetic operators for FVector as it does for float

[misty wharf]

Well that's kinda weird :P

FORCEINLINE bool IsValidLocation(const FVector& TestLocation)
{
    return -InvalidLocation.X < TestLocation.X && TestLocation.X < InvalidLocation.X
        && -InvalidLocation.Y < TestLocation.Y && TestLocation.Y < InvalidLocation.Y
        && -InvalidLocation.Z < TestLocation.Z && TestLocation.Z < InvalidLocation.Z;
}

Yep, this is all it does

InvalidLocation is FVector(FLT_MAX)

so it literally just checks if the vector value is within the bounds of the float datatype

which seems like it would pretty much always be unless something was horribly wrong

[ocean wren]

Probably fails with NAN 🙂

[odd nexus]

Hi there. How can I build a path with a vector array for an AI? So, I have a vector location for each frame (25FPS), the goal is to move the AI along the Path for a 6-7 seconds, but the movement should be smooth and realistic

[mossy nexus]

for each path point, find path between path point and previous path point (unless it's the first point in the array)

[ocean wren]

What have you tried so far?

[odd nexus]

I have tried to make it with Behaviour Tree, with ServiceNode, that update the target location by using the for each loop and with the Task Node MoveTo. But this for loop ends immediately, that target location is defined only throught the last element of the array.

[odd nexus]

Thank you, it looks helpful for the future, when I will find out how to implement this AI-Behavior in general :)

[simple sigil]

Does AIPerception support interval updates? I'm looking at wanting to revaluate what's already perceived to be able to react to changes like distance. So if my AI sees a group of actors it can switch target if who is the closest actor changes. Right now I'm getting OnPerceptionUpdated events when an actor leaves perception, a new one joins, and if there are any fired stimulus events such as damage. I don't see a straightforward solution for rechecking stored actors to update the behavior tree.
I'm not sure if AIPerception is meant to function with PawnSensing (They seem to do a lot of similar things so kinda like it's own implementation. I notice that PawnSensing has EnableSensingUpdates and SensingInterval which would allow for the kind of behavior I'm looking at, but I don't really want to mix & match two systems if they're not meant to be used together.

[misty wharf]

@simple sigil you can use a BT service which goes through the list of currently perceived actors and assigns the appropriate actor to a BB value

Or you can do it on tick in your AI controller and assign to BB, or do it in a BT task whenever it gets executed, etc. - various ways to do it :)

AIPerception afaik supercedes PawnSensing

[dapper spear]

Hi may I know what is the difference between behavior and tasks? I thought they were the same? Some tutorials mix the name. Thank you

[tranquil reef]

hello!
does anyone know how I can get this dynamic age value in blueprints?

[frigid cipher]

This condition is Failing, despite the Key Query being Is Set, and the bool is True. "value: true (fail)"

[celest python]

Do 2nd decorator evaluated if first one fails in BT?

[pine steeple]

no

you can use a composite decorator if you need to make an or

[hearty niche]

Is this the right way to check if pathfinding is complete in cpp?

Controller->GetPathFollowingComponent()->OnRequestFinished.AddUFunction(this,"ReachedLocation"); // inside another function

void UBTTask_FindLocationRelativeToPlayer::ReachedLocation(FAIRequestID RequestID,const FPathFollowingResult& PathFollowingResult){
if (PathFollowingResult.HasFlag(FPathFollowingResultFlags::Success))
{
FinishLatentTask(*BTComponent, EBTNodeResult::Succeeded);
}
else if (PathFollowingResult.HasFlag(FPathFollowingResultFlags::InvalidPath)||PathFollowingResult.HasFlag(FPathFollowingResultFlags::Blocked))
{
FinishLatentTask(*BTComponent, EBTNodeResult::Failed);
}
}

[misty wharf]

Does it work? If yes, then yes, if no, then no

:D

[hearty niche]

I mean syntax 🤔

[misty wharf]

What do you mean? If it compiles it is the right syntax

[hearty niche]

I guess I should test it. Sadly the rebuild also rebuilt the engine again :/

[misty wharf]

ah

I don't use an engine source build so thankfully don't have that problem :P

[hearty niche]

ue5-main so yeah

[misty wharf]

I would imagine there must be some way to exclude it from being built, similar to how the EGL distributed engines work

[hearty niche]

not that experienced so no idea. The problem is when I right clicked my project and clicked rebuild, it should rebuilld my project not the engine -_-

second time this is happening

[hearty niche]

I think the perception settings for the sense in bp determines age duration

[keen crow]

How do I make AI do something when he enters and leaves a zone? For example I'd like to place an area on a map in which AI character starts sprinting when it enters the zone and stops when he leaves it

[hearty niche]

you could just use a trigger and on begin overlap set the variable on your AI to make it sprint and disable it on exit. Pretty sure there are other ways though @keen crow

[keen crow]

Hmm I guess I could but for some reason that doesn't sound right although I'm very new to AI in games. I was hoping there's something to do with smart links in NavLinkProxies or just some existing funcationality in AI components or smth. Like I know you can handle an event when a character enters first point of NavLinkProxy but I couldn't find an event or a function to override when the character leaves navlink. But thanks anyway, I guess I'll try it your way

[lyric flint]

I want to:

• create a scene in unreal that plays a boxing match between 2 computer controlled players
• make camera move the way it moves in real life boxing translations
• record screen of this game, together with CSV file containing 3d transforms of the actors/control rig movement data with time stamps)
• develop a Classifier using this training set, so it could
• if 1080p video of real life boxing game translation recording is given, spit out this stream of floats that is convertable into UnrealEngine animation

How it approach the development of classifier? Say I have a training set - many video files and huge CSV files for each of them

[ocean wren]

Wow, you go for the real easy questions first don't you

Try classifying content from an existing dataset first, there's plenty of pose estimation datasets out there with hand annotated keypoints for still images at least

There's a few from Michael Black's lab for 3D too

But you're talking about a pretty big project. This goes to the heart of a thing called "Motion Matching" if you want to search for that term in semanticscholar or google scholar

CMU also has a 3D pose dataset with video too if I recall (from their multi-camera surround volume)

3D pose estimation from video isn't really a solved thing yet though

https://paperswithcode.com/task/3d-pose-estimation

Ah yeah, CMU Panoptic dataset.. useful for you I think.

[hearty niche]

there is no real right way of doing things and from what I know the navlink proxy just affects pathfinding choices for encouraging or discouraging certain areas for pathfinding

[ocean wren]

NavArea is what you want, its basically a volume, but honestly you could just use that thing that was on the marketplace for this kind of thing?

Navlink's are meant to be just connectors between navmesh chunks, so not really what you're after

You could basically do it with any volume where you can get enter/exit events.. then just change your values on the AIController and feed them to your blackboard as per usual

Navarea is a volume, but it also changes the cost of navigation. But that doesn't really sound like your goal, so just use a trigger volume instead and change a property on the controller to say "I'm inside a trigger" on enter and unset it on leave

[hearty niche]

exactly BP interfaces can make it even easier and save you casts

[ocean wren]

Trying to optimize BP's with interfaces seems a bit redundant.. if you're doing things in BP its going to be slow anyway 🙂

But yeah, why not

[hearty niche]

I mean you would save on trying to cast to the right thing and not be stuck with one parent class

[ocean wren]

So our recommendation is.. use a trigger volume, set a flag (that gets put in the blackboard) and unset when exiting

[hearty niche]

any actor with definition for the logic would do its thing or else the nothing happens

[hearty niche]

what I sad as well you could derive a bp from the trigger class 🙂. \

[ocean wren]

Yeah, that's the usual way

[hearty niche]

I doubt what he wants to do needs to got to the BB

you could just set the walk speed on character movment. he just wants to change speed

[ocean wren]

Well, might as well do things right, throw it in the BB and use your BT properly

[hearty niche]

unless his BTs affect movement speed in some other way as well

[ocean wren]

Yeah, fair point.. I was thinking he wanted a bit more complexity than just setting the speed

You could do the same thing and have an AI queue for a ladder for instance, or wait for a traffic signal. I was kind of thinking of those types of use-cases

[hearty niche]

hmm those are different cases as I said. changing movement speed by going into BB and all sounds kinda more complicated than needed

[ocean wren]

It can be useful to also allow filtering via tags or some such to enable/disable volumes for different classes of actor

Yeah, for just speed you'd just set the movement speed as you said

But when is AI ever just about speed 🙂

[hearty niche]

that's one way of doing it I guess

isn't tag comparing kinda expensive?

[ocean wren]

If you did it every frame, then yes it would be bad.. but once? nah

remember this is simply onenter/onexit stuff

[misty wharf]

Aren't tags just names so they should be fairly cheap to compare

[ocean wren]

If you're worried about speed, you could always use a hash

[misty wharf]

Of course it depends on whether you also need to find the object which contains the tag first and not just look at tags on some ref you carry

[ocean wren]

Well, in this case, you'd get the overlapping actor on the entry, so you could just get the tag value directly once and switch behaviour, as you say, a string comparison although if I recall, the tag values have some optimisation? can't remember if it was the UE default tags or our own

I've used hashes as tags before to avoid string comparisons, so you could easily add your own "tags" container component that simply uses hashes rather than tags (hash the tag string into a hash number and do a number comparison instead)

[hearty niche]

for example, I have a spherecast firing before the enemy shoots to prevent friendly fire so that can be easily 20 plus calls depending on baddie count

[ocean wren]

But in this particular use case, I doubt you'd notice any perf problems

[misty wharf]

iirc a tag in UE (and gameplay tags too) are FNames, which are basically converted into some kind of numeric lookup or something so it's faster than a string comparison

[hearty niche]

yeah saw an article by Tom Looman on those. Might check them out soon

[ocean wren]

zomg: sounds right..

We rolled our own version years back too

Not hard to do, generally useful

Boils down to a unsigned integer comparison, so its damn quick

Well, I guess bitwise comparison in UE's version

Would recommend 🙂

Vegito: I hope you're cache-ing those spherecasts 🙂

[hearty niche]

why would I cache? it is a bttask which checks if there is a friendly blocking the way to the player. If not, shoot or else task fails

tag comparison is something I could optimize though

[ocean wren]

I guess it depends how often you're shooting, but if you have a lot of enemies that'd mount up

[hearty niche]

as of now I am just converting my bttask for ai to get a good location to the player to cpp for performance.

It has an eqs, patfinding and a timer with distance checks which gave 60 fps with 20+ enemies. Gonna optimize that one first

[misty wharf]

Tbh you might not see much of a difference if it doesn't have a lot of BP nodes

[hearty niche]

I do have a lot of distance checks every 0.5 seconds

one enemy to player, one player to eqs origin point (whre the player was when the eqs was called),etc

[keen crow]

@hearty niche @ocean wren my original question with a sprint was just an example. The real problem is I have various mechanics for player in my project like mantling, sprinting, sprinting to sliding, sprinting to wallrunning, proning and say I have a logic in my AI controller that under some occasions AI character just goes from one point to another by AI Move To Location functions. Now I need to know how to teach AI characters to slide under low obstacles, how to wallrun over pitfalls, how to use ladders, etc. I managed to implement mantling for AI but that was easy because I just had to create a custom NavLinkProxy and invoke base character's Mantle function, but things get harder on situations like sliding, wallrunning, sprinting, etc where I want AI character to start some action before obstacle and finish after it so that it can reach destination point set by AI Move To Location. Your idea with trigger volumes is nice but I haven't worked out yet how do I determine if by entering the trigger volume should character really start say wallrunning or if he is doing some other login like chasing character or hiding in cover or whatever

[ocean wren]

Hmm, I see. In that case, I probably wouldn't mark up the world and instead would build a navigation controller on top of the movement component to alter the methods that the movement component uses to traverse. Maybe even call it a traversal component. The idea of this component would be to classify the neighbourhood around the agent to determine how best to move through the world locally. So might do a bunch of sphere/raycasts to determine if it needs to duck etc. Wall running would be a proximity check for a bunch of raycasts or a spherecast and contact distance check to activate wallrunning. They did mention a bunch of this on the Wolfire blog back in the dev phase for that game, so might want to look at that. So the main advice I'd give, is to build it into the characters movement otherwise I could see your world being full of hint volumes and getting a bit difficult to edit.

Usually, what you'd do, is try and identify nearby opportunities for smart movement. So is a wall near me? then wallrun, if there a gap coming closer in front of me? start sliding. The way you discern those opportunities is up to you. You could always mark up the world (they do this for assassins creed games for instance), but it feels like you'd be wise to do some mix of ray/sphere checks and world markup.

This kind of thing is related to the "smart object" concept in game AI, where we annotate world items with potential uses and animation data and lets AI's decide which nearby smart objects satisfy their needs

[keen crow]

The idea of this component would be to classify the neighbourhood around the agent to determine how best to move through the world locally. So might do a bunch of sphere/raycasts to determine if it needs to duck etc. Wall running would be a proximity check for a bunch of raycasts or a spherecast and contact distance check to activate wallrunning.
is it something EQS is designed to solve? I've watched some videos about it but it seems kinda complex and I'm not sure if it really fits for problems like mine

[ocean wren]

EQS could certainly be part of it for sure

[keen crow]

then I guess I just don't know AI in UE4 good enough yet to solve my problem. I'll try to learn EQS more because as you said

I could see your world being full of hint volumes and getting a bit difficult to edit.
that scares me too

[ocean wren]

So here's an example of one thing I've done.. I want to stack a group of AI soldiers up on a door, breach the door and make them enter the room in a "spec ops" manner. So I have components on the door for "stack up locations" where I can edit those per door to make the stacking work. The AI soldier uses EQS to find the stack point that is 1) closest to the door 2) not already occupied 3) on the correct side of the door etc..

Now I don't want to annotate EVERYTHING in the world. But doors and door handles are obvious special cases, so makes sense to mark those up.

But walls? no, I'll sphere/raycast for those

Now I've seen some games (notably some CryEngine ones) where they basically have designers annotate the blueprints for every item that can be placed in the level. So cover locations, mantle locations etc.

Personally, I'd balance it a bit.

Annotated world content looks better generally. Less prone to making errors. But adds clutter to the editor unless you can switch off visualizations etc (you should enable that)

So for instance, you could annotate windows to allow soldiers to shoot out of them by standing in a certain position OR you could just get the soldier to find actors with the tag of "window.open" and go and stand at the nearest middle point of the window and fire and hope it works 🙂

The latter actually does work most of the time, but means you need fairly open windows and kind of fails if you start to have fancy windows where they don't fully open etc.

Always a bit of a tradeoff

You can also have smarts in the objects themselves. My door example, the door controlled the execution order of the breaching because it had different sequences for breacher entry1 entry2 etc.

Ladders were another one where I had more of a smart object approach. Because you want them to queue at the top or bottom if someone is on the ladder etc.

ideally you need your squad to cover the other end of the ladder 🙂

Same for elevators, trucks, carts and the like.

So you hand off some AI from the character to the object to be interacted with. Instead of "use this <door" you go "door, I want to use you, tell me how"

If you were being a proper AI programmer. You'd build an ML model from examples of your own world traversal and apply that as your traversal component 🙂

[keen crow]

honestly this approach with "smart objects" is not really to my taste right now because my previous enterprise background tells me that objects like doors shouldnt really controls actors, or characters how should they interact with it but still thanks for advises. For now I guess I'll fill my EQS knowledge gaps and return to my problem again

[ocean wren]

Well, just know that the smart object approach is what Bioshock uses, and the sims

and plenty of other games

[keen crow]

guess i'm too narrow minded then 🙂

[hearty niche]

there is no right way of doing stuff in gamedev as I said earlier. As long as it works does not stall fps any smoke and mirrors trick can be appropriate

[ocean wren]

The reason they use it, especially in the sims case, is that you can do extra content packs with new objects if you make the behaviour of object usage reside in the object itself.

[keen crow]

good point

[ocean wren]

So they can do loads of new content packs by editing a base "use me this way" script and edit it for new objects. A toaster and an oven for example.

There was a nice presentation from GDC by a guy called John Abercrombie about Bioshock's smart objects in the beach scene.. its on YouTube

[hearty niche]

I think dishonored does it too right?

[ocean wren]

John now works at Epic 🙂

[ocean wren]

Yeah, pretty sure. Probably in that video by Laurent about Dishonored 2?

[hearty niche]

dunno it just feels like it when you interact with stuff in the world I just took an intuitive guess

[ocean wren]

Its a pretty popular pattern, not without its downsides of course.

[ocean wren]

I wouldn't be at all surprised. We were all kind of discussing similar situations back when they were doing dishonored. Most of us were using smart objects in one way or another. If I recall there's some videos about ambient AI's that Ubisoft presented a similar idea too.

[hearty niche]

what are the downsides tho?

[ocean wren]

Well, your world becomes cluttered with annotations. Some poor schmuck has to annotate everything. Some poor schmuck has to check the annotations are correct. Often annotations work ok in test cases but when placed in real levels they fail. Stuff like that.

For example.. I annotate a door to say "you can use me, just stand here and play the door opening animation", then some other schmuck decides to place a barrel in front of the door. The AI finds the door smart object, tries to navigate to the "you can use my stand here" location.. and fails 🙂

One response to that of course is to have an AI whose purpose is to try out all the smart objects 😉 so do coverage testing.

Just have them move around the level trying to use smart objects for stuff.

And log any failures

Automated testing of AI is definitely getting to be more common for the industry. I know plenty of devs who do it, including most of the bigger publishers.

For instance, if you aren't automatically testing your navigation, you're missing a trick

Hell, we did it on worms 2 back in late 90's early 2000's

Not so much for indies though

sigh 🙂

Had a good discussion with a guy from Naughty Dog about it one night in a bar. Apparently they'd not done it fully yet. It was after a presentation by Microsoft where they had like 500 Xboxes testing overnight. I suspect they do it these days 🙂

[misty wharf]

Worms 2

...I was 10 when that came out

probably got it a year or two later but it was pretty great lol

I didn't really play the rest of them after that tbh, but they more or less felt like "worms 2 but with more"

[mossy nexus]

I'm sitting with a problem of defining high alert actions for AI. I've found 3 cases that I'd like to implement (each of which I know the logic for):

Raise Alarm
Attack
Escape```
I'm thinking whether or not I should implement these as behavior tree actions (that are unknown to the AI, but the AI has a type preference between them) or if I should put them directly into the AI as behaviors so that the BT can have a generic `UseHighAlertAction`

Worms armageddon was Worms 2 with better singleplayer content in form of missions IMO. for a long time it seemed like the premium version of worms 2 that us worms 2 peasants never had access to. I was already an adult when I first tried armageddon

[ocean wren]

WA was basically a rewrite of a lot of the frontend part of W2, had some neat weapons and generally was a bit more polished and less cobbled together 🙂

Oh and I rewrote the servers and whatnot for WA

although it was still a nightmare when we got chinese players and half a million of them tried to logon at once

[mossy nexus]

I have a follow up to this. While I have a way to get attacks working, I am wondering if there is a good pattern or something to genericize attacks. I'm thinking right now of having a data table with attacks that essentially store path to a UObject where the logic is contained for each attack... but perhaps there is a better way?

[lyric flint]

Where to read / watch about new Unreal Machine Learning plugin?

[celest python]

Instead of datatable I have a TSubclassOf<UAIAttackActionGameplayTask> (its not a gameplaytask per se but it works too) in my Pawn class and BT creates and runs that

Works for my system, might not be suitable for your system

[mossy nexus]

ok, why have you chosen to do it like that?

[celest python]

I wanted to encapsulate attack actions because while some AI's can only do melee attack some of them has multiple weapons and behaviors, so I needed to decouple this system from BT, otherwise I would need to create BTTask for each behavior

Instead of I just create gameplaytask-like objects now which can be callable from single bt task for each AI

Also since gameplaytask class is just a variable, I can change it dynamically

[mossy nexus]

so I'm thinking for my system to use a generic HighAlertAction that just calls whatever the current high alert action is (this can be an attack, but it can be other behaviors), and let the AI themselves handle it depending on circumstances

like an AI might have their preferred action changed dynamically to ESCAPE if they realize they can't win

however, that would mean I would need a new behavior tree for every type of action

which isn't great

[celest python]

You might not need to. You can do a similar system to mine, but there is one pitfall. For example AI realized it can win, and executed ATTACK behavior. But player did something and they changed their mind and now they want to use the behavior ESCAPE. You somehow need to solve how you can signal this to BT and reset the current running ATTACK task

[celest python]

That's also a problem I'm currently working on now. When AI got damaged and health is lower than X, I want it to use different weapon with different attack task, what happens if this situation happens while I'm running one attack task already?

It's easy to fix but I'm struggling to find the most clever way

[mossy nexus]

I think if it has committed to an action already, then that's that

the next action becomes the important action IMO

I wouldn't attempt to switch actions mid action

but if it's in the process of reaching a place where it can execute that action, then it can switch

equally in my system I don't query the next task until right before it is being used

[celest python]

Alien Isolation solves this by using some kind of states in BT (zoombapup will love this lol), before any kind of action they evaluate decorators similar to "Should Search Rooms" or "Should Attack Player". If "Should Attack Player" returns true, they just run a single task and if something that can prevent this situation happens, its already considered before "Should Attack Player" state so they can cancel it without any problems. For example if some NPC damaged the Alien very badly, they already runned "Should Run Away" before "Should Attack Player" so next tick, before running attack behavior they will execute running away behavior.

If you are not switching them as frequent as me, then its a good pattern imo.

[mossy nexus]

like I said the only crucial moment for me where a switch cannot happen is from when it queries the next action, to after the next action is executed

at that point it will have committed to whatever action. anything up until that point is fair game

but my case is made worse a little (or better, depending on how you look at it), because I've chosen to use a data actor as a generic type to hold all data for actions

and the same data actor is being used for high alert actions as with passive actions

perhaps I just need to subclass it for some extra specific high alert based logic

but I'd rather keep that in the AI component, and not in the action itself

[celest python]

AIController?

[mossy nexus]

(though in some cases it's not possible - for instance, for animations, the animation notifies decide when the event happen with every animation)

[mossy nexus]

I'm inadvertently not using AIController a lot

I've basically created an AI Component that hooks up to the AI director

[celest python]

Nice

Because AIController would be a bad choice

[mossy nexus]

the component handles perception logic, as well as instinctual behaviors (e.g., high alert actions). anything else is handled by the BT

[celest python]

I just store behavior-related things in the class directly, and try to instance/subclass things as possible

[mossy nexus]

I would, but it would remove the genericness of being able to just slap a component on an actor and make it work

I do hookups between components on the actor though... like when health matters

[celest python]

Behavior itself shouldn't be generic in my opinion. BT can be generic but pawns actions and behaviors not

I do my best to instance everything as possible

[mossy nexus]

well like I said, actions depend on what the action data actor has stored

[mossy nexus]

I think I'd need to see an example of what you mean here

[celest python]

As said in there, single BTTask calls pawns stored action. Each enemy type has its own BT, but every BT calls that UBTTask_RunGameplayTask when they need to do something specific

[mossy nexus]

so far I'm thinking:

- Pool of potential attacks (Data Table or UObjectList) on the AI component.
- On high level action -> if Attack -> Query an attack from pool based on conditions.
- Spawn ActionData Actor with Attack at target's location.
- Adjust AI component's sense towards action
- BT handles movement and action once everything has been setup

[celest python]

Though just keep in my mind, I'm explaining how I did my setup for my own specific needs. I don't really know what you need for your system

[celest python]

Is your game something like turn based or similar?

[mossy nexus]

nope, real time

[celest python]

So then you will do something like my system, BT will only signal to X (actor/component/action .. anything) and after that it will handle tracking the target goal etc.? - asking to understand your case

[celest python]

But yeah generally looks good to me

One thing I can recommend is decouple your AIComponent

Don't use it as base of everything

Do AIPerceptionComponent, AIAttackListComponent, AIBlahComponent but not AIComponent

[mossy nexus]

that's why I wanted the action data elsewhere

[celest python]

It still communicates with single component that handles many things

[mossy nexus]

I would decouple, but that means I have to use the actor

and I don't want to use the actor

there would have to be a lot of repeat logic in actors if I have more than one actor with the component

I handle perception in the component, but not behavior generally. Perhaps the AI director should have responsibility for creating the correct action for the AI

[celest python]

Can you make me understand, why it will repeat logic? It will split the tasks, one of them will do X, other one will do Y and owner actor/pawn will act like a base of them

[celest python]

Yeah, but if AI is only targeting player, it can be one of player's components too

[mossy nexus]

yes but consider: any communication between components must be handled in the actor

I make one single exception, because I use one type of component as "eyes" for the perception, and the AI component gathers these themselves

[mossy nexus]

it's not a given. it wouldn't be systemic just to target the player

[celest python]

How do you calculate how AI should react/attack?

Based on which stats

Position, health, direction etc.

[mossy nexus]

I do this instead of carrying a list of transforms for each eye basically

[mossy nexus]

right now I use a "Stress" fluid state that works similar to alertness

any time perception triggers, stress increases

at certain thresholds behaviors become unlocked

[celest python]

So you kind of mapping behaviors based on stress level then? That data actor provides "stress level" and if its higher than X, you're doing Y etc.

[mossy nexus]

yep

[celest python]

Probably an utility system would work better for you

Instead of BTs

[mossy nexus]

the BT takes into account stress levels and tells the AI to do things based on that

I don't think so?

I've used BTs to model all other behavior

the high alert stuff is my last issue

[celest python]

Nah I was meaning generally not for this specific case

You can ignore it

So why don't you literally map your tasks then, I mean literally using TMap<>

[mossy nexus]

I still think I will have to make three different subtrees for the three high alert behaviors I have

[mossy nexus]

go on...

[mossy nexus]

I should say, the perception system outputs the stress level, and nothing else

[celest python]

struct FAIAlertBehaviorWrapper
{
  // Possible alert behaviors for specific stress levels
  TArray<UAlertBehavior> AlertBehaviorArray;
}

TMap<float, FAIAlertBehaviorWrapper> AlertBehaviorMap;

[celest python]

You can add our own level of stats to calculate which one will be considered in that array

[mossy nexus]

ah I see

yeah I do that already, but in a different way internally

[celest python]

Instead of UAlertBehavior it can be FAlertBehaviorData which also holds conditions for that alert etc

And it can be UDataAsset to differentiate it between situations, AI types etc.

So AI director or data actor can only select a data asset

[mossy nexus]

I think the map is a bit overkill as it would make behaviors very unpredictable

I stick to thresholds because they keep things at a baseline

and players need fairly predictable behavior to model their playstyles

[celest python]

How about TMap<FVector2D, FAIAlertBehaviorWrapper> AlertBehaviorMap; then

Adds a range/threshold 😄

[mossy nexus]

I think the BT kinda solves the issue without having to deal with complex mapping

it outputs behavior based on stress thresholds

and it's very easy to see the visual of what's going on

plus I can access passive behaviors as well, when there is no stress

[celest python]

That way you need to handle differentiating that 'actions' -- which leads us back to your first question

[mossy nexus]

well... I would have to anyway

the map solves a problem I've already solved

but I think at least I've understood that the director needs to have the responsibility of the action data, based on the preferences of each AI

and yeah, maybe you're right with the map I wouldn't have to make subtrees, but that would come at the cost of not using a behavior tree

[celest python]

Well each decisions has its trade offs

But I can say, I would never duplicate BTs or use subtrees

[mossy nexus]

I'd prefer using the behavior tree since it works perfectly for all other behaviors

[celest python]

You will still use BT

[mossy nexus]

well you can execute subtrees by creating generic BT subtrees

which I will probably make use of

[celest python]

I know, thats my way of handling things, your one is not trivial

I always prefer to do things like that

BT just signals in my projects

Not handle things

I'm kinda abusing it in some way

I'd need to create toooo many BTTask if I would try to handle behaviors there

[mossy nexus]

same

which is why I think the action data in a separate spawned actor is the best

data separate from logic

I just need to understand how in the end, the action data is generic enough to be used by all AI components and their actors

[celest python]

Is ActionData "Action Handler actor that stores data based on AI's stats" or just "ActionData just stores data" 😅

[mossy nexus]

it stores:
transform
animation
animation notifies
(potentially it stores projectiles)

I would love to make it so I could use it for the AI to e.g. open doors or do other interactions

but I don't know that I can do that without adding logic to it

but basically I spawn it in world, tell the AI "here's your next point, figure out what you need to do with it"

in the case of a projectile, they might need to fire a projectile at the location

in case of an animation, they might have to reach the point and do the animation on the point

etc.

[celest python]

How many AI do you have?

or Max AI count

[mossy nexus]

hard cap around 100, max at any one time will most likely be 20 or so

AIs are generally a massive threat to players that require thorough assessment

you're not going to just walk up and kill one

(but then, they're not actively looking to kill you either necessarily)

[celest python]

Is that ActionData created for each AI or its simply an AI director?

[mossy nexus]

it's an actor in the world

each actor in the world has different data

for instance they are also used as patrol points and points of interest

they're placeable in editor as well

[celest python]

Then having a seperate AActor is a bad design plan. It is similar to your example of yesterday, like you're giving players a defensive objective but you're not preparing them enough for that. It's not a disaster, but its not good either. I know it's an intentional choice but either way its not a good thing. Anyway I know thats not a subject but just wanted to point it out

[mossy nexus]

I'm not sure what you mean?

[celest python]

I mean using actors for this is not a good choice

[mossy nexus]

what would you use?

[celest python]

Depends, probably it will be something against your design but not actor. Probably I would just code it inside pawn or into a component

[mossy nexus]

but... that doesn't work

[celest python]

If it requires you to use actor for any case its bad

[mossy nexus]

it needs a separate transform

because otherwise you can't use it as a position in the world

likewise the points control animations because animations are pertaining to the world locations

if it was a struct in a data table, there'd be a disconnect since you'd have to find out where to place them, then go in and create a struct in a data table to represent them

but you'd have no visual control over how the data would work in the world

now you could argue that it just needs to be an actor at design time, and that once the game begins, the actor self destructs and sends its data to e.g. the AI director to create structs, but I'm not sure that would be preferable

mostly because there is more complexity once you start including stuff like conversations between two actors who are scripted to speak to each other. maybe once I know the full scope of complexity that would be a viable solution

and that solution will only work insofar they stay as pure data actors. if I start introducing logic, I won't be able to convert them.

[celest python]

From what I can see you sorted out things based on your way and thats working but I'm not able to see your whole system, I'm just sharing what I've seen around the industry: Usually not having seperate AActors how handling things is a general rule. For example, once upon a time we refactored a whole sense and suspicion system because I used UObjects instead of structs. It's not something "never do it", but if there is a way that you can do it without actors, even if its ugly, do it that way, in my opinion. Even with working indie studios using actors idea was not welcomed.

I'm just sharing what I've seen around the industry:
I just pointed it out to make you consider again, totally not to imply you should change it etc.

Also having seperate spawning (also frequently destructed and recreated again?) AActor can be taxing for memory

It depends on hardware and your project

[mossy nexus]

well it's not a mobile project, haha

[celest python]

We were developing for PC and we refactored whole UObject based suspicion and perception system to USTRUCTs 😄

[mossy nexus]

I've taken this system out of companies that I've worked with that worked on stealth type games. the only difference is that none of those companies used UE. the concepts, however, remain the same.

[celest python]

Yeah, I was going to ask that too

I sensed that non-UE concept 😅

Maybe thats why I'm disliking it

[mossy nexus]

well consider this: there needs to be an atomic class that you can place in world, that can carry data

unless there is a more lightweight version of actor, then actor is just that class

if there was a FStructWithLegs that carried a transform around, I would of course use that instead

but given that limitation, there aren't really great options inside the engine

I also can't imagine the nonsense it would be for any designer wanting to use a non visual system to place down patrol points or model behavior points for AI in world

but perhaps the self destruct option has merit. the actor gets killed off come game time, but passes on its data

in essence being a proxy for the data

[celest python]

We solved that with doing some very low level things because we had too many waypoints both design and runtime but I did not involve with that process, so I don't know how it should be handled

[mossy nexus]

every level may have some ~50-150 points in total

[celest python]

But I don't know, even with that project, we were not explicitly placing waypoints, we were defining 'areas', and we were handling the behavior based on that

Waypoints were just single FVectors

[mossy nexus]

right, but those can't be placed in world... at least not by default

and FVector also doesn't signal direction

[mossy nexus]

like I said, if this was possible, I'd be the first to throw away the actor

[celest python]

I understand

[mossy nexus]

now you might be right that I don't need to spawn them through cpp for action data though. perhaps I need to make a FActionData struct and then just use the actor as a wrapper for that data

[uncut rune]

Why does the Agent Radius setting in the recast manager always get reset to 35 whenever I restart the editor? Is that supposed to happen?

[celest python]

If its marked as Transient in C++, yeah

[mossy nexus]

thanks for the conversation btw, I've made some notes. I'll see how can wrap my thoughts around what you've mentioned

[uncut rune]

Doesn't seem to be marked at Transient. I also changed the AgentRadius parameter in the DefaultEngine.ini to 75 but still does not work. It's weird because the Cell Size parameter saves, but not the agent radius.

[celest python]

That's very weird

Is it happening with every recastnavmesh in every level?

or just this one?

[uncut rune]

Its happening for all the levels 😕

[celest python]

Then we're missing something

But I dont have any idea currently, maybe its intended

There was a agent radius config on project settings

Maybe its using that somehow

[uncut rune]

hhmm, I have that set to 75 as well

[celest python]

Sadly I don't know then. Maybe someone else had similar issue can help

[final loom]

How taxing is it do you figure to have a few dozen AI perception systems running at once?

[mossy nexus]

depends on how taxing each of the components inside AI perception are individually

if they're well optimized a few dozen shouldn't be a massive issue

[final loom]

Hmmm, how might I optimize?

I mostly have sight, sound, and touch.

[mossy nexus]

I don't know your systems

so I don't know how you'd optimize them

but generally: fewer checks, cheaper checks. avoid square root, avoid traces

avoid normalize

[final loom]

I was using the built in AI Perception system.

[mossy nexus]

I don't know the built in system

I made my own

[past pelican]

woah

[celest python]

https://tenor.com/view/astonished-woah-gif-5025843

[final loom]

What? It's not that crazy.

Not sure what needs the default one didn't fulfill for them though

[celest python]

I liked the reaction 😄

[mossy nexus]

I basically still needed to extend it and the EQS queries. so I just didn't want to bother

[final loom]

Alright. I don't know if it is too limited for me either, but for now I think it will work.

[cold trout]

anyone know if NavLinkProxies work for CrowdFollowingComponent somehow?
I've got a custom NavLinkComponent that notifies NPCs when a smart link is disabled (when being used) and Invalidates() their navigation path, but they dont actually stop and repath.

Definitely had it working ok when npcs just had PathFollowingComp

[celest python]

Their pathfinding systems are very interesting: https://youtu.be/yqZE5O8VPAU

[hearty niche]

I am getting a crash on my BTTask node for some reason any ideas why @misty wharf?

Error: Failed to find function ReachedLocation in BTTask_FindLocationRelativeToPlayer

this is the dispatcher I want to bind my function to

Controller->GetPathFollowingComponent()->OnRequestFinished.AddUFunction(this,"ReachedLocation");

the function :

void UBTTask_FindLocationRelativeToPlayer::ReachedLocation(FAIRequestID RequestID,const FPathFollowingResult& PathFollowingResult){
if (PathFollowingResult.HasFlag(FPathFollowingResultFlags::Success))
{
FinishLatentTask(BTComponent, EBTNodeResult::Succeeded);
}
else if (PathFollowingResult.HasFlag(FPathFollowingResultFlags::InvalidPath)||PathFollowingResult.HasFlag(FPathFollowingResultFlags::Blocked))
{
FinishLatentTask(BTComponent, EBTNodeResult::Failed);
}
}

[misty wharf]

@hearty niche my guess would be the delegate add is wrong. Have you tried AddDynamic(this, &ThisClass::ReachedLocation)? Tbh haven't used non-dynamic delegates much so if it's not a dynamic one you might have to ask on #cpp how to do it correctly

[hearty niche]

hmm

[hearty niche]

Unrecognized type 'FPathFollowingResult' - type must be a UCLASS, USTRUCT, UENUM, or global delegate.

Controller->GetPathFollowingComponent()->OnRequestFinished.AddDynamic(this, &UBTTask_FindLocationRelativeToPlayer::ReachedLocation);

[misty wharf]

Is it actually a dynamic delegate?

[brittle lark]

Hi
How can I disable all AIs in level ?

[brittle lark]

I want to do this while a sequencer is playing and then I want to re-enable it

[celest python]

Way 1: Have a boolean on your game state something like "bPlayingSeqeuncer" and check that boolean first before evaluating other nodes in BT
Way 2: Loop all AI Controllers and disable logic, then do reverse when sequencer ends
Way 3: If you're with C++ maybe there is more advanced way with UBrainComponent but not sure

[mossy nexus]

I coulda sworn I posted a question here but

AI touch sense without collision, what's the go to?

[mossy nexus]

best answer: AI director subsystem

other answers: gameplay tags

[queen karma]

anyone has a good tutorial for behavior tree? followed few tutorial but i think they are outdated

[misty wharf]

Outdated how?

I've not noticed there being that many changes to how BT's work, maybe the UI has changed just a bit but the method of operation is still the same

[main heart]

Try Inside Unreal videos!

[timid condor]

hey guys, I'm trying to set a location for an AI to move to through an EQS. My level is huge so I have to use a navigation invoker. However, if the point my query selects is outside of the navigation invoker's range, my AI won't move there. First screenshot is my nav mesh range and my EQS range. The second is another game that I'm attempting to emulate while I learn. When the second game does a "Move to" in the behaviour tree, it seems to generate a path for it to move through while mine doesn't and subsequently can't move to the point. How can I do this?

[odd nexus]

Hi there. How can I move an NPC from one point to second point and then to third point without stopping? That NPC is continuously moving toward third point location, without stopping at first and second location.

[timid condor]

alright well I'll figure out how to do this at some point. For now I just conceded to losing 20fps by massively increasing the nav invoker range.

[undone hull]

maybe you could check on the Nav range and if the position the EQS sends it's out of the Nav range then clamp it to the nearest point inside the Nav

i've actually been looking for something like this, do you have any doc on Nav Invokers? 💙

[timid condor]

yeah I basically did that by adding a test to ensure its only inside the nav range, but of course that meant it would not move far at all. I've seen little to no documentation about anything to do with this haha, so what I've learnt to do I've learnt by trial and error and figuring it out myself from the names of everything

[undone hull]

exactly, what you could do is take the point that the original EQS gives, save it like your main "Target Location" but move your agent by another location that is clamped to you nav and when he reaches that mid-point check again if you have to re-clamp the Target

[timid condor]

Ah yeah that could totally work

[timid condor]

What I was trying so hard to achieve, was what you can see here. Where the ai has a nav invoker, but also a new nav mesh that is set up along a path to the target location. I could not for the life of me figure it out. very frustrating

This game is going to have mod support soon though, so when it does I should be able to take see how it was done

[undone hull]

the GDC yt channel just released how the AI on Death Stranding tackled this kind of issues dude, hope it helps 💙 --- https://www.youtube.com/watch?v=yqZE5O8VPAU

[timid condor]

Oh awesome I'll watch that now

thanks

[undone hull]

💙

[ocean wren]

FWIW, I think they added the grid based pathfinder to get around the navigation issue for large terrains? I think Miesko was using it for Fortnite at one point. Although I don't remember where they showed that. It was prior to them ripping off the PubG gameplay, so pre battle royale back when Fortnite was a sort of tower defency thing.

[mossy nexus]

is it worth it to make a service for (slow) projectile avoidance?

I'm a bit uncertain that a service can abort tasks in favor of other decisions

so I was thinking decorators instead

[misty wharf]

Services can't affect control flow afaik

I guess you could use one to set a BB value which is then watched by some decorator which then affects control flow

[mossy nexus]

I mean a a decorator is going to be cheaper I guess

so if services can't affect flow control, I have no issue using decorators

[pine steeple]

you can use a service to detect for projectiles

which then changes your BT to avoidance via decorators

[split grove]

I already asked this in the blueprint server, but this question might be best suited here. Does anyone know of a tutorial on how to make AI "Road Network" spline? I want to be able to make paths that AI can Transverse and have a choice from different paths not just one "straight path". If what I am asking is confusing, which it might be as I am a half wit. lol Assassins Creed GDC talk on "Meta AI" is what I am looking at replicating

[final loom]

Is there any event or function that gets called per each end of a offMesh Nav link whatever it's called? That would be super useful

[sullen pier]

Hello have a question, so we have BTTask and PawnAction who can explain for what we need Pawn Action?

Is there any method to use PawnAction inside BTTask? because i found that some implementation is copied in both of them

And looks like Pawn Action may listen brain component messages

[celest python]

Pawn Action was used before gameplay tasks

Zoombapup knows more about them

AITasks or BTTasks should be able to listen brain components too

@sullen pier

[celest python]

Check BTTask_MoveTo

UE5 probably (personal opinion: %99) will use AITasks (though UE4 also using AITasks right now, you just need to enable a flag on project settings) and not gonna use pawn action anywhere

I dont recommend using them if you dont have any specific use case

[sullen pier]

@celest python thx, the actual use was just reimplement to use interfaces for both PlayerController and AIController

with brain component

rts/fps like mode

and found that PawnAction may be instanced inside cpp

Is there any way to abort All BTTasks? for example move to location if i switched to manual controll?

[celest python]

Probably you can lock the brain comonent

It should reset the BT

[sullen pier]

Ok, thank you!

[stray violet]

Hey there,
I'm in a BTT and want to update a variable on my ABP.
That alone is easy enough, but the changed boolean begins a transition into a new state, which involves a montage playing during that transition in the state machine.

Is there a neat way to do this in a BTT and either wait for that completion?

[misty wharf]

You would need to listen to some event that triggers when the animation completes

and trigger the Finish Execute from that event

[stray violet]

Thanks, thats what I ended up on. Now I just gotta debug why my LeftStateEvent isnt firing

[mossy nexus]

Ensure condition failed: !SearchData.DeactivatedBranchStart.IsSet() [File:D:/Build/++UE4/Sync/Engine/Source/Runtime/AIModule/Private/BehaviorTree/BehaviorTreeComponent.cpp] [Line: 1592] 
There should not have more than one deactivated branch. (Previous start:[0:4], New start:[0:4]
[2021.10.27-21.48.17:230][493]LogOutputDevice: Error: Ensure condition failed: !SearchData.DeactivatedBranchStart.IsSet() [File:D:/Build/++UE4/Sync/Engine/Source/Runtime/AIModule/Private/BehaviorTree/BehaviorTreeComponent.cpp] [Line: 1592] 
There should not have more than one deactivated branch. (Previous start:[0:4], New start:[0:4]

[2021.10.27-21.48.17:230][493]LogStats:             FDebug::EnsureFailed -  0.000 s
UE4Editor-Win64-DebugGame.exe has triggered a breakpoint.```

I'm getting something like this, and I'm not sure why

[gloomy scroll]

Hello, how do I make the playerAI avoid enemyAI?

I have a player AI and an enemy AI. I would like the playerAI to avoid the enemyAI so I setup the enemyAI with an obstacle volume affecting navmesh. The problem is that if I do this the enemyAI pathfiding becomes finnicky/jittery, because it is having trouble navigating with a volume around itself as obstacle. I want playerAI to avoid enemyAI but I don't want enemyAI to avoid playerAI (hence I am not using rvo avoidance).

[misty wharf]

I would imagine there must be some way to configure either RVO or crowd avoidance such that only the player character uses it

[pine steeple]

there is

RVO has flags on who should be avoided or not

1 could be player, 2 could be enemy

1 can avoid 2, but 2 should not avoid 1

[worthy python]

Hi! How can I get "OverlappingActors" from custom "EnvQueryTest" ? Now I have empty array (but it's 100% must find overlapping actor). Why logic to find overlap is not working from this class?

It.IgnoreTimeLimit() is just experement. Same result without it

[devout dove]

Hey anyone know how to fix this issue with the hand

[pine steeple]

you need to use IK

to keep the hand attached to the gun

#animation

also looks like you imported the animation for a different skeleton and never had the pose setup properly.

[devout dove]

lk?

i use the default UE4 skeleton

[misty wharf]

inverse kinematics

[keen crow]

How do I make some logic to execute even when a branch is aborted by some decorator on a branch with a higher priority? Is a service with OnBecomeRelevant and OnCeiseRelevant overriden a good place to have this logic? I'm kinda confused that services are kinda for updating constantly while a branch is executing and not for obligatory in and out logic but I can't think of any better place

[mossy nexus]

can you sketch a diagram of what you're trying to do?

[keen crow]

Something like this for example. Decorator from left branch will abort right branch. And in the right branch I have an entry node SetPatrolMode(true) and an exit node SetPatrolMode(false). As I understand if the patrol branch is aborted by left branch, SetPatrolMode(false) wouldn't get fired so I tried to place this logic in a service StartPatrol which has OnBecomeRelevant and OnSeizeRelevant overriden. Is it ok to do so?

[misty wharf]

Does it work? If yes, then yes, if no, then no

:D

It seems like a reasonable way to achieve this type of modality to me

[keen crow]

I'm still new to BT so just wondering if there are better ways. What bothers me in such solution is that services are designed for regularly ticking and updating state of AI. This OnBecomeRelevant and OnSeizeRelevant actually come from BTService ancestor (UBTAuxiliaryNode)

[misty wharf]

It might be possible to also use a decorator for it... I don't remember for sure if they had that function though

 *  Other typical use case is creating a marker when specific branch is being executed
 *  (see OnBecomeRelevant, OnCeaseRelevant), by setting a flag in blackboard.

btw it says that about it in the code so I think your usage is perfectly fine

[keen crow]

they do as they also inherit from UBTAuxiliaryNode but for some reason I refused to test decorators. I think I thought that if decorator check fails OnCeaseRelevant won't execute

but I think I should check it

[ocean wren]

Pawnaction was the first attempt at making sequential actions, that eventually got outdated by gameplay abilities system (gas). Basically pawn actions are just sequences of enter->update->exit classes that you can add to a queue of actions. Which is fine for some things so I guess they left the code in. But the message at the time was that pawn actions were being replaced by GAS.

[sullen pier]

@ocean wren yes thx, already investigated that

[ocean wren]

I used them, they do work and all 🙂 so hey, why not?

Personally, I modified the classes to allow channels of actions and used them for things like tank control where actions were at different levels (gunner, commander, loader etc)

[sullen pier]

its quite simple, but im tried to use it in PlayerController as a top down brains

after possess

[ocean wren]

I think the use case was more just a simple queue that was "fire and forget"

[misty wharf]

It feels like there's a bit of overlap with pawn actions and bt tasks too, since at least in my case I have a lot of bt tasks for those types of "actions" my npc's do and I sequence them in BT's :P

[ocean wren]

Yeah, that makes sense.

I used a channel-ized version because I like the simplicity. Just shove an "action" into a queue and it gets run in order. Add some queue shuffling/clearing/locking etc and I guess it is a bit like the GAS.

Made sense for a layered AI I was building at the time

for controlling a tank

[misty wharf]

Yeah

[sullen pier]

i refactored pawn actions to brain component with behaviour tree, but again found bttask and aitask using controller->move fly etc

[misty wharf]

bt's are great for sequential stuff but they're not the best if you need something to happen simultaneously I think

[ocean wren]

Its the kind of thing that's actually fun to program 🙂 doesn't feel kind of braindead like the blackboard implementation 🙂

Yeah, BT's really aren't designed for that layered stuff. No "walk and chew gum" 🙂

[misty wharf]

you mean kick ass and chew gum lol

[ocean wren]

well, yeah 🙂

Man I'm wiped today.. did 4 hour workshops and then 2 hours of project meetings. I need a weekend off 🙂

[celest python]

Until a few months ago my everday was like that 😄

[misty wharf]

PSD and ICC file format investigation

[ocean wren]

ICC the colour thing?

[misty wharf]

yeah

[ocean wren]

I'm covering someone else's classes as well as my own, so timetable is pretty brutal, means I have to spend all weekend in prep

At least I get to play a bit with the class. doing sentiment analysis on multiple chat streams next week 🙂

[misty wharf]

nice, was just watching some CS on twitch and thinking they really need something like that to get rid of all the shitposters

[ocean wren]

I had a final year CS undegrad project to make a teams joke bot, but nobody took it 🙂

I was dissapointed

Although, I've got a couple doing lip sync and text-to-speech-to-text for metahumans 🙂

with dialogue generated by GPT2/3

which if you ask me, is fun, but then I'm a bit warped 🙂

[misty wharf]

I'd love to be able to use something like that for my game as I generated stuff like movies and movie synopses procedurally

but I have no idea how any of that ML stuff works and I bet it would be way too heavy for it anyway lol

[ocean wren]

naaah man. Thats what I work on in my own research. PCG cinematics and behaviour

ML based.. by learning from movies

Stealing all the ideas from Google and Facebook and NVIDIA 🙂

[sullen pier]

brain component has background tasks or something, just started surfing sources, probably will add new parallel tasks

[misty wharf]

Yeah never really looked at that in much detail other than seeing that Behavior Tree is a subclass of that

[sullen pier]

i mean implement new logic

[ocean wren]

Its about bloody time someone just rewrote the whole goddam AI system for UE. I mean I'd do it, but I don't think Epic care 🙂

Its so far from "good" 🙂

I have bigger fish to fry right now too.. what with consultancy and research projects and the like

[misty wharf]

it seems to do the job pretty well at least in both projects I've used it in

but neither of them really required super complex logic

[ocean wren]

Its ok. It's just not "good" 🙂

[misty wharf]

lol

[ocean wren]

I mean more about the usability and documentation than the code. Although the code has some bad smells too

[misty wharf]

documentation is a problem with almost everything in UE :P

[sullen pier]

i started to use unreal a while ago and i know about it only little) But never the less already extended lots of things

[ocean wren]

I just feel so bad for all of the people coming into the channel and not having a clue where to start on things 🙂

[sullen pier]

code is quite good, with lots of comments 🙂

[misty wharf]

I'm sure Epic could afford to hire a couple of folks to fix the docs

[ocean wren]

I worked on an AI system for MMO's called StoryBricks where we actually thought about how inexperienced users might think about AI

zomg: and yet, many years later, here we are 🙂

[misty wharf]

Thanks Tim.

[ocean wren]

He's too busy rolling in his mountains of money to care about AI and documentation 🙂

[celest python]

Why would people pay for UDN if there was a good documentation

/** An Octree */

[misty wharf]

Mmhm

[ocean wren]

?

[misty wharf]

You'd never know half of the stuff in C++ exists unless someone tells you about it on #cpp :p

[ocean wren]

My favourite part of the codebase is the node editor code 🙂 I'm twisted that way.

[celest python]

They namespaced something I cant remember and for that reason I was not able to automatically rename assets when they created, after that I never opened editor code again

[ocean wren]

Editor coding and plugins are the real powaaah!

🦾

[pine steeple]

@celest python UDN is for dedicated Epic support, not documentation

[gloomy scroll]

This is not giving the expected result. I cranked my enemy ai rvo avoidance radius to 5k and the ai player still doesnt go around it. AI player has rvo avoidance activated.

[uncut rune]

I'm using SimpleMoveToLocation to move my AI character along a navmesh. However, when the goal location is outside the nav mesh, the AI character stops moving. I want him to get to the closest point in the navmesh instead. Is there a way to do this?

[mossy nexus]

AIMoveTo and acceptance radius I suppose

or you could map a path maybe

I'm having an issue where a decorator is being skipped despite being true. I'm not sure how I'm supposed to handle higher priority tasks that are transient in nature that then also allow it to return

[humble hedge]

Checkout FindDistanceToWall in RecastNavMesh.h

[uncut rune]

hey, so I'm not sure how I will get the closest point on the navmesh with the FindDistanceToWall tho. I tried using the ProjectPointOnNavMesh, but I'm always getting a zero vector back so that's not working either (Picture of implementation below).

[humble hedge]

I am no expert in Navigation and EQS, but I believe there are two solution and it might work(For chasing player): 1. Use MoveTo LastKnown vector Location https://answers.unrealengine.com/questions/413063/ai-walks-outside-navmesh.html 2. Use a EQS query to find position that 1. Get closest position near player 2. Get furtherest position away from itself 3. Point is on nav mesh

[humble hedge]

I think you have to trace through code to see why it is not valid. There are lots of settings that could go wrong for nav mesh. Try delete all RecastNavMesh in your level and move NavMeshBounds a little bit to generate new nav mesh (For changes to take effect).

[celest python]

Check Visual Logger to see why it is being skipped

[keen crow]

Where can I read about running behavior trees from behavior trees? Currently I don't understand how do I pass some data from parent BT blackboard to child BT blackboard and how do I stop child BT from parent BT (I assume just by decorator) and from inside child BT itself?

[pine steeple]

you can not run bt's from bt's

but you can run child bts via Run subtree

slightly different as your main tree is still running.

and you can't have different blackboards

Child BT must use same blackboard as parent bt.

[keen crow]

then what does this node do?

[pine steeple]

runs a sub tree, but parent tree is still running, and they can not have different blackboard than the parent

[keen crow]

ok I got it thanks

btw is it a good practice to run subtrees or should I avoid that?

[pine steeple]

i use them regulary

i use the Run Dynamic

and i inject different trees based on ai behaviour

makes the BT's cleaner for complex AI

http://thegames.dev/snaps/UE4Editor-Win64-DebugGame_SXvPGYiHUV.png

used alot inside our "Hunter" monster

[keen crow]

what do these tags mean in run behavior dynamic task? i didn't understand a thing -_-

[pine steeple]

Gameplay tag

you pass into AIController->BehaviourComponent->RunDynamicSubtree

[keen crow]

but what does it mean to run a behavior dynamically with an injected tag?

its like in the context of a running subtree I can somehow read those tags?

[pine steeple]

you choose what BT to inject into it

http://thegames.dev/snaps/UE4Editor-Win64-DebugGame_V4PMdzQZPE.png

like this

then the BT will ibject BT_DogEvade into it

and run that

[keen crow]

still don't get it. I choose what BT to inject in regular Run Behavior task also. What are those DefaultGameplayTags.ini about?

it seems like I don't know a lot about BTs because none of that makes any sense for me yet -_-

[pine steeple]

http://thegames.dev/snaps/UE4Editor-Win64-DebugGame_UxHqYCYCqa.png

so for example the above, by default this node will run Basic monster evade

but with my injection above

it will change it to use DogEvade

[keen crow]

oh wow. I guess I understand it now but can't imagine a usage for this yet. I'm currently only learning with just 1 type of enemy

[pine steeple]

yeah its more for complex stuff

[keen crow]

seems like some BT-level polymorphism

[pine steeple]

i keep my bt's generalized as much as possible

and inject things in when needed

but there is a limit

(works fine for basic monsters, but gets less manageable on complex ones)

so for them, i make unique behaviour trees

[keen crow]

cool. its rather clear for me now, thanks again

[pine steeple]

yeah BT's are hard to get used too

but i have around 100ish bt's

for 30 monsters and human AI

lot of them are small subtrees which gets injected into the main tree

[keen crow]

@pine steeple can you help me with another question? Before I started learning BTs I was playing with custom NavLinkProxies. I managed to make a simple one for jumping on a ledge just by subscribing to OnSmartPointReached event. But I couldn't find out how to set up a custom NavLinkProxy where a character needs to start one action and the beginning of the link and stop doing it when it leaves the link. Like for example I have a mechanic for a player to crawl or crouch to go under low obstacles and I can make character crouch or prone when it enters first point of NLP but I don't know how to make character uncrouch once it gets through the obstacle. I asked this question before and someone suggested just using custom triggers which makes the bot uncrouch once it leaves the box collision but after discussion we agreed that it would become cumbersome to maintain as there are more and more NLPs and harder actions like sliding under obstacles, running walls, etc. So is there any better way?

[pine steeple]

you just need to check if AI location is nearly equal to target location (end location)

that is the only way

that is how we handle vaulting

though the vault calls in to the AI controller

and tells the AI controller its finished

and this then ends the link

but same principle

[misty wharf]

I have a similar setup for going through doors although in my case the AI controller figures out when it's reached the end by itself and informs the link

it's a bit annoying because you can't really pause its navigation pathing, give it a new nav target, and then resume the previous pathing later

or rather, you probably could, but if you use f.ex. AI MoveTo, changing to a new nav target for traversing the link would cause it to consider the pathing finished

[mossy nexus]

thanks I'll check it out

[keen crow]

Is it possible to somehow group or categorize blackboard keys?

[misty wharf]

nope

[keen crow]

so everyone is fine with having dozens of blackboard keys in one blackboard?

or if I have dozens of keys in one BB then i'm doing something wrong?

[misty wharf]

No idea tbh. I have some BB keys and a bunch of decorators which access other data that's not in the BB

[meager bobcat]

how do you get a blueprint bttask to tick?

[meager bobcat]

oh i had Lock AILogic ticked in my task and it stopped tick

now my shit ticks

[uncut rune]

I gave up using the ProjectPointOnNavMesh function. I honestly think that function is broken. I noticed that they used it in the MoveTo function in the AIController class and doesn't seem to work there either after doing a couple of tests. I just ended up using EQS and made a custom context for the player. That seems to be the only viable solution for this problem without changing the source code.

[pine steeple]

it works fine for me

i use ProjectPointOnNavmesh a lot.

[golden owl]

Hey everyone, I'm doing a gamejam and time is running out on me, I'm trying to do a barebones quest system, I already finished the dialogue system and need some help
How do I make an AI decorator that checks wether or not you have a quest item (bool on char)

[uncut rune]

oh, then its probably not broken then 😂 . For some reason, it always returns 0,0,0 no matter how large or small I make the extents. I even mimiced the same implementation from the MoveTo function in the AIController and got the same buggy results. So idk 🤷‍♂️ . Im not too worried about it now tho.

[gloomy scroll]

Can someone explain the difference between a task with "AI move to actor or location" and a simple "move to"? With "AI move to actor or location" whenever there's a "wait" in the BehaviorTree the AI executes another "AI move to" task during the supposed wait.

But if I use "Move to", it respects the "Wait" lol

[lyric flint]

hello, how can you add collision to ai for the whole body?

mine is kind of giant

and i want player to be able to cross between it's legs

[gloomy scroll]

increase capsule radius and capsule height size. When you click on the capsule component in the viewport, it shows on the right.

[lyric flint]

yes but player cannot go between it's legs which is the one i've wanted to do. enemy ai is giant

[gloomy scroll]

add 2 capsules for legs, leave a space between them

[lyric flint]

haven't seen that option yet. how can you add another capsule?

[gloomy scroll]

are you on bp viewport or mesh/skeleton editing?

[lyric flint]

bp viewport

i already have the skeleton collisions

[gloomy scroll]

On the viewport, to your left there will be a list with the components. Above the list there will be a dropdown menu. Clik on the menu and select capsule collision.

[keen crow]

what's the usage of Push pawn action node in BTs? I tried googling about it but it seems just like the tasks so why ever use it?

[grizzled bolt]

Anyone here have any idea on creating ai that can do flanking?

[ocean wren]

Yes, use EQS and make the dot product calculation bias to the sides of the player

But... I'd recommend against it, players will be getting shot in the head and dislike that they don't know why 🙂

Other than that, its just the usual setup

You could bias the EQS to choose routes to the flanking position that are visible from the player, but that just looks a bit haphazard

[grizzled bolt]

Don't worry, the Player is not gonna get shot in the head by the enemy

I tried using EQS, it didn't worked out as I expected since its responsible for generating position and doesn't really care about the pathfinding. So, most of the time, the enemy will go in front of the Player and then proceed to the generated position, which makes the enemy look stupid 🤣

But you mentioned using EQS to choose routes, can you share a bit more on that?

[ocean wren]

Generally, you can use the EQS to bias the chosen location in a direction. So use perpendicularity to choose "off to the side". But yeah, if you wanted the pathfinder to not pass in view of the player, you'd need to modify the cost of the navmesh to score against any visible navmesh polygons

But my experience is that flanking comes across to the player as enemies just running away.. then later shooting you from behind/side

We had this exact same problem a looong time ago on one of the first first person shooters on the amiga

What we ended up doing is making the enemy take cover.. then wait a bit, then try and move up.

But always in view of the player (aside from when its fully in cover of course)

[mossy nexus]

anything the line runs through ticks. but decorators are conventionally branching points

[grizzled bolt]

Ah yeah, thought so too.
I am using a box collision marked as dynamic obstacle to affect the area costs at the moment 🤣

[grizzled bolt]

It does, which is why for mine I have the enemies signal to the players that they are gonna flank the players. Depending on what situation it is of course, in some situations where the intention is to make the players jump, I won't have them signal to the players.

Thanks for your input!

[quasi glade]

hey, does anyone know why 'ai move to' can suddenly stop working or work incorrectly? I had this setup and it was perfectly okay and some day AI just stopped moving (thought it still tries to but stands still)
my built nav is ok
no obstacles
no additional components except volume trigger, pawn sensing and some fancy smoke (all component have 'affect nav' turned off)
did not alter any movement component settings

[mossy nexus]

you can check the Enums you get as a result from the AIMoveTo and set them to print strings on fail

most likely it can't path to one or more of your locations

[quasi glade]

debugged all locations, it behaves the same everytime, feels like the pawn is stuck, it doesn't even rotate towards next destination

[mossy nexus]

did you read what I wrote?

[quasi glade]

y

[mossy nexus]

so what did I say?

[quasi glade]

says success

though in debugging it does not do the thing after 'on success' and keeps playing like it didn't

even when I make a loop after 'on failed' it keeps looping, but says success

[mossy nexus]

then do as I said and put the enum on OnFail

[quasi glade]

k

says aborted 😭

[mossy nexus]

aborted most likely means that it's outside of navmesh

or it received a different order

what it could also be is

because you have a delay

it triggers the delay and starts moving to another point within a second

your setup needs wokr

[quasi glade]

unfortunately the issue remained still even after I turned off delay

recreated navmesh, rebuilt it and yet AI still does not move

what can that red spot be?

[mossy nexus]

try and see if there's a path to your location on the navmesh

there's a node for that

[quasi glade]

I checked all locations with 'navigation raycast' it said true and named the final point for all of them correctly , so I guess it does work

[mossy nexus]

navigation raycast just tells you the location is valid on the navmesh, not that there is a path to it

there is a node called find path on navmesh or something, use that to check

[quasi glade]

okay thanks I'll check it

[keen crow]

I'm having a problem with understanding trace test in EQS. Why are the spots behind walls aren't filtered out by trace test that requires to hit the context?

And here's the EQS query

what am I doing wrong?

[misty wharf]

@keen crow it's kind of a confusing system - only blocking hits with bool match enabled means that it will count hits that hit anything as success

[keen crow]

What do I do then -_-? I tried disabling only blocking hits but it didn't work

[misty wharf]

iirc you need to have it count things that don't hit anything as success

[keen crow]

How do I do that?

[simple crest]

What's in the drop-down trace settings?

Advanced

[keen crow]

Default values, I never touched them...

yoooooo

[simple crest]

Raise up the trace so it doesn't always hit the ground with those settings. And probably disable "only blocking hits"

[keen crow]

Thank you thank you

[misty wharf]

Yeah it's kind of confusing how it works at first lol

[keen crow]

not perfect but at least seems correct

still curious though what can be wrong with these ?

Current trace test settings

[queen olive]

A quick silly couple of questions. I have navmesh set up and ai bp controller and blackboard with a look for new location move to wait system in place working well. My model has a walk/run with a walk left/right. how do you go about making it not so wooden I.e gets to xyz rotates 13* then walks straight to xyz2 in a direct bline of sight and how would you go about adding the walk left right as it turns a curve of sorts not just spinning on the spot. Second one is I only have 100 in scene and the FPS drops dramatically iv seen 1000s used in a niagra crowd. Not sure if I can have them as physically gotta Le things with ballistics or not and if not I’m guessing as the bullet gets close it should be turning it from a niagra to a physical mesh…..

[dull loom]

Hey guys, do you know if there's any way to avoid the hard deceleration when aborting a move? I'm using dynamic navigation and when the path is recalculated the previous move is aborted and the acceleration drops to 0. This sudden change is breaking something in my Animation Blueprint.

[silent cove]

Not sure if this belongs here or in #cpp. But is there an easy way to completely reset all blackboard key values?

[mossy nexus]

neither quick nor silly questions. this is something that's hard to pull off. basically you'd want to create a custom move to node that takes a bezier style input curve per point

[mossy nexus]

not as far as I know. what would they be reset to?

[mossy nexus]

this can then dictate speed along the resulting spline, making speedups and slowdowns as necessary

I'm actually wondering if this would be possible in BP now...

but my initial reaction is no, since you require latent functionality

[silent cove]

Just empty values. I just want to start up the blackboard completely fresh is the goal essentially

[mossy nexus]

given that they are wrappers, there is no good way to flush them

[silent cove]

crap :/

[mossy nexus]

it might be worth considering how many values are dependent on being flushed

in my general AI BT I have 1 variable that I flush

[mossy flame]

Hi, does anynody know if Hearing in AI Perception registers events from Report Noise Event adjusted to walls and other geometry. For example if there is a wall between Pawn and Actor who reports Noise Event does he hear it in a same radius if there was no wall? Thank you

[misty wharf]

Hearing doesn't have any kind of occlusion

you'd have to implement that yourself

[mossy nexus]

if you got the chops you can use a navmesh for it

I've done that

[queen olive]

thankyou. least i have some keywords to go do some research.

[mossy flame]

thanks a lot, yeah pathfinding is probably the way to go.

[mossy nexus]

so with debug icons or what have you installed, this actually bugs out and I get an error: cpp Ensure condition failed: !SearchData.DeactivatedBranchStart.IsSet() [File:D:/Build/++UE4/Sync/Engine/Source/Runtime/AIModule/Private/BehaviorTree/BehaviorTreeComponent.cpp] [Line: 1592] There should not have more than one deactivated branch. (Previous start:[0:4], New start:[0:4] [2021.11.01-22.31.28:633][586]LogOutputDevice: Error: Ensure condition failed: !SearchData.DeactivatedBranchStart.IsSet() [File:D:/Build/++UE4/Sync/Engine/Source/Runtime/AIModule/Private/BehaviorTree/BehaviorTreeComponent.cpp] [Line: 1592] There should not have more than one deactivated branch. (Previous start:[0:4], New start:[0:4]

I'm tagging you just in case you would know how to resolve this?

[celest python]

I've never seen this error before but I'll give it a look on my free time and let you know. iirc you had some similar error posts here related with BT, something probably went wrong with nodes

[mossy nexus]

this has to do the skipping

[celest python]

Well, wait a second, "Target Acquired" decorator runs first and "Cast Intent" runs for 2nd one?

Set a breakpoint to this line and check values of .Describe() functions' returns

If new end is lower than previous end, it complains about you're returning back to a previous node on a selector

Which is not a valid thing, afaik

Also, can you share the visual logger screenshot?

[celest python]

I dont have the same error logs used on my source code for the same line (4.26 version)

[mossy nexus]

oh I see. wait so if I change the big selector to a sequence, it's ok?

[mossy nexus]

I didn't actually use the visual logger

[mossy nexus]

and yes, that is correct

I figured that if it aborted lower first, then came back it would retry lower due to being a selector

[celest python]

Honestly since I never tried to do this, I never got a similar error too, but just having a guess what you're tried to do is probably against the design of BTs. Maybe someone with wider experience can say something with more confidence

[celest python]

Let's try, but I'd suggest trying Visual logger first, it gives more detailed info

[celest python]

what you're tried to do is probably against the design of BTs
Btw, do BT continue to execute other nodes when those two skipped?
if it returns back to previous node, it might end up in a infinite loop, and thats might be the case why BT put an assertion to that line

[mossy nexus]

yeah it returns to the default nodes. all the other connections work flawless. I seem to recall this used to work flawlessly as well, I think the bug may have come with an update

like it stops play because of the error but I can resume and it'll not pop up again and never cause an issue again

[celest python]

Thats how ensure() works

Lets you know something is wrong and never bother you again

[mossy nexus]

fair enough

but yeah it's only between those two nodes

I have 5 nodes springing from the same selector

the others have no such issues

trying to understand why the relationship between these two nodes is like that

[celest python]

Is there any other nodes that returning to previous node?

And working flawlessly?

[mossy nexus]

yep

[celest python]

Hmm

Maybe the reason is when the 2nd one (the one on the left) failed, it should execute the 1st one again, and its also failed

If not, I think you should see visual logger if it will provide more info. I don't think I can brainstorm more about this without knowing further informations since I never saw the same error before

[mossy nexus]

just for reference

the lowest three have no issues and work perfectly between each other

one of them isn't implemented so it always returns false

so it remains untested

the two other high prio nodes have issues between each other

the visual logger has no info

and sequence breaks the order so I'd have to rearrange logic to trickle down

but perhaps that is the answer?

[celest python]

Do you still get the error with sequence?

[mossy nexus]

no but I can't get it to the point where I would get the error

like I would have to change the logic of the BT tasks

I don't think it would work

hmm now strange the third one doesn't trigger like it used it

very weird

[celest python]

That might not be pointing anything since the error was relevant with selector node

[mossy nexus]

well consider that with a sequence every task has to succeed

[celest python]

SearchData is a selector thing

[mossy nexus]

so in order to pass it down the chain I would have to succeed each of them

[celest python]

No I'm not saying you shouldn't use selector, its just error was related with selector node itself

So when you removed, error can be gone

It might not be giving correct clues to find the error

Maybe @ KaosSpooktrum knows more about this situation, but I'm afraid to @ him, he might be busy 😅

But he usually checks #gameplay-ai

[mossy nexus]

I might give him a prod the next time he's around

thanks for the rubberducking at least

aha, the problem happens with the cast intent node regardless

perhaps it doesn't like that it is aborting itself

it has no problem aborting lower prio, but it has a problem aborting itself... which doesn't make any sense. it should search for other nodes then, and there are other nodes. maybe it's because there need to be two separate decorators

hmm

[celest python]

. (hi Kaos :D)

I'm an idiot, did not see your "abort both" flag at the first stage

[mossy nexus]

yep

that's intentional though

or at least, how I have understood the "aborts both" observer

for as long it stays true, it aborts lower prio

as soon as it turns false, it aborts self

[celest python]

So if Cast Intent returns false, it will skip the right one even the right one is true

[mossy nexus]

well it shouldn't

it should restart the tree no?

if it aborts both at the same time when it's false, let's say (worst case scenario)

then the tree restarts, and it reevaluates every node in order

[celest python]

No? It will just continue to execute other nodes

[mossy nexus]

then why would it not check and execute the next node in order?

[celest python]

You're aborting 'both'

[mossy nexus]

yes

[celest python]

self and lower prio

[mossy nexus]

yes

[celest python]

which prevents the execution of next node

[mossy nexus]

so worst case, that means it restarts the tree because the selector fails

[celest python]

But selector does not fail

Every sequence or node should fail

[mossy nexus]

it does, all children aborted

[celest python]

You said it continues to execute other nodes flawlessly?

[mossy nexus]

it skips that one node

but in the case of it aborting the tree

that's fine too

because it restarts

and finds the right node

the BT doesn't just die

it reevaluates in case of failure

my point is just

in either case, it skips a node with a decorator that is true

if it restarts the tree, then cast intent is now false, the next node has a decorator which is true (and aborts lower prio), of which path it should then take

but it's definitely something to do with the aborts both. even with just two nodes (the other have no decorators), it still causes the same issue

and I know it didn't use to be an issue

this is logic I hadn't changed for a year or more

interestingly, other aborts both work as intended

[celest python]

Is there any possibility previous decorators affecting the condition of your skipped decorator?

Quick example, first decorator sets the "targetplayer" and since all of them checks targetplayer is valid or not, whole execution stops

Or maybe something is wrong with notify observer settings?

[mossy nexus]

hmm I only have one, on a separate selector one level up. it basically initiates the BT values

[celest python]

Have you ever breakpointed the BT? Maybe some insteresting stuff come out

While doing that you can examine the values

on editor

[mossy nexus]

I have.

there isn't anything

other abort both nodes are working as intended

this is starting to seem more like a bug

[mossy nexus]

I don't do things based on BBvalues

(at least not here)

[celest python]

Can you both debug in BPs and inside of the IDE to pinpoint exact location that assertion happening?

put breakpoint to that ensure on line 1500~ and put another breakpoint to start of the BT in editor

also there should be a function or a condition to call that decorator on that function body

it should show you why it is not being called after that assertion

Is it just me or our messages are being deleted lol?

I was trying to find the logs you posted

[mossy nexus]

ok so now... I changed some of the low prio decorators on the nodes that otherwise worked fine

and the entire tree works as intended now

except I still get that ensure issue

[celest python]

Literally everything got deleted on me

[mossy nexus]

wait what?!

[celest python]

what the hell 😂

[mossy nexus]

ok so now... I changed some of the low prio decorators on the nodes that otherwise worked fine
and the entire tree works as intended now
except I still get that ensure issue

[celest python]

@mossy nexus are messages deleted on your side too or am I seeing things? I really need a sanity check on this

[mossy nexus]

nope I see all our correspondence @celest python

have you tried restarting discord perhaps?

[celest python]

Yeah restarting solved, that was a very weird issue though 😂

[mossy nexus]

perhaps a discord server failure somewhere

and they had to get a backup

[celest python]

Another thing biting me is I don't have same ensure() function like you