#gameplay-ai

1 messages · Page 12 of 1

unborn jungle
#

I can kind of work around it if I make multiple nav links that each step towards the final desired point (this is for a strong jump pad that jumps you high on the map) but if there is a way to extend the default range while keeping navmesh build settings intact, it would be great!

boreal tundra
#

Is there a way to reset navmesh settings to default ?

unborn jungle
#

😄 I tend to have a blank project to reference the default values

#

I wish they had those yellow arrows to reset

boreal tundra
#

Aahhaahah true

#

but good idea

#

i'll use your method

#

thank you !

lethal helm
#

is there an easy way to disable the obstacle check on the AI sight sense?

lethal helm
#

nvm

#

created a new sense that just does overlaps

celest python
#

Today I thought about a crazy idea like parsing FMath and FVector and converting each util function to ST conditions

#

So I wont have to write functions like this manually

image.png
shadow furnace
#

I wonder if you could just have a little dinky textbox that raw edits the condition of that statetree?

#

Probably not worth the effort

celest python
ocean wren
#

Aren't all nodes magic?

harsh storm
#

@celest python Trying to do a task in C++. I keep getting an error on compile for the state tree. "missing instance value". I thought implementing this virtual const UStruct* GetInstanceDataType() const override { return InstanceDataType::StaticStruct(); } would've handled that. What is the minimum needed in order to do this?

# 
USTRUCT()
struct FMoveTaskInstanceData
{
    GENERATED_BODY()
    
};

USTRUCT()
struct FMoveTask : public FStateTreeTaskCommonBase
{
    GENERATED_BODY()

    typedef FMoveTaskInstanceData InstanceDataType;

    FMoveTask() = default;

    virtual const UStruct* GetInstanceDataType() const override { return InstanceDataType::StaticStruct(); }
    
    virtual EStateTreeRunStatus Tick(FStateTreeExecutionContext& Context, const float DeltaTime) const override;
};

This is what I have. I copied from DelayTask

ocean wren
#

Why all the love for StateTree recently? something happen?

harsh storm
#

Ooohh shiny pretty much

ocean wren
#

Fair enough 🙂

harsh storm
#

Jokes aside - I am doing it 'cause Eren is doing it. No real reason otherwise 😅

ocean wren
#

See, you're making ME want to play with it

#

and I don't have time 🙂

harsh storm
#

I just randomly decided to do it last night. We're both making zombie games. So I was just like, ehhh - screw it. I'll join in on this journey with you

ocean wren
#

Isn't literally the whole channel making zombie games?

harsh storm
#

Are there other types of games to be made?

ocean wren
#

Oh my 🙂

harsh storm
#

wtf - now it works

#

🪄

ocean wren
#

I've got to spend more time with Houdini, more time with NLP, more time with geometry scripts.. more time with ML code.. I guess can't hurt to add StateTree for giggles 🙂

harsh storm
#

Yeah - sounds like you have the time too 😅

#

I do enjoy BTs though ❤️

ocean wren
#

BT's are fine.. I'm not a fan of the UX though 🙂

harsh storm
#

I guess in order to actually update the task, you have to remove the task from the state and then add it back (after changing some .h stuff in C++ that is)

harsh storm
ocean wren
#

It could do with a usability overhaul is all.. the basic concepts of BT's are easy enough

#

although hardly anyone seems to get them 🙂

#

And personally, I'd have wanted better parallelism

#

But the amount of pain people seem to go through with BT's in UE is probably more a UX issue

#

I think thats true of UE in general though.. wonky naming, weird conventions, inconsistent UI etc.

#

The fact that different node editors do different things because they were made at different times by different groups of developers..

celest python
celest python
ocean wren
#

The dream team dun messed up already??

#

I am dissapoint 🙂

celest python
#

oh yeah its statetree's compiler

#

it checks !TaskNode.Instance.IsValid() && TaskNode.InstanceObject == nullptr

#

if returns true, throws that error

#

i have two tasks with exactly same format - except my instance data struct has properties

celest python
#

after doing nothing

ocean wren
#

One does not simply do nothing.. one must BE nothing for it to work

celest python
#

real question is, zoombapup... 😄
did you get any ideas about how verse will work as AI scripting language?

#

from that.. very scientific paper and slides

#

I'm curious of your opinions

#

FP as an AI scripting language

ocean wren
#

No, I ran away from it hard

celest python
#

🤣

#

reasonable

ocean wren
#

functional languages hurt my feelings

#

Besides, I've got other fish to fry.. I doubt I'll be adopting a new language anytime soon

#

hard enough putting up with the smell of python

#

Look at their examples though, its going to be a cold day in hell before most UE users adopt it.. unless someone manages to convince Tim S that documentation is a good thing out of the blue 🙂

#

Which seems unlikely

celest python
#

i mostly wonder functional programming itself can be useful for AI scripting

ocean wren
#

Well, prolog was one of the initial AI languages, so yes it can

celest python
#

verse inherits some prolog features too

ocean wren
#

But I don't think it has the right mindset.. games are largely event driven.. functional languages are mostly data driven

#

I guess we'll see.. I'm sure it will work, but work nicely?

#

If literally every other game engine adopts it as a language then MAYBE

celest python
#

which scripting language is your favorite for game AI? I also know you dont like lua

ocean wren
#

but.. given the ML world literally adopted the worst language ever as its defacto standard (python), I'm not hopeful for rationality to win out

celest python
#

you mentioned AngelScript a while ago

ocean wren
#

AS was ok

#

Lua was mostly OK, but didn't like the tables

celest python
ocean wren
#

There was one called GameMonkeyScript that was fun 🙂

#

I think honestly my favourite language is C#

#

so if someone made an unmanaged C# with pointers.. I'd be on board

celest python
#

I guess everyone with experience in this server favorites C# 😄

ocean wren
#

The reason I like C# is mostly to do with the lack of header/cpp and the like

#

the language itself is near enough to C/C++ but without the shitty syntax

#

$&^^ crap

#

you get first class delegates, properties etc..

#

linq is also hella cool for inventories

celest python
#

even with .NET7 you get to ship to console platforms

#

AOT compiler

ocean wren
#

and lambda's in C# don't break my brain the same way they do in C++

#

Don't get me wrong, I still hate it, but I hate it less than all the other choices

celest python
#

do you like it more than python? 😄

#

that would explain all

ocean wren
#

python is an abomination

celest python
#

😄

ocean wren
#

Seriously.. its so bad 🙂

celest python
#

scientist like it... CB_pepe_hands

ocean wren
#

Scientists aren't programmers obviously 🙂

celest python
#

you have a point sadcatup

ocean wren
#

Btw: Houdini indie is free for the next few months.. if anyone wants to try a node based geometry thingy

#

as part of GGJ they're giving a license

#

Do recommend playing with it

#

Cascadeur is also v1 now, worth a play

wary ivy
ocean wren
#

True, but it isn't "good" its obscure crap because of the language.. making different choices for language design is the point

wary ivy
#

not sure if obscure is the correct term

ocean wren
#

You know what I mean though..

wary ivy
#

it just sounds like you don't know about how it works 😄

ocean wren
#

Its more that when you work with students, you feel the pain of someone new to it

wary ivy
#

so yeah, C# is easier because it manages stuff for you

#

yea there's a reason why java was the freshman language instead of C when I started in university

ocean wren
#

Java has its own can of worms too 🙂

wary ivy
#

this was 13 years ago

#

a lot has happened since 😄

ocean wren
#

I was trying to explain aspect oriented programming to some second years struggling with pointers a few days ago

#

got me thinking about a bunch of these pain points

#

given we're teaching OOP, but the game industry is desperately trying to ditch it for perf reasons

#

well, not OOP per-se, but the idea of large monolithic classes (AActor etc)

#

Unity just made their DOTS thing a proper thing.. UE introducing Mass etc.

#

Its yet another concept that people have had to learn and then learn its not good 🙂

#

"Here's how to write a class"

celest python
#

the game industry is desperately trying to ditch it for perf reasons
finally we see the light.. slowly

ocean wren
#

"oh by the way, classes are way too slow now, so lets break the class apart" 🙂

#

You can forgive people being a tad confused

celest python
#

classes are way too slow now
data is slow? Thonk

ocean wren
#

Maybe I should teach Mass and StateTree in my graphics class 🙂

#

large classes that don't fit in cache 🙂

#

sloooow

celest python
#

but its about how you evaluate rather than size of the class 😦

ocean wren
#

Its how you architect things

#

check out aspect oriented programming

celest python
#

even with oop, cpu should try to fit things you access over and over afaik

#

so rather than class size itself the amount of data you are evaluating should matter

ocean wren
#

the point is to try and package up one aspect and iterate it in a tight loop all at once in-cache

#

and ideally in parallel across cpu threads

#

along with parallel GPU versions of same

#

its not an OOP problem per-se

#

but it kind of stems from the way that OOP is taught

#

i.e. "bind all your functions to your data"

celest python
#

oop makes things difficult to thread

ocean wren
#

big blobs of oopy goodness

#

unfortunately big oop blobs mean terrible perf 🙂

#

I could generate some big oop blobs for a lecture 🙂

#

turns out that the C guys had it right..

celest python
#

rust

echo lark
#

that's why usuall OOP design is bane of peformance

celest python
#

i though the "instruction"s matter

#

rather than object itself

echo lark
#

i simplified it

#

in general

#

imagine memory as cells

#

cpu fetches X cells forward

#

it doesn't care what is inside those cells

#

if there was nothing usuful for current context, well though luck, evict cache and fetch next cells

dawn schooner
#

is there a way to "breakpoint" CPU reads? would be super useful as a learning resource

celest python
#

you can monitor cache hit rates

#

and debug them

#

but never accurately

#

btw @echo lark since you are here..
as usual I was roaming around yours and vblanco's message history and found you mention something about "vectorVM" -- but you refer it as "VM behind the BP", isnt it specifically created for niagara?

#

this night spent a few hours with vectorvm's source code

echo lark
#

you could be right, I honestly don't remember

celest python
#

ah, alright. I'm just generally curious about BP's VM and other VMs used in the engine

#

just wanted to ensure I'm not missing anything 😄

echo lark
#

the other one

#

is Solaris

#

but it's not in engine yet

celest python
#

do Epic ever upload those things to P4?

echo lark
#

probably will drop in sometime next year

#

no

#

it is developed on fortnite stream

#

sometimes something leaks to github

#

;d

celest python
#

😄 so they were not intentional

echo lark
#

, yeah well sometimes you need to make change to main engine it is just auto merged

celest python
#

to be honest I'm very hyped about Solaris, it was first mentioned at 2019 without a name

#

2019 also when epic had a deal with shookumscript

echo lark
#

it's for Verse

celest python
#

i hope verse/bp mix VM comes out of it

#

vblanco said BP might be transpiled to verse

#

for nativization purposes

#

the idea of not switching to C++ each time I need to loop an array more than 10 element excites me a lot

dawn schooner
celest python
#

valgrind equivalent works for windows 10

#

and it.. something like emulates PS4 cpu

#

so you'll debug your hit rate on lowest common denominator

#

it also can display where hit rate is low on your code

#

etc

#

existence of tools like this proves to ship PS4 successfully you need to take care of your cpu caches and threads, tbh 😄

#

because all related tools come out from console developers

unborn jungle
#

How can I get nav link proxies to work over large gaps?

#

It seems that the range is very low, surely it is possible to account for large launch pads and such that AI can use to traverse a map?

harsh storm
harsh storm
plush cargo
#

why do i run into a lot of problems trying to spawn an enemy with AI after a time

#

and it works fine if placed before run

#

i solved it

#

weird stuff

lethal helm
lethal helm
#

classes with thousands of lines are so much fun to comprehend

ocean wren
#

So its only a bit of a side effect that the code often seems a bit easier.. I guess its a natural evolution of components

#

But I read about this kind of thing years ago.. when looking at data oriented programming and then aspect oriented programming

lethal helm
#

seems like it would be harder for developers to me

#

human brains think in objects

ocean wren
#

You'd think.. but no 🙂

#

Its a bit weird.. because we still teach old school OOP, but that's still the "standard" method if you're not about performance

#

But then, people use python, so who cares about perf 🙂

lethal helm
#

for the applications python is used for developer time is much more expensive than cpu time

ocean wren
#

technically true.. but imagine all the crappy python code out there in the real world.. slowly killing all of the CPU's and making Santa cry... all... slowly... like molasses

lethal helm
#

actually let me remove those conditionals lol

#

developer time is much more valuable than cpu time

ocean wren
#

pfft, I'm not a fan of that idea.. so you're ok with everything just getting slower and less responsive as long as some no-code idiot can do it?

#

that's where that heads

lethal helm
#

nah. people have a tolerance for performance that doesn't change

ocean wren
#

eventually it'll be your mother "coding" your life support machine and it'll only be able to pump once a minute

#

then we're in the realm where html is "coding" again 🙂

lethal helm
#

the industry simply tries to hit that equilibrium

celest python
lethal helm
#

processing power goes up -> optimization goes down -> fps/responsiviness stays within the range people can tolerate

verbal shore
lethal helm
#

course AI coding threatens to upend this whole thing

#

cheap/fast developer time producing highly optimized code

verbal shore
#

Injecting values, running behavior trees were not that difficult but dealing with slate was a nightmare.

lethal helm
#

wat

#

why would you want that

ocean wren
celest python
lethal helm
#

but that sort of thing shouldn't even be in the behavior tree

celest python
#

no? it exactly how BT should have been at the first place

#

1/2 of my obsession with ST comes from the flexibility of how you control the data and params

lethal helm
#

walk speed is a character property and attack on sight is a controller property

celest python
#

its composed to AI tool that way

#

easier to work with imo

#

i exactly follow same pattern

#

that way you only get to care about the AI graph

#

so BT somehow becomes the "controller" of the AI

lethal helm
#

thats disturbing but you do you I guess

verbal shore
ocean wren
#

Yeah, I'd rather BT was decision only.. certainly wouldn't try and tie in animation control and such

lethal helm
#

that feels exactly what you were complaining about earlier zoom lol

ocean wren
#

But hey, whatever floats your boat

lethal helm
#

big monolithic classes

#

except behavior tree flavored

verbal shore
#

I can create behavior trees for combat, for patrol behaviors etc. and reuse them as many times as I want while being able to tweak the settings for each AI character

lethal helm
#

thats what a task is for

verbal shore
#

Yes, but then you need some object to feed those values

lethal helm
#

this is what my behavior tree looks like lol

celest python
ocean wren
#

I've always had a seperation from my logic and my execution, so that I can feed the execution from outside the logic

lethal helm
#

if it ain't broke don't fix it

#

this horror show needs to go though

image.png
celest python
#

i also have something similar to this, but i use data assets instead. state of the BT affects the character movement properties and state machine on anim graph that way

image.png
ocean wren
#

PaintRoam?

lethal helm
#

poorly named

ocean wren
#

PaintRome?

lethal helm
#

WhileRoamingPaintEnviroment would be too verbose

#

probably should be a task as well but I dislike using parallels

verbal shore
celest python
#

Are decorator variables also exposed to parameter groups?

verbal shore
#

Yep, you just need to replace them with the plugins base classes and it does the job

celest python
#

nice

verbal shore
#

It takes a little bit setup but I tried to explained them as best as I can in tutorial

#

Native properties are not exposed but it's not too difficult to include them

celest python
#

Native properties = UPROPERTY(EditAnywhere) float Something;?

verbal shore
#

Yep, it only detects variables created in blueprints

celest python
#

hmm, interesting. I thought it should be easy to query all reflected properties

#

is it because of the BlueprintBase classes?

verbal shore
#

No, I was going to go in that route but it would be too limiting for the future. I actually query them but there was a flag I was checking that basically preventing native properties to be exposed.

lethal helm
#

I recently discovered that blackboard checking decorators are actually effectively free performance wise since they are purely event based so I think I'll be merging my three BTs into one and just toggling BB values using my controller

verbal shore
celest python
#

dont you just need to add || It->HasAnyPropertyFlags(CPF_Edit)? thingken

verbal shore
#

If my computer wasn't really this slow... I'm checking it right now.

wind surge
#

how can I make my npc go through from the door when its open? Navmesh isnt generating a navigation for it(using navmesh proxy works but its buggy, I dont know why)

image.png
celest python
verbal shore
# celest python dont you just need to add `|| It->HasAnyPropertyFlags(CPF_Edit)`? <:thingken:313...

Sorry, it took too long for me to test it out, everything went wrong while I was simply trying to edit a single line.
I completely forgot that it was actually a choice I made. I didn't want to include native properties so I decided to filter them out, that's why I preferred that. It was just a plugin for my project before I decided to publish it. Without filtering them something like this happens:

celest python
#

I completely forgot that it was actually a choice I made.
😅

#

makes sense though

#

if you didnt want to filter them

#

LogicDriver also goes with BlueprintVisible method

#

to avoid filtering code

verbal shore
celest python
#

what I can recommend though, if you ever decide to add filtering, add two arrays to project settings via developer settings and exclude properties from name based on that array, possibly with mapping to specific uclass. I had some serious issues with logicdriver because this feature didnt exist and had to wrote on my own

celest python
verbal shore
celest python
#

Good night salute

verbal shore
#

You too!

rich veldt
#

Hi. If i create a UNavigationPath, say using FindPathToLocation etc, is there a way to know if a nav link proxy is a part of the path, and get a reference to it perhaps (or call its interface functions)? Or is that solely done through the use of PathFollowingComponent?

Also same question but get the nav area of a segment of the NavigationPath instead

#

Trying to learn more in depth the code for pathfollowingcomponents, navlinkproxy, etc by reading through the source code and just had this curious thought.

uneven cloud
uneven cloud
rich veldt
lethal helm
#

I guess the better question is why are nav modifiers cheaper then regular navigation regen?

uneven cloud
lethal helm
#

ah I see. so you get most of the benefits of static while still being able to have stuff like dynamic doors/abilities

uneven cloud
#

Yeah. There are also a lot of tricks you can do with them to avoid using full regen.

lethal helm
uneven cloud
lethal helm
#

alright. and then how do I set different modes for different levels

#

just set it in the recastmesh?

celest python
#

If anyone thinks the same, could mention it or put an emoji to the post

harsh storm
#

@celest python Woohoo - my PR got merged-ish

celest python
#

you beat me into it

#

i was too lazy to clean my ssd and do a pr (and get accepted) earlier than you pensive

#

though since ST is stable I dont think I'll ever do any PRs for BT anymore since I wont be using it much 😄

harsh storm
#

Oh - this was for the ST 😈

#

I'm talkin' a big game - but I just added a setter method 😅

hearty niche
#

Any useful AI perceptions functions I can expose to BP? Need to assign my junior dev

lethal helm
#

pretty sure blueprint only gets SetSenseEnabled by default

#

setters for modifying sense properties during runtime would be a good place to start I guess

hearty niche
harsh storm
#

I do think, at this current time, BT's are more straight forward than ST's.

celest python
#

never scales well as ST though

#

but with Emrehan's plugin it might be easier for large scale AIs

crystal hatch
celest python
#

well cant beat the developer itself in this argument, but at least from my experience in terms of UX and state handling ST helps way more 😄

crystal hatch
celest python
#

I like how ST itself becomes the "ai controller", meanwhile BT is heavily coupled to the AAIController

crystal hatch
#

yeah, that's one of the lessons we learned from UE3/4 days.

#

Mass is another expression of that lesson 🙂

dawn schooner
#

Are state trees something like a singleton that processes state changes for a given entity, or you have one ST instance for each entity?

celest python
celest python
#

you have a "context" parameter, which is the actor ST is bound to

dawn schooner
celest python
#

it is

#

it even calculates the memory its using

dawn schooner
celest python
#

since anim graph can run on another thread i think you shouldnt stay away from it? Thonk

#

it exactly does what ST does, works on top of the data bindings

dawn schooner
#

you know, for example, show walk+attack animation if unit velocity is bigger than x and unit is attacking

celest python
#

only difference is you can also bind ufunctions unlike ST 😦

celest python
dawn schooner
#

also, I am supporting two unit types - skeletal mesh rendered, and Niagara rendered

#

so the animation handling itself is bound to those

#

I am just interested in the ST to set a logical state, not to directly change the animation

#

not sure I am explaining myself very well xD

celest python
#

I think anim graph is still more performant and convenient

#

in a sense ST is a fancy data structure and a few modules evaluating things each tick

#

meanwhile anim graph also is a state and runs on another thread

#

and you cause less instructions on it's NativeTick function to set values

#

then binding things on anim graph

dawn schooner
#

however I'll need something to process the state and set the Niagara animations on tick

celest python
#

do it in some uobject

#

handle state in it

#

and bind to animgraph

#

and feed niagara from the same uobject

lethal helm
#

an ai is just a bunch of state trees all tied together lol

#

animation state, movement state, pathfinding state, behavior state, perception state

#

better it all be compartmentalized instead of all being in one sphagetti class

celest python
#

For state tree context, is external data == schema params?

harsh storm
# celest python never scales well as ST though

Well - sure. So far, I don't really like how the state tree events work. Only being able to process the event in tick is a bit wonky. I imagine it is specifically for cache stuff.

It really is more convoluted to get perception working with ST than BT. At least imo.

I'm not 100% sure on when I should be using the InstanceData or not. The two default tasks show both ways. My assumption was that if you want a value to be unique per-actor (or entity), put it in the InstanceData struct. If you want it to be global, put it in the task.

dawn schooner
harsh storm
#

I'm not 100% sure what you mean by that. Maybe I just haven't encountered that kind of functionality as often? Or maybe I have and just haven't realized it 😅

dawn schooner
harsh storm
#

Oh - I'm pretty sure it has to be on tick because of cache stuff

#

I haven't dealt with #mass specifically. Too early for me.

crystal hatch
dawn schooner
harsh storm
#

@crystal hatch I saw that my PR was merged-ish (and I think Mikko added one or two things to it). I'm not sure how ya'll actually do it, but 🥳. 2 days 🙂

I can now confirm that AI team is receptive to PRs 😅

#

I think ya'll merge internally somehow (or just copy paste the code 😅) and then just close the PR on github?

crystal hatch
harsh storm
#

Oh - I'm fine with touching up. Wasn't trying to convey I wasn't if that's how it came across.

crystal hatch
#

No worries, I'm just saying that's pretty much the regular process.

unborn jungle
#

Anyone know why warnings such as LogNavigationDirtyArea: Warning: Skipping dirty area creation because of empty bounds (object: FoliageInstancedStaticMeshComponent are spammed in the logs for certain painted instanced foliage meshes?

#

In this case many dense ones that do get culled at distance, I haven't seen this warning for other painted foliage

#

In this case the foliage doesn't have collision but it appears to have a valid bounds when viewing in the static mesh editor. Is it fine to just ignore these warnings? Using 4.27

sullen escarp
#

with State Tree is there a way to get the owner in BP Tasks? FStateTreeExecutionContext is not a blueprint type, do I need to make a cpp task or is there another way to get a handle on the context in BP

harsh storm
#

In your task, create an Actor variable and put it in the Context category. Then when you use it, you can bind to the Actor context variable that you selected the class of for the state tree

#

@crystal hatch Actually - real quick if you don't mind. How would you (or Mikko) look to use perception data in the State Tree? I'm currently sending a ST event when the perception is successfully triggered and then switching to the state on reception of said event. The event contains the actor that I sensed. The first task gets that actor and sets it so following tasks can use it.

It feels kind of weird to have a task for that part (receiving the perception data that is). Would an evaluator be a better fit?

crystal hatch
celest python
#

Also since Mikko is tagged - do you guys plan introducing the feature of being able to bind UFUNCTIONs to properties for data binding system? Its a default feature of animation graphs and lack of it in ST forces us to rely on evaluators more to fetch data from an external source

celest python
#

since every function is const you and mutate a member property

#

and there might be some data locality / cpu cache trick behind the instance data thingy

harsh storm
#

So - we can't mutate properties that are not in the InstanceData?

celest python
#

yeah

#

you cant even bind a delegate

#

you gotta keep it in instance data

harsh storm
#

How is that part enforced?

celest python
#

by making all 3 functions consts

#

enterstate, exitstate, tick

harsh storm
#

Maybe I'm not being clear

#

I have bProcessedEvent in my Task. It is not in InstanceData

#

I can freely change it in the task.

#

I lie.

#

Forgive me Eren

#

I see what you are saying now

#

It seems to auto-const properties

celest python
#

i was about to ask about your magical power to breach const correctness and mutate that property 😅

harsh storm
#

It is in fact, not a const bool in my .h

celest python
#

rider might be being ambigious

#

its about function that you are in is const

#

not the property itself

harsh storm
#

Yeah - it's rider shenanigans

#

Just tried in a non-const function

#

Didn't know a const function did that to be honest

#

All of my const functions are pretty much just getters though 😅

celest python
#

there is an optimization trick that I dont understand yet behind that decision

harsh storm
#

Wait - I'm silly. Of course I knew that about const functions. Silly brain. Brain fart moment.

#

I'm just going to go about my merry way

unborn jungle
#

What is the correct way to make use of nav link proxies over large distances? For example if telling an AI that a teleporter can connect navmesh location A to navmesh location B?

#

Is it possible to lift the nav mesh proxy allowance distance? (As currently it seems to fail (the dark green segment doesn't show) after quite short distances)

keen onyx
#

hello, how would i make an enemy follow a path, however that path can have a width and they can choose a location inside that width? (sorry if that doesnt make much sense idk how to explain it)

haughty coral
#

@harsh storm did you create non-instanced BT node before in C++? If I understand correctly, the instance data is just what instance memory is for BT nodes. Same concept. The node is not instantiated, so things you would usually store in class vars and set during runtime, should go in instance data.

#

Unless you make the node instanced, then you can use class variables as usual

rich veldt
# uneven cloud You might be able to, but I haven't been in that code for awhile. I'd look at t...

good call on that function!

FNavMeshNodeFlags(PathPt1.Flags).IsNavLink()
and
if (PathPt0.CustomLinkId)
{
UNavigationSystemV1* NavSys = FNavigationSystem::GetCurrent<UNavigationSystemV1>(GetWorld());
INavLinkCustomInterface* CustomNavLink = NavSys->GetCustomLink(PathPt0.CustomLinkId);
StartUsingCustomLink(CustomNavLink, SegmentEnd);
}
(where pathpt0 is an FNavPathPoint getted from the FNavigationPath instance)

#

pretty much tells me all i need to know

#

not immediately seeing anything about navarea but i am aware of rama's method of getting the info from the actual nav mesh, from the position of something etc. can probably learn from that too if i end up wanted to know more about getting navarea info.

uneven cloud
rich veldt
#

ohh ok cool

harsh storm
harsh storm
#

@celest python Have you tried messing with actually processing ST events yet?

celest python
#

not yet

harsh storm
#

😭

#

I'm trying to figure out if the event is just being dropped because it is definitely not processing it.

celest python
#

it must be similar to GAS's event listener thing

#

because schema itself has some bounds to event system

#

are you sending an event and it doesnt reach or not in the events received so far?

harsh storm
#

Sending it and then transitioning to a new state, which has a task - that on Tick is supposed to process the events through the ForEachEvent function

celest python
#

where do you send it?

harsh storm
#

StateTree->SendStateTreeEvent()

#

I'm not sure I understand your question though.

celest python
#

it was probably just not relevant anyway - was wondering if something could come up from that additional info

#

when you step into ForEachEvent, does container has any events stored?

harsh storm
#

I'm still trying to get a semi-clean way to get perception data into the tree 😅

#

I don't like some of the approaches I've done thus far

celest python
#

Completely ignoring the optimal solutions - a ST way of perception would be just an evaluator imo 😄

harsh storm
#

Yeah - probably. But I don't want it to be always ticking 😭

#

Which is why I want to do it in the Task. Because it is only relevant in a certain state

celest python
#

I thought of enabling/disabling it by events

harsh storm
#

Yeah - break point isn't even triggered 😭

#

Yeah - it just...isn't getting the event 🤔

celest python
#

gotta debug SendStateTreeEvent() then

harsh storm
#

Yup

harsh storm
#

Didn't even know you could in the context of an evaluator

celest python
#

it still has access to ForEachEvents and instance data struct, i'll just set a boolean by a tag

#

I'm thinking about creating evaluators for each sense

#

so they respond to their own tags

#

so zombies will be able to see but not hear sometimes etc

#

though -- my zombies has very minimal requirement of this

#

maybe a large scale perception system needs to be more planned upon

harsh storm
#

So yeah - putting it in the evaluator definitely receives the event.

#

There is probably something, somewhere in the process, that is clearing the event prior to my Task getting it perhaps?

ocean wren
#

Bloody zombies

harsh storm
#

Okay - so transition order does matter. Seems to evaluate from top to bottom

harsh storm
#

But it seems like our consensus is that they are a good way to get outside data to the state tree

celest python
#

hmm

#

lmao Mieszko's reaction to the message 🤣

image.png
harsh storm
celest python
#

and when move to finished (my moveto system is external - a subsystem)

celest python
celest python
celest python
#

even with statetree reflection data corruption hunts you at somewhere 😔

image.png
#

I wonder if its related to BP corruption though.. Why IsScriptStruct doing a CastChecked? Thonk

#

is property added to schema parameter is essentialy a struct or something?

#

oh it was StructUtils function

rich veldt
#

Hmm...I'm in a bit of a pickle. My doorways don't fully connect the navmesh when open, but I can use simple link to allow AI to walk through. when closed, smart link is enabled and they will open the door first. all works fine.
However if they see the door is open and start to head through and I shut the door in their face, they still try to keep walking through along the simple link path, since the smart link is enabled after they entered the link. They think they can still walk through.

#

Just brainstorming better ways to do this or something I'm maybe missing...any advice/opinions would be helpful

#

maybe im stupid and just adjusting the nav mesh size settings so that it connects through doorways and deleting the simple link is the best answer ahaha

#

but i'm more interested in other interesting answers for this here, for learning purposes.

ocean wren
#

Have the agent store the smartlink object reference and broadcast a message when you close a door that lets and agent using the smartlink know of the change in state? usually I have a resource locking mechanism for things like doors, elevators, ladders etc.

#

there's probably a resource locking mechanism in gameplay abilities or something you could use

ocean wren
harsh storm
rich veldt
#

or something like that.

#

but yeah i think broadcasting change in state to interested parties is the way to go here

#

can also play a "stumble backwards" animation too for fun like you slammed it in their face

harsh storm
#

So - it seems like the ST property binding is one way. I wanted to see if I could alter the value in an evaluator, from a task. Did not work.

ocean wren
#

I added text to image generation into my voice control plugin and didn't quite realize how much fun it would be

#

it feels.. strange, but in a "how does this not work this way" way

#

I don't know why, but being able to point to a painting and tell it to change to a random picasso is quite liberating 🙂

ocean wren
#

Oh crap.. added the default UE text to speech plugin and its so wonky that its almost endearing 🙂

ocean wren
#

Can highly recommend it for shenanigans

#

Its like a slightly less depressed Marvyn 🙂

#

Will have to integrate amazon polly though I think.. just because

haughty coral
#

It was mentioned before that SM is a superset of what BT can do. It led me to thinking yesterday whether potential BT v2 could be just a different editor for ST and under the hood it would be still ST just with autocreated transitions to mimick BT flow of events. That way it would be basically still just a single system and improvements to ST would improve also BT v2 under the hood. It would look like a BT only because of editor frontend.

ocean wren
#

I used to write my BT editors in the form that ST uses.. basically its a listcntrl

#

I always preferred it to be honest..

unborn jungle
#

If you don't need to make use of the events that smart links can provide, should you disable them on any nav mesh proxies to save performance?

uneven cloud
unborn jungle
#

As watching a video on them it seems both the smart and simple links are both enabled and overlapping each other by default

#

No need for smart features in my case so may as well disable

rich veldt
#

Hello again! More smart link/nav link proxy questions.
https://forums.unrealengine.com/t/how-ai-use-smartlink-equally/403612

Second half of this person's post applies. Wondering how to utilize broadcast enable/disable for nav link proxy

Epic Developer Community Forums
How AI use SmartLink equally

Hello. I am developing AI jump using NavLinkProxy with smartlink. for example, There are lot of bot. may 10 bots. A lots of bots are using only 1 navlink. like this picture. I think AI use a low cost path. I referenced RoboRecall jump proxy. That is amazing way. They are only thinking move to one location. So they only release and lock i...

#

line 375 of NavLinkCustomComponent is the commented line he's referencing (at least in the code version i have)

#

i dont care really whether that's commented out or not, but still wondering what event is actually getting called during the broadcast (is it on the AI Pawn?)...supposed to be using SetBroadcastFilter to choose the delegate but i havent found where that's actually called yet (if at all? suppose i could just call it myself)

#

it seems the delegate should receive the link and a list of all the path following components that are in the radius, in order to call what i assume is something like that commented out line (PFC->OnCustomLinkBroadcast) (which i cant actually find on path following components)...

#

this seems unfinished/like broadcasting doesn't actually do anything?? lol or im really missing something

mental basalt
#

I know this is a complex question with no easy answer, so I am more looking for suggestions. I have a building system sorta like Rust or Ark and I want an ai to pathfind through it in order to reach the player. I want them to be able to hit walls, go up steps, break doors, and do this all strategically. I am currently using a modified version of dijkstra's algorithm. This works ok for the tileable builds like foundations, walls, and even ramps to travel to another floor. However I have no way of checking for floor placeables like a storage box and it is staring to feel very limited. Any pointers in the right direction would be great.

celest python
#

A naive input I can give with my limited knowledge is:
path finding just provides an array of FVectors with some other utils, and you can actually change how engine finds path by default instead of using custom pathfinding methods -- there are a few functions engine checks corners, does traces from navigation etc.
so maybe if you can analyze how its done and find a way to purposefully make AI walk towards an obstacle you could handle the rest of the interaction via a component in the character by interpreting what kind of obstacle you are hitting manually (i.e. if you are hitting a door execute break door action or if its a wall play hitting a wall animation etc)

#

@mental basalt

#

PathFollowingComponent follows the path from point A to point B blindly in a straight line

#

so all you have to do is make pathfinding ignore some obstacles create path directly through them

#

or you can maybe make this contextual -- let obstacles provide path points etc

mental basalt
#

@celest python Yeah I was definitely considering something along the lines of just modifying the nav system. Definitely a big undertaking tho and I have thought of multiple edge cases that would cause issues. Still something that could provide more insight. Thank you

near jetty
#

i have been a bit off the ue5 stuff for a bit, how is state tree going? any updates on EQS and navmesh stuff?

celest python
#

StateTree became stable at 5.1

#

NavMesh and EQS pretty much same other than small tweaks afaik

#

There is a new NavCorridor thing related with pathfollowing, probably will form a shape in 5.2

#

but initial implementation exist in 5.1 too

near jetty
#

whats state tree abot? why should i use it instead of cpp state machines or similar

celest python
#

It has a flexible bottom-up state system (nodes determine which state to transit without any limitation) that you cant achieve with other methods without creating a huge sphagetti and there is a data binding system with custom linker & compiler that lets you bind other properties from arbitary sources to properties inside of your ST nodes, just like animation graph

#

ST also makes you get rid of AIController since its just a struct, so your ST graph becomes the AIController itself with the flexibility of data binding system because you dont rely on tightly coupled systems like BT<->BB<->AIController

ocean wren
#

Does it actually replace AIController? or act as a component within one?

near jetty
#

oh so ST is just an uobject/struct? that is pretty sweet

#

for AI inside mass i guess

celest python
celest python
#

you just use AIControllers to get pathfollowing comp etc and make AIs possess pawns

celest python
#

it literally carries everything to graph and you can just work with having generic nodes with input/output parameters instead of specific nodes dedicated to something

harsh storm
#

Huh - seems like parameters for the ST aren't being passed in the way one would think. Created a parameter, edited the value in the component, checked the value that got passed in and it is the default value.

harsh storm
#

@short aurora I know you've dabbled with ST's. Have you ran into this issue with them yet? 👆

short aurora
#

nope, never had an issue with that

harsh storm
#

Okay - sooooo apparently I just had to restart the editor? Interesting (no, not live coding btw)

short aurora
#

did you forget to compile a tree?

harsh storm
#

Nah.

short aurora
#

some weird stuff happens if you don't compile after making changes

harsh storm
#

This isn't the first time I've encountered this with the ST

#

I've changed a BP task before (different exposed variables) and it didn't update.

#

Just seems a little buggy for me at the moment. I've had it also flat out not allow me to delete a task until I've restarted the editor as well.

harsh storm
pulsar igloo
#

Trying some "Dynamic" nav mesh and some "Dynamic with modifier" navmeshes. But its just not picking up whenever I create a pit.
Tried giving turtwig a modifier and "Can edit" and it works. So It must be something to do with the dissplacement map

#

I am using a static mesh that gets modified based on the dynamic mesh in order to cast this, so not sure if thats causing issue but cant find anything online.

I all else fails I'm gonna try create an inverse of the pit as a really cheaty way to fix, but gonna keep looking for anything online

image.png
harsh storm
#

Converted this simple BT to ST. Here is how they look side-by-side.

So far, my big gripe with ST is some of the data flow processes. Otherwise, I do think it is much easier to write a native task/condition/evaluator for ST than it is to do a native thing for BT. Not saying it is much more difficult in BT mind you.

The parameter binding is nice - but I would like for it to be two-way instead of one-way. It is probably just my way of approach that needs to be addressed more so than needing two-way binding. But I was trying to clear out a value in a task and it was bound to a value in an evaluator. Clearing it out didn't change anything in the evaluator.

It was mentioned before as well, but having some visualization like how the BT does it, would be nice as well. The VisLog is great though 👍.

image.png image.png
tawny raven
#

Do NavModifierVolumes support multiple volumes that are disjoint or have holes, properly considering the disjoint area and holes when applying this to the navmesh?

#

I am trying to construct this case with the geometry brush but so far no luck, I only get odd results from that

#

I created a tool in UE5 that generates navmesh areas from painted landscape terrain layers (simplified boundaries, triangulated and then extruded) but it doesn't seem to work and i am now trying to construct an minimally viable example based on a NavModifierVolume, but so far it seems it expect a connected and hole-less volume

tawny raven
#

Ok so the NavModifierVolume when set on Box or Cylinder with "Hollow" on allows to create a hole and it works absolutely flawlessly (altho the Cylinder crashed while changing the parameters at some point)

#

that's pretty sweet!

#

My generated volumetric mesh however doesnt work, i suspect it could be the triangle edge directions or something funky like this

image.png
harsh storm
#

So how does the Parameter type work in ST?

#

Invalid - pretty sure this can be safely ignored in most cases. Used mostly internally I believe.
Context - This is for your context stuff
Input - ST won't compile unless this property is bound
Parameter - ????
Output - Says that the node will always set the value

image.png
celest python
#

Parameter is just like void something(float asd);'s float asd

harsh storm
#

Doesn't look like Parameter is actually used much at all.

celest python
#

it lets you also type literal values

#

in editor

#

unlike Input

harsh storm
#

🤔

celest python
#

Input forces you to put a binding, parameter can take a value that you can hardcode

#

and output probably copies the value to set to bound property

harsh storm
#

Simply doing like int Age; with a EditAnywhere UPROP does the same thing though, no?

celest python
#

yep

harsh storm
#

Seems redundant. I thought it would've meant more 😅

celest python
#

You dont ever end up needing to pass parameters to tasks? Thonk

harsh storm
#

I just do EditAnywhere

celest python
#

or the idea of "Category="parameter" sucks

harsh storm
#

Instead of Category

#

UPROPERTY(EditAnywhere) vs UPROPERTY(Category=Parameter) - I just do the former

celest python
#

does it let you bind data?

harsh storm
#

Yeah

celest python
#

nice

harsh storm
#

Actually, wait. Idk

#

I'm silly

#

I'm doing this:

UPROPERTY(EditAnywhere, Category = Parameter)
bool bUseActor;
#

🤣

#

I'm literally doing both.

#

This was when I was first writing stuff and I just guessed that Parameter was a valid Category

#

Let me take that off and see if I can still bind

harsh storm
#

So yeah - even without the category, I can still bind

celest python
#

i assumed if I dont put it, it would consider as Invalid

#

so went with it so far

harsh storm
#

So, like I said, seems kind of redundant at this point in time.

#

While you're here Eren, do you know if it is possible to actually alter information somewhere else? So, I want a task to clear out a value in an Evaluator for example.

mental basalt
#

Hey does anyone know where I can find the source code implementation for FindPath()? After going to its definition in VS it just returns (*FindPathImplementation)(AgentProperties, Query). Going to definition of that just takes me to FFindPathPtr FindPathImplementation. This is where I get stuck and confused.

celest python
#

AbstractNavMesh.cpp iirc

#

its a function ptr declaration, FFindPathPtr is is the type of that func ptr

#

its being set to a specific function

terse panther
#

hey there everyone, i was just going through the AIPerceptionComponent header file and found that there are many functions that are c++ exclusive, so just wanted to ask what are some of the c++ exclusive function that you guys use to make your Ai workflow easier?

celest python
#

I think you should use output category?

harsh storm
#

I don't think that'll actually do what I'm asking. That's there so you can't alter it in the editor, but you can still bind to it because the node is supposed to be altering it.

celest python
#

Hmm, output category doesnt work as I thought then?

#

i thought it was suppose to act like a reference to property that you bind and it should modify after node executed in ST's update loop

harsh storm
#

I mean - I can still try to see. Not like it'll take much effort 😅

#

Yeah - doesn't work that way @celest python

celest python
#

hmm... well it sucks then, i was expecting it to act more different than an input parameter

harsh storm
#

Well it kind of does

#

Input param, you can alter in the editor and bind it and all that

#

Output is meant ONLY for other stuff to bind to.

#

Because the node is going to set the value

celest python
#

what "other stuff" covers in that context?

#

schema parameters etc?

#

oh wait.. i just understand what you were actually asking btw

#

you were asking if its possible to alter a different node's property from another node

#

i dont think it is possible, you need to centralize data into schema parameters and work with in/outs i guess

harsh storm
#

That is the purpose of output

#

It should be an output because the task itself is going to alter it. So it should not be edited by outside sources (which is what binding or changing it in the editor would do)

harsh storm
celest python
harsh storm
#

Yeah, that's where it exists.

#

The reference itself is protected. No public way to get it.

#

So I'd have to inherit and expose it more or less.

#

But then I need to figure out how to just update like...a single variable.

celest python
celest python
harsh storm
harsh storm
celest python
harsh storm
#

Oh - yeah, that. Yeah, it felt like that to be honest. But hey, ya' never know 😅

celest python
#

Only if events provide an FInstancedStruct sad

#

it'd be nice to interact with other nodes

#

via ST events

celest python
harsh storm
#

It would be nice to be able to at least update the parameters at runtime though.

harsh storm
celest python
#

oh i get it now

#

it becomes something like a local variable scoped to all tasks in a node?

harsh storm
#

To w/e can bind to it really

celest python
#

so output params becomes something like schema parameters? Thonk

harsh storm
#

If it wasn't an output param, in the editor, I could bind it to something else - but that would be silly, because the task is going to overwrite it anyway.

#

Look at it like "output data" more or less. Your task is going to do some work, something else may be reliant on that data. You don't want to let anyone else mess with this data accidently (manually editing it, binding this data to something else, etc...). Mark it as an output param. It lets OTHER things bind to the output param, while not allowing anyone to edit it.

celest python
#

I see

#

pretty cool tbh

harsh storm
#

So - Input forces you to bind. Parameter for free editing or binding (or just EditAnywhere). Output for non-editable stuff, but let others bind to it.

celest python
#

yeah perfectly makes sense now

#

thanks

harsh storm
#

Now, my next goal is to see how I can actually update the parameters in the schema through code.

harsh storm
#

The answer is in PropertyBag.h, FInstancedPropertyBag has a nice comment above it.

#

I've tried setting the value but it seemed to zero it out 🤔

#

It's like the change isn't being saved after leaving the method 🤔. I've tried going through the various Sync methods as well.

haughty coral
#

nvm, I see you already use it on the screenshot

harsh storm
celest python
#

@pallid mica ChillBar_wave

pallid mica
#

But I guess we can talk here too

celest python
#

so tl;dr you create two USTRUCTs

#

one for the node itself, derives from relevant FStateTree.....CommonBase

#

either task, condition or evaluator

#

and one "instance data struct"

pallid mica
#

I assume this is advanced usage, since BP only people don't have access to C++ Structs?

celest python
#

its the default workflow for any C++ node you want to create rather than advanced, but in BP yeah its just a matter of adding a new variable and settings its category name

#

linker or compiler goes through property list and does its job behind the scenes

pallid mica
#

Ah, one sec, then I maybe kow what you mean

#

Okay so I can only use Struct or Objects

celest python
#

hmm, what do you mean? thingken

pallid mica
#

I'm trying to understand how I link up more context

#

Creating a boolean in a Task and putting it into Context doesn't work

#

A Struct seems to work but now it's angry about bindings

celest python
#

it should be, ensure you're giving proper names based on available category names

pallid mica
#

How do I populate more values into this from BPs

#

"Cound". Lovely. Someone wants a free PR for UE badges?

celest python
#

add a boolean to ST Test Task and set its category to Parameter

#

if it doesnt appear something is wrong

#

this struct shouldnt be context

image.png
#

thats why its complaining

#

it should be either input, output or parameter

pallid mica
#

Ah

#

So... Output of something can be Input of the next ?

#

Difference between Parameter and Input?

haughty coral
#

yes, it can be input for substates or following states at the same level

pallid mica
#

And what should all be Context? Without modifying it via the C++ Struct I assume I can't add more to Context?

#

Not that I need to, just trying to understand the basics here

celest python
#

Context is equal to parameter, its just a helper thing to make it match with your actor type in the schema

image.png
#

it just adds a rule to the ST system that helps people to avoid issues and bind context actor parameter to nodes

#

schemas are also inheritable, you can override from base schema class and make params promoted to desired type like how Actor param is promoted to another type

haughty coral
#

Context is like a hardcoded base params of the given type of ST.

In case of AI, you get a Context which is your owner actor.

If you would create own type of ST, called Interaction State Tree Schema, you could define own context vars like Iteractor, InteractedObject.

pallid mica
#

Alright, but that's C++ only?

celest python
#

creating new schemas are c++ only

haughty coral
#

But most of the time you just want to run your ST within ST Component of some actor. And for that you can use the default schema.

pallid mica
#

Yeah totally, again just want to know what I can do

#

The "Parameter" category, how is that working?

haughty coral
celest python
#

its literally like a function parameter, it copies the data from binding or the value to instance data struct of the node -- if you are on BP its just setting the relevant property without messing with instance data

pallid mica
#

Right, what I'm missing is where variables in the Parameter category are meant to show up

celest python
pallid mica
#

Ah, yeah they aren't

#

Let me see why

celest python
#

hey exi wanna check lounge 😄

pallid mica
#

Default Category or Parameter Category both seem to not make it show up there

celest python
#

we tested yesterday -- or one day earlier

pallid mica
#

Maybe something obvious?

#

Does it need to be Instanced Editable or so?

#

Ah yeah it does

#

ajkfalösfgjaklöfjkaös

#

WELL, okay

#

So Params are something I can set on the Task, similar to how that worked in BehaviorTreeTasks.
Context is bound to the Context, by default only Actor.
Output is something other same level or children can bind to with Input, so I can forward data from upper tasks.

#

I assume that goes also for other things than Tasks and is interchangable as in a Task can Output something a Condition can Input?

#

As well as Evaluator?

#

Seems straight forward, but the Context, Input, Parameter, Output AND INSTANCED EDITABLE part is something they could have better explained in the examples.

#

Typical "Just do it." tutorials.

pallid mica
celest python
#

normally every task had its own local evaluators like tasks and enter conditions

#

but then they removed it

#

evaluators are only for schema globally now

#

and it practically has zero difference from a task runs at root node

pallid mica
#

So I can ignore them?

#

What were they meant to be used for?

#

Condition and Task are straight forward

celest python
#

for example a pathfollowing component or perception component can be replaced with an evaluator

#

or an EQS that runs every 0.2s

#

can be an evaluator

pallid mica
#

But that doesn't work/exist now anymore?

#

Only globally for the whole tree?

celest python
#

it exists and works, its just it was available for every node before, now its just only globally for whole tree

#

so its activated and ticks until ST is stopped

#

you dont have control over it much like tasks

#

siliex created a task that runs an evaluator for this 😄

#

just to hack it

pallid mica
#

Okay, but what does one put in there now? Unless I have something that should tick for the whole tree it seems rather useless?

celest python
pallid mica
#

Is that basically what a Service in Behavior Tree was?

celest python
#

this is what Mikko said - developer of ST

celest python
#

though this one was more influence over graph

pallid mica
#

Okay, so I can put some stuff in there that should push some data (Output?) to Tasks globally, if I ever need it.

#

That's fair

#

And that is, on a task level, replaced with the task doing it itself?

celest python
#

yes

pallid mica
#

Alright

celest python
#

for me, evaluators are used to set external data (schema params etc)
and tasks for working with things in local context

pallid mica
#

Yeah I'm still at the "try to recreate that AI Behavior Tree in State Tree" level

#

Cause I HATE BehaviorTrees

celest python
#

I was migrated to FSM completely because I didnt like BTs... i hated that too

#

then unwillingly came back to BTs

#

but ST is a blessing for me

#

literally made me remove 5k lines of code today

#

because of how it abstracts things away with data binding system

harsh storm
#

I currently use my evaluator as a way to process state tree events

#

Putting something in the "Parameter" category seems to be completely optional honestly. Just need to be instance editable or EditAnywhere UPROP.

cedar mirage
#

https://sourceb.in/2udqv4nVjZ

I'm trying to figure out how to get GraphAStar working for my purposes, I already made a struct based on TGraph and one for TQueryFilter as well as initialise them (I think, unless I'm doing it improperly, is that what's causing the issue? FGraphAStarDefaultNode<TGraph>,false>), but I keep getting this error on line 278 on the sourcebin code:

[
    TGraph=FAStarGridGraph,
    TQueryFilter=FAStarGridGraphQueryFilter,
    TResultPathInfo=TArray<FIntPoint,FDefaultAllocator>
]```

I've been using:
https://github.com/ZioYuri78/GraphAStarExample
https://crussel.net/2020/08/16/builtin-a-pathfinding-in-unreal-engine-4-25/

as examples and although I'm not following them exactly (I'm just trying to make the character move according to square cells, not hex grid, but I think I'm at least getting the principles correct so I'm not sure why it's giving me that error? Thanks if anyone replies, I've tried looking up information but there doesn't seem to be much asides these two.
SourceBin
GraphAStarNavMesh
GitHub
GitHub - ZioYuri78/GraphAStarExample: Unreal Engine 4 generic graph...

Unreal Engine 4 generic graph A* implementation for hexagonal grids. - GitHub - ZioYuri78/GraphAStarExample: Unreal Engine 4 generic graph A* implementation for hexagonal grids.

crussel.net
crussel
Builtin A* Pathfinding in Unreal Engine 4.25

A few years back I wrote Neatly replacing NavMesh with A* in UE4 and ever since I’ve had a vague notion that it’s probably gone wildly out-of-date. As it happens I was recently working …

dawn schooner
#

Any good docs/resources about state trees?

#

Besides the official docs themselves 🙂 Doesn’t need to be Unreal’s, can be other implementations/info about the same concepts

harsh storm
#

There's like 2 more pieces that I am trying to figure out.

dawn schooner
#

I'm mostly trying to understand what it does and what usecases does it support

#

Bcs it could be a good solution for our combat system (maybe even replacing our AI controllers), it seems to map well with our current flow diagram

harsh storm
#

Supposed to be a combo of State machines and behavior Trees.

dawn schooner
#

mmm other 2 concepts I am not very familiar with, will look into it :D

lethal helm
#

what if I don't trust the nav system to select it for me

harsh storm
shadow furnace
harsh storm
#

@shadow furnace Have you messed with the ST much in regards to #mass?

shadow furnace
#

yep

#

I am making some source changes to make it less painful

#

I made it actually log missing fragments now

harsh storm
#

I already had a PR merged for ST 😈

shadow furnace
#

frankly, it ensuring on a missing fragment is really weird as I mean, the whole idea is to just not run if you don't have the query fragments

harsh storm
#

Anywho - have you been able to figure out how to actually change a parameter in the FInstancedPropertyBag?

#

I have changed it but it doesn't seem to actually stick

shadow furnace
#

...which FInstancedPropertyBag?

harsh storm
#

In StateTreeRef

#

It has a method, GetMutableParameters(). This is the parameters for your state tree.

shadow furnace
#

I assume it is not syncing

harsh storm
#

Yeah, that's what I thought too. But even after setting it, I tried each of the Sync* methods, still didn't seem to actually stick

#

So I'm not sure what I'm missing here.

shadow furnace
#

seems like it needs to be a FStateTreeExecutionContext

#

FStateTreeExecutionContext::SetParameters is used at least once

#

:/ I have no clue how this is stored

#

everything is so complicated

harsh storm
#

Yeah - it is struct upon struct upon struct upon struct 😅

harsh storm
#

Pretty sure that is only meant to actually set the params for the execution.

#

idk though. Just so many structs 😅

shadow furnace
#

I really dislike making a new struct just for this one thing to have an int that can also be invalid

#

saving one line of code by making 3 new types for me to have to worry about

harsh storm
#

For loop go brrr though

shadow furnace
#

I guess if it's to save on every bit I get it though

#

definitely lots of uint8 etc

harsh storm
#

I try and look at all this ST complication as just a thing for ECS stuff honestly.

shadow furnace
#

It's completely divorced from the ECS module

#

GameplayStateTreeModule has one just for regular actors

harsh storm
#

@shadow furnace @celest python I did it! Changed the parameters at runtime of the state tree

#

Oh my jeebus is it convoluted though

#
  1. Get StateTreeRef from the StateTree component (you have to manually expose this with a getter method, just inherit from the component)
  2. In the StateTreeRef, get the state tree. The StateTreeRef should have a reference to the running state tree. The one specifically for this STR.
  3. With the running state tree now, you can get the parameters - which is a FInstancedPropertyBag.

Now you have the actual params for the state tree. Now you just do the set methods that you can find in the FInstancedPropertyBag struct.

#

So, it seems like what happens is the asset has its own property bag. Then when the state tree actually gets created at runtime, it copies those params into its own property bag.

#

I don't know if this is the correct way though. But it does work.

shadow furnace
#

once you have a context you can just SetParameters?

#

StateTreeRef.SetStateTree(MyUStateTree);

harsh storm
#

Is that the right context?

#

I wonder if that is the expected way to do it?

shadow furnace
#

I have no clue...

harsh storm
#

To set the parameters, it has to go through all of them.

#

So if you're just changing one value...

shadow furnace
#

Also a lot of the parameters are just null for my case

#

I think the parameters being one giant blob is to pack them in memory?

harsh storm
#

This is what set params does

image.png
shadow furnace
#

I assume the data view index is how you target one at a time?

harsh storm
#

Yeah, I'm assuming the same.

#

But I don't see how it increments in this method

shadow furnace
#

FStateTreeIndex8 exists only to make MAX_uint8 the invalid value...

#

or maybe this is the index of every kind of statetree

harsh storm
#

One weird thing that happened though.

auto StateTreeRef = PawnStateTree->GetStateTreeRef();
UStateTree* ActualStateTree = StateTreeRef.GetMutableStateTree();
const FInstancedPropertyBag& Params = ActualStateTree->GetDefaultParameters();

float AttackRange = Params.GetValueFloat(FName("AttackRange")).GetValue();
GEngine->AddOnScreenDebugMessage(-1, 3.0f, FColor::Yellow, FString::Printf(TEXT("Value: %f"), AttackRange));

Params.SetValueFloat(FName("AttackRange"), 300.0f);
AttackRange = Params.GetValueFloat(FName("AttackRange")).GetValue();

GEngine->AddOnScreenDebugMessage(-1, 3.0f, FColor::Yellow, FString::Printf(TEXT("New Value: %f"), AttackRange));

This is the code that I'm using. The very first time I ran this code, AttackRange printed 0. Then I set it to 300. The New value was correct at 300.
Then I closed the game and re-ran it. Ever since that first time, the first print prints 300.

shadow furnace
#

ah, this is a compilation thing br_fear

harsh storm
#

So I'm wondering if there is some behind the scenes thing that I did by doing this approach

#

I have this in my OnPossess function.

shadow furnace
#

FStateTreeCompiler::Compile touches this

harsh storm
#

I'm just not sure why it prints 0 the first time, then the correct value every time after.

shadow furnace
#

I assume you are encoding the value into it after it flushes something

#

also are "DefaultParameters" what we want here?

harsh storm
#

Might also be PIE shenanigans. When I do it as a standalone, it prints the 0 the first time each time.

harsh storm
#

/** @return List of default parameters of the state tree. Default parameter values can be overridden at runtime by the execution context. */

The comment above the method does seem to imply that it should be overridden by the execution context 😅

shadow furnace
#

where's your execution context?

harsh storm
#

What do you mean? StateTree comp and StateTreeRef don't expose that

#

I am trying to change the parameter value from outside of the StateTree

#

The context is a local variable in StartLogic()

shadow furnace
#

seems you might want to construct a FStateTreeExecutionContext

harsh storm
#

Yeah, that is one route. But I'm not sure if that will be the same execution context that is actually being used

#

I guess to keep it all the same - I'd also need to expose a way to get the StateTree Component's InstanceData.

dawn schooner
harsh storm
#

@shadow furnace So I just did a quick test doing it the way StartLogic() does it (with creating Context and what not). Didn't seem to update.

shadow furnace
#

I have no clue then

#

my take is that you probably have to set the defaults before you actually start it then

harsh storm
# 
// inherited ST component
FStateTreeInstanceData GetInstanceData() const { return InstanceData; }

// code updating things (this is in a different cpp file)
auto StateTreeRef = PawnStateTree->GetStateTreeRef();
FStateTreeInstanceData Data = PawnStateTree->GetInstanceData();
FStateTreeExecutionContext Context(*InPawn, *StateTreeRef.GetStateTree(), Data);

const FInstancedPropertyBag& Params = StateTreeRef.GetMutableParameters();
float AttackRange = Params.GetValueFloat(FName("AttackRange")).GetValue();
GEngine->AddOnScreenDebugMessage(-1, 3.0f, FColor::Yellow, FString::Printf(TEXT("Value: %f"), AttackRange));

Params.SetValueFloat(FName("AttackRange"), 150.0f);
AttackRange = Params.GetValueFloat(FName("AttackRange")).GetValue();
GEngine->AddOnScreenDebugMessage(-1, 3.0f, FColor::Yellow, FString::Printf(TEXT("New Value: %f"), AttackRange));

Context.SetParameters(StateTreeRef.GetParameters());
shadow furnace
#

FWIW the only statetree stuff I have done is making a new task

#

nothing like this

harsh storm
#

hahaha

#

I'm going to have to wait until Epic gets back from holiday. I have like 2 questions for Mikko 😅

#

But as it stands, my first approach does work.

#

Just don't know if it is the "correct" way

shadow furnace
#

A big part of me wishes that the stuff intended to be used in mass most of the time WAS in Mass

#

like a lot of the surrounding Mass stuff stores state in a data structure and tracks it for later

#

when... they could just be mass entities

harsh storm
#

Maybe it's because they're trying to have access points for the actor system though?

shadow furnace
#

I suppose, but the typed element framework just uses mass

#

so I don't think it's that unreasonable to have stuff stored in it

#

code/polymorphism can be regular uobject functions etc

#

that reminds me, I have to finish this PR

harsh storm
#

Oh well - I accomplished what I set out to do tonight.

shadow furnace
#

are you calling sync?

harsh storm
#

What does that actually do?

shadow furnace
#

ah wait,

harsh storm
#

Actually don't know if that is even relevant for this to be honest.

shadow furnace
#

FStateTreeReferenceDetails::SyncParameters

#

ah, this might be asset level

#

nevermind

#

I swear something like this is called

harsh storm
#

Yeah - I was reading that as something for editor stuff

shadow furnace
#

FStateTreeReference::ConditionallySyncParameters ?

harsh storm
#

Nope. Still didn't work

#

Tried both SyncParameters & Conditionally

shadow furnace
#

okay I'm out of ideas

#

later 😄

harsh storm
#

Well - at least I did find a way that works though!

pallid mica
#

Tried to follow what you two discussed, but I wonder what this is even for, or rather what you are trying to accomplish in the end.
(trying to learn here)

#

Params.SetValueFloat(FName("AttackRange"), 150.0f);

#

I assume this returns Success?

#

Regarding what the idea behind this is: Isn't, in this example, AttackRange something you can pull from the Owning Actor from the Context?

#

Just still wondering what the idea is behind setting those params (and where one even uses those then).

#

@harsh storm

keen crow
#

can I make debug build-only BT nodes? I tried wrapping entire content of logging decorator inside of #if UE_BUILD_DEBUG, but the compiler says that my UPROPERTYies myst not be inside preprocessor blocks

celest python
#

its better if you dont

#

prone to many bugs

keen crow
celest python
#

you can only have #if WITH_EDITORONLY_DATA

#

node itself will stay in BT graph but context wont

keen crow
#

yeah so far I'm fine with node staying, I just don't want it to do anything. I think I could also wrap contents of functions in #if UE_BUILD_DEBUG in this case

#

thhanks anyways

harsh storm
# pallid mica Just still wondering what the idea is behind setting those params (and where one...

Uhh - yeah, you could 100% just bind to variables that live on the actor and handle it there. This was more a curiosity journey than anything honestly.

What I was messing with was the actual parameters of the state tree. So if you're looking at the state tree asset and look to the left side, there is a section "Parameters". I was trying to work out how to change those values at runtime.

What this does open the door to though is a more generic ST. If I have two unrelated classes, but for w/e reason they both have AttackRange, I now have a way for them to share a state machine. I know I know - I'm kind of stretching here 😅.

#

So the context variable might not be one of those two actors. It might just be the generic Actor class.

pallid mica
#

Wouldn't an Interface do the job then?

harsh storm
#

Can an interface be the context variable?

pallid mica
#

Not saying your idea isn't worth exploring

#

Well what I mean is, your two Actors can have the Interface

#

You can just use the Interface on the Context I guess?

#

But bindings then won't work I guess

harsh storm
#

Yeah - no bindings.

pallid mica
#

It's strange that there are no "from outside the tree" accessors for this

#

Did you check if the SetParam thing returned success?

harsh storm
#

I don't think so. Didn't look honestly.

pallid mica
#

Would be worth checking I guess

#

Cause then you can explore why

#

Ultimately this is just a Set call on the FFloatProperty pointer

#

So if it does fail, then you can check there

harsh storm
#

But this whole thing started because I was trying to work out a good way to handle integrating perception data with the ST.

  1. Use evaluators (which is what I currently do)
  2. Bind directly to some variable on the actor
  3. Alter the ST parameters
#

So - SetParameters inside the ExecutionContext has no return.

#

Setting things on the Bag though, does return something that we can see if it was successful or not I believe. Returns TValueOrError<Type, EPropertyBagResult>

#

So, it would be a good idea to check the result and what not if you're operating on the bag directly. This code is just exploratory code anyway, so I'm not really doing that.

#

The part that was failing was the SetParameters() part. And that returns void.

elder perch
#

Is it possible to get the nav mesh to not build for certain static meshes (like rocks) without getting rid of that rocks collision so that AI doesn't get stuck on tricky nav mesh generated areas. Ideally I just want to not generate navigation around rocks so the AI just ignores it and walks around it.

image.png
uneven cloud
uneven cloud
south gale
#

im having an issue where Nav modifier volumes only work if the "mask fill collision underneath for navmesh" flag is checked.. ive already confirmed that i dont have anything above it with 'fill collision underneath', so im not sure why it would be relevant

#

any ideas about possible causes?

echo turtle
#

Hello.

#

I don't think the issues with the Sight Radius.

image.png image.png
harsh storm
#

PawnSensing is old. Prefer Perception.

#

That said - you are breaking line of sight.

echo turtle
#

ok

harsh storm
#

@crystal hatch Speaking of which - is it high time we just remove the PawnSensing stuff?

echo turtle
#

How can I get it to not break the line of sight?

harsh storm
#

Don't go behind the wall

echo turtle
#

Thanks - I've got some parts of the wall removed and it works.

young lintel
#

Alright, my AI debug overlay says that I have three active smart objects in my map, which is the expected number, but Find Smart Objects returns 0, every time

#

I'm not doing any filtering, the query box is 200 meters across and all the smart objects are within 20 meters of the character who needs to find them

#

I'm not getting anything in my log about smart objects

young lintel
#

Welp, nevermind, it finds them and claims them, now it just doesn't want to navigate to them

celest python
#

@shadow furnace ChillBar_wave
apologies for pinging

since you are familiar with the concept from #mass , do you know how I should return TConstArrayView<> from a function with multiple elements?

TConstArrayView<FStateTreeExternalDataDesc> UZGStateTreeAICharacterSchema::GetContextDataDescs() const
{
    // MakeArrayView(&BehaviorDataDataDesc, 1), MakeArrayView(&ContextActorDataDesc, 1)

    TConstArrayView<FStateTreeExternalDataDesc> ArrayView(std::initializer_list(&BehaviorDataDataDesc, &ContextActorDataDesc));
    return ArrayView;
}

this doesnt work -- compiles fine but crashes and array view is Invalid

elder perch
shadow furnace
#

I'm honestly clueless if this is doable or even desirable

celest python
#

its calling MakeArrayView in base class

#

though its not possible to pass multiple elements with it

pallid mica
#

Can StateTrees be nested similar to BehaviorTrees?

#

E.g. could I have a generic StateTree that, when reaching a specific State, executes some Dynamic inner one?

#

I see some "SubTree" and "Linked" Type options for States. Not sure how they work though

fickle cedar
#

Seems like StateTree doesn't come with any tools for navmesh navigation, does it?

pallid mica
#

Depends on what you need

#

The Nodes in Behavior Trees weren't the most fleshed out once either. Probably stuff you can just recreated for State Tree

fickle cedar
#

I would like to make a character/actor go from location A to location B

#

(on top of a 3D landscape)

#

Maybe I should still go for behavior tree instead?

pallid mica
#

You can do that with the Tree too though

fickle cedar
#

Yeah? Any tips welcome 🙂

pallid mica
#

Just write a Task that performs the movement

#

This doesn't seem any different to Behavior Trees in that regard

#

I'm moving our BT to StateTrees right now too. Doing the Roaming Task atm

fickle cedar
#

Oh cool. Willing to share your approach? 🙂 Cause I'm afraid of having to implement path finding / obstacle avoidance, which sounds complex

celest python
#

those implementations exist in components

#

BT just calls relevant functions

#

you can do the same

#

MoveTo is just AIController->RequestMove + FindPath

pallid mica
#

Yeah you don't have to code all that stuff

#

Just the Task the calls MoveToTarget fwiw

celest python
#

if you are willing to go with hardcore way you can rewrite those components as tasks/evaluators but embrace good amount of boilerplate code writing process 😄

pallid mica
celest python
pallid mica
#

You just have to code the Task that does this for you

#

Like calls that for you

#

And ours is a bit more involved

#

Cause we have a GA that performs the movement

#

The Tree just sends a GameplayEvent to the Owner with all Info and waits for the Ability to end.

#

It's a bit more involved so not worth sharing for basics

pallid mica
fickle cedar
#

ok ok

pallid mica
#

The Enter Condition UI is quite sexy I have to say

#

Editor Property UI that allows Condition Building via some simple clicks is quite nice

#

That was so much more vague in BTs

dawn schooner
#

Each ST task is just a call to whatever function you want, right?

pallid mica
#

Yeah, a Task is just there to execute logic for you

#

What logic that is is up to you

#

Can be Finding a Location

#

Moving to one

#

Dealing Damage

#

Destroying something

#

Hell, even just changing the Music that is playing

#

(StateTree for Music State handling in that case)

#

It's one of those things where it's surprising no one ever made a proper implementation of State Machines before

#

I always had to code them myself

#

(much less complex of course)

fickle cedar
#

Btw a small side question... the current 5.1 docs suggest that in order to get a reference to the Character/Pawn/Actor within the task, I should create an "Actor" variable of the type "Actor (reference)", set it to the "Context" category, and the actual reference will be just somehow magically passed in? or did I miss something?

pallid mica
#

Yeah

#

There are 4 Categories that you can use. I don't know why they don't point this out.

#

Context: Binds to Variables from the "Context" of your StateTree. Without further C++ changes, that's just a variable named "Actor" with the Type of "Context Actor Class", which you can select in the StateTree asset at the top left.
Parameter: Those show up as parameters when you select the Task/Condition/Evaluator(?). The variable has to be marked as "Instance Editable" though. It's a variable you can define per Task Instance, so if you use the Task in two places, you can pass in a value. E.g. a boolean, and integer, etc. Like a float for how long a Delay should be.
Output: Provides this Value as an "Input" option for other Tasks/etc. that have a variable in the Input Category with the same Type I assume.
Input: Requires you to bind another Variable to this when using it. Example would be a Task that provides a "TargetLocation" as Output and a second Task that requires a "TargetLocation" as Input. When using them both in the same State (not sure about this, but I couldn't bind to anything from a different State on the same level at least), you can then bind the Input of one to the Output of the other, forward variable. Basically like a Function with Outputs and Inputs.

#

--
So here I can't seem to bind the Input of Roam to Point (or rather Enemy Move To) to Find Point (or rather Output of Enemy Find Roam Location).

image.png
fickle cedar
#

Ok, that's valuable information, thank you!

pallid mica
#

Even though they happen in order

#

This would work, but that's not what I want

image.png
#

So instead I moved both tasks into one State, after each other. That should theoretically do the trick (: it compiles at least

image.png
fickle cedar
#

Gonna try and remake exactly what you've got there xD

pallid mica
#

At least I think that's correct

#

Honestly, try to recreate your BehaviorTree first

#

Mine will look wildly different eventually I assume

#

Good thing though is that this removes this nasty global Blackboard with TargetLocation in it

fickle cedar
#

Yeah I've got the Tree set up roughly, now I'm trying to implement this MoveTo part (either move to a user-set target point in the level or a random roam point around that target point once the unit arrives)

pallid mica
#

Also I think BlueprintBase things should have a NAME property, so I can overall rename them to something else than that nasty long STT name that my BP is called

#

Well FText FriendlyName w/e

fickle cedar
#

You can rename them in the Tasks section afaik

#

Just rewrite STT Enemy Move To to e.g. Move To