gameplay-ai

[crimson galleon]

quick cash concepts may change, but its the same allure

why work for it when you dont have to?

[ocean wren]

hate them all 🙂

[crimson galleon]

you want to be slave to corporate all your life? dont you hate that idea? wouldnt you rather make money without any sort of effort? arent you lazy? dont you want things and have no sense of exchange? you mean you're not so selfishly motivated?

baseball cards.... beanie babies... NFTs

[ocean wren]

no, I hate corporate too 🙂

[crimson galleon]

lol

[crimson galleon]

also those funko pop figures

[ocean wren]

right, night all

[green vault]

why would someone use a black board key to get the variable over getting it from the actor itself in a task?

[celest python]

You can't have modularity and re-usability if you bind your variables to your actor class

For example in Crysis, both Korean soldier AIs and flying alien AIs are using same BehaviorTree

[ocean wren]

You could have a blackboardinterface that the AI used to get its values from the actor?

[celest python]

That could be a workaround but why would you do that? You can just store the common key names and pass values to BB via BlackboardComponent. Inheritance should stay between blackboards imo while actor/character classes can be completely different

Not doing that would not be a disaster or a very bad thing but having a proper BB that holds variables is a good practice imho

[ocean wren]

I'd argue that UE implementation isn't a proper BB in the first place

[celest python]

Never saw other implementations but UE's look like a fancy UDataAsset

[ocean wren]

well, the fact that you cant use arrays pisses me off

[celest python]

Yeah, also using raw FNames instead of FGameplayTags also annoying

[ocean wren]

the fact you can't share a blackboard with other actors pisses me off

[celest python]

Sometimes I keep variables inside AIController for this

Since AIController is less likely to be seperated between inheritance branches

[ocean wren]

Yeah, its just a bullshit limitation that never should have been

again, you could add an interface to AI controller I guess

but still, it pisses me off 🙂

[celest python]

Sure, we're paying %5 of our income to use the engine in the end 😄 It sucks UE ignores features like this

[ocean wren]

only past the first mill.. I can live with that

and frankly, its our responsibility to make these things, not epic's

[celest python]

They rely on "developers able to PR or implement their own solutions since we provide source code for free" mindset

[ocean wren]

And you know what? they're right

[celest python]

Since they provide premium support to AAA by UDN no one complains either

[ocean wren]

its what sorts the men from the boys (and girls)

[celest python]

It's right, but lets say I'm a project manager instead of a developer, I'd need to pay extra money to my developer to do what Epic had to do in the first place

[ocean wren]

I doubt UDN offers much value to be honest, other than "we're really screwed, give us a hand" stuff

[celest python]

For very specific and low-level engine things they really provide god-level support, from what I've heard of

[ocean wren]

Well, its OUR responsibility to make our games right? not theres. If this is a game-required feature, then get on with it

Yeah, they're the knowledge provider of last resort. But I suspect that's mostly only useful when you're in the shit with the innards

usually UObject and blueprint's grossness 🙂

[celest python]

Rather than 'lack of features' I don't like they expecting us to PR for issues, like we are getting paid for it

That makes them comfortable and dismiss the game/project-breaking issues

[ocean wren]

well, yeah, I'd berate them for not having stable features

but having not implemented something you need? do it yourself!

[celest python]

Yeah 👌 😄

[celest python]

💯

[ocean wren]

at least you can, unlike with Unity say

[autumn ibex]

I have a problem with my AI actors unable to walk past each other ( even though I set their collision capsules to NON-SOLID to each other). Is there a setting where I can tell the Navmesh pathfinding to ignore the collision capsule of the actor?
https://www.youtube.com/watch?v=AcT3InyqQDU

I'm using the "MoveTo" node in the behavior tree.. and it works fine if I set the collision capsule size to "1 unit" as the AI are able to easily walk past each other and don't get stuck. but I don't want to set the capsule to 1 unit as they'll clip into the world

[waxen saddle]

It looks like you have avoidance on?

I think its RVO something, a setting on the actor

[autumn ibex]

lemem check.

Nope, I have it disabled ( by default )

[waxen saddle]

hm

[autumn ibex]

these are my Character Movement settings

[ocean wren]

You've got your capsule settings wrong

[autumn ibex]

my capsulse settings

[ocean wren]

Check it by standing one on top of the other and you'll see, it won't fall

[autumn ibex]

hmm..

[waxen saddle]

Is it set to dynamic obstacle on the capsule?

[ocean wren]

your capsule is set to block pawns

and it is a pawn

[autumn ibex]

no, it's a PAWNAI

I made a new category

PawnAI is non-solid to each other

and they CAN go through each other

[ocean wren]

is it derived from pawn?

[autumn ibex]

my Actor is yes

My AI actor is derived from pawn

[ocean wren]

The collision is going to be on your actor, not your ai actor

[autumn ibex]

oh hmm

[ocean wren]

aicontroller

[autumn ibex]

ah lemme check that

so I need to put the Capsule collision on my controller

NOT on my Actor

[ocean wren]

well, you want it on your actor, but if you have it on your AIcontroller too, then it might be that its colliding between the two

[waxen saddle]

Dynamic Obstacle is probably on

[autumn ibex]

hmm i checked it's off

[waxen saddle]

hmm

[autumn ibex]

is that setting on the Capsule collision?

[ocean wren]

did you derive from detourcrowdcontroller?

[waxen saddle]

yes it is

[autumn ibex]

no, It's derived from AIController

[ocean wren]

Yours definitely look like rigid body collisions

[autumn ibex]

this is what my AI controller look like

[ocean wren]

and not just avoidance

and your collision is huge 🙂

[autumn ibex]

for testing purporses

just to show the effect

[ocean wren]

and RVO is off?

[waxen saddle]

Heres my capsule collision

[autumn ibex]

yes, RVO is off

[waxen saddle]

it really looks like RVO

its off in game too? not turned on with code somewhere?

[ocean wren]

does doesn't it..

[autumn ibex]

lemme check

I dont even know what RVO is.. so I'm sure it wasnt turned on accidentaly

is there a global setting for it in the Project settings?

[waxen saddle]

It makes the things adjust their path to not touch other things on their way to places

[waxen saddle]

hm

[autumn ibex]

ah shit.. looks like someone did enable Avoidance

[waxen saddle]

all good

i could kinda tell right away lol

[autumn ibex]

thanks for your help guys

gonna try disabling that

[waxen saddle]

👍

[autumn ibex]

boom! fixed

god.. thanks for your help

[waxen saddle]

yaay

[ocean wren]

I'd use DetourCrowdController instead of RVO anyway, it deals with blocking a lot better, but needs some tweaking from the defaults

[autumn ibex]

didn't realize it could be enabled in code

[waxen saddle]

yeah the inspector is annoying sometimes that way..

but you could ask them to do that stuff in the class constructor

then it wont mismatch at begin play

[autumn ibex]

ill look into that detourcrowdcontrooler, thanks

but tbh, I kinda prefer having the AI non-solid as I'm used to that in Source games like Left4Dead

[ocean wren]

just be aware the default detour settings Epic did are.. kinda wonky

[autumn ibex]

they have some push-away code when non-solid actors are too close to each other

[waxen saddle]

yeah they were always non-solid with your settings

just actively avoiding eachother

[ocean wren]

yeah, I usually turn off rigid collisions and do a force-penetration seperating force myself

[autumn ibex]

and don't have to run expensive detour code.

[ocean wren]

expensive? I guess it depends on the scale of your crowd

but yeah, simple seperation force works well enough, works in all of valve's games and players live with it.. so yeah

[autumn ibex]

yea, plus I have these prety narrow corridors

where I want tons of AI running through

and I'm not sure any detour algorithm will work well in those spots

[ocean wren]

well, it'll definitely work better than RVO, but narrow corridors are always going to look wonky unless you're happy for agents to pass through each other

[autumn ibex]

yea, I definitely can live with that.

it's not so noticable when they're moving..

[ocean wren]

its one area of focus for crowd agent research

[autumn ibex]

only when they stand still where it looks odd, and that's where I run the seperation code

[ocean wren]

yeah, I do allow my agents to pass through, but have them swap if they're stuck after a while

and use a seperating force to push them apart

just in case the seperating force causes a blockage

[autumn ibex]

ah, good point

[ocean wren]

the old switcheroo is pretty common 🙂

[green vault]

is it okay to have multiple services on 1 node ?

[crimson galleon]

i got multiple on one node 🤪

[green vault]

ok

[mossy nexus]

it's ok, but not necessarily efficient

but you're not breaking any laws by doing it

also never expected to see gooseman on this discord

[ocean wren]

after he died in that plane accident with maverick?

[runic flare]

Is it possible to set up an Observer Aborts with a boolean value change?

I can't figure out how to do this...

seems like I can only use IsSet or IsNotSet rather than the condition itself changing

I see I can observe value changes for numbers or strings

[ocean wren]

doesnt isset use a bool?

[misty wharf]

I'm fairly sure this was possible to do in BT's with the Blackboard decorator at least? 🤔

[autumn ibex]

any way to prevent the navmesh from creating areas on top of certain static meshes. I don't want a nav area on top of my cars 🙂

[ocean wren]

navarea set to navarea_null

navmodifiervolume that is

[autumn ibex]

ah thanks ill look into that

[runic flare]

I assume IsSet is true when the key has a value, false when it's null

but maybe IsSet is false when the key itself is false, even if the key was explicitly set to false

that would be kind of confusing but I'll test it out

[crimson galleon]

I belive is not set is null and is set is true or false, except for bools

[runic flare]

ohhh

interesting

[crimson galleon]

Zoombapup will come down with the axe if i am wrong though 😋

[ocean wren]

Yeah, I think you're right, I'm a bit vague on the details, but it sounds right to me

[runic flare]

Setting up a test now, will report back

[ocean wren]

probably easier just to check the code 🙂

[runic flare]

I started to, but it's a bit of a jungle

[ocean wren]

get used to it 🙂

[runic flare]

Might be this, in which case you'd be correct

[ocean wren]

seems so 🙂

[crimson galleon]

trial and error wins me another 😝

[ocean wren]

You win this one, but I'll be back!! muhahahaha!

[runic flare]

thanks guys

[crimson galleon]

sure thing, good luck

[rustic lark]

My ai is not moving on loaded levels

[cerulean girder]

did anyone ever help you with this, or did you figure it out? becuase im having the exact opposite issue...does path exist is returning true even though point b is in the air...as if the decorator is not taking Z into account at all

[crimson galleon]

yeah, ill get true even if the player is on a separate island of navmesh from the ai, clearly failing to move. I ended up making my own task using the move to function and then base it off of that

[cerulean girder]

yeah this node is completely wonky, im just going to EQS it...its not a prblem when my ai is ranged but if the flying character is above a melee ai...the ai just spins around in circles underneath...atleast I can fix that fairly easy....i wonder how exactly this node is intended to be used cause it clearly doesnt work in my case

if it works at all

[crimson galleon]

I feel like i had it working at one point but yeah a bit tricky

[crimson galleon]

it can be in either place

but a task is a self contained bp graph (at least for those of us that dont cpp)

i have a few tasks which cast the controller to the specific ai class so i can call functions within it

uhm

letting the bt do as much work as possible is what i try to go for, but sometimes i have to do some things in the ai class

probably cleaner to manage these aspects in one place if possible

but behavior trees are not as state based as i thought, they can only run one task, but its more of.... deciding which task to be performing

ive been applying my previous familiarity of fsm concepts and its not too different for the most part though

[crimson galleon]

🤦 just checked my BT, I am using it after all!

[rustic lark]

I have a problem with my ai. If i play my game from a loaded level the ai dont want to move.

[marsh aspen]

So, i wanted to know more about navigation. And in the project settings you can set differents agents, you can also create new nav area and make them only for a type of agent. My prob is, once i've set up the agents in projects settings, i don't find how to tell an actor that it's part of this agent class. The result is, no one move anymore, any tip ?

[misty wharf]

afaik you have to just set the character's settings so that it matches one of the agents you've configured and it'll use the appropriate navmesh

[marsh aspen]

that's the thing, where exactly do you configure the character's settings so it matches the agent you want ?

[misty wharf]

Wasn't it in the character movement comp or the character class itself? I forget but it was definitely in there somewhere

[marsh aspen]

my prob is, i can't "easily" assign each actor to be from a part of a city and another actor to be from another one. I thought the agents were just a little enumeration to set up, but it seems way more complexe

[misty wharf]

There is more exposed in the C++ side

I was also looking at this at one point and you can get the agent names and such but it didn't seem to have any really obvious way to say "use this navmesh"

other than getting certain agent props and then returning them from the actor itself

[cerulean girder]

test it to make sure its giving you real results...for me it thinks theres a path to a enemy ai thats flying 2000 units in the air...sure, theres a path to the point directly below it 2000 units but i wouldnt consider that a "path" for a land based character lol

[crimson galleon]

Ive had jt on for awhile and was testing so i do believe it all works but i can check again. I think when i thought it was not working right was because i did not set it up correctly. But it checks navmesh path (fastest) from self to target

[cerulean girder]

make a target for the ai that uses it and place it high enough where you wouldnt consider that "land reachable" see if your ai thinks there is a path anyways

place it above nav mesh area that can be nav'ed tho

and see it return true

[crimson galleon]

That is how ive been testing it and need to work anyways. I want the ai to know when the player is too high up and unreachable

[cerulean girder]

instead of eqs i just created a condiational check that checks the distance in Z

[crimson galleon]

That is also something i do :)

[cerulean girder]

oh thats probably why it seems like its working for you

cause youre checking the Z distance after

[crimson galleon]

Well ive tested with that btd

Without*

But ill check it all again, it should simply not work if it cant find a path

[cerulean girder]

idk then, i debugged it for like an hour and no changes in the settings made the ai realize that 2000 units above a nav mesh is not reachable

[crimson galleon]

Oh i believe it is finicky

[cerulean girder]

so i made the condiational check and it worked out great

[crimson galleon]

Some level of magic is at work

Yeah but i also dont want the ai to try navigating if the player is simple unpathabke

[cerulean girder]

eqs knew right away tho, but adding an expensive call for something i can do with a distance check seemed like overkill

[crimson galleon]

Navmesh check is fast

Eqs for just a valid path shouldnt be that bad

Anywho, ill get back to you when im on computer

[cerulean girder]

have a nice holiday 🙂

[crimson galleon]

Back atcha

[ocean wren]

Can't believe its almost New Years

Been busy prepping some lectures for early next year and thinking about plans within plans 🙂

Thats if we survive the C-vid situation here in the UK

[crimson galleon]

just checked it, working fine 😉

[crimson galleon]

with it putting more people at home, im sure holidays arent necessarily as jolly as the norm

[ocean wren]

lots of killing off grandma and grandpa, lovely 🙂

[crimson galleon]

already seen more turmoil around me than prior - domestic/relationship problems with people and neighbors for example, lack of working class, more unemployed beggars - its risen drastically now that its holidays

eh yeah that too

had i think 3 or 4 visits from the fire dept here and not to the same place

[ocean wren]

I'm just wondering if we'll be having another lockdown at work.. that'd be pretty unpopular with the students

Likely be early in the new year though, so we'll have to wait and see

going to wear an N95 for a bit though 🙂

[crimson galleon]

lol my campuses just opened up a day before finals were over for the holidays

[ocean wren]

Going to have to pull my finger out and get some papers published this year, which is a tricky task at the best of times

[crimson galleon]

Hell yeah! Modular BTs online, now to hook em up to the data tables 😁

[crimson galleon]

i tend to lean towards a manager class that solves locations, sort of like formations

i havent tried, but i suppose EQS could work, but im pretty sure it wont help for being in air

[ocean wren]

What kind of behaviour do you want? can you find an example on YouTube in another game?

I've worked on flight sim AI before, but I'm guessing you're not talking about that?

You probably want to look into steering behaviours

What you've basically done is add a seperation force 🙂

makes sense collision wise

[crimson galleon]

trying to make my own BT decorator node and i dont get how to make this for my own (in bp)

[ocean wren]

You don't, its in the base class?

[crimson galleon]

after scouring, i guess its cpp locked from me 😦

[ocean wren]

Well, it'll just be there by default

[crimson galleon]

here's my custom cooldown decorator

lol shit, hang on, bad example

[ocean wren]

I saw a presentation where they suggested putting stuff like cooldown times in a BB key also, can't remember the reasoning though 🙂

[crimson galleon]

here's my custom blackboard decorator

here's the ue one

i just made tick check, to see if that works the same way

[ocean wren]

Oh I see, you mean the Notify observer is missing?

[crimson galleon]

yes

i think its centered around the events that i can override.... since thats the only thing i can really alter

[ocean wren]

I'd have thought it would be available by default in a decorator... maybe you derived from the wrong class?

[crimson galleon]

pretty sure thats the one

nothing below that

[ocean wren]

Looks right

[crimson galleon]

ive been making tasks and services no problem

this is more elusive

[ocean wren]

You'd have to lookit the code 🙂

[crimson galleon]

well, my cooldown works fine

yeah...

i guess ill check the git

[ocean wren]

I might have to break open UE5 today and make something

Need to record a new years video 🙂

[crimson galleon]

writing a better nanite renderer eh?

[ocean wren]

Kind of want to mess around with some control rig proc anim stuff

[crimson galleon]

id be down to check that video out

[ocean wren]

Got to make an annouce video for new year for new consultancy business and maybe do some discussion on the multiverse, because its on my mind

[crimson galleon]

looking like it is tick

this shit is so fragmented

[ocean wren]

oh right, so override tick and it shows up?

[crimson galleon]

oh im still reading the code....

i doubt im going to get that to show up

i made the same entries as in the original

dont think there's a way to expose any of it

[ocean wren]

yeah, honestly it needs a proper re-write, but that's never going to happen 🙂

[crimson galleon]

seeing that they're doing something else right?

kinda feel they stopped on the perception system just like pawn senseing

[ocean wren]

yeah, I saw the massAI thing that Mikko Mononen was working on

I don't know if they're redoing the tooling for behaviour though

seems more like navigation stuff

[crimson galleon]

i saw some of Mikkos blogs, its over my head, but they sound like more that what is in engine

lots of navmesh stuff indeed

[ocean wren]

Well, Mikko did the open source navmesh generation library that's in both UE and Unity, so he's got form in terms of navigation

he did detour system too

And to be fair, the navigation stuff could certainly use a boost to functionality

but I see way too many problems with people struggling to make basic behaviours, feels like having some work on behaviour specification is needed

[crimson galleon]

gotcha

the man is the maker of navmesh it sounds like

owning the concept at least

makes sense why eric referenced his solutions for death stranding

i actually have the same kind of functionality of my BT all in BP but i wanted to learn BT so i recreated it and aside from some friction, its a bit better than my BP version

[ocean wren]

Well, navmesh as a concept has been around for ages, but Mikko did the recast library while he was at CryTek and open sourced it.. everyone then used it 🙂 because he was kind enough to use a permissive license

[crimson galleon]

i was thinking the debug view for BT would be helpful... but it really isnt

[crimson galleon]

yeah, thats all i mean, i know he didn't actually come up with it, but he's leading the charge

[ocean wren]

You can always write a better one

I've extended the visual logger a few times, its kind of fun to do

[crimson galleon]

probably not with 0 cpp skill

[ocean wren]

ah, yes, that would be a problem 🙂

[crimson galleon]

i could go back to BP and continue that route, but still, this is a minor snag

ive already considered workarounds, but if i could, id like to have my own blackboard decorator for condition checking

need to remake most of the ue nodes so i can drive them from data table lookups

[simple crest]

the worst time to start learning c++ is never

[ocean wren]

correction, its ALWAYS the worst time 🙂

so might as well get on with it 🙂

[crimson galleon]

And deadlines

But yes one day

Im not trying to be or get work as a programmer though

[ocean wren]

Best not, its for the birds 🙂

[crimson galleon]

riiiight

they said that about art

[ocean wren]

art is definitely for the birds 🙂

[simple crest]

the only truly valuable job position to be skillful at is ideas guy

[crimson galleon]

right right.... i dont know if i can come up with half-baked ideas that i cant see them through

looks like my custom dec is firing on tick, but its not updating when it goes false...

[ocean wren]

ideas guy 🙂 hahahah yeah, I've seen plenty of those. No real skill but want to be in charge because they are the "ideas guy"

[crimson galleon]

your quick 😜

[simple crest]

[ocean wren]

Its pretty much every "job" post on most of the forums I've been on over the years

wanted: team of 15 to make my game. I don't have any skills, but I'll be the owner and the ideas guy.

payment: zero 🙂

[crimson galleon]

will be in credits

[simple crest]

[ocean wren]

yeah, "we'll add your name to the credits" is a good one

[simple crest]

exposure

[crimson galleon]

they wont

thats someone elses job anyways

[simple crest]

i've de-AI'd #gameplay-ai enough for one night haven't i

[ocean wren]

worth quitting a job eh?

[crimson galleon]

no thank you

[ocean wren]

please dont

[simple crest]

oh fuck yes gimme that hot free nitro

[ocean wren]

it says dicsord 🙂

[crimson galleon]

i mean, there's a target on them already...

[ocean wren]

that's a new one

[crimson galleon]

.ru hahah nice one

[simple crest]

<@&213101288538374145> you aren't all online/green yet so maybe they're going slow with this one

[ocean wren]

totally legit

[crimson galleon]

speaking of derailing AI

[simple crest]

dicksword.ru

[ocean wren]

AI is for the birds

[fair zealot]

:no_entry_sign: MC Reaper#2506 was banned.

[ocean wren]

I mean literally, I've worked on bird AI before 🙂

[crimson galleon]

thanks Roy! your awesome!

[night hazel]

yeah, that was the first ping about it

[ocean wren]

Thank you kindly, sir!

[crimson galleon]

oh does everyone not ping mods?

[night hazel]

you cant ping at-everyone

but at-moderators works fine

[simple crest]

sometimes they only post in one channel every minute or so... you know, for extra authenticity

[night hazel]

just, immediately ping moderators if you see that

[crimson galleon]

the offender pings at-everyone, so i figured no need

[night hazel]

only moderators can do at-everyone

that ping did not go out

[crimson galleon]

oh shit, ok

[ocean wren]

Man, that's a job i don't miss. Having to moderate the old indiegamer forums.. ugh, it was like 99.999% spam

[crimson galleon]

lol, i see, thats a use

user*

[night hazel]

if someone who cant ping the role tries to, it'll hilight like it did ping, but it wont ping

[crimson galleon]

gotcha gotcha

[simple crest]

we could really use some #gameplay-ai to autoblock the bots

[night hazel]

saves all y'all from getting notifications all the time 🙂

[crimson galleon]

thanks for the info Roy

[ocean wren]

bots are actually quite easy to make for discord

[crimson galleon]

explains all the scammers

[night hazel]

the scammers are compromised accounts

[crimson galleon]

oh that sucks

[ocean wren]

I did a livestream for my students and wrote some python code to do a discord bot based sentiment analysis

[night hazel]

the reason we use manny is because it gives them a way to appeal the ban. we get a lot of "my account was compromised, i didnt mean to scam" appeals

[ocean wren]

then fed that into unreal engine in the last stream

[crimson galleon]

mhm

[ocean wren]

next stream, we are making a crowd battle sim

[crimson galleon]

TABS?

[ocean wren]

like that yeah

using fun particle system based AI

[crimson galleon]

oh, wasnt someone asking about that?

[ocean wren]

you can use a skeletal mesh component and anim blueprint in particles now, since 4.26

[crimson galleon]

hey, thats the version im on 🤔

ive had the idea of doing some kind of swarm like enemy

[ocean wren]

its actually kind of easy now

niagara is kind of weird, its massively powerful, but really complex and of course the documentation sucks and the usability is a bit.. weird 🙂

[crimson galleon]

yep, havent dove in after the last time

[ocean wren]

I keep giving it a go, but sometimes it just doesn't do what the hell I expect it to

[crimson galleon]

lol

[main heart]

What are the freaking laws of AI?

[ocean wren]

Law #1 NEVER ask what the laws are. Law #2 See Law #1

[ocean wren]

25% cat, 27% dog

[ocean wren]

thimble the hamster?

[crimson galleon]

Theyre still asimovian iirc

[cyan sigil]

Merry christmas. Has anyone been able to implement GOAP in unreal? How?

[simple crest]

Tried. Got annoyed by it. Built a utility ai instead.

How? 3-12 months of effort I guess? 😄

[misty gale]

I havnt dove into utility ai yet but o the surface it reminds me of alot of the same stuff goap was doing

[midnight scroll]

I'm wondering if anyone has any insight on AI behavior organization. Behavior trees initially seem nice until you need to start doing multiple things at once. For example, a simple bot that runs and shoots at the same time while staying faced to their target. This seems to require a few parallels, and a lot of copy pasted nodes to slightly alter behavior for conditions. I find myself wanting to convert a lot of this over to tick based actor or component behavior to simplify the logic pattern.

Should I maybe consider writing much larger task nodes that encompass this more complex behavior? I just find the large huge sprawling graphs, or having to make several large smaller trees to subdivide logic in very disruptive and ugly to follow.

[crimson galleon]

i have that sort of thing, there's a look at and jump check that I have in BP, the BT can talk to them (start/stop) but they dont exist in the tree. Run Dynamic might also work to manage calling the same kind of ~~task ~~ tree under different branches in your main BT.

[celest python]

I usually code those things in Character and just signal "start" "stop" states based on the behavior. I also like to sequence actions in a single BTTask. My personal attitude against huge Behavior Trees is negative
In any case: <#gameplay-ai message>

I rarely do parallels, not because I don't like them, just most of the times my BT usage is not above "moderate" complexity and whenever things get complex I try to decouple it from BT

Lately I was thinking about a project specific FProperty (for bools) listing on decorators so I could set them to true/false in the editor without writing a custom decorator for each boolean in my character

So I don't have to signal bWantsToShoot bWantsToRun manually by writing etc.

[midnight scroll]

My first and real use of BTs was when I was messing around creating "Customers" for a shop style game. This actually worked out incredibly well because these pretty much did one thing at a time. Spawn, find store, run to store, find shelf, navigate to shelf, decide to buy or not, repeat last couple steps with a chance to decide to leave the store. It was very linear and behavior trees made that incredibly easy. But I'm really finding that complex behavior trees where the AI needs to handle multiple different states all at the same time is just.. stupid convoluted. And I'm only at aiming, moving and shooting. Already considering how to work a lot of other states in that I need to consider later is making BTs look like a terrible idea.

[celest python]

I wonder how Epic did their shooting bot AI on UT -- I was never be able to compile it since it requires vs15 😄

[celest python]

If you are going to design the BT with states, maybe you can see Alien Isolation's AI. Their AI ultimately relies on states and they have a great example of it

[crimson galleon]

i do suppose thats what a parallel is for, but its limited to a task. I had jump checking as a BT branch, but it would interrupt other behaviors and i needed conditions and services to determine the series of tasks. I offloaded that from the tree, only because i always need it running, so it made sense to just do it in BP since it interrupts other states. Since it fires regularly, i check if the AI is moving (not attacking and some other stuff) to allow this check to perform. I have both melee and ranged attack logic, and tried to keep that all together, but found it easier to just have a look at task call the event in my BP, the animation notify calls to stop the look at.

so i do have this stuff spread around, but i feel thats not unusual when it comes to communication

overall, there's less masses of bp graphs, and more concise little things

[celest python]

Alien Isolation BT source code (requires Alien Isolation installed): https://github.com/MattFiler/OpenCAGE
How To Use BTs properly GDC talk (this one explains the rules against Alien Isolation's methods, but this one is more correct): https://youtu.be/Qq_xX1JCreI
Demonstration of Deadalic's RTS AI: https://www.youtube.com/watch?v=1Dm1G6fUuFs&t=884s

[simple crest]

Epic's ai module has some resource class somewhere but I don't know what it's for

When I built my thing I set it up so an ai has resources (head, arms, legs) and each action requires one or more resources so it becomes possible to "move" and "reload" and "look at" independently

I never figured out how to do anything quite like that with ue bt's

[cyan sigil]

From what i understand, GOAP is like Astar but it navigates differentq states instead of nav points. Utility is where you give all possible actions a score and pick the one with a higher score?

[simple crest]

Yeah that's the most basic summary

[misty wharf]

The resource system afaik just allows you to lock/unlock some resources which does things

As far as I can tell the two builtins are logic and movement, which lock/unlock BT logic and movement respectively

[main heart]

You sir, are a gem!

I need to find eight more

Or five more, depending on the context

[crimson galleon]

Lol it was meant as a joke. But believe there are eight facets to AI

[main heart]

Names?

[crimson galleon]

Attributes?

[main heart]

OF the facets, I mean

Sure

[crimson galleon]

I dont recall them but i remember reading that there are 8

[main heart]

ah, my manners need polish

Nevermind!

[shadow sierra]

Anyone know if its possible to factor in a nav agent's radius when projecting a point to a navmesh?

I'm trying to spawn characters onto a valid nav area and using the appropriate query to project random locations onto the nav mesh, but as you can see the character is spawned right over the edge of the nav area

[mossy nexus]

Just to throw my hat in the ring: I generally try to think of how real life behaviors would be and model trees based on that. I think starting from a point of a FSM is a fallacy (as pointed out by others, repeatedly, in this channel). If you're finding yourself doing multiple complex behaviors then perhaps it's worth rethinking if those behaviors truly are relevant at exactly the same time. I use very few simple parallels, and usually only in edge cases

I find that many things we think of being parallel in life often happen in sequence

[celest python]

For AI states its alright but I have those vars in my character class

Some of them are replicated

[ocean wren]

states? ptooey!

[crimson galleon]

its so clean 😁

[bold dove]

Hello guys! I'm working on a fnaf game. I made a ai in blueprints, but i can't solve how i can make the jumpscare because my ai is taller than the player, and i want to made it, if my player get jumpscared, he move up a bit. Anyone can help me?

[celest python]

I think this is a game design question rather than UE's AI framework or AI behaviors

I didnt play fnaf but this could be solved by moving camera to upside or via animations?

[bold dove]

Hmmm. Maybe moving the camera is the solution. Thank you!

[bronze yacht]

The most dreaded part of my project has come, an AI

[simple crest]

ez just code an ai to write your ai for you

[bronze yacht]

This is actually the one thing I've never done before

[bronze yacht]

genius!

I was gonna watch Ryan Layley's AI overview, anyone have a resource they really liked?

[crimson galleon]

you plan on using BT?

[bronze yacht]

Yes

[crimson galleon]

https://www.youtube.com/watch?v=iY1jnFvHgbE

i wouldnt expect from any one video to achieve your entire AI, that one helped me understand each area

[bronze yacht]

Thankfully, it's a simple AI

Remember Slender: 8 Pages?

[crimson galleon]

i know of the game

i know that there are many variants

[bronze yacht]

It's a simple AI with simple rules. Teleports when you're not looking, gets closer with each page or as a function of time

[crimson galleon]

i know not of the actual AI, i always thought it would simply move to player constantly, didnt know it teleports

and i thought it only got faster after each page

but hey, sounds like you know it

[bronze yacht]

It's probably more complex than that, but simple goals first

[crimson galleon]

true true

[bronze yacht]

I'd like to make it more complex, but considering I've never made an AI before

[crimson galleon]

well, complex does not necessarily mean its better

test a simple AI, see how it feels

youll get a sense for what it may need, or ideas of how to make it interesting

its all problem solving as usual though

[quasi pecan]

I am having a weird issue and I can't pinpoint what is causing it, I have a companion AI setup that you can switch to and it's working all fine but sometimes if I switch to the companion, the other character's AI stops working. If I use the debugger then to see, the BehaviorTree shows paused and I don't understand what's causing that because I have called the RestartLogic from brain component to restart it.

[ocean wren]

Are they both using the same BT by any chance? if so, I wonder if its that you need to run the BT as instanced in some way?

Usually it tries to share BT's amongst AI, so I guess if you pause one, maybe it pauses them all.

Just speculation though

[celest python]

Dont watch Ryan laley

Conrad's link is awesome

[quasi pecan]

Different BTs and Different AI controllers as well but that is an interesting point.

[quasi pecan]

It confuses me further because it works 1 time and refuses to after that first time. Meaning, when I try it once I am able to switch between the characters just fine and the other character does it's AI thing, but as soon as I swap a second time, the one I swap from, stop running it's BT.

I am able to switch between the two just fine but I can't have their AI's running.

[ocean wren]

I guess you'll have to break out the debugger and take a look at the code

[quasi pecan]

Yeah it seems like that.

[quasi pecan]

I have managed to solve it, so apparently, the pin on the MoveTo tasks that says, "Lock All Logic" if that is ticked to true and you swap in between, what happens is that tasks gets to never finish, even if you possess that pawn with the same controller that started it and as a result the AI freezes, forever @_@

[ocean wren]

Interesting, thanks for the info

[celest python]

Anyone tried to run BT without an AIController?

After several days of brainstorming I decided BT is a good choice for some elements of my game director but using AIController makes the code more messier than ever

[ocean wren]

Why does it? its just an actor

[celest python]

Game Director needs to spawn AIController for each player, just accessing it and storing references (and maybe communicating with it) makes the game director subsystem sphagetti

But not much of an issue, just wondered if anyone ever tried it

[ocean wren]

?

Well, yes, AIController for each AI... and the problem with that is?

[celest python]

Not each AI, I want to use BT to help Game Director prioritize player's behaviors (for example if player is not fighting during a combat BT will notify game director and game director will focus to this issue). For they will communicate with a UActorComponent that added to PlayerCharacter

I need to spawn an AIController for each player character in the game

[worthy python]

hello. I created my own EQS generator but i can't debug found points with EQS Testing pawn. What i need to do to make it possible?

[ocean wren]

Why can't you use the pawn?

[mossy nexus]

I use it without an AI controller. But that will only work if you're not using it for AI. I use my secondary BT for camera movement

Even for my AI directed pawn I use a BT component rather than through the AI controller

but, an empty default AI controller is necessary for AI BTs to work

[celest python]

Hmm, alright. Thanks for letting me know

[crimson galleon]

You should see the eqs when assigned to the eqs pawn so long as the context is querier. If you have a custom query context, such as for the player, it will not show. I think there is a way to do so but i just change the query context to preview then change it back.

Morning AI ppl

[misty wharf]

I think custom contexts work just the same, but obviously Get Player Pawn will return nothing if you're not in play mode

so you could add some logic to accomodate use of the testing pawn and have the context return something else if player is null

[ocean wren]

Yeah, I just made something derived from EQS Test pawn and added a "target" to it, that I use as the target for the query etc..

with contexts and whatnot, basically the same way I do for AI characters

[cedar quartz]

does taking a sha256 hash of a number and then the hash of the result over and over accomplish the same thing as using an rng with a seed multiple times?

[mossy nexus]

an excellent AI question I believe

!8ball does taking a sha256 hash of a number and then the hash of the result over and over accomplish the same thing as using an rng with a seed multiple times?

[fair zealot]

@mossy nexus, :8ball: My reply is no.

[mossy nexus]

well then

[celest python]

Cranz lmao 😂

[celest python]

Why not #cpp instead? 😅

[cedar quartz]

roger

[mossy nexus]

👉😎👉

[celest python]

Did you handle blackboards as how AIController handles it?

[mossy nexus]

blackboards can be handled directly through BT assets

I haven't actually taken a look at what AI controller does with BBs as I at the time just wanted to separate BT from AI

[celest python]

Hmm, so you just created the BehaviorTree component in the actor and did not implement anything related BBs on AIController?

[mossy nexus]

I have a BT component that handles starting a BT outside of the AI controller. BB is routed in through the BT asset pointer

And... No, my AI controller currently is blank

I'm debating whether or not to move logic into it atm

[celest python]

Alrighty, thanks

[mossy nexus]

that being said I use no AI controller features even in the BT. If it wasn't because pawns needed an AI controller to function I'd not need them whatsoever

[celest python]

I'm thinking to get rid of AIController for some of my pawns too

I realized removing BTs completely and some of the AIController logic could provide ~%15 performance increase

[mossy nexus]

that's interesting

[celest python]

Thats for my case though, you know my game mechanics

It's mostly because of zombie horde

[mossy nexus]

sure, I was just thinking how removing the AI controller would affect other things. you'd need custom MoveTos

[celest python]

You can use built-in AITask for MoveTo

UAITask_MoveTo

What MoveTo task node is doing just creating it and running it if you enabled bUseGameplayTasks on project settings

[mossy nexus]

doesn't that require a controller?

[celest python]

iirc no

It requires pawn

[mossy nexus]

huh

[celest python]

It might trying to access controller inside through by using pawn reference

Its been a while since I analyzed it

[mossy nexus]

I'll have to look into it

I tried using MoveTo previously without a controller and it hard crashed every time. Took me a while to figure out what even was wrong

[celest python]

disregard - it asks for AAIController

[mossy nexus]

well then

[celest python]

It needs pathfollowingcomponent which stored in ai controller

[mossy nexus]

that's the one

[celest python]

If PathFollowingComponent does not require any AI Controller; what moving the AI pawn is pathfollowingcomponent anyway, you can easily refactor UAITask_MoveTo it to pawn instead of AIController refernce and get your own pathfollowing component inside the initialization function

AITask_MoveTo just subscribes to a delegate in pathfollowingcomp and waits for the signal

then ends the task and notifies the required places

[mossy nexus]

sure but it does seem like the binding between controller and pawn was created for a reason

which is why I pivoted on the idea of not using the AI controller

[celest python]

PathFollowingComponent seems to be not asking for AIController specifically at the first glance

[mossy nexus]

but I am interested in how this could potentially work without a controller

and if it indeed is cheaper

whole idea behind a pawn is to have a controller though, so if there is no controller any actor could be a pawn

[celest python]

From what I can see PathFollowingComponent asks for AControllers on virtual functions that you can override and implement your own logic

[celest python]

I don't think it would make any difference in your case

For me its usually BT tick makes the overhead

~100 pawns running and BT keeps ticking just to check if signal received from pathfollowingcomp

~70 of them are just blind dead zombies that has no actual AI implemented other than running to target goal

Removing BT on them would mean getting rid of 70 additional ticks and some for loops to search data on BT for each of ticks

[mossy nexus]

yeah I'd love to consider BT tick to not happen as frequently

or get activated only through delegates

and it doesn't affect any of the navigation of your zombies?

[celest python]

What do you mean? Removing BTs?

[mossy nexus]

removing the AI controller

BTs were always optional

[celest python]

I'm not considering to remove the AIController completely yet, I'll just delete the some of the custom logic I implemented to AIController

[mossy nexus]

ah, fair enough

[celest python]

But if I would remove AIController too I dont think it would affect anyway

[celest python]

Based on this assumption

Looks like Epic gave a way to alter pathfollowingcomp to your own system

[mossy nexus]

makes sense

I'll have a look around the next couple of days. Although it is a bit premature optimization on my part

I'm yet to implement factions properly

[celest python]

Although it is a bit premature optimization on my part
I'd go brrr if I was on your situation 😄

[green vault]

why does it go into this node ?

[celest python]

My next game is never going to have big amount of pawns/AIs 😦

[celest python]

Interesting

[mossy nexus]

I was never going to aim for more than 20-50 really. I like balancing towards AIs you have to think laterally about

also my AIs are mostly what you would consider "bosses" in other games with satellite AIs acting as support

[celest python]

I thought making a horde shooter would be easier in sense of indie-spirit but that was the biggest mistake I guess 😄 I'll never reach more than 30 for my next games. I also want to try other types of game designs since I earned some experience with this current one

[mossy nexus]

my biggest issue has been trying to reconcile the fact that one AI could be spread out over a vast area, and how to handle this

[celest python]

I really couldnt point out to anything by looking at to screenshot, have you checked Visual Logger?

[green vault]

what a visual logger 😄 ?

[mossy nexus]

most likely BB condition returning true

[celest python]

It's a logging tool made by Epic, you can find it on View > Developer Tools

Also you can just check CurrentTarget on decorator instead of bHasTarget

It can understand if its nullptr or not

[green vault]

i cant find a view tab ?

[mossy nexus]

I have had issues with the default BB based decorators though so I understand the issue

[celest python]

Isnt EQS literally made for this 🤔

[green vault]

ok thats maybe handy

[celest python]

On the left *up side of the editor

I dont have UE open at the moment

Fighting with bugs in Rider 😦

[mossy nexus]

I don't know, I use my own custom perception system. Eyes and ears are now components ;)

[celest python]

Ah, nice. Though EQS does not care about perception directly, if you havent yet, you might wanna take a look

[mossy nexus]

I did take a look before I decided making it try to do what I wanted to do would be too much work

so I opted for 600 lines of combined perception component/AI director code to solve it instead

[celest python]

Haha like a boss 😎

[green vault]

ok i found it lol

[celest python]

Sorry both chatting and coding right now, I dont know how did I wrote "left right side" instead of "left up" 😂

https://tenor.com/view/head-bang-pissed-stressed-upset-ugh-gif-16972393

[green vault]

np dude

[crimson galleon]

because no matter what, TargetInAttackRange is Set. I don't know what your BTS does exactly, i assume it checks distance, but your decorators are only checking if it is set with a value and not checking if it s greater than or w/e

[mossy nexus]

greater than only works with arithmetic anyway

[crimson galleon]

its my guess on the service without seeing the logic

if its a boolean, then true and false are still considered Set, and instead it should be cleared

[mossy nexus]

usually the solution is a custom decorator. I would expect that to work here as well 😎

[green vault]

the service looks like this

[crimson galleon]

clear the key instead

[green vault]

i thought is set / is not set is exactly what i think it is

[crimson galleon]

maybe bool is different, but is not set means you've cleared it

[mossy nexus]

ah yes, the good old ambiguous wording of the BB condition decorator. I suggest either looking at the code for it or making your own

is set/ is not set has no meaning for bool afaik

[crimson galleon]

i think you are right, and is not what i assumed that it was distance based

or float, but indeed a bool

[green vault]

didnt change anything still running into that node

if i did it right ?

oh wait im stupid lmao

forget to pin the condition

ok still not working 😄

[crimson galleon]

thats a bummer

aside from the advice of writing your own, i would mess with some of the decorator settings

[celest python]

what visual logger says though

[crimson galleon]

maybe also print the distance result from the service

[green vault]

@celest python it looks like this dunno if that helps

[celest python]

when you click blue things

it will provide a text below

that are logs

each blue one is timestamp with logs

[green vault]

@celest python this is the last one before the other pattern

[green vault]

i dont know why but the service doesnt get executed anymore?

maybe its stuck in the task ?

[crimson galleon]

seems weird, service should run regardless

[green vault]

oh no

i think i found the mistake

[crimson galleon]

i think you need to be using RecieveTick so it runs

lol yeah

[green vault]

im assuming its not a tick

[crimson galleon]

but you can do both

looking at mine, i run ReceiveSearchStart and ReceiveTick

search start is for the casting and variable setting so tick can reference them

[green vault]

if i use tick it does fire but its delayed or some shit so the condition is true but it starts the task like random some run further some stop

[crimson galleon]

?

[green vault]

i mean it doesnt get stuck on the node anymore

when using the tick

[celest python]

try clearing the key on ondeactivated node

[green vault]

but some run close even tho they should stop when the distance is below 400

[crimson galleon]

they all look stopped in a picture

[green vault]

the 2 in the middle should be that close

all should be on the edge

[crimson galleon]

shouldnt*?

ah

[green vault]

oh yeah sorry

shouldnt

[crimson galleon]

the service runs every .2 is that what you mean?

[green vault]

nah but have to look again maybe its some logic issue they think the target is not in range that why they go in to the move node

im assuming you have some experience would you make a state system for the thing i try to do like an enum with walking/attacking/idle and use that as a decorator?

[ashen mica]

Why won't my "Set Blackboard Value as Bool" work? I've ran into this problem before but now I can't figure it out.

[crimson galleon]

seeing that the ai gets its target, sounds possible

[crimson galleon]

uh.... no i dont think so

[crimson galleon]

the structure of BTs (though im sure to be roasted once again) is you can treat each branch as a state decision - essentially a hierarchy of the most important to the least important state to be in, though in some cases those branches become a bit more complex

[green vault]

but is it right that the one on the right is always the "fallback"

[crimson galleon]

the highest priority is on the left, correct

the tree you have setup looks right, when its the custom stuff you have though, there might be some problem, or the data being fed into it

[green vault]

when an actor gets destroyed its auto null right so i dont have to check if its alive with a service or ?

[crimson galleon]

i suppose, though checking if it is valid is always good just in case

[green vault]

is there a way of running the bhaviour tree slower ?

[crimson galleon]

increase the check rate

its at .2 right now

oh, my bad, the tree

[green vault]

yeah

[crimson galleon]

dont believe so

but you could put a cooldown on the branch you specifically want to iterate less on

[green vault]

i think its working now damn

satisfying feeling

[crimson galleon]

🤞

[green vault]

but it feels like the outcome who wins is random 😂

[thorny ether]

https://tenor.com/view/happy-pretty-please-wagging-tail-dog-cute-gif-15323902

[stark oyster]

is AI usually derived from the Pawn class, or Character class?

[crimson galleon]

what do you mean by AI?

AIController comes from Controller

[stark oyster]

I mean a character that can move and shoot but is operated by the engine it's self, not an actual player.

[crimson galleon]

AIController is what you want, but you need to make your logic for those behaviors

you also need a pawn or character for this AIController to posses

[stark oyster]

Ok thank you. That helps out a lot.
Also, Don't know if your knowledgeable about blueprints but I had a pretty interesting question on Structs and AI characters.

#blueprint message

[crimson galleon]

i dont understand that question

[stark oyster]

Don't understand as in I didn't explain it well, or Don't understand as in you don't have the knowledge?

[crimson galleon]

i dont understand what you are asking about

the former

[stark oyster]

I have recently Dipped my feet into Structures in UE4, soon after starting, I created a struct that could hold character data.

I know full well that if I'm goin to have multiple of the same character. (multiple creepers / multiple storm troopers / whatever) that I'm going to need to instantiate something for each character.

I'm just starting to realize that this struct which is supposed to be used a database for all character data, will need to be instantiated since it holds an original copy of a characters data.

now I'm not sure this is a good way to clump together variables for a specific character. In other words, I don't know if structs should be used to hold all character data, (since I don't even have logic setup yet)

Because of this, Would an instance of class not only be better but a replacement for this struct idea?

(I feel like my implementation of structs is starting to go outside the scope of structs capabilities)

[crimson galleon]

class and struct are different things, that doesn't make sense to solve the issue. I would not recommend using structs for what sounds like a large collection of data and instead check out Data Assets.

Data Assets would allow you to work with them in order to customize settings. You could also use Data Tables instead of Data Assets, which will need a struct to define its data columns. But Data Tables cannot be manipulated, only referenced - in general it will get messy if you want to try setting values in struct - so look into Data Assets or Data Tables to see which align with your needs.

[stark oyster]

You say that "Data Assets would allow you to work with them in order to customize settings"

By this you mean that data assets could be used to update a specific characters variable in real time like health? Or is that sort of operation not supposed to be used that frequently with data assets?

[crimson galleon]

manipulating variables throughout gameplay would be better as their own variable reference and not through manipulation of the data asset or struct

you could have health, which defines the max health, but when taking damage, that should be a different variable, like CurrentHealth, so you can do checks/changes with it without fetching the data set

[stark oyster]

and this variable (CurrentHealth) would be local to the exact actor blueprint that represented the character?

therefore instead of fetching the entire dataset for a stormtrooper's health (lets say), instead it would be fetching the variable of the character whose original reference was from the data asset?

Hopefully that makes sense and I understand this right

[crimson galleon]

depending where you want to store it, it would live in one actor (pawn, character, or controller)

fetching the data is not that bad, it's assuming that you should manipulate the parameters within the data asset during gameplay - there are ways, but its not very pleasant

local variable is also a term for a variable that lives in a specific function that cannot be referenced outside of that function - even in the same actor

[stark oyster]

Ok, I'm definitely starting to understand it now.
And that's may bad with the misinterpretation of the local variable, I had no idea it was that specific

I think that's it though. (going to start researching all of this more to get a better grasp)

I can't even articulate how much this info helps.
Thanks for the help man, Means a tremendous amount.

With help this good, why have google! :)

[crimson galleon]

yeah, sure thing, its stuff you learn over time, now comes wrapping your head around it for your needs, good luck and good night

[stark oyster]

You have a good night as well,
And awkwardly late Merry Christmas! lol

[mossy nexus]

yes but it would require a perception system

[keen crow]

is it an ok situation if a blackboard key which I set on a task node has selected key name correct but the selected key id incorrect (255) which results in .IsNone() check return false even though the value (at least by name) is set?

[celest python]

Anyone knows what "execution forbidden" means?

When I spawn my enemies runtime BT does not execute and give this error

Doesnt happen when placed to level

[misty wharf]

if the can.... is a decorator then it probably means the decorator blocked execution

[celest python]

Interesting, all decorators should return true since values are either true or not nullptr

[shadow vortex]

so im trying to get Ai to form a line, or a queue system, where line up and take turns getting to a point. how would i go about something like this?

[crimson galleon]

array, offset location behind each

just a thought

[shadow vortex]

im not sure how i would use an array in that fashion

[crimson galleon]

alright then dont 😛

[shadow vortex]

so like, an array of blank actors as the places to wait or actual vectors? cause right now i have only the front point in line and they just gather around it.

the front point is a blank actor

[crimson galleon]

not what i was thinking. more that each actor gets behind the other

[shadow vortex]

me either, thats just what i got so far lol. so how would the array factor in?

[crimson galleon]

the array would be of the AI you want in the line 😛

index 0 is front of the line

then just reference each, generate an offset location behind it for the next one to move to and so one

could have them look at the AI if you want to (when they finish move)

[shadow vortex]

mhm i see, and how would the ai know when to move up the line?

and to the next point

[mossy nexus]

AI director is a neat way to solve issues like these

[celest python]

I wanted to do same thing a few years ago, most difficult thing was handling the situation when there is no space in the nav mesh to draw a line

-> prevent AI from moving when AI needs to go there with a line formation
-> send some of the units only and let the others try a random point in the nav mesh

I just cancelled the project instead xD

[misty wharf]

I probably need to do something about my queue system for that at some point

Right now it just goes in a straight line so it would put the npc's outside the building if it got long enough lol

I'll probably just try to find a way for it to keep adding queuers to the left and right side of the last valid position in a straight line

or just have them stand nearby I guess because I can imagine that being a pain to implement

[celest python]

If the queue is a static location in the world, its fine

Querying the dynamic locations is hell

[misty wharf]

Yeah it's 100% dynamic player selectable and there can be other player-placed objects in the way too :P

[celest python]

Maybe a grid system could work in your case though

You were developing a supermarket game iirc?

[mossy nexus]

a linked list type queue could work

[misty wharf]

video rental store but close enough lol

[mossy nexus]

even without an AI director

[shadow vortex]

What I am doing essentially is creating "lemonade stand" I want to form a line whencustomers make the choice to get Into line and after 1 st in queue leaves the line moves forward. That's what I am looking for

[shadow vortex]

Where would I look into that

[crimson galleon]

sounds like the above convo talked about it

[shadow vortex]

That's the reply to my convo

[crimson galleon]

i think one good point was about the situtation where the line may be long and needing to account for it

the idea is to place actors in a line, generically speaking, which there might videos on, or possible something similar like formations. You'll probably find these tutorials use a separate actor to handle the formation logic, i.e. an AI director/manager

[ocean wren]

I did door stacking for my AI squads, by creating a component I could add to things where I wanted them to stack. If I issued the squad a command to stack on a door or use a ladder, the AI director would look at where I'm pointing (this is an FPS) and would check if there was a component that could be used for stacking/waiting etc. So the AI director would pass that component + object to the squad members. the component handled the choosing of squad members to stack and where to stack (component had a list of positions which had gizmo's you could move around the level), so the logic of stacking/climbing and sequencing was in the actor being used.

So basically, create yourself a queue component (or maybe something a bit more general) and have it contain a list of queue positions. Have methods on the component to request being added to the queue, which would return a position to move to. The component can then shuffle things along the queue etc.

You can make properties have custom gizmo's and stuff, which allow you to add locations when you select the actor the component is on

stuff like that

Just had an interesting thread from Borut Pffiefer about his work on Wierd West on twitter.. nice thread about using tag-based utility style decisions based on world tagging

so the TLDR was that he's tagging pretty much everything, locations, objects, events etc. But the objects can have multiple tags. Then at decision time, he's using a utility style system to filter and evaluate the tags the system has collected

so for example, standing at a bar, in a saloon, in a busy town, might have "TownBusy","Saloon","NearBar" tags for location and then tags for time of day, if a barkeeper is there etc.

He then filters things based on tag priorities etc. Filtering and scoring utility style (bit like EQS if you like)

That idea of having everything use that same tag interface is pretty nice. Might have to play with it.

I guess you'd need some UI to be able to debug what tags are being evaluated per actor

and probably want to be able to visualize the tags on each object somehow

[worthy python]

hello. I added leaning/peeking around cornees (procedural spine bones rotation). And It broken AI aiming. How to fix rotation of bones back to AI control rotation? I'm using ALS

any tutorial or example of leaning for AI will be very helpful

[mossy nexus]

don't think this is an AI problem. sounds more like an animation/skeletal issue

that being said AI aiming should probably be independent of any kind of aim offset

also AI control rotation isn't a thing. control rotation implies camera. AIs don't have camera

[worthy python]

it's breaking my mind

[mossy nexus]

yeah but it isn't an AI issue

[worthy python]

it's base function inside AIController

[mossy nexus]

yep but ask yourself what it does

[worthy python]

as can i see it's updating ai control rotation for a focus

[mossy nexus]

more than likely it exists virtually only

in either case it's not really anything to do with AI IMO

[worthy python]

Ty, i can understand it 😅 But only people who working with AI know how to resolve it, imho

[mossy nexus]

not really no. I'd say ask in #cpp

your main issue seems your approach of control rotation, which is abstract from AI

[worthy python]

thank you. I will try, eh

[ocean wren]

Usually, you'd want to have your weapon location controlled outside of your skeleton, then add a control rig for your skeleton so that it IK's to the weapon position. That way you can move your weapon around properly and the skeleton moves to match it. You can add some lean into the basic skelton then, but have the IK still solve for the final position/rotation

Think I might implement that tagging stuff in UE, see what it feels like to use

[modest tusk]

So I have a question about this, can I use two different languages for a project? I couldn't find any results, but I want to use python for AI, and C++ for the game, is that possible?

I heard that unreal is the best for VR development so I wanted to see if I could do that for better AI.

[celest python]

UE just supports C++ and Blueprints system out of the box, Python is supported only for editor modules

You can implement your own system, some people did that on #cpp

[modest tusk]

Alright, thank you!

[celest python]

https://github.com/Tencent/UnLua Tencent has a Lua system for UE

Though its not battle tested

[celest python]

VR is nice with UE but never try with multiplayer

You'd need to rewrite a lot of things from scratch

Since Character framework does not work well with VR movement

For a competitive game it'd be a hell to work with

[ocean wren]

Epic are working on a new language too

Personally, I use http requests for AI stuff, if you're talking about ML use

in game AI we tend to use C++

on the other hand, you can create ML models in pytorch and use torch C++ library

[modest tusk]

I want to make a game like Pavlov shack on the quest, maybe even rift and other platforms, but I want to sell it for free just like Pavlov for a while, then make a different game but in a different style like a battle royale or a mission based one.

[celest python]

Was Pavlov multiplayer?

[modest tusk]

It is

But not mass, 10 player

[celest python]

Rather than player count, UE's default framework limits you very hard by orienting everything to ~2010's shooter mechanics

You wont be able to get prediction out of the box if you dont use CharacterMovementComponent

And it's not suitable for VR at all, afaik

[mossy nexus]

to be fair VR is still getting to grips with many things. it's going to be a while before anything gets normalized for it. HL Alyx was probably the first real attempt

[ocean wren]

Naw, plenty of VR experiments out htere

there

[mossy nexus]

yes, experiments

[ocean wren]

So its not like there hasn't been a fairly normalized setup. Just depends on the game genre as you'd expect

[mossy nexus]

it's too early to have anything normalized

there isn't even a real market for it outside of some early adopters

[ocean wren]

for VR stuff?

the market for VR is mainly business to business enterprise stuff.. not really games

[mossy nexus]

case in point

[ocean wren]

I guess the Meta thing will make it more consumer friendly and add more of a market, but its still mostly about digital twins and collaborative tools etc.

[mossy nexus]

VR from a consumer perspective is still largely at a fad level

[ocean wren]

I wouldn't call it fad, I'd call it "early adopter"

[mossy nexus]

sure, semantics

[ocean wren]

It honestly can't be fad, I've been working on VR stuff for over 20 years now 🙂

[mossy nexus]

it's been fad for at least half that time

if not longer

[ocean wren]

And there's some serious $$ for VR dev in the B2B market

[mossy nexus]

people want it, then forget about it, then want it again

[ocean wren]

I'd still not regard it as a fad

its just not mainstream is all

[mossy nexus]

plus it's not like any other market. it requires expensive hardware to run

it requires people to be able to stomach the VR experience

so the market can only ever grow so large

[ocean wren]

Yeah, there's quite a difference between the consumer VR thing and the military stuff for instance, pretty much a different thing market-wise

[cerulean girder]

Meta terms of service for assets placed in thier "world" will most likely cause a lot of asset vendors to blacklist use in Meta...just food for thought... apparently the meta terms make some statements about ownership that doesnt sit well with asset creators

[mossy nexus]

I have 0 concern for meta

[ocean wren]

yeah, I don't really care much about meta, but at least them putting money into VR can drive the hardware forward a bit

[cerulean girder]

Same...its secondlife with FB logins

[mossy nexus]

... and NFTs

[ocean wren]

second life drove some things forward though, even if it was a trainwreck in the end

SL was a very fad based thing to be fair

[mossy nexus]

who would have thought I'd turn into a luddite in my later years

[ocean wren]

I'm working on holodeck stuff, so only tangentially related to VR, but I still appreciate the money going into it

[cerulean girder]

To be fair SL got me interested in code when i was younger 😂

So in a way i owe them some respect

[ocean wren]

SL was one of those short term fashion things that probably ended up as a gateway for some.

I just saw it as an annoyance because of all the media 🙂

[mossy nexus]

AR I can sort of get behind

I think there's potential for neat stuff

[ocean wren]

AR hardware isn't there yet.. but yeah, I really think AR is the future

[modest tusk]

So what engine do you recommend then?

[mossy nexus]

and it doesn't necessarily require a massive helmet screen

[ocean wren]

if they can fix the optics to render no background

The first time I tried a hololens, I understood that eventually AR will win out against VR and monitors

[mossy nexus]

one can hope

[ocean wren]

even though the hololens is trash 🙂

[mossy nexus]

#deprecateVR

[modest tusk]

VR is on top

[ocean wren]

I was an early VR guy too.. like REALLY early VR, which was even more trash

[mossy nexus]

well yeah VR 20 years ago I can imagine was terrible

[ocean wren]

AR is basically a superset of VR

[modest tusk]

AR is ok, but I personally prefer VR.

[mossy nexus]

cool, thanks for letting us know 😉

[ocean wren]

lets see.. first headset I tried had a resolution of 180x120 or something 🙂

[modest tusk]

👀

[mossy nexus]

sounds all kinds of awful

[modest tusk]

Agreed

[ocean wren]

high end was like 320x240 or something

[mossy nexus]

I had an interesting experience with a VR movie almost a decade ago, but I'm not going near that stuff again

[modest tusk]

🤣

[ocean wren]

yeah, I think mostly at that time, it was people who knew 3D and sort of imagined the future stuff of VR

same with the hololens now.. its actually usable, but not great. But give it 20 years and it'll be perfected

[mossy nexus]

I can definitely see that

plus AR can be on phones etc.

so I think there's something to be had there

can't wait for the geocaching games

[ocean wren]

AR with a phone is a bit of an odd one

We've done a few projects using that. But I'm not a fan of the form factor

[mossy nexus]

give it 20 years 😉

[ocean wren]

Yeah, don't think I'll be bothered by then 🙂

besides, I'm aiming for a holodeck before that happens 🙂

[mossy nexus]

sure

I'll probably pick something like that up and see if there's going to be some applications for LD

[ocean wren]

Where we're heading, there aint no levels! 🙂

[mossy nexus]

there'll always be levels because people will always need direction

[ocean wren]

well, not levels in the sense that you'll create them. Maybe you'll describe them or select them

dammit, I need a movie to watch 🙂

[mossy nexus]

I'll create visual markup in some sense or another

it's what I do

[mossy nexus]

I can recommend "don't look up"

even if it's sad in a frustrating kind of way

[celest python]

I'd suggest research more about Multiplayer VR in UE and especially ask people who tried before, if you can't get a positive results try Unity. Because there are other frameworks do easier networking in VR (not 'better' maybe but easier)

My knowledge about VR multiplayer is based on some friends comments I've met in the industry

I'm just sure CMC is not going to work in your case, but I tested that back when latest engine version was 4.23

Maybe someone made a workaround, or even Epic provided something new

[ocean wren]

There is a new VR template

Already saw Don't Look Up. Good film.

[mossy nexus]

it's frustrating because it was too real

[ocean wren]

Yeah, too realistic

[celest python]

"Upgrade" (https://youtu.be/36PDeN9NRZ0)

[ocean wren]

Seen it

I've seen everything 🙂

[misty wharf]

how about Ex Machina

it's somewhat topical to #gameplay-ai as well :P

[ocean wren]

Hahaha.. good catch

seen it 🙂

on the subject of AI, anyone got any recommendations for games with good AI? I mean I'm after some interesting games to play. Probably going to buy Weird West, but that's not out yet

[celest python]

Have you played F.E.A.R.?

"Thief" was good too

and MGS V

[mossy nexus]

ex machinima was great

[mossy nexus]

on the off chance you haven't played it, hitman

[ocean wren]

yeah, played all of those, although I guess I could play fear again, but it was pretty simple

hitman I didn't really get along with

and yeah, played thief 🙂 and mgs V

I'm looking for a new game really

[mossy nexus]

I got nothing

[ocean wren]

I'll probably try out Ark Raiders when they ship it

[mossy nexus]

wait until I release something 😉

[ocean wren]

When's that going to be?

🙂

[mossy nexus]

TBA

I'm gonna try and see if I can have a VS ready end of next year for realsies though

just on my personal project

but I did say that half in jest

[ocean wren]

I kind of want to work on a big AI project now.. but have to complete this latest research really

[celest python]

CE'a game reminds me Dishonored, just came into my mind when he mention his own project

If you havent played Dishonored, its a nice candidate too

[mossy nexus]

dishonored is just modern thief

[ocean wren]

Yeah, I loved dishonored, althought they spelt it wrong 🙂

[mossy nexus]

(give or take)

[ocean wren]

its definitely in the vein of thief like games

[mossy nexus]

it's definitely interesting going back the family tree on immersive sim games

speaking of, Prey 2017?

SOMA?

[ocean wren]

Prey hmm

[mossy nexus]

soma will live in your head rent free for years

so proceed with caution

[ocean wren]

that the sort of horror style thingy?

[mossy nexus]

yes

[ocean wren]

yeah, not into horror 🙂

[mossy nexus]

and possibly the best writing any video game has had... well, ever

[mossy nexus]

if dishonored is modern thief, prey is modern system shock

except better tbh

[midnight scroll]

What is generally the preferred method of pawn's walking around each other? There's a few that I've read on so far, but there are also debates on them. Wondering what people here might use.

[celest python]

Probably there is a better way but one thing comes to my mind is checking the closest navigation affector in a distance threshold and query the environment with given distance and store the locations and queue them in an array, then use MoveTo

If there is any object blocking the navigation is a 500m, do a trace or overlap somehow (gonna get expensive 😦 )
get the closest one from overlap check
if its not closer than minimum tolerance you desire set it as circle radius
query the environment
create pathpoints (fvectors stored in an array)
loop MoveTos

[ocean wren]

Detour Crowd is currently the best method in Unreal Engine, but you have to tweak it some

They set it up a bit wrong by default

You can set the sample distance where it chooses to avoid other agents, how many agents to account for etc. Its pretty tweakable

[steady elm]

i don't understand how the navigation system calculates the agent radius, is it just from the center or is it around the model?

[ocean wren]

Pretty sure its in the navigation system properties

if you go into the preferences and look for navigation system, its probably in there somewhere

[steady elm]

i know but for some reason it stops and says it has reached its target before a trigger i gave it reached the target even tho the trigger has a bigger radius so i'm confused why that is

@ocean wren

[ocean wren]

Oh, thats in the moveto node itself, if that's what your using

there's an acceptance radius value on the node in the BT

[keen crow]

Are AI controllers expected to be unique for each character no matter what BT they are running? I have 4 characters on my level and 2 of them are running the same BTs but I'm seeing something strange - AI controllers names are duplicated in BT debug selector

[misty wharf]

Yes, each AI pawn has its own controller

(You certainly could create a controller which manages multiple pawns, but it's probably somewhat complicated - but I'd imagine in some niche usecases it could offer performance benefits)

[mossy nexus]

each pawn has its own instance of a controller

[keen crow]

then why are they duplicated in the dropdown?

[ocean wren]

Only the gods know

And they aint tellin

[midnight scroll]

@keen crowEngine version?

Wondering if they're not just named similar. But still four different instances.

[keen crow]

4.26.2

[midnight scroll]

Hmm. 4.27 version of that looks a little different.

[mossy nexus]

you could double check by seeing what the print screen tells you

[midnight scroll]

I can't break it like he did. In 4.27 they're named with no doubles, and it actually tells you where they were spawned.

[ocean wren]

Man, its so frustrating not having artists 🙂

I need a machine learning artist

[misty gale]

That'd be very usefull!

Specifying inspirations, fidelity and whatnotz and just get a procedural mesh output !

[ocean wren]

We can do a bit of that, but mostly in 2D

I wonder if we could do the same but using a nodegraph (thinking about it)

Given there's now differentiable renderers out there

We can figure out object positions and types.. but actually generating new ones? not yet at least

[misty gale]

It shouldnt be that hard should it

Generating complex procedural faces sounds harder

[ocean wren]

Well, the 3D structure has more challenge because its got things like occlusion and a higher dimensional space

That said, we've got a ton of data in terms of videos and whatnot, we should be able to eventually learn the whole goddam thing. I'm just frustrated because I've not got enough GPU power to do my work on it right now.

[misty gale]

And meshes generated by actual people etc 3d scanned ones and whatnot

The archive should be more than sufficient

[ocean wren]

The problem comes when you're trying to create something that isn't from real life, we can scan everything sure, eventually. But what of art?

I guess it needs to have some notion of what art and creativity is and then use that in the generative step

You kind of need a differentiable PCG system really

[mossy nexus]

if you get it working sign me up

bonus if it comes pre rigged and you can do the same with animations

[ocean wren]

Yeah, my aim is motion based behaviours, so that's always going to be a focus

But the idea of using existing scene understanding approaches gives way to the issue of being able to generate novel ones

i.e. we can learn stuff like where different objects are in a scene (this is well studied), but what of generating novel scene layouts with the same object distributions? or with novel objects that still make logical sense

And in a way that we can properly evaluate and interpret

I guess treating scenes as a class of SDF's might work

[cerulean girder]

is there another test i can preform in eqs other than "does path exist" to determine if im selecting a point on the current nav mesh. check if the path exist is hitting my test way too hard in some cases

almost 2ms on just that check

[misty wharf]

It seems you could just not use EQS for it?

If you're checking a single point for a single thing you gain nothing from EQS

If you have some reason to do it with EQS, you would probably have to write a custom test

[ocean wren]

There's a project point to navmesh you could use for sure

[cerulean girder]

the eqs query im doing determines a point that an ai can summon back up units that is furthest from the player characters, that currently doesnt have a unit, and is not visible by the players. one of the test in it determines if the point is reachable...meaning if i spawn the ai at that point can it get to the player...my nav mesh is already hand catered to only allow the nav where i want it (i didnt just slap a huge square in an area and call it done)...but the path test is 2ms long...the entire rest of the query is running at .4-1ms...

[ocean wren]

That sounds pretty expensive, given how low res the navmesh is

[mossy nexus]

doesn't help with navmesh islands 😔

but I do do the same

[cerulean girder]

exactly...2ms for a path test is disgusting

[mossy nexus]

well no

it's expected

[ocean wren]

it does string pulling on the nav generation, I suspect if you didn't switch that on it'd speed up a fair bit

[mossy nexus]

pathfinding algorithms are expensive

[ocean wren]

they're not THAT expensive, the navmesh is low poly after all 2ms is too expensive, but its doing string pulling which he obviously doesn't care about

the other thing is, you're probably doing it for each point?

[cerulean girder]

thats why i was wondering if there was another way to check if im selecting a nav mesh point

[mossy nexus]

nlogn sounds expensive to me 😉

[ocean wren]

what I'd do, is use EQS to query points for the visibility and proximity stuff, then use a dyjsktra to the points and just return the first point when you hit a navpoly with it inside