gameplay-ai

[uncut slate]

my these navlinks are not linking like the black lines

the one works

but this one isn't working ?

if i move it over by the first one it's fine

but anywhere else its not showing up

[slow bobcat]

Is the green blob on the left nav? It's difficult to see

[uncut slate]

yes the green is nav

i got it to work at different locations, but i'm still confused as to why it won't work for some of these when it looks like it should be fine

[slow bobcat]

Can you show the nav tiles? In the recast object you can select an option for it. It will paint some but yellow squares

[uncut slate]

see the green one

it should be valid imo

[slow bobcat]

I don't see the tiles. Can you:

• paint the polygons as you are doing.
• do not paint filled polys
• paint the nav tiles
• zoom out a but to see a more top-down view

[uncut slate]

if i zoom out it doesn't show them

only around where my camera is ?

[slow bobcat]

It's should allow you to zoom enough. How far are the nav link sides from each other? Doesn't look too far, but still worth checkign the tiles

[slow bobcat]

All that there.... Are those nav polys? What is that blue border/dots? What's going on there?

[uncut slate]

its the water blocking the nav

the links are connecting the green

or the bottom of the navmesh

is what it is

its so you can't nav into the water

[slow bobcat]

Let's see the tiles by the nav links failing

[slow bobcat]

Ok, try moving it to another tile. Just one side. It might be an issue with how the tiles are organised

Also, if you delete the nav link and re-create again, does it fail with the new one?

[uncut slate]

yes i tried that a bunch even copy/paste a good one didn't help

[cyan sigil]

I am working on a character that hides from the player. I can already detect when they are visible. What's the best way to set up in Behavior Tree to change behavior when a decorator has been true for more than X seconds?

[uncut slate]

so i can move them and align them like they are, but its weird i can't put them exactly where i want them

[slow bobcat]

It's maybe an issue with tile indexes. Obscure but known problem.

I can say 100% but without checkign the nav at the locations etc... Also try restarting the editor. Sometimes it bugs out

[uncut slate]

first thing i did

[uncut slate]

ok thats a pain lol

[uncut slate]

it looks like i can't span more then two tiles ?

[slow bobcat]

Yeah, usually that's the issue

[uncut slate]

i just increased the tile sized lol

seems to have solved the issue, thank you

what does a larger tile size effect ?

it solves my problem but does it have any side effects ?

ig i'll find out

[cyan sigil]

Any help for my task above?

[slow bobcat]

add a service that changes the decorator value back so the decorator aborts execution and re-evaluates for example

In code you can force a tree restart from tasks too

[slow bobcat]

More polygons per tile, larger times to rebuild tiles, more area covered, so more memory used for it I would assume

[cyan sigil]

I found a solution:

The second one checks on tick

Now I need to be able to detect if the character has taken damage or not. I already have a damage event but am unsure how to have a decorator or service react.

[slow bobcat]

What is that? You have a mod to use bt's horizontally. They quite difficult to read

What the heck is an IF in behavior trees?

[cyan sigil]

Hierarchical Task Network. Very similar to BT.

[slow bobcat]

Aaaaaaah

Yeah but no. This are technically finite state machines. But OK that's cool. Is that for. Unreal?? Never seen it before

Or is it a plugin?

[cyan sigil]

It's a plugin.

Main difference is that it's capable of "Planning" with a simulated blackboard, instead of only looking at current world state.

[slow bobcat]

Yeah. Those are for goal planning strategies. What's the name of the plugin?

[cyan sigil]

IF node is what they have instead of Selector.

https://maksmaisak.github.io/htn/front.html

[slow bobcat]

Old AI Programmer rant: selectors are not ifs.

But I get it, I was confused because I thought it was a bt

[slow bobcat]

This looks fantastic. Wonder how performance is. Goal planners tend to be expensive on cpu

[harsh storm]

Look up the HTN plugin on fab. Actually. I can just do that for you, lol https://www.fab.com/tr/listings/1423ad9b-9c53-43be-b4c8-af1b655377bf

@brittle lynx is the author - he can speak on the performance.

[brittle lynx]

I have been summoned.

To be more precise, the Prefer node is what this has instead of Selector.
https://maksmaisak.github.io/htn/docs.html#/prefer
An If is its own thing.
https://maksmaisak.github.io/htn/docs.html#/if

Performance-wise it's very much on par with Behavior Trees. Rendering becomes a bottleneck far before AI does. For characters that are just capsules (so no complex rendering or animation), people have benchmarked hundreds active at a time. If memory serves someone on this server brought it up to ~8000 with some tick rate optimizations.

Goal planners tend to be expensive on cpu
It's not a goal planner, unlike GOAP. HTN just forward-plans through the branches until it finds a plan that ends in a node that has nothing after it. You define which tasks can go after which tasks, so you don't get this combinatorial explosion of possible plans that you see with GOAP-like systems where any action can follow any other.

[undone trellis]

nobody ? 😢

[dense owl]

https://tenor.com/view/tumbleweed-highway-blue-sky-gif-14068616

[slow bobcat]

Sounds like you want to abort the execution flow to interrupt sleep.
One solution would to have a decorator checking a bbk bool (canSleep). If true the sleep task will execute. If false, it will not. The key part is to set the decorator to "Abort Self". Then all you need is to change the value of CanSleep to false and the task sleep will finish

[slow bobcat]

Yeah this is the plugin that was linked after my comment. Looks great

[slow bobcat]

Thanks a lot for the explanation. Amazing work you have done here

[celest python]

I really like we can keep track of the "future" in HTN

[slow bobcat]

@harsh storm my friends pointed to another plugin, which has almost the same name?
https://www.fab.com/listings/5ac2e509-9918-4eeb-90f8-f257e5cd230f

[harsh storm]

Yeah - that's the same thing

[cyan sigil]

What's the best way to make a branch that is entered or exited based on if an Event has occurred that frame?

In behavior tree, or equivalent.

I could make a service that sets a blackboard value enabled for one frame but I feel like there's a better way.

[misty wharf]

🤔

[misty wharf]

I would try creating a decorator which aborts on the frame when the event occurs or something like that

Not really sure but seems worth a try

[cyan sigil]

So the decorator would bind, and then what happens during the broadcast? Is there a function a decorator can all to automatically abort itself?

I wound up making a service that adds the number of times the event has fired to a blackboard key.

[slow bobcat]

Yeah you subscribe to the event delegate from the decorator
(we usually have the function callback withing the memory of the decorator). Use on be come relevant/on cease relevant to register / unregister. Then, in your callback, you call the abort logic. You can check the blackboard decorator to see how is it done.

We do this for gameplay tags and gameplay events

[paper basin]

Hello. I have a small problem with positioning my AI after it gets up from the chair.
When the AI sits, I adjust the location, disable collisions, and set the movement to flying. It works, so it stayed like this.

But I have a problem with teleporting it back to the navmesh. As you can see on the screenshot 1, the AI gets placed between the chairs, even though there is no navmesh there. There is a small bit of navmesh directly under the chair but I'm not sure if this is the cause - other tables that have the same chairs but further away work great.
I tried to eliminate this small square by adjusting the chair simple collision and setting the value of blocking navigation under the mesh on mesh component - but it didn't change anything. The chairs have BlockAllDynamic collision with QueryAndPhysics.
Second screenshot shows what I use for the teleport. Right now I'm still messing around with the height of the Point and the Extent but it didn't change anything yet.

Any tips how could I fix/improve this?

EDIT: Turns out making the navmesh higher fixes spots under the chairs... but adds navmesh on top of them. Which is the same kind of problem again.

[stone coral]

Hello. I have a small problem with

[upbeat hamlet]

Hi guys, I want to run an EQS query for 8 different actors as querier. Then I want to store the resulting locations in a map, with the actor as the key.
Now obviously this doesn't work as the EQS query runs async and I can't access the for each loop's actor anymore.
How do you solve something like this? How do I pass the actor through the async event?

[vagrant quarry]

Did you ever fix this? I think I'm experiencing a similar issue

[near condor]

I believe when I upgraded to 5.6 it was all good again

[oblique surge]

anybody have an idea why State Tree breakpoints would not be working (5.6.1)?

I have tried setting on enter/exit state, I have tried setting on tasks. The state tree is running but the breakpoints are never hit

[slow bobcat]

I have seen this with someone else here and turned out the state tree was not running. If you open the debugger for your state tree, is it recording and doing things? Can you see the state with the breakpoint running in the debugger?

[oblique surge]

it is moving the object that the state tasks tell it to, and if I add a PrintString to one of the tasks, it prints at the appropriate time

I'm just moving an object around through a series of states, all of which execute correctly

Hmm I put breakpoints in ST_Shooter in the FP variant and they are also not firing, but the bots are running around and shooting me

[slow bobcat]

Check with the debugger to be sure things are working as expected

[oblique surge]

normal BP breakpoints are working (I can break inside the tasks themselves).

just yeah the state tree breakpoints entirely fail to trigger

[slow bobcat]

again, check with the debugger

[oblique surge]

I'm unclear what that means

oh you mean code debugger

[slow bobcat]

I mean the State Tree Debugger

[oblique surge]

oh ffs

it's an entirely separate debugger?

lol

PEBCAK

[slow bobcat]

I strongly recommend reading the documentation of State Trees before using them

[oblique surge]

with all respect, "the debugger tab" is not the clearest instruction

[slow bobcat]

dude... there's a picture about how to open it and a text explaining what is it

[oblique surge]

"Use the dedicated State Tree debugger" would be much clearer, particularly since it looks exactly like the visual debugger

[slow bobcat]

it does not

[oblique surge]

yes, me and lots of other people not getting this is entirely our fault, not bad documentation.

[slow bobcat]

It even says the menus you need to click...

[oblique surge]

it does not say you need to be on the state tree "blueprint" to get that custom drop-down

[slow bobcat]

I suggest reading this in case there are some other things to be learnt
https://dev.epicgames.com/documentation/en-us/unreal-engine/statetree-debugger-quick-start-guide

[oblique surge]

if you are on your level window, and click the Window tab, there is no debugger

I have read the quick start guide. I have a functioning state tree

also, if you do not have the State Tree Debugger open, it does not break

given that you are first told to set breakpoints and see it stop before you are told to open the debugger, I maintain this is a documentation issue

in any case, thanks for your help

[slow bobcat]

Do your breakpoints stop now with the debugger open?

[oblique surge]

yes, they work with the debugger open

[slow bobcat]

That might be a new thing. I recall breakpoints stopping without the debugger in earlier versions. I think? Can't remember

[cyan cairn]

Guys do you have any idea how i can abort all task in patrol sequence, when my state changes to chase. When it changes from patrol to chase Move to Task is still executing.

Same with when AI is in chase mode, state changes on patrol and move to in chase sequence is still executing

[slow bobcat]

Call StopMovement

That will cancel any movement task running

[cyan cairn]

Yeah, i mean thats one way for movement itself, but what when i have a different longer task changing state key will not be sufficient enough because it will not just abort itself and still go on

[slow bobcat]

Not sure I follow. Correct me where I'm wrong.
You have 2 issues: you don't know how to abort the current task in the BT + you don't know how to also stop movement.

To abort the flow you need your decorator to be able to have abort conditions.

To abort the movement, you need to call StopMovement on your character when you get our of Patrol, chase etc

[cyan cairn]

Yeah, my overall problem is how to stop/abort an task node (for example Move to, but it can be something totally different) or entire branch in my case when i change State (My ai State) for example from Patrol to Chase when Patrol sequence is executing, My Behavior tree wont execute Chase sequence, it will end Patrol sequence first. I want to abort sequence of Patroling and momentarily change it to Chase sequence

When my AI sees me and its in patrol state, it changes state key yes but it does not fail Patrol state

[harsh storm]

Honestly - with the way you're using the Behavior Tree, you shouldn't be using it. You should be using the State Tree. Behavior Trees are not State Machines.

But you are writing your stuff to be that way

To abort those branches, you need to set the Observer Aborts stuff in the Decorator for Is In State.

Decorators have a way to abort whenever the value of w/e it is observing changes

[cyan cairn]

Tried setting them, but even when value changes it is still executing

[hallow compass]

i always hated BT design

glad ST is getting upgrades

[cyan cairn]

I will probably change to state machines, see that BT have problems with this kind of stuff

[harsh storm]

https://tenor.com/bVG8n.gif

[harsh storm]

Show your decorator. BTs are pretty old and haven't received much love over the years, so they're pretty solid.

I find BTs to be much more simple than STs personally. And the data handling is also much more straight forward.

[cyan cairn]

I don't know if i understand this correctly, if Observer Aborts is on self will decorator abort when state will change

Like will it abort this particular branch?

[harsh storm]

Pretty sure you need the blackboard key property to be the one that has the actual value you want observed.

The "Observer aborts" stuff probably just pays attention to that.

Instead of any BB key

[cyan cairn]

Okay that makes sense

[harsh storm]

Check out the decorator for "blackboard key is set"

[dense solar]

The apostrophe (') is working and when i click the 4, 5 any other it won't working.

[slow bobcat]

those are numpad numbers. You can't use the top row numbers on the keyboard, only numpad (in case that's the problem)

[storm needle]

if "Is In State" is your custom native decorator, check the BlackboardKey Decorator when it observes the Blackboard key to have changed to have the decorator evaluate itself and use the "Abort - Self or Both" to take effect.
If you could set the state on the Blackboard and just use the blackboard decorator node would work out of the box.

@cyan cairn your custom decorator would need to listen to your state changes in "OnBecomeRelevant"
then call this once the state has changed

        const EBTDecoratorAbortRequest RequestMode = (NotifyObserver == EBTBlackboardRestart::ValueChange) ? EBTDecoratorAbortRequest::ConditionPassing : EBTDecoratorAbortRequest::ConditionResultChanged;
        ConditionalFlowAbort(*BehaviorComp, RequestMode);

to have the decorator evaluate your raw condition value of this decorator.
in "OnCeaseRelevant" you have to unregister your decorator again from state changed.

[dense solar]

😅Yeah now its working. just wasted 1 week for find that problem.

[slow bobcat]

you can re-map the buttons if needed in the editor options

[dense solar]

I'm trying to get into Unreal C++. Do you have any tips or advice for a beginner? I'd really appreciate it.

[slow bobcat]

I would look for a course in youtube most probably. There's probably something for beginners. I would checdk Tom's Looman courses, heard they are super good

[dense solar]

Thank you brother💖

[slow bobcat]

these ones
https://www.tomlooman.com/

I would avoid chatgpt and the likes for a good while because they are missleading. If you use them, double check the code, check the source etc

[signal island]

guys the state tree ai component schema is not exported? i cant use it as a base to override a new context

[harsh storm]

You can just look to see if it is

If it isn't - its not that big of a deal to be honest. All it really does is add the AI Controller

[misty wharf]

It isn't, but yeah you can just copypasta it

[slow bobcat]

so how's people supposed to extend the Schema? I thought that was something expected to happen

[harsh storm]

https://tenor.com/b0gIS.gif

[misty wharf]

Yeah I thought so also, but the schema just defines what stuff is valid to use in that particular schema, and if you copypaste it, you achieve the same result

Might be an oversight I guess

[harsh storm]

Same thing exists with the DetourAIController. It isn't exported. So you have to just do the same thing that class does.

Sure, less of a headache - but meh

[slow bobcat]

yeah but... then the engine source changes... you didn't notice and... re-copy...

[brittle sapphire]

**Training a neural net for full skeleton control as a way for players to control characters with zero animation. **
Using: https://github.com/AlanLaboratory/UnrealMLAgents (port of Unity ml-agents) + a custom tick system to speed up unreal.

• Clip was achieved in 6 hours (24 hours simulated time)
• Boxes are center of mass bounds (moves with feet)
• White balls are reference animation skel bones, green balls are neural net skel bones.

*Training Test 2 - Goal is to learn forward locomotion. *

• Animation matching system is working as intended, but the network is just learning the best average pose to hold to reap consistent reward. Which is cool because it's a lot more stable standing in place, but that wasn't the desired outcome.
• Custom foot shape working well, toe pressure providing stability as intended, heel ball shape allows foot to rotate to counteract lean.

Need to:

• deep dive leg motor power, too weak (probably audit all motor drives)
• check to see if foot friction is too high... it probably is...
• reduce foot contact sensor reward
• add strong reward for forward velocity (but penalize for too fast to avoid reward via falling)
• maybe slow down the walking animation, or put greater reward on leg position matching over the other bone-match-points.

Anyone else doing ML/RL training in unreal? would love to collab or brainstorm.

[slow bobcat]

I strongly recommend joining AI in Games discord. Lots of. People doing research for ML in games, doctorate and profesionals. Very good place to ask for help / colaboration
https://discord.com/invite/SUhttg6

[brittle sapphire]

oooh thank you!

[slow bobcat]

https://tenor.com/view/it-takes-two-therapy-no-gif-25267782

Co lla bo ra tiooooon

[signal island]

is there a way to expose functions?

[harsh storm]

#gameplay-ai message

Seems to be the way

Pretty much - inherit from FStateTreePropertyFunctionCommonBase

[signal island]

really cool

[signal island]

wtff this is soo powerful

[hallow compass]

https://notes.hzfishy.fr/Unreal-Engine/Extra/Learn/Learn-C++-for-UE

[harsh storm]

@slow bobcat Well well well - it looks like you're giving a talk at UE fest next week?

[hallow compass]

[slow bobcat]

https://tenor.com/XWxh.gif

[harsh storm]

https://tenor.com/2XnZ.gif

[slow bobcat]

stop spying my Only Devs accounts

[harsh storm]

I was actually just curious to see if there were any talks at this UE fest that I'd look forward to being put up on youtube. And noticed a familiar name....

[slow bobcat]

It's not me. It's an impostor

SUS talk

[celest python]

likely api forgotten again

[mortal spoke]

Did you figure it out?

[iron roost]

kind of, I stopped building map and just packaged. Also made sure to use a new folder each time. I use to build, cook package.

[slow bobcat]

Are you using sublevels with World Composition or World partition?

[hallow compass]

arent sub levels broken with world partition ?

also, whats world composition ?

nvm found it, legacy system using level instances

[slow bobcat]

Yep

If using world composition there's one reason that could explain the issue

[mortal spoke]

If you are using world partition, make sure you have correct main RecastNavMesh.

My problem (navigation working in PIE, but not in packaged build), was caused by other recast being pulled from level instance actor and being set as main, causing NavigationDataChunkActors to never register because the RecastNavMesh was different.

[onyx basin]

Hi, im making a procedural dungeon. do u think is possible to turn off the dynamic nav bound to save cpu since i dont need that after the dungeon has been generated?

[slow bobcat]

Yeah, there are function to lock and pause nav generation in the nav system

[onyx basin]

how?

are they accessible in bps?

[slow bobcat]

In Bp's... I doubt it but no idea. You will have to check in code what's exposed. Or grab the navigation system in BP's an check what's there

[onyx basin]

yeah i did but i dont think its here

do u remember what they are called in c++

[slow bobcat]

I think you do it with the Update Locks in the navigation context

TSharedPtr<FNavigationLockContext> NavUpdateLock;
UWorld* World = CurrentComponent->GetWorld();
if( World )
{
  NavUpdateLock = MakeShareable( new FNavigationLockContext(World, ENavigationLockReason::MaterialUpdate) );
}```

Not sure if that's the right way or this one

    FNavigationLockContext Lock(NavSys);

    UE_LOG(LogTemp, Warning, TEXT("NavMesh updates are now locked by FNavigationLockContext."));

    // 2. Perform your batch operations
    // ... code to spawn or move many actors ...
    UE_LOG(LogTemp, Warning, TEXT("Batch operations complete."));


    // 3. No need to manually unlock!
    // The 'Lock' variable will be destroyed when the function ends,
    // automatically calling NavSys->UnlockUpdates() in its destructor.```

Being the lock within a function scope

ok I asked chatgpt and seems to agree with me?

#include "NavMesh/NavMeshBoundsVolume.h"

void YourClass::SpawnStuffSafely()
{
    // Lock the nav system — navmesh updates won't happen until this object goes out of scope
    FNavigationLockContext NavLock(FNavigationSystem::GetCurrent<UNavigationSystemV1>(GetWorld()), TEXT("PauseNav"));

    // Perform any navigation-affecting changes here
    SpawnYourActors();
    MoveThingsAround();
    AddNavRelevantModifiers();

    // Once NavLock goes out of scope (end of function, or when destroyed),
    // the navigation system will resume and rebuild as needed
}

this other way so you can control the lock lifetime yourself

TUniquePtr<FNavigationLockContext> NavLockPtr;

// When pausing nav updates
NavLockPtr = MakeUnique<FNavigationLockContext>(FNavigationSystem::GetCurrent<UNavigationSystemV1>(GetWorld()), TEXT("PauseNav"));

// Do your operations...

// When ready to resume nav updates
NavLockPtr.Reset();  // Unlocks and triggers dirty area processing / rebuilding

According to chatgp you have to pass the NAvigation System to it, I have seen in code they pass the world, so not super sure

Ah ok it doesn't really matter, it's a world context object

so I guess in you case it will be

• create context lock
• build the world proceduraly
• On World generation finished, release the lock

but this is definitely not possible in BP's

[onyx basin]

wow thx, too bad im not familiar with c++ . i have to work on it for a bit

[rigid sage]

Hey all, quick question about State Trees, is it not possible to read variables from actor components? My 'current health' value lives in a HealthComponent I made that lives on the character blueprint, but said component is apparently invisible to the state tree 🤔

[slow bobcat]

The component has to be public and use a specific meta tag IIRC

[rigid sage]

Ahhh, interesting. I couldn't find that in any in any of the writing, you wouldn't happen to know if that's documented anywhere?

[slow bobcat]

Can't recall. Not in front of my pc so I can't check if I'm tripping like bad AI about the tag. Check the actor/controller in the schema. I believe you can bind to their components?

But we definitely access components in our tree. Worst case scenario I can help you tomorrow

[rigid sage]

Yes, I saw that some of the object variables and some components like CharacterMovement are exposed, just not my Health component. I'll keep looking tonight but failing that any help would be appreciated 🙏

[slow bobcat]

Check in code how is the character Movement component declared in the character actor class you use in your schema.

[slow bobcat]

In the schema itself there must be a tag where they store the pointer to the actor

[merry totem]

Hello, AI MoveTo does not work with Nearest BP_Envy
im using find nearest actor node and it finds that correctly because I tried to print it
also casting is working correctly with printing
but the problem is this actor is not moving towards Nearest BP_Envy

actor is character, have character movement and default AIController Class
when I try to use AI MoveTo with get player character [0] it goes towards player correctly

[misty wharf]

Is it printing Fail or Success?

[merry totem]

Fail

[misty wharf]

Visual Logger is generally helpful in debugging movement, you can find it in the Tools menu I think under Debugging

You first open it, press record, then press play, and let the game run the logic you wish to debug, and it should show info on it in the visual logger window

At this point my guess is that the navigation fails to get a path for some reason, either the target is outside the navmesh, or it can't project the target's position to the navmesh

[merry totem]

yep, problem was actor can't reach destination cause near actor there was no navmesh volume
I just disabled "Can Ever Affect Navigation" inside actor bp and it works

thanks a lot that I learned visual logger stuff

[rotund peak]

i've been wondering, can i add aiperception to the player's actor? or is there something better? the goal is to have the server push info to the player when the player's character perceives something. i was thinking that it is effectively the same thing as a npc perceiving something. (but it needs an aicontroller to access the aiperception, so yeah, it'd need both a playercontroler and an aicontroler?)

[slow bobcat]

But do you need all the senses, factions, debug etc the Perception system gives you? Maybe you can do a simpler version within a player component. Moreover if it's a 1st or 3rd person game, where you will need to consider the camera FOV etc

[rotund peak]

probably will need all of it, displaying server validated conditional info based on LoS, multiple fov angles, multiple distances, and other conditions. displaying sounds through walls.

[harsh storm]

I'd check if the actual perception system needs an AIController or if just the component does.

If the perception system doesn't - just write you own player perception component that hooks into it.

Will probably still be kind of awkward to use - but w/e

[rotund peak]

thx will look into that

[rotund peak]

!! going to try adding the UAIPerceptionComponent to the player's pawn. aicontroller isnt necessary

[steel cloak]

I have a question regarding some issue I'm having with my AI. I'm currently trying to debug this and I'm at a loss. I am currently making an AI that can follow the player, I've got it working sort of except that when the AI reaches the player's last known location (which is what I want) it fires a OnPerceptionUpdated event and then runs the logic that follows, which is not what I want. I hope this makes sense? Basically how do I maintain the perception sense so the AI doesn't lose the player

[shadow furnace]

Maybe just have the perception updating know about how close it is to the goal? Does it suddenly lose all context inside of the perception update?

[steel cloak]

Once the move to function runs in my behavior tree it runs the TargetPerceptionUpdate event again. So basically it sees the player, moves to a location, then the perception event runs again even though the AI had line of sight the entire time. I'll try your suggestion tomorrow when I'm working on the project again and report back

[misty wharf]

Perception updated only runs when the perception state changes, so only when gain sight or lost sight

[steel cloak]

Yep! And I don't want it to run again, the issue is that it is after the movement ends, its like it loses sense for some reason

[misty wharf]

Check rotation and any blocking objects

Eg if it moves up to a point and at the end quickly rotates to face some direction, this could cause loss of sight, and similarly if either of the characters is holding a separate actor like a weapon or such, those could block the sight traces as well

[burnt wraith]

Do state trees work? I am adding a very simple state tree that should just go between 2 states with a 5 second delay task between them and it just continually iterates between the 2 states

With no delay

[misty wharf]

They work just fine. You probably configured something incorrectly with the states or their transitions

[burnt wraith]

It seems like when I issue a Status of Success in any of the tasks, it immediately executes the next task even in the "EnterState" task

Do I need to make it so that it only succeeds in the "good" state inside my tick functions?

[misty wharf]

Yes, if any of the tasks finish (succeed or fail), this will trigger the related transitions. I think 5.6 introduces a feature to allow you to choose whether this happens or not, but in previous versions you have to design your tasks with this in mind.

[burnt wraith]

Gotcha. Almost feels like I need a specific task for setting success or something weird. I just assumed that if I set an "Idle state" and it doesn't fail, it's a success instead of setting it to running

So if I for example create a delay task (I know UE already has a built in one), and I set it to running, but I have a second task that is "Set AI Behavior State" and it sets an enum to "Idle" then returns success (instead of running), this will trigger the entire State of the tree to succeed?

[misty wharf]

Correct

[burnt wraith]

Weird. So it's not specific to the task succeeding, but the entire state

[misty wharf]

Yeah, I guess this is why they introduced the feature in 5.6 that lets you choose whether the state requires a single task to finish or all of them :)

[burnt wraith]

So if I have a state that has a bunch of tasks, how do I make sure that it only succeeds on the very last one?

[burnt wraith]

Oh, I don't see that option. Am I blind? haha

"On State Completed" I assume?

[misty wharf]

Uhh... I'm pretty sure it should be in there somewhere, folks here have talked about it. I've not updated to 5.6 yet so can't really say where exactly

[harsh storm]

It'll be under the tasks part of a state

On the right

[proud timber]

is it possible to make an eqs with actors AND locations?
I even tried using the composite node but it still only returns either actors OR locations and never both

[slow bobcat]

You could create a context that gathers the actors and uses their locations + generates other locations. It will return only locations, but some will match actor locations. Another option is that you create a custom type, a struct with an array of actors and an array of locations inside. Your generator will generate one object, then your tests will be customised for that struct. The result of the eqs will be that struct. But it all sounds like too much work fir something you can probably handle another way.
What is it exactly what you are trying to do? Why do you need both?

[steel cloak]

I just wanted to pop back in and say that it was an issue with my behavior tree and I was very very confused. I appreciate everyone taking the time to assist though!

[misty wharf]

Good that you got it solved :) They can sometimes be kinda annoying to debug yeah

[proud timber]

thanks
I want to make the npcs decide depending on their sizes if they should attack or run away
so I thought if the eqs returns an actor it should move to that actor (attack) or if it returns a location it should move to that location (run away)
I already have a custom eqs test to compare the sizes so that is not a problem

[slow bobcat]

Maybe you can create the concept of "target" and have the info there. A target to approach would be a location or an actor etc

[proud timber]

ig

thanks

[warped tiger]

Hi, can someone tell me how to deal with this kind of looping behavior in state tree? I'd like to transition from the 2ndFindResources to ClaimResource when a resource has been found, but ClaimResource won't call StateEnter on its tasks because I have set bChangeStateOnReselection = false . If I don't set it to false StateEnter would be called when the transition DoSomethingWithResource1 -> DoSomethingWithResource2 happens.

[unreal wolf]

https://cdn.discordapp.com/attachments/85338836384628736/1418906188795346986/image.png?ex=68cfd2dd&is=68ce815d&hm=b2e01a4c01a458aefdd73756e742ee9c784a27bde0abd01bc2486fe350be05cb&

how would i go about figuring out nav link proxies to make this escalator one way?

it's fine if the ai walks up it, because i can just switch the animation manually upon entering the escalator and since it's not root motion it'll look fine

but i can't have them walk up and down the 1 way escalator

[slow bobcat]

Nav link proxies have an option to allow navigation only in one direction. Check the details panel

[sonic yoke]

does anyone know a good workaround for the bug in state trees that makes tasks finish before a move to is excecuted?

[slow bobcat]

Don't know that bug. Can you show your tree with the state and task failing?

[sonic yoke]

Well basically when I use a move to node it never gets to finish execute even when the pawn reaches the end. It's the same if I use the built in node or a custom one

[slow bobcat]

Sounds like an issue with your state tree. If you post an image we can try to help out

[hallow compass]

is avoidance what i need to make the AI try to find another root ? or should i place a nav modifier on each AI so it blocks

[slow bobcat]

You could use avoidance but that narrow space with so many enemies will not be easily solved

[hallow compass]

im trying with nav modifiers, but i dont know how i can have the AI to ignore its own modifier

[rotund peak]

ok so mass moves processing of homogenous ai actors off the game thread. at a high level, i want to move all ai off the game thread.

@small meadow I'm very interested in following and learning from your work on this. is there a git or write-up on your multithreaded ai framework?

[slow bobcat]

This is a bad idea. You are forcing re-build the affected nav every frame the nav modifier moves

[hallow compass]

mh, can i nudge the path finding to give him some points to follow or start with ?

feels like not a uncommon issue for close range fighting AIs so they have to be some not super heavy and long to implement solution

[slow bobcat]

I would try with crowd control first and see what happens

You could always implement a heat map, calculate how crowded the area is and pass that info to a navigation filter somehow

[hallow compass]

ill have to dig how to do that

[slow bobcat]

You will need to change the nav poly flag to indicate the "level of crowdness" and exclude said flags

[hallow compass]

ill come back in a few days or weeks for this, rn i have some higher task to do

but i appreciate the directions

[cyan sigil]

I am getting started with state tree. It seems like if I have a Delay task it still makes all tasks in a state run at the same time instead of in order. Is there a way to make them run in order?

[harsh storm]

Tasks are meant to run concurrently

If you want to delay stuff - make a sibling state

[slow bobcat]

I recommend to read "your first 60min with state trees". It's a bit outdated but the core principals are still current and well explained.

[cyan sigil]

I'm following that now.

[cyan sigil]

I managed to try that. I made the first sibling have a delay, then an explicit transition to the next sibling state. What happens if the parent runs and the first sibling fails? Will the parent try to pick the second state instead?

[slow bobcat]

There's a way to control the flow. By default it is set process children in orden. If the child state is not one, but two sibiling states, first one finishing will trigger the transition I believe, whether it's a failure or a success

[paper basin]

Hello. So I'm designing my behavior tree for a customer in a tavern management game. This is my first project with a behavior tree, so the first one was pretty bad. Now I'm in the middle of watching a course on behavior trees and I'm trying to design a better tree.
The first one I made looked like a state machine, so I think it is kinda hard for me to break out of this approach. I don't want to use state machines since I want to have more emergent behavior later (like customers reacting to weird things happening around them) and behavior tree seems better for this - at least to me. And it just feels better to make personally.
Anyway, I have a slight problem with the part highlighted on the screenshot. SitDown sequence can both fail and succeed during normal gameplay (fe if the ai got blocked or interrupted). So I added a decorator for "IsSitting" with NOT and "IsSitting" on the next branch.
But it seems.. weird. Is there a better way to design it here or is using the same decorator like this a normal practice?
I squished the tree a little and threw away unnecessary nodes. I hope the descriptions on nodes are understandable

[tranquil meadow]

what is the best practice for pathfinding in combat zones for large maps? I presume separate navmeshes on each combat zone right?

[slow bobcat]

Yeah it is common to use it like that. Usually you would have a selector with a branch gated by "is sitting" and then another branch that doesn't check (which is the same as "is not sitting")

[slow bobcat]

Do you mean in between combat zones?

[misty wharf]

I have a bunch of complicated BT's in my video rental store sim for NPC logic with some vaguely similar sounding ideas like Verael... I'm honestly thinking of revamping it in state trees because it feels more suitable at this point 🤔

It works but it's a little bit clunky to deal with error handling in particular, but also redoing them would be a lot of work so not sure if I'm going to feel like doing it lol

[tranquil meadow]

no. Basically parts of the map will have combat zones. I just want enemies to stay within their own combat zones.

I would've assumed it's just multiple navmeshes (one for each zone) but it seems that only one navmesh can be active in a level

[slow bobcat]

How are you handling your level? World partition or World composition (sub levels)? Are you using Dynamic navigation or static?

You should just need to add nav bound volumes where needed and have nav where they are

[tranquil meadow]

static navigation, i'm not using world partition (i think).

basically wanted a small level to explore, nothing big like open world

[slow bobcat]

Then just drop nav bounds volumes where you need nav and generate it. That's all you need

[night relic]

I wonder if there is built-in functionality to make an ACharacter to move through the navmesh following a spline

[night relic]

Just want to make sure I'm not reinventing the wheel

[steel imp]

The run eqs doesn't seem to run at all on my enemy character.
I have also added print string task, it just outputs the result of query.

When I press ' key and numpad 3... nothing shows
if I remove the print string task and debug, then it shows me eqs grid but only where the character spawned and not where it moved currently

Got it to work. Had to remove print from same state and add different state for it

[uncut cloak]

Im pretty new to behavior trees and state trees so im just practicing to learn the systems better. does anyone have any suggestions on videos or content that helped them learn how to make better ai?

[proud snow]

Hello, is there a way to achieve this sort of navigation for the ai? Instead of walking in a straight line, the ai would try to smooth it out, or interpolate it or something.

[slow bobcat]

You can try checking rotatin rates. Set a low rotation rate and calculate a path. The AI will slowly turn to face it, causing a curve like the one you painted

[proud snow]

How would i calculate a path? I'm using ai move to in the behavior tree, along with a target actor as a destination

[slow bobcat]

That calculates the path for you

Is what happens in the background without you noticing it

[proud snow]

Okay i feel like i misunderstood something. By rotation rate you mean the one in the character movement component, right? Thing is i already enabled use controller rotation and set it to 180~. The thing is, it really only affects the rotation of the actor, the ai itself still takes sharp/straight corners

[final prism]

Has anyone done leashing in Unreal? The only way I can think of properly doing it is by marking up nav polygons, otherwise the pathfinding could path outside of the range, or is there a better approach?

[slim egret]

Hi all, I'm trying to create a state tree that doesn't tick (UE5.6), so I set the Scheduled Tick Policy to Denied.
I have a global task to query the week day and day time (morning, evening etc) and want my state tree to evaluate every day time change.
I have a root that I tagged so I'll be able to SendStateTreeEvent to "force" the state tree to evaluate every day time change.
Is that the correct approach? (This currently doesn't work for me for some reason)
Also, would the global task evaluate every time I send such event, as I understand, global tasks evaluate only on tick, so I'm not sure what happens behind the scenes in a case of sending state tree event... does it considered as a tick or makes the tree tasks to tick for 1 frame?

[paper basin]

Hello, it is me again. I still have some trouble with understanding the correct flow for the behavior tree, so maybe someone with more knowledge could point me in a good direction - it can be a video or a blogpost, anything that could help.
Basically I keep being worried about Tasks failing in sequences. Especially the Move tasks. And it makes it hard for me to think around it. I attached my current design for the tree (It is for a customer npc at a tavern) - the place I'm talking about is marked with a red circle. Can this MoveTo fail? If yes, it would fail the sequence, which means my tasks for reputation and cost could run more times? How else could I make it work? Or am I just imagining problems?
General criticism is welcome too if I'm doing something completely wrong.

[slow bobcat]

180 is quite high. That means your AI can rotate up to 180 degrees per second. Try setting it to 35 por example

[slow bobcat]

I have use (usually combined) 2 flavours: one measures distance from the spawn location, the other whether the AI is within a volume. When too far or out of the volume, we interrupt the BT (call stop movement) and make it go back to the spawn location for example

[slow bobcat]

Denied means the tree will tick every frame (always). When enabled, the system will handle when to tick it or not

Even if you send an event, not ticking will not work. The events are processed in tick, same as tasks, transitions etc. You need your tree to tick to function

How it works is:

• you send an event
• the event is added to a collection (IIRC an array)
• next tick all events are processed (then transitions, then conditions and tasks)

Before everything else ticks, the global taks tick. They are the first ones

[slow bobcat]

Have you considered using the decorator Force Success?

You can add it to your move task and, even if it fails, the sequence will continue the execution of the flow

[paper basin]

Hmm, I didn't think about it, I'll look into it. But in this case it wouldn't fix the edge case problem but it could prevent the worst case scenario which is a good start

[slow bobcat]

I would suggest you star looking into organising things in sub-trees. That will help

[nimble smelt]

I just realized something basic I dont understand. I'm "launching" a missile by calling SpawnActor, and supplying the BP for AI missile pawn, and this works fine. But it must have a controller too, right? That's where all my detection logic lives.
So when an AI pawn is spawned like this, what creates the controller? How does it get assigned as the controller for this particular pawn? Where is that logic?

(This is all in aide of extending my save/load system to AIs. There's a lot in the controller I'd like to save but it's tough to restore when I don't know how they come into being)

[slim egret]

mmm I see.. So I need a different approach altogther. Thanks!

[tranquil meadow]

how do i generate? Does it do it automatically?

[final prism]

Oh I guess that's a lot simpler and makes sense

[slow bobcat]

Put a nav bounds volume on the level. Scale it to cover the area you want to have nav (you can use several volumes). The in the editor menu "Build -> build paths". I suggest you check a tutorial about basic concepts for nav in unreal. It will solve many questions

[paper basin]

You configure this in the Blueprint of the Pawn/Character. There are two things there: possession setting and the default AI controller. You need to assign it your controller and set possesion of AI to "Spawned" or "Placed and Spawned".
I don't remember the names of these setting, but if you type in the blueprint details posses and controller it should find them

[tranquil meadow]

Gotcha

[nimble smelt]

That's not the question though. If I set the pawn to be possessed on spawn, and then spawn it, where does the controller come from? What calls its constructor?

[paper basin]

Still not sure if this is what you are asking for but:
https://dev.epicgames.com/documentation/en-us/unreal-engine/API/Runtime/Engine/GameFramework/APawn/SpawnDefaultController

[nimble smelt]

That IS what I'm asking! So the pawn builds its own controller. That's politically dark, but very useful for what I'm trying to do

[slow bobcat]

Rather than make your missile a pawn and spawn an AI Controller (which is overkill if we are talking about a projectile), I would spawn a simple actor with a Projectile Movement Component.
You have a tutorial here about this approach
https://dev.epicgames.com/documentation/en-us/unreal-engine/implementing-projectiles-in-unreal-engine

[nimble smelt]

Eh, it's a guided missile that also exists as a standalone actor in level sometimes (which activates and pursues if the player gets too close). It needs cornering and wayfinding. I think the pawn and controller make sense

[slow bobcat]

Nothing you can't do with the approach of the link, but no wrong answers here. If it works for you as pawn... A pawn it is

[cyan sigil]

Can anyone provide an example of correctly declaring a new FStateTreeTask in C++?

[slow bobcat]

Just check how unreal does it. It's your best guide

If not... Check zomg blog about state trees

[cyan sigil]

I've managed, thankfully.

[subtle totem]

Hello guys should I use Behavior Trees for my AI or State trees . I am planning to have 15-20 alive enemies on average on my game. Which solution do you think offers the best performance ?

[harsh storm]

Either would be fine for that scale. Your issues will come in other areas.

That said - BT's are more stable and have more information about them and ST's is the direction Epic is going.

[celest python]

if you are just prototyping and has no experience at all BTs are easier to start with

BTs are designed with prioritization of behaviors, ST allows arbitrary jumps between states and has a nice hierarchical execution system and utility feautures embedded into it

[hallow compass]

Honestly State trees are better now

Super hard to keep a clean BT over time if you got a not simple AI behavior

[harsh storm]

That's just a skill issue

[harsh storm]

@slow bobcat Any neat info around AI from UE fest?

[harsh storm]

A bit late, but the reason this happens is because this line FStateTreeReadOnlyExecutionContext Context(GetOwner(), StateTreeRef.GetStateTree(), InstanceData); in the method. It is the first line in the method. If your component doesn't have a state tree, the StateTreeRef.GetStateTree() will return a nullptr. But the type FStateTreeReadOnlyExecutionContext requires a non-null pointer.

So that's why it crashes. It is at least fixed in 5.7

(In case you never actually figured out why it happened)

[slow bobcat]

Umm some.
SlateIM is getting there but light years from Imgui COG plugin.
Missed the network movement talk, but heard the thing has moved forward quite a bit, still far from being production ready.
Nothing new on AI, Siggi had to cancel his trip due to personal reasons and we never met.
Saw a talk where they profiled a game live with insights + the new Chaos Visual Debugger and that was very good.
And my talk was good it seems. Got people stopping me to ask for questions and advice during both days. Will post the video here as soon as it's up (might take months)

[hallow compass]

Never heard of ImGui COG

[harsh storm]

https://github.com/arnaud-jamin/Cog

[hallow compass]

Never heard of it

[restive imp]

Hello everyone !
I need some help with a basic task for my AI but i'm having some problem
My IA is able to see all my player and i want it to be able to sort it out by distance and always choose the closest target to him
(See Screen) can someone tell me where i did wrong ? https://blueprintue.com/blueprint/t7w3xgob/

[slow bobcat]

Almost nobody has and it the very best set of debug tools you will ever find for unreal. Unbeatable so far

[slim egret]

Hi, if I want my state tree to be purely event driven, does it makes sense to set the root tick rate to something like 9999999? or is there a more reasonable approach for this?

[paper basin]

Uh, your loop body is not connected, you use completed instead - which means it will chose the last actor that was in the array, since that will be left over in the foreach in most cases. Also your branch checking for Player tag probably shouldn't be connected out of False pin. Also why do you even have two loops there, the first one also connected only on completed...?
Also why do you name variables confusingly - if a variable contains a singular actor, and is called "ActorsSeen", it makes it confusing to read through

[restive imp]

The tag actor is connected on false because i didn't define the tag atm i've just put in for future purpose, i though completed mean it should take into account every player seen | it's shoud be better like that https://blueprintue.com/blueprint/ok7f6ofq/

[paper basin]

I recommend going back to a Blueprint course, because errors like this suggest a lack of basic knowledge - and I'm not saying it to be mean but you will have problems on each step without understanding simple things like loops and functions.
The problem now comes down to the return node. You return as soon as you find a distance closer than the default. SO basically the first actor in the array.
Return itself should be on completed, so the loop runs through all actors before returning.
Return in general means "Exit the function here". Even if it is inside a loop, the loop doesn't continue.

And you don't change the best distance after finding a closer actor, you need to also set it where you set the best target

[restive imp]

thank's for the input, i'm not taking it the wrong way don't worry but i'm here to learn so that why i ask which part i didn't understand well i'm not trying to create some AAA i'm trying to do a project for myself and learn i didn't use loop before so that why i'm having trouble understanding everything

everything working as it should now

[paper basin]

Of course, but the number of errors suggest that you are stumbling in the dark. I know it is popular to say "don't get stuck in tutorial hell" but at first it is important to go through a course or two so you don't waste time on said stumbling around - so when you need help it won't be 6 errors but 1 for example

[restive imp]

i understand but usually tutorial explain one precise things and when you want something a little more specific it's hard to find a clear explanation and i don't really know where to seek help, (english not being my first language also make some things harder to understand)

[warped tiger]

Is it possible to create grand-child classes of FStateTreeTaskBase that override the parent 'sFInstanceDataType? Context: I want to create a state tree task child that has different instance data

[misty wharf]

Isn't it just a matter of overriding GetInstanceDataType as you would need to do anyway?

[warped tiger]

I don't think so. FStateTreeExecutionContext::GetInstanceData(const T& Node) is reading T::FInstanceDataType which is an alias defined in FSTateTreeTaskBase subclasses and as far as I know, it can't be overriden in grandchild classes, right?

[misty wharf]

I use this and never had any problems with it

using FInstanceDataType = FMyInstanceData;
 
virtual const UStruct* GetInstanceDataType() const override
{
    return FInstanceDataType::StaticStruct();
}

(Similar to how the builtin tasks do it)

[warped tiger]

But how do you access the instance data in StateEnter, StateExit and Tick functions? Don't you use the function I mentioned before? 🤔 GetInstanceDataType() will only return the data type, not the actual instance data.

[misty wharf]

FInstanceDataType& InstanceData = Context.GetInstanceData(*this);

[warped tiger]

Is that working for you when you have ST tasks that inherit from other ST tasks that have different FInstanceDataTypes?

[harsh storm]

@chilly nebula Any neat things coming for AI in 5.7? The roadmap is pretty much void of any gameplay framework stuff.

[misty wharf]

I don't think I've actually tried that... I could imagine it being problematic 🤔

[harsh storm]

As long as you override the GetInstanceData - you should be fine

Because that returns a UScript*

So it doesn't matter what the parent overrides. Unless you're trying to do some parent call

In which case, then just have your instance data struct also inherit from the parent's instance data struct

[warped tiger]

That's the problem, if the child InstanceData does not inherit from the parent one, the returned instanced data will still be the parent one because I can't override the
using FInstanceDataType = SomeType; alias in the child class, right?

template <typename T>
typename T::FInstanceDataType& GetInstanceData(const T& Node) const
{
    static_assert(TIsDerivedFrom<T, FStateTreeNodeBase>::IsDerived, "Expecting Node to derive from FStateTreeNodeBase.");
    check(CurrentNodeDataHandle == Node.InstanceDataHandle);
    return CurrentNodeInstanceData.template GetMutable<typename T::FInstanceDataType>();
}

[harsh storm]

You can

If it doesn't work - just don't inherit from a task that is actually filled out.

[robust veldt]

I have output parameter AttackRange in ST, but chill state can not access it, always use default value. Does anyone has idea.

[warped tiger]

You're right, I can. Unfortunately if I call a parent ST task function that uses InstanceData it will use the parent's InstanceDataType rather than the child's

[harsh storm]

As expected

[harsh storm]

It depends on your tree setup. Keep in mind that all tasks in the active tree will run. So, the parent state's task that actually calculates the range may not complete before the child state accesses it

And it depends on when the bound properties are actually copied

Which I've never looked into

[harsh storm]

TIL that if you fill out the description in a State Tree condition - the ST will display that for the If text in the tree.

[dim gull]

Getting some mixed responses on the performance of state trees, I can profile myself but I'm hoping somebody here might know -

For 20k+ units with potentially different behavior (different targets, different attack states, etc.) is StateTree an appropriate AI solution? All single player, if that matters at all.

[harsh storm]

I'd imagine it should be doable. Considering STs were kind of designed for #mass. So that's the channel that would probably know more on how to handle that.

[robust veldt]

https://tenor.com/view/guy-long-hair-mustache-magic-shia-labeouf-gif-17741293

Thanks. I unbind/rebind property several time and it is working somehow.

[cyan sigil]

When using State Tree DebugText, where does the debug text appear?

[harsh storm]

It used to show on the actor many moons ago - but it has been borked for a couple of versions now.

I think its like in the corner of the screen or something now.

I can't remember. I don't use it anymore.

[cyan sigil]

I found the problem. Somehow the State Tree Component is saying the State Tree reference is invalid, even though I have it set to a state tree. 😠

{
    STATETREE_LOG(Log, TEXT("%s: Start Logic"), ANSI_TO_TCHAR(__FUNCTION__));

    if (!StateTreeRef.IsValid())
    {
        STATETREE_LOG(Warning, TEXT("%s: Trying to start State Tree component with invalid asset."), ANSI_TO_TCHAR(__FUNCTION__));
        return;
    }

    FStateTreeExecutionContext Context(*GetOwner(), *StateTreeRef.GetStateTree(), InstanceData);
    if (SetContextRequirements(Context))
    {
        const EStateTreeRunStatus PreviousRunStatus = Context.GetStateTreeRunStatus();
        const EStateTreeRunStatus CurrentRunStatus = Context.Start(&StateTreeRef.GetParameters());
        bIsRunning = CurrentRunStatus == EStateTreeRunStatus::Running;
        
        if (CurrentRunStatus != PreviousRunStatus)
        {
            OnStateTreeRunStatusChanged.Broadcast(CurrentRunStatus);
        }
    }
}```

Don't know what I'm doing wrong. Does a State Tree Component have to be attached to an AI Controller? I have it on my player character because I just want actor reference.

[harsh storm]

It does not have to be - no. Unless you're using the AI schema

[cyan sigil]

Any idea why StateTreeRef is always Invalid?

Reference is set.

[harsh storm]

Turn off start logic automatically

Then start it manually

(Though, it should still work with it on)

The only reason I could think at the moment that it would be invalid is if you're spawning the wrong class

[cyan sigil]

Seems to have worked after clearing and resetting the reference in editor. However, the print is still not appearing attached to the character, and the delay is being skipped entirely 🙁

Any idea what I'm doing wrong?

InstanceData.ReferenceActor is always null in the DEbugTextTask

[robust veldt]

Check Visual Logger

[cyan sigil]

OKay this is weird. Every time I restart my editor, the state tree is considered invalid compile.

It worked properly the first time I ran it, but not after.

And the instance actor is still null!

Ahhh I figured it out. I had to actually link the Actor parameter.

Still getting the compile error.

[cyan sigil]

I'm not quite there yet, but I will need this: In state tree, I want the character to be able to respond to an input "P" or "K". In addition, it should have a mounting counter of how many Punches and Kicks it has done since going back to the root state. What's the best way to track that counter?

I guess it would be a service within the Punch/Kick parent state.

And a variable at the State Tree level.

I'm starting to SEE

[wide robin]

I would probably:

1. Send a callback when you throw punches and kicks (either when you press the input or when you actually execute the punch and kick action)
2. Make a global task that:
   2.1 On enter listens for that callback and every time increases the counter, that are exposed as Outputs
   2.2 On exit resets the counters
3. Assignt that global task to the subtree you'd want to use

[chilly nebula]

Theres a few new things for state trees, including debugger integration with rewind debugger and finally debugging a specific instance.
Also, multi config support for auto navlink generation

[cyan sigil]

Auto navlink generation?

[slow bobcat]

Why wasn't this mentioned at all at Unreal fest?

Watched the 5.7 talk and it was all render and animation stuff pretty much

[shadow furnace]

it's exactly what it says it does... it's in project settings for ai stuff

instead of manually placing navlinks it tries to find edges and test for nice places to make new navlinks

[hallow compass]

im working on some AI perception system for my game, but it has some common grounds with what you would have in combat games.

here is a very broad graph that shows what I plan to have, im interested in any suggestions people might have from their experience.

this is for an open source plugin on github, im not working on a paid product.

[static raven]

Hey Guys, I'm looking for a good example of how to setup NavMesh for big world partition maps. Is there any Epic demo project I can take a look at? Valley of the Ancient implements nav mesh?

[mighty geode]

Can somebody help me understand and explain how can I achieve what days gone did, with their hordes?

I wanna implement horde of zombies such as 200, 300 or more

[harsh storm]

That is a very broad question

[mighty geode]

Like?

[harsh storm]

It involves optimizing logic, animation, bones, models, CMC, so on and so forth

It is a very very broad question

Use navmesh walking in CMC
Look into animation sharing potentially
Don't give full logic to every zombie until alerted - have them "follow the pack leader" pretty much
Minimize how many bones the model has

Just a few things off the top of my head

[mighty geode]

I was thinking it in such a way.

1. Reduce Lod

2. Have a sphere radius, it'll work as a pool and other zombies will share the animation from that lead zombie (Problem: if all zombie would have sphere, who'd they follow)

3. If they overlap player (like acceptance radius) the zombie would be in full alert, independent state.

4. Use dynamic NavMesh.

Problem: choose player target (multiplayer)

[static raven]

having a SphereComponent in each zombie will kill your performance

[dim gull]

Currently, there is no way to dynamically add states to the state tree or anything like that, is that right?

or, I guess a better question could be, what dynamic elements currently exist for a state tree, if any?

I guess you can always make a state that represents some abstracted thing and modify runtime values / settings in order to have more dynamic behavior

[chilly nebula]

Probably too much material to cram into one presentation would be my guess?

[slow bobcat]

Booo! Give us the AI! We don't care about how many polygons and lights you can use! Booooo

XD

[harsh storm]

I just want performance out the wazooo

I want an empty scene to be able to run at 2k fps like it does in Godot 😅

[slow bobcat]

Imagine... Moving 100 enemies without praying to the gods

[celest python]

Best we can do is 25 characters bottlenecked by animation overhead while playing montages

UAF will save us.. in the not near future

[harsh storm]

I reckon it'll be a UE6 thing sadly.

[chilly nebula]

it absolutely will 🙂 you saw what it can do in the Witcher demo, and that wasnt even finished

[cyan sigil]

Seems like a Service for the required state scope would be ideal, right?

[heady silo]

How do I pass tags into the AI Controller? I'm trying to do a detection system where an agent has the ability to recognize other agents by faction. So for at least the initial test, I want to filter perception to only members of a faction other than its own to select as a target for combat. But I can't seem to find a way to pass that information into the AIC to influence the perception update. So everyone just shoots at each other unless I have them only able to percieve a player tag, which would be fine if all factions are hostile to the player, but I want something a bit more complex where enemies can prioritize each other for combat and only attack the player if they pose a more direct threat (they're closer, or damaging them more directly, etc.). Anyone have a good way to do this? What am I missing?

[static raven]

I would use GenericTeamAgentInterface. Its already integrated with PerceptionSystem.
You can configure senses to only process stimuli coming from actors with specific relationships (for instance: hostile, neutral or friendly)
AIC already implements the interface.

I don't know how much in-build blueprint helpers are for this system, but yeah..., I would first use what is already there

[mighty geode]

That's the problem

[keen crow]

if my NPCs can vault over/climb obstacles under 2 meters and safely fall down from 2m height, is it a good idea to change navmesh agent properties to set max step height to 2 meters and then make a MoveTo task child that, while active, would do shape sweeps to decide if the NPC needs to jump down/vault/climb? An alternative I see is to try to use auto generation of navlinks, but I'm kinda worried that my project is an open world game and I doubt hundreds of navlinks would go unnoticed performance-wise 🤔

[hushed crag]

any idea why slots ina uniform grid panel are spaced unevenly?

the padding is all the same

inconsistent spacing

very annoying

[misty wharf]

Are you confusing AI with UI? :D Try #umg

[hushed crag]

OOPS

[dusk condor]

Following this tutorial to set up Mass Traffic
https://youtu.be/RRWr_Hnn5Bg?t=938
Where I bookmarked the video the author says you must set the following cVAR
ai.traffic.parallelize 0
to avoid crashing.

Problem is, this CVAR does not exist in 5.6!
And of course I crash on simulate with the following error
Assertion failed: ZoneGraphStorage [File:D:\build\++UE5\Sync\LocalBuilds\CitySample\Windows\Plugins\Traffic\Source\MassTraffic\Private\MassTrafficInitInterpolationProcessor.cpp] [Line: 47]

Any Zone Graph Mass Traffic AI experts able to help?

Possible solution
https://forums.unrealengine.com/t/mass-ai-traffic-crashes/874813/7

[dusk condor]

Anyone experienced with AI based traffic tools in Unreal?
After spending a few days exploring the native MassTraffic AI plugin (part of City Sample) I've given up.
Prone to crashing, almost no documentation. The only Youtube video I cam across is several years old.

I came across this in FAB and was hoping someone could comment
https://www.fab.com/listings/0c8bafd7-aca4-4b8c-a11d-38741d744b54

Also a more expensive option, probably way more features than I need
https://www.fab.com/listings/b4aedd9f-2ca9-499d-b283-0d9864520a0d

[near condor]

Built a custom logging system for the state tree system. I have learned how the current system works but I wanted to try out how a more "legacy" approach to logging the state changes would work for debugging.

[ember pilot]

hey guys, i have an enum and try to use it in my blackboard but i cant find it:

[harsh storm]

There might be. Look at StateTreeTest.cpp and you can see how they're adding states and transitions in pure C++. I don't know if it persists and all that either.

@slow bobcat Tagging you as well because it might be something you're interested in as well

[cyan sigil]

It's time for me to add a way to have my state tree track player inputs. Any ideas?

[heady silo]

How do I access it? Every AI tutorial I've seen so far has been like "check these boxes for detecting and don't think about them" because to actually do stuff with this it's still locked behind C++

does StateTree handle this sort of thing better?

I've mostly been sticking with BTs for now because I at least can get the AI moving and firing, but I still feel like I'm not getting anywhere

[terse coral]

I just have a issue anoying :
I have a Main Level Persistant that contains several Sub Levels,
while I disable a level's visibility there is automaticaly a building navigation,
then while I play my game and load stream that level, its like the navigation (nav) mesh do not exist,
Its why all of AI in that level wasnt moving : if (!NavSys->GetRandomPointInNavigableRadius(TargetLocation, 0, NavLocation))
that line is supposed be always true, but cause that building navigation after disabling level's visibility, it come always false because there is no navmesh no more,

In project Settings, I could set RunTime Generation to Dynamic, it fixe all navmesh issue, but the game is stuck under 10 fps, I dont understand why people tell to put it to Dynamic, and Unreal do not give a function to Rebuild navmesh by uself at runtime or Idk,

but for now one solution to fixe my issue is to enable all my sub levels visibility to true, but that's not the real good way to fixe it, if I have 100 sub levels with very large map, I would have 1 fps

[cyan sigil]

I'm having some trouble. I'm making a State Tree Task that iterates through a map of Input Actions and binds to the input component, hoping to send a gameplay event to the tree once the action is triggered. However, I can't seem to figure out how to properly bind a struct like FStateTreeTaskCommonBase to input action, especially because the callback will not include the proper instance data.

[slow bobcat]

Yeah saw this some months ago, but it was not very useful for us

[slow bobcat]

What for? What's your use case?

[cyan sigil]

I'm using a state tree to handle how my player character performs different melee combat moves based on button presses and state.

Example: They have a Punch and Kick button. There are 4 punces and 4 kicks. Pressing each button uses the ability at that point in the combo. So you could do Punch 1, Kick 2, Kick 3, Punch 4.

Or Kick 1, Kick 2, Punch 3, Punch 4.

State Tree seems like a good way to accomplish that.

While also integrating other kinds of moves.

[slow bobcat]

For level streaming with nav:

• add a nav mesh bounds volume in the persistent level (doesn't matter where, we add a small one at 0 0 0)
• load all your sublevels in editor and place nav mesh bounds volumes in the sub levels where you need them
• build navigation as static (pure static or Dynamic with modifiers only)

What will happen is:

• because you places a nav mesh bounds volume in the Persisten level, you ensured that the Recast object will be generated in that level and will be always loaded
• because you places nav bounds volumes in the sub levels, as you stream in/out sub levels they will stream in/out the nav data they own (each sub level has something calles Nav Chunks where that data exists). The data will be added/removed to/from the recast object

[slow bobcat]

But wouldn't this be a problem because you will interrupt the state for kick 1 to play punch, 2, causing the anims to stop before they should? Don't you need to accumulate the buttons pressed over the last X frames and play a sequence of animations based on that?

[cyan sigil]

That's easy to avoid via transition conditions.

[slow bobcat]

Ah I see what you want. My suggestion would be something through the enhance input, register the input happening in a global task and expose it through an output array that contains all the buttons pressed in the current frame

[cyan sigil]

I was anticipating using a Global Task for the tree. I can see the value of using an array.

However, there are advantages to using tags that I want to pursue as well.

Though perhaps I can go input -> Array, and then array-> tags?

[slow bobcat]

Massive fan of g.tags here so yes, go for it. Much easier to handle since you can do tag queries etc. As an example, we have over 1200 tags in our game (counting all parents). Impossible to handle everything we do any other way

[terse coral]

Thanx for answer,
but I'm not sure, I added a nav mesh in the persistant level,
then I added a nav mesh to each of my sub level (that need),
I load all my sublevel (visibility true), that make building navigation,
but if I unload (visibility false) my sub level, that still make building navigation,
so its still the same issue in game : no nav mesh,

do we are supposed always keep all of our sub level loaded (visible) before playing in editor, and also before build the game ?

[slow bobcat]

• make sure you use static or dynamic with modifiers only
• disable rebuilding nav automatically in project settings
• build the nav with the whole game loaded

That's should work. Are you using streaming volumes to load/unload sublevels?

[terse coral]

Thanx,
I fianlly got it,

• actually stop navmesh from building automatically in Editor Preferences->Leve Editor-> Mescellaneous
• then manually Build -> Build paths

[lyric zodiac]

Not sure if this is the best place to ask as it's a question involved with state tree tasks but also GAS. I want to create a state tree task that takes an ability class and dynamically populates properties of the task to reflect the properties of the ability class itself so that they can be edited/bound from state tree.

The goal is to be able to grant and activate abilities with custom values throughout state tree for enemy AI.

The ability params would be dynamically constructed based on the ability class passed in. I am able to see the variables reflected but their types are not truly reflected and I cannot bind to them like you normally would. I'd like to just expose blueprint class specific variables (not base CDO parameters such as bMarkPendingKillOnAbilityEnd, for example) and have these variable types as boundable properties in state tree

Has anyone does something similar for state tree and granting/activating abilities for NPC? Any advice would be appreciated 🙂

[lyric zodiac]

The closest that I've come, that works, is having a matching struct parameter in my ability and in the task. The task is responsible for taking the property values, and copying them over to the Ability CDO when its granted/activated. However, this method requires that I either:

• Create a holistic struct that can satisfy every possible need across all abilities in the game.

• Create a unique struct and unique state tree task for each ability so that it can be read from state tree in the same way.

Neither is ideal and I would prefer that a single state tree task read the parameters of the ability class passed in and dynamically expose them as boundable state tree parameters that would then be passed to the ability itself when granted/activated.

[slow bobcat]

We do this in Bt's. What we do is:

• we have UPROPERTY's designers can tweak in the task (in your case this will be params) that are things the GA will need
• when the task runs, we pack all those values in the GA payload's optional object using a custom struct
• within the ability code, we grab the payload optional object, cast it to our type and read/use all the data

To give an example, we have the Task Approach Target, where designers set an Actor bbk representing the goal actor and things like if the AI should walk or run, the anim to play, the min distance from the target to reach an many other things.
We pack that data designers set (plus a bunch of internal data we gather in code) and send it in the payload of the GA. When the GA_ApproachTarget runs, it reads the bShouldWalk value from the payload and changes the character velocity based on that. It read the montage it should play for the approach. It launches a MoveTo async tasks and for the movement request distance we pass the Min Distance to Reach from the payload

[slow bobcat]

As you can see it's pretty much what you described here. I don't think it is possible do a reflection system that will read a GA and automatically create a class task with all the params. At least not "on the spot". You could definitely have a script that generates c++ code for a tasks when passing a specific GA

[lyric zodiac]

Yeah that is what I am doing the example that I have working; I just wasn't sure if its the best way to go about it; I guess I could generalize categories of abilities like Grant Reaction Ability, Grant Melee Ability, Grant Projectile Ability, etc; having structs that contain parameters that we know we'll need in each ability category and then copy those over in the optional object of the payload

[slow bobcat]

We kinda do that. We have general ga's we inherit from for particular cases. General Ga's have a Payload struct we inherit from for the the children ga to add their specific stuff.

[lyric zodiac]

Thank you for the insight 🙂

[hallow compass]

for some reason, once i transition to ChasePlayer to ChargePlayer, ChasePlayer tasks seems to not be executed at all

the transition is evaluated on tick, but it can only be suceed once (because the charging bool becomes true once ChragePlayer state is executed)

[hallow compass]

also, how can i make a On State Suceeded transition on leaf state take over the Tick Transition placed on parent ?

[slow bobcat]

Are you waiting for all the task to finish or is the first one finished triggering the transition?

[hallow compass]

im not sure what you mean by that

[ember pilot]

heyyy i have problems with implementing a C++ Enum into my Blackboard someone able to help?

[slow bobcat]

In a state, by default, the first task to finish will trigger the state transition, which can cause other task to never execute or be interrupted. In 5.6 you can define if you want that or if you want all task to be finished before triggering the transition

[slow bobcat]

You posted code right? Can you link it again please?

[hallow compass]

well here the transition is done via checking conditions on tick

so ChasePlayer runs, then at some point ChargePlayer will, but tasks in ChasePlayer arent re-entered anymore

[ember pilot]

yep

[slow bobcat]

What does the state tree debugger say about your problem?

[ember pilot]

.

[slow bobcat]

Umm code seems fine. Could it be it didn't compile? Had that in the past. Try using the enum in a cpp file somehow and re-compile, just to be sure

[ember pilot]

for context - i made today a new file called VillagerJobEnum.h : ```#pragma once

#include "CoreMinimal.h"
#include "VillagerJobEnum.generated.h"

UENUM(BlueprintType)
enum class EVillagerJob : uint8
{
Farmer UMETA(DisplayName = "Farmer"),
Lumberjack UMETA(DisplayName = "Lumberjack"),
Soldier UMETA(DisplayName = "Soldier"),
None UMETA(DisplayName = "None")
};``` but it didnt change anything

[slow bobcat]

But if you use that enum in a cpp, does it work?

[ember pilot]

and if i try to use the enum in my Villager Blueprint i see every Job

[slow bobcat]

Ah ok

[ember pilot]

ye

i just cant see it in Blackboard

[slow bobcat]

Umm weird. I don't recall doing anything special to use the enum in a bbk...

Can you see any other enum declared in code from the engine?

[ember pilot]

i dont know why i cant see it

[slow bobcat]

Check how any of those enums are declared in the engine source code, in case there's something missing

I'm not going to be in front of my pc until next week so I can't double check on my side

[ember pilot]

where do i find them?

[slow bobcat]

The reaps you see it here is because you are declaring the property with your enum type. Bbk 'a are a bit different

[slow bobcat]

Engine source code. If you are compiling your own code for the game, you should have it

[ember pilot]

ye i only need the path

[harsh storm]

There is a special way to get Enums to show up in the BBK

I forget how though

[slow bobcat]

If you use ctrl+shift+F the enum name and reuskts will be thrown, check files

[slow bobcat]

This is buzzing in my mind but I can't recall either

[harsh storm]

Type in the Enum Name @ember pilot

[ember pilot]

im a little stupid to find it lol

[harsh storm]

Right below the type

[ember pilot]

like that?

[harsh storm]

Yeah

[ember pilot]

ok and now?

[harsh storm]

You're done

[ember pilot]

ok it works thank u so much

love u guys @slow bobcat @harsh storm

fcking magic

[slow bobcat]

Wait what?

Why couldn't you see it before and can now? You already searched for your enum name before no?

[ember pilot]

i had to enter the name manuelly in the enum Name thing and then you can see all enums in ur tree

the enum type window is ass

so stupid

dont know what epic games thought when they programmed it this way lol

[harsh storm]

It was done like 11 years ago

UE was much different back then

[slow bobcat]

Umm I don't recall writing the name there but haven't looked into this for a very long time

[ember pilot]

it works so we dont ask xd

[paper basin]

I thought I'm going to be smart (a'ka lazy) with the Behavior Tree Tasks and just make a task parent with all the key names, so I don't have to create them in each task separately.
But then I noticed it does this during play and debug - basically displaying every variable even from the parent task:

Sigh. It's going to be bad if it continues like this. Can I turn it off somehow or is my lazy approach doomed

[harsh storm]

I'd say your lazy approach is doomed

And your overall design is doomed

Requesting the chair shouldn't need so many keys

Tasks are meant to be small, reusable things

[paper basin]

It is not going to be anything big nor complicated, I just didn't know it will obscure the debuggin with all of the parent variables though

[lavish vault]

I'm trying to make a blueprint interface between two states of two different actors, but it's not triggering. Both have implemented the interface, but I suppose the interface must be implemented on actor as well? Not sure what to use as Target on the bpi call either (how does this matter?)

[harsh storm]

The target should be what implemented the interface

So it should be some kind of reference

[lavish vault]

hmm yeah that is difficult because how do i reference the instance of a specific state tree task of another actor? if i was able to i'd probably not use an interface to begin with.

the problem is basically that I need a way for two different actors/AI's (with state tree) to coordinate with each other, acknowledge each other and execute logic in turn. Like "ActorA: hello, can you move to this location" -> "ActorB: acknowledged, moving to position x" -> "ActorA: thank you. (exit state)"

[harsh storm]

What is your end goal? Like - what problem are you trying to solve with this approach?

[lavish vault]

I have two cats that are meeting each other. They need to figure out the following:

• Is the first meet positive (become friends or enemies)
• As this "befriending"-process (states) are on-going, they will circle each other, meow etc
• Depending on the result (of befriending process) they will react differently in terms of movement, animations etc.

Since they are running separate states which might be triggered at different times (seconds difference) and with an unpredictable deviation of distance, I can't estimate exactly where cat x is in the process. I basically need a way to have them talk/update each other

so far i've gotten away with just having sub/child states etc, like for when they're fighting. But anyway, I want to figure out a way to do this communication more directly

[harsh storm]

Get a reference to their state tree and send a state tree event

[lavish vault]

yeah like when you send with a gameplay tag? i do that with some things, but feels like kind of a nuclear option/reserved for major state changes.

but guess i might as well

[misty wharf]

You could always abstract it behind an interface or something so you're not directly calling it from the other NPC

[harsh storm]

Live Property Binding: Tasks can now push value updates to bound outputs dynamically during execution, resolving issues with outdated or static data in async/tick workflows.
👀

[wide robin]

I also support the idea of using tags. Was away during the weekend so didn't check replies here :)

Did you manage to make this work?

[cyan sigil]

Havent tried yet. Still converting to instanced task

[pine quarry]

Hey all, I was adviced to maybe and try to ask here, but if there is someone here from Epic that can have a look at a PR I've created for the UE Source Code, I would really apprecaite it 🙂
https://github.com/EpicGames/UnrealEngine/pull/13719
It's a really small QOL improvement to the StateTreeDebugTextTask class.

[slow bobcat]

Maybe @chilly nebula

[lone heart]

Hey!
I been using State Trees recently and I was wondering if there is a way that an state is marked as completed only when all its tasks succeed? They way it seems to be working is that when the first task is marked as succeded will determine if the state succeeded.

[wide robin]

Yes, if you're on 5.6

This is what you're looking for in the detail panel when clicking on a State

[lone heart]

Great, that's what I was looking for. Thanks!

[runic flare]

Hello, I'm trying to use an enum match condition in my tree. I have defined a public BlueprintReadOnly member variable on my AIController. In my schema my AIController class is set to my custom class, and it's available as Context. However, I don't see my enum variable showing up to bind to in the condition. Is there anything else I need to do to get it to show up? (I'm on 5.5 currently)

[runic flare]

Ah I made it BlueprintReadWrite and EditAnywhere, works now. I tried using a provider function but seems like that might only be available in 5.6...

[signal island]

is this correct to pass a struct to the context?```cpp FStateTreeDataView Data{AIInstanceData.StaticStruct(),AIDataInstance->GetAIInstanceData()};

ContextDataSetter.SetContextDataByName(AIInstanceDataName,Data);

[signal island]

you can just ```cpp ContextDataSetter.SetContextDataByName(AIInstanceDataName,FStructView::Make(*AIDataInstance->GetAIInstanceData()));

[near condor]

@dim gull can I DM, I had some questions for ya

[dim gull]

Here is fine, DM is fine, np either way

[paper basin]

Hello. Could anyone point me to something (text, video, example) that would help me learn how to properly implement Blueprint Behavior Tasks that take time to complete? Like the default MoveTo or Wait.

[regal creek]

anybody have any ideas why AI doesn't work only in Standalone game mode? in PIE i can click my interact button and the npc will start Moveto'ing and the "walking" print sting will fire, but in standalone "not walking" fires.

I spawn this actor in via button press and i have of course set the auto possess to place in world or spawned

[wide robin]

Did you check if there's a navmesh?

[regal creek]

Without it it wouldnt have worked at all

[wide robin]

I'd use the Visual Logger then to see why the move to it's failing

[lyric flint]

hey guys how can i differentiate between senses like for example hearing and sight

[misty wharf]

You can use get sense class or something like that from the stimulus value and compare it

[lyric flint]

i tried thatt but cant seem to find the function or wtvr to get the class

[misty wharf]

It should be called "get sense class for stimulus"

[lyric flint]

ah im in cpp ill see how to call it in cpp

thank you

[misty wharf]

Oh, yeah if you're in C++ then you can probably just check it from the stimulus struct directly

[regal creek]

man what the hell, i woke up, turned my pc off, started my project, wanted to mess around for a bit, spawned my npc, clicked on it and now it walks, all in standalone mode...

i mean im glad but man why cant it save me the frustration and just work without a pc reset

[sinful geode]

Hello guys!
May be anyone had something simmilar before. AI pawns tend to select not the shortest path in some cases. There is no crowd avoidnace or RVO enabled. I'm using basic MoveTo task in BT.
Some times it desides to take the red path over a yellow one (witch will obviously be shorter)
It is because AI don't want to use another navmesh cell. Is there an easy way to fix it?

I know that I can make cells bigger but still in some places that will remain an issue.

[sinful geode]

Lowering the Heuristic Scale of a NavMesh helped a bit. But still in some cases it happens

[worn mesa]

Is there a reason that a behavior tree can be inactive despite being connected to a pawn? I’m not sure if it’s conflicting with something but everything seems like it should be connected

I’ll send screenshots in a few minutes

[worn mesa]

What does steps back: -1 mean?

[dense owl]

Did you use run tree node on it ?

[dense owl]

I think that just shows you how far down the tree it is. Root is before the first step

[worn mesa]

Hm

Alright

[worn mesa]

I don’t believe I did

[dense owl]

Well for the tree to run you have to tell the AICon to run behaviour tree

[worn mesa]

Had to run out for a bit but I’ll be back with more screenshots of what I’m doing

[dense owl]

Np. Also make sure the pawn is using the correct AIController

[worn mesa]

They should be….. I think😂
Pretty sure they’re connected where they should be

This is a local 4 player game by the way. But I’m also running into an issue where my behavior tree simulates correctly when there are no players connected (all ai players), but if I connect a player it breaks and stops running

[dense owl]

Ah… #multiplayer has its own other set of rules tho

You could be looking at issued stemmed from that

[final prism]

Does anyone have any ideas what could cause a custom navlink to disappear at runtime but generate fine on the initial navmesh build?

I'm not even sure which part of the engine I should debug to see what on earth is happening lol

[wise basin]

hello can anyone help me solve the issue where the AI are looking at eachother's feet instead of their face?

i assume this is because they are focusing on eachother's root bone

i tried spawning a invisible actor at their face but it messes up their perception

[static raven]

If I remember correctly, AIController uses AActor::GetTargetLocation() to get the actual location to use as FocusLocation

nope, I was wrong, it uses AIController::GetFocalPointOnActor()

what I did was to override it and call GetTargetLocation of the target actor inside GetFocalPointOnActor

default implementation of GetTargetLocation is the ActorLocation, which are the feets

so I overrided GetTargetLocation to return whatever I want to be the optimal target point for my characters

AIController::GetFocalPointOnActor is exposed to BP

but AActor::GetTargetLocation doesn't

[wise basin]

Oof, i dont know how to write cpp

but thank you for pointing me in the right difection

[static raven]

you should be able to override AIController::GetFocalPointOnActor, thats the important part, you can use your own BP function instead of GetTargetLocation to ultimatly define what is the optimal location for each of your characters

I hope I'm not overcomplicating my explanation

[wise basin]

this is exactly what Ill try to do

thank you very much

im kinda ok with bp

ill try to make something like this with timers and calls from behaviour tree

[runic flare]

Can someone explain to me the use case for Output variables in State Tree? Are they meant only for global tasks (or Evaluators)? What I would love to do is have my Patrol state run its FindNextLocation task and Output the location, and then have my subsequent MoveTo task bind its Input location to that Output. I technically CAN do this by running these tasks "sequentially" within the same state, but I've read this could result in unexpected behaviour. I can simply have Task A write to a property ref that Task B can read from, but I find that obscures the data flow and I much prefer the idea of explicitly binding an Input to an Ouput. Any thoughts are appreciated.

[slow bobcat]

I would recommend reading Your First 60min with State Trees, it's in the pinned messages

It clarifies use cases and base functionality

[slow bobcat]

What type of navigation are you using? I would assume dynamic?

[paper basin]

What sort of AI system would you guys suggest to use for a simple AI that has a rigid schedule? Like an AI that is supposed to be at different places depending on the time of day and physically move between them.

• coded through a controller? It will work, but maybe not the first solution I would go to
• behavior tree? Seems kinda pointless for it, since the schedule will be sort of static
• state tree? I didn't use them yet, so I don't know
• something else?

[misty wharf]

This is a surprisingly common question I've noticed :)

[misty wharf]

I don't think it really matters how you implement it. Any of those solutions would work for it. I would probably look at state trees at this point, because it sounds kind of stateful - eg. states like "moving to scheduled location", "doing scheduled task", "idle" (eg. after finishing scheduled task but schedule hasn't changed yet), etc.

[paper basin]

Hmm, maybe I'll use state trees for it then, just so I can learn what they are about at the same time.

[final prism]

modifiers only, but I've sort of figured out it's an issue with the projection. The custom navlink component projects via geometry instead of navmesh, which causes some issues with our custom geometry

[slow bobcat]

We should ping a FAQ where one question is that and the other just says "have you check visual logger?"

[misty wharf]

haha

[harsh storm]

No one reads pins anyway

[slow bobcat]

A pin about reading the pins

[dusk condor]

Is anyone expereinced using the new AI 'Production Assistant'? What are its strengths and weaknesses? Would you use any of the suggested code in production?

[odd kestrel]

Hello everyone. I'm using StateTrees to control enemy logic, and when an Enemy dies, I call StopLogic in the state tree to, well, stop the tree. Sometimes, however, I get the following errors. Does anyone know what is causing it? Is there another way to stop a StateTree?

[wide robin]

Where are you calling the StopLogic from?

[odd kestrel]

I'm using GAS and have an attribute listener that tracks health. When health reaches zero I call a function to cancel abilities, remove effects, etc. At this point I get the controller for character and do Controller->BrainComponent->StopLogic(FString("Character Died"))

[wide robin]

Have you tried to fetch the StateTreeComponent and run StopLogic there? Instead of doing it on the BrainComponent

[odd kestrel]

BrainComponent is the StateTreeComponent instance in this case

[wide robin]

I remember doing that with a cheat for our project and it worked just fine, so I don't think I have other ideas about what might cause it

[odd kestrel]

It's a bit of a weird issue. I have other state trees that are just fine. Not sure what is different about this one and why this is happening. I was looking at the code and is almost as it expects the state tree context to be in an specific state so the tree can be stopped or something

[runic flare]

I want to create a state tree task where I can bind to a delegate and update task instance data when that delegate fires (or at least queue an update). This isn't possible (afaik) with FStateTreeAIActionTaskBase or any other struct parent because I can't ensure UStruct validity when the delegate fires. I'm thinking that overriding UStateTreeTaskBlueprintBase makes the most sense since it's a UObject but I've seen some comments suggesting that's not ideal -- thoughts?

[slow bobcat]

You can do this. The delegate should bind to a function in the task instance data struct itself instead to a function in the task object. I do this in some places

[runic flare]

oh interesting, had not considered that… thanks will give that a shot

[grim reef]

Hello everyone. I wanna ask some questions. Does anyone here use NVIDIA ACE or NVIGI in Unreal Engine. Not many instructive vids about this so I want to consult the way to implementation in UE as well as you guys’ feeling when using this tech from NVIDIA. Thank you

[fair raven]

Hello All. Why does my enemy AI’s perception fail to detect me when I’m standing still? It only detects me when I move into its perception area, but not when the enemy moves or looks in my direction while I’m stationary.

[slow bobcat]

Which function are you using to check for detection?

[fair raven]

UFUNCTION()
void PawnSeen(AActor* SeenPawn, FAIStimulus Stimulus);

[slow bobcat]

Is that an engine function? From which class?

[misty wharf]

Ugh why is this so complicated... I'm working on creating a task system for my game's NPC's, so that for example, when a "storage slot" becomes empty, the game will generate a task that an NPC must fetch an item to put into the slot

But... if someone else puts an item into the slot, the task must be canceled... so it needs to pass around info about who put something into the slot (eg. like the instigator in some events), because if the item being put into the slot is from the NPC working on the task to fill the slot, then it shouldn't cancel the task...

But...... other NPC's will sometimes temporarily pick up the item in the slot. And it wouldn't make sense for it to generate a task when this happens. So it needs to somehow also handle that scenario...

ugh lol

And I guess to make it even slightly more complicated, it's possible the player might pick up an item from the slot to just look at the item and then put it back... so it also might just need to have a small delay between generating a task when the slot becomes empty so it doesn't immediately try to fill it, because there's no way to determine whether the player intends to put the item back or not :P

[agile eagle]

I need help.. the binding option for actor is all gone is the newest version of ue5 when using AI State tree.

it suppose to look like this

[misty wharf]

have you tried removing and readding the task?

(assuming it worked in an earlier version)

[spring inlet]

also i kinda doubt that it was ever possible to bind to an actor there

best you would be able to set is likely a class reference for the player class and then fill it in during runtime 🤷

never used StateTree tho

[misty wharf]

You can bind actors there but not arbitrary actors

For example, if you have a task which returns an actor as its output, you could bind to that, or bind to actors from the context

[bleak drift]

how can i disable a AIPerceptionStimuli?

[misty wharf]

By stimuli do you mean a sense? Because a stimuli is the data from a particular sense, and I'm not sure what it would mean to disable the data

[bleak drift]

AiPerceptionStimuliSource - The component that is percievable

is what i meant

[misty wharf]

It should have a function like unregister from system or something like that

[bleak drift]

yea that somehow didn't work 🙁

[misty wharf]

🤔

[bleak drift]

But it should?

[misty wharf]

What type of actor do you have the stimuli source on?

[bleak drift]

A palyer pawn

[misty wharf]

Have you disabled the automatic registration of pawns as stimuli sources?

[bleak drift]

but maybe I'm overseeing smth, if it should work i know where to look

[bleak drift]

noo how do i do that?

[misty wharf]

https://zomgmoz.tv/unreal/AI-Perception/AISense-Sight#turning-off-pawn-autoregistration

[bleak drift]

ohhh wow ok

[bleak drift]

Awesome now it works as expected!

Thanks a lot! 🙂

[misty wharf]

👍

[hallow compass]

im still interested for soem critics for this system

[misty wharf]

Hard to say, it seems like it could work. It might be helpful to expand on why you're designing it this way, rather than using the builtin bits

[hallow compass]

im not rebuilding an AI perception systems, i will use UE's one in mine

for example stuff like perception params such as range will be used from the AISense configs

[misty wharf]

Ah

[hallow compass]

what im doing is more like a wrapper

with little features added to it for now, but if i ever need more stuff later i can easily add it

instead of redoing same stuff over again on each new project

[misty wharf]

"in case I need it" sounds like a danger zone lol

you probably need to redo them at least a couple of times before you've figured out most of the potential issues hidden in there

at least this is my experience with things I haven't built before :P

I'm working on the task system stuff I posted about and I'm already starting to think the approach I'm using is wrong

(#gameplay-ai message)

[hallow compass]

yeah i got my fair share of that to

ill just build it in a "general" way as much as i can, test it with my game, and over time ill see how i refactor it if i see some issues as i use it with other projects/systems

[lunar schooner]

https://streamable.com/vd46ia anyone have any idea why my nav mesh in this particular line is behaving this way?

to clarify, that's a nav mesh modifier i'm moving in the video

and there's no other collision causing this

[lyric flint]

so i have two nav mesh bounds, when playing in game, they are floating above the level, even tho in editor they are placed properly, how could i fix this?

[grim tartan]

Are you sure theres no other collision affecting this?
Cuz, you bring the modifier pretty far back, and the line doesn't follow it anymore.
Suggesting theres possible weirdly slanted collision there.

[lunar schooner]

no other collision that I can see.. i've also gone through each object placed in the level and found nothing that appears to be affecting it

https://streamable.com/uio9ps here's another example

[lyric flint]

this is where is placed in level,

but this is where its being placed when i play the game

[cyan sigil]

I'm having some trouble with StateTree. I have an array of structs that is being maintained by a global task. I want to have a Condition check the values within that struct. However, performing a parameter binding is causing the struct to be erased every frame. What should I be doing instead?

[cyan sigil]

OKay, I found that I have to NOT mark the variable as parameter or output. If I leave it as default category the value is preserved.

Next step: I have a task within a state that activates an ability and ends when the ability ends. Is there a way I can have the state tree continue to evaluate and potentially interrupt?

[cyan sigil]

Now I'm having a problem where the task seems to be evaluating four frames in a row.

How can I force the state tree to re-evaluate on event or condition?

[misty wharf]

Not 100% sure but I think you would have to trigger a transition to the state which condition you'd want to re-evaluate

Since I think they only evaluate when the conditions are "relevant"

[paper basin]

Hello, I'm not sure how to figure out what is happening here or what I could do about it. My Behavior Tree MoveTo task fails on the chairs in the middle. Even when I set the acceptable radius to be very high, MoveTo just refuses to go near them.
What setting could help with that if any? I want it to only move close enough to the chairs, not to literally stand on top of them, I thought that is what the acceptance radius was supposed to do. But I think it just doesn't even start the pathing and this is failing the moveto.

Or is the only solution just finding a navigable location in radius by myself instead of relying on the position of the mesh component?

I managed to get it working by unticking these two boxes on the moveto. But I guess I will have to find a better solution for this in the future anyway because it is still a little weird

[misty wharf]

Based on this and how you got it working, it sounds like you're trying to navigate to a point that's not in the navmesh. Radius affects how close to the point it needs to be before stopping, but the point still needs to be reachable for that to do anything.

What you can try is using project point to navigation (or something like that) which can project a point from outside navmesh to the navmesh

[paper basin]

Yeah, I figured as much. I misunderstood the acceptance radius

The chair origin is in the middle at the bottom of it. Seems like it works for the side chairs since the agent radius reaches this point from the side, but it doesn't from the back

[misty wharf]

Yeah iirc the default projection option uses the agent radius or something like that

But you can manually project it as far as you want

It's worth noting that depending on your geometry and navmesh shape, it's possible it could end up projecting the point to the wrong place, so you may need to consider some other options, like having specific entry points that you navigate to, rather than the specific point

[paper basin]

I was thinking about adding a scene component to the back of the chairs and using that to pathfind instead of the chair model itself

[misty wharf]

Yeah that could be one way to do it

[hallow compass]

can i request a move (or more low action) to the AI with given path/points ?

if yes, what functions should i look for

[misty wharf]

maybe you can do it through the path following component? That's what I'd look at for starters at least

[hallow compass]

i will do that

i guess i can also see what the move task does

[lunar schooner]

no one knows?

[hallow compass]

at runtime can i tell the state tree to no longer call tick ?

bShouldCallTick can only be set in ctor since state tree functions are const

[misty wharf]

Hard to say, could be some artifact with how the navmesh gets generated. If it's problematic you could try tweaking the navmesh generation settings like resolution and such and see if it has any effect

[hallow compass]

in 5.6 if i try to bind a task property to my custom func one i get this laggy behavior

it only does it on my custom one, other engine func bindings work correctly in UI

USTRUCT()
struct FCAIBGetIdleComponentInstanceData
{
    GENERATED_BODY()

    FCAIBGetIdleComponentInstanceData();

    UPROPERTY(EditAnywhere, Category=Parameter)
    TObjectPtr<AActor> Input;

    UPROPERTY(EditAnywhere, Category=Output)
    TObjectPtr<UCAIBIdleComponent> IdleComponent;
};

USTRUCT(DisplayName = "Get Common Idle Component")
struct FCAIBFunc_GetIdleComponent : public FStateTreePropertyFunctionCommonBase
{
    GENERATED_BODY()

    using FInstanceDataType = FCAIBGetIdleComponentInstanceData;

    virtual const UStruct* GetInstanceDataType() const override { return FInstanceDataType::StaticStruct(); }

    virtual void Execute(FStateTreeExecutionContext& Context) const override;
};

once i add this func binding, all slate bindings lags in the state tree until i remove the task with the "bugged" func binding

[cyan sigil]

That's frustrating. One advantage of BTT is that it can re-check higher decorators constantly.

[cyan sigil]

Or maybe I can put the transition on a parent state?

[misty wharf]

You should be able to yes

[hallow compass]

Still stuck on that

[hallow compass]

update: this UI glitch only happens when i use my component as the output type

and when i use UActorCOmponent or any child as a the type

it doesnt happen if i use some other non UObject type or AActor

[lunar schooner]

do you mean these settings?

[misty wharf]

Yeah. There should be some others too

Cell size I think it was

[lunar schooner]

yeah there's cell size options inside of each of those drop downs

for low, default and high

[amber lark]

Hello, I currently working on my AI Behavior, and a made a "Roaming Query" for my State Tree to use. My Issue is that the result of the Task doesn't give any Output. When I start the game the AI is Running to the middle of the Level and start roaming there. Does someone have a clue what the issue is or what I'm doing wrong?

[wide robin]

Roaming in State Tree with EQS

[lunar schooner]

are these settings more expensive? seems like it takes forever for nav mesh to rebuild after making changes

[hallow compass]

what can cause the transition to fail ?

okay, seems like its because the parent state "Idle" was blocking it since the state selection mode was "None"

[cyan sigil]

Does anyone have experience subclassing a statetree state instead of only tasks? I have a situation where I frequently have a state represent an ability, and it should test CanActivate when deciding to enter, then call Activate when entered. It seems more reliable to combine those into one state instead of an enter condition and a task.

[hallow compass]

im curious, what is the merge thing ?

[slow bobcat]

You might get a different path and you want to merge it with your current path instead of just using the new one. Useful if you calculated parts of the path manually that you want to conserve

Never used it myself

[wide robin]

<@&213101288538374145>

[hallow compass]

dunno if this would interest people, im currently working on a generic open source "Common AI Behaviors" plugin, right now the "planned" content is the following:

example for the static idle stuff

you use a State Tree tasks to run it

internally a world subsystem ticks the stuff

currently working on the patrol stuff

it all runs using polymorphisme, so the idea is that you and I can add any kind of behavior (new states, or childs of exisitng ones)

almost nothing pushed for now, but here is the repo https://github.com/hzFishy/CommonAIBehaviors

(this has a dependency on my FishyUtils plugin)

[misty wharf]

Seems somewhat game-specific tbh

Could be useful for those that need those types of behaviors I guess?

Also, I don't know how good of an idea it's to use anim notifies for game logic stuff - there's some discussion about it on #cpp regarding how they're unreliable

I think there was some way to make them a bit better but there might've still been some gotchas on it

Although I'm currently using them in my game just fine, not sure if there is potential for issues down the line

[hallow compass]

im not using the regular UE notify pipeline, i do what Vaie suggests which is registering a callback using trigger time in code

you add the notifies in your animation, in c++ i get them all and have 1 callback for each, which can run w/e

this way none are skipped

[misty wharf]

Ah

Yeah that was the proposed solution that I saw also I think

[hallow compass]

i wrote down what Vaei said here: https://notes.hzfishy.fr/Unreal-Engine/Animation/Concepts/Anim-notifies

[hallow compass]

yeah, thats how AIs are in games.
you can still use the core system which is behavior agnostic

current setup

and plop in your AI behavior

[sacred stone]

Has anyone here managed to get references to individual automatically generated navlinks?
They are not treated as individual smart link actors and are actually treated as collections that are part of the navmesh.

Functions from INavLinkCustomInterface treat them as a collection.
The CustomNavLinksMap from UNavigationSystemV1 contains the IDs of the collection, not individual links.

[topaz inlet]

hello! Using random reachable point in radius, is there a way I can make sure it's using thte collision radius of my player. I have a super simple setup, using floating pawn, I have it pathing to the positions on teh way to the target, but it gets stuck sometimes at smaller meshes that do not block the navmesh, but well, do block my "character". It's currently just a small boxy robot that moves around.
Edit: Actually, perhaps my move to function is wrong.

There is this concept of nav agents, but i'm unsure how it showuld be used with that function.

[slow bobcat]

You say an ID points to a collection, what's the collection itself? Is it an array or what are we talking about here?

Because the ID's are part of a t-map (CustomNavLinkMap), what's the object used as a value?

[sacred stone]

the ID of all the navlinks generated by the JumpDown config

you have 5 navmeshes in the level that are generating navlinks? you have 5 IDs

They are basically the IDs of the owner (something that implements INavLinkCustomInterface

The whole system is very counter intuitive for me atm. Especially since a lot of the functions in the interface make it sounds as if they're for individual links

[slow bobcat]

I'm not following and I don't have the code in front of me right now but you have a TMap<FNavLinkID, customLinkOwnerInfo>

The TMap has one entry per nav link generated no? Then, per nav link, you have the owner info.
Can you post here what does that struct look like?

[sacred stone]

One entry per NavMesh that generates Navlinks, so to speak. But the entry owner value isn't the navmesh itself, but the handler that implements that interface I mentioned above.

{
    FWeakObjectPtr LinkOwner;
    INavLinkCustomInterface* LinkInterface;

    FCustomLinkOwnerInfo() : LinkInterface(nullptr) {}
    NAVIGATIONSYSTEM_API FCustomLinkOwnerInfo(INavLinkCustomInterface* Link);

    bool IsValid() const { return LinkOwner.IsValid(); }
};

Link owner here is the UBaseGeneratedNavLinksProxy instance

If anyone here is more experienced with this feature, do correct me if I am wrong please.

[slow bobcat]

Yeah that's the proxy you can set for custom behaviors with the nav link. Through the proxy and the interface you should be able to get some data and access the nav link detour information.
If I find time tomorrow I will take a quick look, see what can I find

[tranquil meadow]

was late to the reply. The Build -> Build Paths thing was what i was missing

[slow bobcat]

Solution for your issue

[slow bobcat]

ok I might be wrong. I missed this comment

{
    ensureMsgf(false, TEXT("Should not be called on a generated navlink proxy since it's not representing a single link."));
}

Didn't look at the implementation.

Ummm weird

[chilly nebula]

FYI - we just published an article on tick changes/scheduling in State Tree. Hopefully clears up some of the confusion around this feature.

https://dev.epicgames.com/community/learning/tutorials/z3km/unreal-engine-tickless-statetree-changes

[misty wharf]

Nice :)

[celest python]

Hi Siggi! Are there any plans to create property access bindings for utility weights in nodes?

[hallow compass]

crazy how little virtuals the perception system and component has

Cant do anything

[regal patio]

https://youtu.be/8Zs_RzNGDGI?si=YlSZOflbRjCQjYmP

[tranquil meadow]

question, if i want an AI enemy to look at the player constantly while moving around, what do I need to do?

So far I have them set the player as their focus and disable orient rotation to movement but that just has them not rotate while moving

[slow bobcat]

Usually is solved in animation with head IK's.

[tranquil meadow]

ah so focus doesn't do anything

[slow bobcat]

Focus is related to movement and capsule forward orientation. Looking at is a different thing

[tranquil meadow]

ah well

i guess for my case

when i said looking

i basically meant that

[slow bobcat]

For example, if you set the focus to the player and you move the AI to the left, it will rotate around the player. But if you want to move in a straight line to the left and look at the player, you need a head IK in your animation

[tranquil meadow]

eye sea

when you say rotate dough

do you mean they move around the player?

the capsule itself doesn't rotate?

[slow bobcat]

It's not great but you can do lots of things. Most of the cases is a matter of gathering the information provided by the system and process it yourself

[slow bobcat]

Yes, they will move around the player, looking at it, in a circle trayectory. The capsule rotates toatch the focus.
We have different ways to do this in our game, one is by having "move to the left / right" root motion animation where we force the focus on the target (the player)

[tranquil meadow]

The capsule rotates toatch the focus.
ok so the capsule is supposed to rotate to look at the focus

look at again just means to rotate towards it

[slow bobcat]

Yeah.

[tranquil meadow]

hmm, let me see if the capsule actually does cause atm, it doesn't seem to do that

[slow bobcat]

That's what focus does: aligns the forward vector of the capsule towards a point or a direction

[tranquil meadow]

hm

yeah atm it's not doing that for my end and unsure why

atm setting focus via decorator

but unsure if i should just be setting focus every frame

oh ok

now i know why

cause this was checked off (and had to check off orient rotation to movement)

[slow bobcat]

Yeah, that's the key part. Sorry I didn't mention

[bleak drift]

Is there a way, I could make a AI patrol a dynamic path, meaning I'm just adding path points on the go and in proceeds the points I'm giving it?

I have this setup now with move to, but the issue is, that when i wait until the move is done and then continue, the ai stops for a short amount at the point until continuing. I also have set: MaxAcceleration = 100000, BrakingDecelerationWalking = 100000, BrakingFrictionFactor = 0 and GroundFriction = 0.
But there is still a tiny stop.

Can I do something with the path follow component, where I just push points to it or what would be the best way of doing it?

[slow bobcat]

You would need to merge the new points into the path before you reach the last one, otherwise the movement will end

[bleak drift]

Can i do that with the PathFollowComponent?

[slow bobcat]

You could yeah, but not sure at what moment. Is this because you are adding random points or what?

[bleak drift]

Yea it's like a new point is chosen when the one is reached or it's in quite close proximity.

It started because AI with move to hugs the walls, and I want it to walkt in the middle of the path.
I can't really create a spline and make it follow along that, because there is quite a lot of branches and on top of that, some paths might change at runtime. So i decided to make it just grab a potential point in range and move to that.

It should just be like patroling and constant movement, but i feed it the locations.

[slow bobcat]

then I suggest this:

• have your own AI Move To task (like the one the engine uses)
• upon finishing calculating the path, move the path away from the corners and replace the path calculates by the engine
• Then start moving on the path (whyich will start the logic in your path following component).

It's better to post-process the path to move it away from corners than do it on the go

There's an explanation on how to move the points away from corners here. The talk will be available online at some point (youtube)

https://archive.org/details/unreal-fest-25-scalable-aisouls-like-games

[bleak drift]

But that just gives me a static path, I need it to be dynamic.

[slow bobcat]

why?

what's dynamic about it?

[bleak drift]

So you can look at it two way, either I'm already doing what you are saying, by having my points being the path it moves.
But then the issue is still that it is not smooth, because it stopfs for a few frames.

Or I would create a longer path, which then has the same issue at the end when i generate the next one.
The other thing is, that the Path might not be valid at some point anymore, because it gets blocked and is not navigateable anymore.

[slow bobcat]

the way I see it:

• calculate a path to a patrol point. Then to the next and so on.
• merge the paths into one (they are just arrays of points)
• post process the path to move it away from corners
• start moving.

in your path following component you can deal with points not being reachable when a new path segment gets set and re-calculate just that one path point

if you generate points on the go, generate 2, so you have some marging before the path ends

[bleak drift]

Yea so basically just push points dynamically to the to the path follow component, what i was asking at the beginning?

Is that possible and how would I do that?

I basically just want to avoid move to so that it gets a constant forward movement

also it appears there is not to much documentation on the path follow component out there 😅

[bleak drift]

But if you are suggesting to just move some partial points at a time, I'm doing that right now with move to, but then you always have the small stop between so thats not an options with move to.

Hope what I'm saying makes sense

[slow bobcat]

The issue is that you are waiting to reach the last point to add a new one. You need yo add points to your path within the same move request before reaching the last one. Only way I can think of right now is to do that in FollowPathSegment checkign if you are Xm from the last point (since you are technically always moving to the last point) and add another point to your nav path

[slow bobcat]

by your explanation I would imagine you are doing something like this:

• the AI is the blue dot and you calculate a path to the first point (the line to the obstacle corner)
• then you want to pick another point (any of the 3 red dots).
• you will need to decide where to go while at the yellow cross, before reaching the end of your path (that only has 2 points), which will cause the move request to finish.

you will need to calculate a path to the next point and merge the new path points into your current path

in our project we use our own extension of NavMeshPath and we have a function to insert points into a path

{
    if (PathPoints.Num() > 0)
    {
        const int32 maxIndex = PathPoints.Num();
        const int32 indexToUse = index != INDEX_NONE ? FMath::Clamp(index, 0, maxIndex) : maxIndex;

        uint8 areaID = RECAST_DEFAULT_AREA;
        uint8 flags = indexToUse == maxIndex ? DT_STRAIGHTPATH_END : 0;

        FNavMeshNodeFlags navPointNodeFlags(0);
        navPointNodeFlags.PathFlags = flags;
        navPointNodeFlags.AreaFlags = areaID;

        FNavPathPoint point;
        point.Location = locationToAdd;
        point.Flags = navPointNodeFlags.Pack();
        PathPoints.Insert(point, indexToUse);
    }
}```

another option is to use the merge path function in the AI controller and replace the path you are using in your current movement request from the path following component

    AIMODULE_API virtual void MergePaths(const FNavPathSharedPtr& InitialPath, FNavPathSharedPtr& InOutMergedPath) const;```

[hallow compass]

can i get the UAISense live object instance from a FAISenseID ?

[slow bobcat]

Do you mean the thing you sensed for the ID sense?

[hallow compass]

yes

there is some statics, but it gets the default UObject

[slow bobcat]

Ah you mean the sense object itself

[hallow compass]

oh wait, does the engine use only CDOs ?

or are the senses instantiate

[slow bobcat]

Yes, and then you apply sense config onto it

[hallow compass]

oh nice

[slow bobcat]

you can grab your currently use sense config, change its params (radius or whatever) and refresh the sense with the new params calling OwnerAIPerceptionComponent->ConfigureSense(yourSenseConfig)

[hallow compass]

im doing a tracking thing

where i get a tag for each sense for storage

for user friendly tags

[bleak drift]

yes you are right, to add the point before it reaches the actualy point would be not issue, not just fully in advance have the whole path

[bleak drift]

ok thats sounds really awesome, but how do you tell the ai to move along the path?
I think thats the only thing I'm really unclear about, how I can say follow that path?

And what you are saying is that when i add a point to the path before it reaches the last point, the ai will continue right?

[slow bobcat]

You don't have to do anything. Change the path and the move to task will simply handle it. That if course assuming you are changing the "future point paths". I have never tried this but maybe the path observer has an event to deal with such case, I would imagine that OnGoalChanged event might trigger

In any case it should be fine

[bleak drift]

but where is the paths array on?

which object

I just know move to location, i think thats where my confusion comes from in that case 😅

[slow bobcat]

That's your homework. Check what does the AiMovetTo async task do, how does it pass the move request generated to the path following component and what's inside

[bleak drift]

ahh ok i see

[slow bobcat]

Btw all this can only be achieved in code. I have assumed you are doing c++

There's no way to do any of that in BP's

[bleak drift]

yes I am

[slow bobcat]

Cool. Then... Happy debugging

[bleak drift]

ok so you are saying I'll find some truth in the MoveAITo function right?

[slow bobcat]

That's your start. Or you can set a breakpoint in SetPathSegment in path following component. Both will tell a different part of the same history

[bleak drift]

Ok thank you a lot for all the info, I'll try to figure something out there 🙂

[slow bobcat]

You have some clues here

[torpid juniper]

I'm trying to have a guy get next to an object from a specific orientation.

The red line traces from the character to the object. the green/pink line offsets it from the nearest point in the collider back towards the character. The blue line is the result of trying to find a navigable point from the pink / green intersection point. Anyway I can better identify blue points that correspond to the green/pink than doing an iteratively larger "random point in radius" check until it succeeds? I want the angles to be as consistent as possible.

[misty wharf]

I guess you could just take a point at a certain distance from the object's origin in the direction of the trace?

[bleak drift]

yes thanks a lot! :))

[hallow compass]

still got this issue

[slow bobcat]

Had that issue once. It is still there sometimes. To this day we have no idea why it happens

[hallow compass]

on a side note, does tasks has unique ids ?

im trying to cache on an external object the "static state id" wit hsome data, so when its re-entered i can get the mapped data back

[torpid juniper]

The answer was Project Point to Navigation which is apparently Get Nearest Point on Nav Mesh

[mighty geode]

How did Witcher 4 inside UE5.6 achieved such 300 crowd with all complex AI along with optimization?

And can we use that technology too or not?

[slow bobcat]

Mass + a shit ton of custom optimisations with the help of Epic. But just using mass you will get good results

https://youtu.be/F-v6Xz6-pf4

[mighty geode]

Tbh it's only single thread with no much gameplay. And I've lot of gameplay mechanics, heavy world. And would require multithreading too, so I thought using mass would be more complex work into it. But I'll give it a try then

[slow bobcat]

Mass is multi thread. That's how it works. You don't have to do anything

I can recommend couple videos
https://youtu.be/eJR82WyIl_U

https://youtu.be/vagDihyY1pk

[mighty geode]

thank you for this

[slow bobcat]

If you check YouTube you will find other examples with animation, AI etc

[chilly nebula]

Many things, some of which you can learn about from talks given at UEFest Orlando and Stockholm.

In short: Mass, State Tree, new animation system, Mover + a lot of work 🙂 It was a hybrid actor/Mass entity setup.
For AI specifically, we optimized State Tree a lot to be able to run tick-free for actor components, made navmesh work in Mass and leveraged the new Mass-based avoidance system that was originally created for City Sample.

[slow bobcat]

The mass avoidance system, is that something available as a feature for other projects? Is it Boid Flocking or closer to ORCA/RVO?

[mighty geode]

Navmesh isn't supported for mass?

[slow bobcat]

It uses zone graphs, splines and other stuff but not out of the box

You can still generate nav but you don't have all the hooks in code you have with a Movement component etc

[mighty geode]

The rogue entity says mass not good when to use less simulated entities.

Mine is, hordes. Zombie horde, but trigger at certain mission

And they have a simple job

Move to Soldiers

And kill them

When horde is not present, but zombies are coming. Those are fully simulated zomhie

Searching for food, eating dead bodies, moving around, screaming

[slow bobcat]

Yeah mass is only for crowded situations. If you have just a few entities, it's not worth the effort, there's little gain there

[mighty geode]

Have AI preception

[mighty geode]

My current Target is 100 atleast for a horde

[slow bobcat]

What's your average number of zombies at any given point vs a horde?

[mighty geode]

20-50 around the military base

And 100 specifically from mess (yk food court area) the horde triggers when the soldiers fix generator

[slow bobcat]

Over 20 fully working characters will require a decent amount of optimozations already.

[mighty geode]

Yes

Actually we had a 200 Zombie target but we reduce to 100 looking at our the game, optimization and minimum hardware requirements

[slow bobcat]

Honestly knowing you will have over 30 I would just go mass, that covers your horde cases

[mighty geode]

1080p 6gb Vram
8gb ram
i5 6-thread processor
Windows 64-bit

This is our minimum hardware requirements target along with 60fps

[slow bobcat]

If it was soemthing like having 20 enemies at all times, it's not worth the implementation effort, but in your situation seem a reasonable approach

[slow bobcat]

Ufff good luck having 20 enemies working properly with those requirements. Will depend a lot on the type of game but... Character movement component and animation alone will already be an issue

[mighty geode]

Let's see, perhaps could be reduced if the level gets small

This is a blockout

[slow bobcat]

Keep an eye on things like the significance manager, animation sharing (can't recall how is it actually named) and the likes

Also keep a close eye with Chaos Physics, because their overlaps are not cheap precisely

[mighty geode]

I was looking at days gone approach

A pool of zombie with lead that will animation sharing

And when at certain range, they activate their complex AI simulations

[slow bobcat]

That's what you would do with mass I think

[mighty geode]

Oh, gotta check more into the mass and planning

[slow bobcat]

Keep me posted. I'm very interested in these approaches (didn't have the chance to work in massive crowds or open world projects)

[mighty geode]

Yeah sure

[slow bobcat]

Appreciated

[chilly nebula]

yes its in the Mass experimental plugin. Its a simpler force based approach.

[chilly nebula]

it is since 5.6

[green loom]

Hey all!

I finally decided to take the plunge and get back to developing the AI for my third-person combat game (I am going for Khazan type combat) and I am struggling a little with the simplest of things.

I am starting basically with a 1-on-1 setup but want to understand how I would operate crowds as well to facilitate smooth natural movement for the enemies.

I have a system running that does some EQS queries to try and idle move between attacks but it looks a bit clunky (i.e. it picks a point and walks to it regardless of how the player moves rather than adjusting the movement in a smooth curve) and I would run into issues if there was more than one enemy (clumping etc.)

Are there any resources I should look at in terms of implementing this sort of natural “surrounding” or “keep your distance while waiting to strike” type behavior?

Furthermore, AI move to by default is very abrupt, I assume if I want acceleration, I have to manage that myself by adjusting the max walk speed or by overriding the default node logic?

I found some references to things like UCrowdFollowComponent but I am not sure if thats what I would want to use in a case like this. The AI itself is using StateTrees.

I would take any guidance on how to make this kind of movement look and feel natural. I would offer Malenia’s movement from Elden Ring as a reference point.

[harsh storm]

It depends on the zombie honestly. I can regularly have 100+ zombies without mass. ACharacter btw.

[slow bobcat]

Yeah but you have been working on optimizations for how long now?

Yours are quite good for what I heard

[harsh storm]

Honestly, for zombies? Not too terribly long. I pretty much do the same stuff that Kaos does to achieve those numbers.