🤖┃ai-navigation

[violet wyvern]

is it an ok practice to use NavMesh only to calculate the path, and then move a Character Controller along the path? the current NavMeshAgent seems to be a bit slippery for me (it often hits walls because it doesn't slow down fast enough), and it's also rotating very slowly. i had an idea to move the Character Controller along the path, because this way i will have more control on movement details e.g. coding my own acceleration system or rotating the character to face the current waypoint.

[molten sequoia]

Using your own movement to follow the path is fine.

[violet wyvern]

ok, thanks for answer

[light saffron]

there is a fantastic tuto made by the navmesh component team, it shows more advanced way to control character with the navmesh agent parented to the root and some form of lerping to control the rig. can't find it anymore. anyone knows the URL for it?

[sharp nexus]

Hello, I'm trying to get a random position on a navmesh but I cant find any resources to do it, I don't want to make it with a radius, I want to get a random position on the whole navmesh.

[light saffron]

this avoids a sampleposition, might work for u

    {
        NavMeshTriangulation data = NavMesh.CalculateTriangulation();
        var                  idx  = data.indices;
        int                  tri  = Random.Range(0, idx.Length / 3);
        Vector3              v1   = data.vertices[idx[tri * 3]];
        Vector3              v2   = data.vertices[idx[tri * 3 + 1]];
        Vector3              v3   = data.vertices[idx[tri * 3 + 2]];
        // random barycentric coordinates
        float u = Random.value;
        float v = Random.value;
        // barycentric -> cartesian
        Vector3 pos = v1 * (1 - Mathf.Sqrt(u)) + v2 * (Mathf.Sqrt(u) * (1 - v)) + v3 * (Mathf.Sqrt(u) * v);
        D.raw(new Shape.Sphere(pos, 2f), Color.yellow,2f);
        D.raw(new Shape.ScreenText(pos), Color.black, 5f);
        return pos;
    }

[sharp nexus]

A bit late but thank you it works

[light saffron]

make sure you cache the triangle array

[sharp nexus]

The NavMeshTriangulation data ? I did it

[light saffron]

cool

[violet wyvern]

how can i make the navmesh agent always come to an object from behind? i am making a monster that stalks the player, and i don't want the monster to suddenly walk in front of the player. i had an idea to create an invisible "shield" that will be in front of the player, forcing the navmesh agent to find a path to reach the back of the player. are there any better options of doing that?

[dreamy sentinel]

not sure if this is the right channel but ```!pip install pyngrok
from pyngrok import ngrok

port = 5004
ngrok.set_auth_token("{ngrok authtoken}")
public_url = ngrok.connect(port)
print(f"Public URL: {public_url}")

!mlagents-learn --force``` anyone know why this isn't working? I'm using google colab for my mlagents. Should I use something else and what? I have all the dependencies installed and mlagents does tell me it's running on port 5004 but my local machine cannot find that port.

[light saffron]

targetPosition = -player.forward*5?

[lost plume]

All right, i didn't really know where to put this but i have a coded ai (a chasing one for a first person horror game its roaming ai with the phases idle, chasing, and searching) for some reason, the ai seems to spas out and walk and flail all over teh place while staying in one spot. And after the search phase ends adn they go back into idle they just go straight for a wall and try to go through it. Ill post the ai controller code too just in case it might be the problem

(the image with the wall hierarchy is the same for most walls BTW)

[violet wyvern]

I think this will just offset the agent target position, but won’t really force it to take another path to remain unseen.

[light saffron]

right. carving forward could work if you interpolate your character to agent position, because otherwise it'll pop

[violet wyvern]

does setting an area cost to mathf.infinity completely discourage and agent from going there? i have an idea to just check what the player is looking at, and the area it's looking at will just have a mathf.infinity area cost, so the agent won't go there and will chose another path instead. does it work like this, or am i wrong?

[light saffron]

that's a good way if your level is small because you need to pre define your areas at bake.

[violet wyvern]

do i have to re-bake the area after changing the area cost?

[light saffron]

unless navmesh surface component does area weight... then you're set and can do area at runtime

no you can change weight at runtime

[violet wyvern]

alright

thanks for help

[light saffron]

mmm unless i'm confusing with anypath and astar

[violet wyvern]

is anypath a pathfinding algorithm?

nevermind i searched it up and it's an asset

[light saffron]

check out the navmesh surface runtime baking sample, if you add a volume modifier then you can have enemy avoid player front

[empty sonnet]

How would I use the outlined mesh as a navmesh? When I click bake, it just bakes for what seems an eternity. And after looking into it, I think it's trying to bake everything?

[signal ridge]

I don't remember where the exact setting is, but you can limit the navmesh bake to certain layers.

[empty sonnet]

Thanks, I think I need to try and understand the navigation stuff better even though it's but a small part of the game 😦

[light saffron]

later is in the surface component

[alpine glacier]

Hey uhhhhh i need help with something

[light saffron]

interesting findings with surface navmesh component: the navmesh doesn't seem to get included in a build

[light saffron]

no that's not it, it's just the behavior of navmesh is different in build... never seen that b4

[light saffron]

they changed the UI again. open object collection and select children

[high hedge]

read the rules for help #854851968446365696, dont ask if people are avalible just ask your question

[lavish river]

So I'm trying to re-learn ML-Agents after a couple years. Using the latest ML-Agents package and Unity 2021.3.11f1

I'm following the "Get Started.md" from the github, and after installation I need to navigate to assets/ML-Agents/Examples/... but this does not exist. Under Packages, there is no "Examples" folder in sight. Is this a problem with the unity version I am using, or did it lie to me saying to use the package manager window?

Swapping to unity 2023.2.20f1

[lavish river]

It still does not exist,

None of my packages appear in the windows file explorer either???

I can't read. Did not read note.

[lavish river]

Cloned the github, copy/pasted com.unity.ml-agents to my assets folder, but cannot open the example scene. I'm just gonna give up again for now.

[whole comet]

For some reason the side of my mesh is getting baked as navmeshsurface but the top isnt?

[uncut tide]

Hello. Is there a way to optimise weighted A*? Because without weights, my A* searches path really fast, but when weights are added it slows down a lot and I have no idea how I can optimize it. For Open list I use binary heap and HashSet for closed list

[jaunty raft]

How and when do you calculate your path costs? How do you access the weights? Potentially you are loosing a lot of time by following pointers into random memory locations.

[uncut tide]

                    int tempG = q.g + Mathf.RoundToInt(getDistance(q, neighbour) + neighbour.movementPenalty);

                    if (tempG < neighbour.g || !openSet.Contains(neighbour))
                    {
                        neighbour.g = tempG;
                        neighbour.h = Mathf.RoundToInt(getDistance(neighbour, endNode));
                        neighbour.parent = q;

                        if (!openSet.Contains(neighbour))
                        {
                            openSet.Add(neighbour);
                        }
                        else
                        {
                            openSet.UpdateItem(neighbour);
                        }
                    }

    int getDistance(Node a, Node b)
    {
        int dx = a.x - b.x;
        int dy = a.y - b.y;
        return dx * dx + dy * dy;
    }

[uncut tide]

I can provide whole class if needed. When weights are low, it calculates this very fast(like max 1ms) but with larger weights added calcation spikes to even 130 ms for long distances

I thought about precalculating some paths, but I need to be able to change which nodes are walkable during my game

so not sure if that will work

[jaunty raft]

This doesn’t tell much but in any case, first level of optimization would be to figure out if you can avoid any extra work (calculations) you are doing. If your graph needs to be mutable though, most truly effective optimizations will be unavailable to you. How long is such a 130ms path… that seems like an absurdly big graph? Typically the most important first optimization in a* is making a really really great priority-queue/heap. If your nodes are classes or otherwise not stored with data locality in mind, you’re likely loosing a lot of time in cache misses.

[uncut tide]

Map size is 256x256

Nodes ofc. I was thinking about changing from arrays to to lists in my heap, that could improve some things.

So nodes as struct could help?

[tulip wren]

If you haven't it's also worth implementing some kind of debug display that can show you all the nodes it's visiting

Something like everything that's on the closed and open list by the end of the search

130ms is definitely abnormal

On an aside, depending on whether you need perfect paths, it also never hurts to have a limit to the number of nodes it will explore, and an early out where it just provides the best incomplete path it could find

Then just generate a new path as your agent approaches the end of the previous one, often the new path will then make further progress

Won't necessarily be a perfect path but is usually in the realm of "good enough", especially if you tweak the cutoff point

[uncut tide]

I mean, I don't need perfect path, I just need for the agent to avoid high cost areas, and get to distance b with reasonable time

[tulip wren]

I'd say that's the way to go then

The rest of the gains are mostly from generating less paths and multi-threading it as much as possible

In my own project my current setup is to make a request for a path, which is then put on a queue for a job system, and to then retrieve it on the next frame

Oh, and using burst for it if you can, it will also double the performance easily

[uncut tide]

My path calculations are also split between multiple frames, and I do calculations on a seperate thread. When I tried to do multithreading, my cpu usage spikes to 100% no matter what. I think I would have to learn multithreading better first.

[uncut tide]

what's a burst?

[tulip wren]

You'll find it if you google unity burst. Basically it's their framework for writing code which gets compiled down to fast native code.

The caveat: you have to use only a subset of C#, primarily everything has to be structs

No references, no managed data

It's a learning curve but well worth it if you're doing performance critical stuff

In your case you could just apply it to the job which handles the path generation

And the rest of your codebase could stay as is

[uncut tide]

aight, will take a look. Thank you!

[oak osprey]

hi everyone, I have a gameobject setuo like this and im trying to bake a navmesh onto just the sidewalks, so i added navmeshmodifiers to the road to remove it, that works fine, but the navmesh does not bake onto the sidewalk

does anyone have an idea why?

[oak osprey]

nvm got it, for anyone wondering the sidewalks are too small to fit the agent so making the agent radius smaller baked them

[uncut pagoda]

how do i bake navmesh in only a certain region

[radiant palm]

You can use the navmesh surface component

[uncut pagoda]

thx

[lost plume]

pulling this down

[lavish river]

I'm finding out that CodeMonkey's "How to use Machine Learning AI in Unity" is not a good tutorial. There is so much just skipped over, like what text editor he's using to simply "go to definition", or how he got that scene when he first started on the unity portion.

I struggled trying to find a version of python that worked, not realizing that the installation guide lied about which one to use, as python 3.10.12 and higher have no available installers, and the most recent previous version I found an installer for was 3.9.13, which works seemingly well.

My main issue is; what do I do after starting unity?? He seems to open unity with no context of how or where he got the sample scene shown, a "now draw the rest of the owl" situation. I tried to clone the git repo and drag/drop it into assets but I just got conflicts, this was after installing ml-agents v1.0.6 just like in the video. He mentions the git repo, but never what to do with it.

Is there anyone here that has experience with ML-Agents, and is capable of explaining what to do? Because from what I've found on youtube and various forums, nobody has any clue how to explain anything or give any context. This is something I have been struggling with for about a year's time total, across the last 3 years. A few years ago, I managed to open an example scene and watch the agent move, but that was about it. I never found anything about how to actually create a new ragdoll agent like the crawler, and I can't figure out for the life of me how to even get back to that point.

[molten sequoia]

Move this to #1202574086115557446

[lavish river]

Will do

[tight prism]

why while using navigation obselete there is no ai path (there is in other places)

i baked but still no

[craggy zenith]

Can anyone tell me why my NavMeshAgent objects hover off of the tilemap by 0.58 whenever I hit play?

[feral frigate]

I have a very large navmesh terrain. Is there a way to only update a small section without rebaking the entire surface? Or avoiding the cost of rebaking the entire surface.

[radiant palm]

navmesh surface component

[feral frigate]

yeah thats what in usign on my terrain

[half hull]

how to make the blue area take up the entire surface of the mesh please?

[jaunty raft]

Set the character radius = 0

[half hull]

character radius on the nav mesh agent?

[stark delta]

im using mlagent trying to teach ragdoll to balance itself to stand and im using headlevel to reward however after the ragdoll goes below the headlevel of minus point which is ending the episode my ragdoll is not resetting but only just lying down on the ground im using ResetRagdoll(); beginning the episode can someone help me? if codes are needed i will send through dms

[silver quiver]

Anyone having issues with NavMesh v. 2.0.3 ? In unity 6.0.12f?

I'm having a NavMesh that constantly gives 'Failed to create agent because it is not close enough to the NavMesh' forever.

Like
I've already gone through about 3 or 4 forum threads, baked the NavMesh, tried placing the agent on a cube or something normal, instantiated it via script, etc

[silver quiver]

Solved! There was an NPC that was lost in the middle of nowhere (completely off the map) and I think the NavMesh was baking for it.

[half hull]

How to remove bake navigation on a scene or update it? There is a bug in my scene. When I move objects, the navigation does not update and the characters walk in space. I'm using Unity 6.

[vestal mica]

So my nav mesh navigation was working fine yesterday but has gone really weird. It should be doing a single navigation to a single accessible point but gets jammed and snapped back. As you can see I seem to be getting strange results from neighbours and avoidance as these debug lines show. Any clues as to where I can even look? I removed any new objects and re-baked the navmesh which looks fine.

[vestal mica]

Phew, found the culpret. Adding a Nav Mesh Obstacle to a ProBuilder object you would think would default to a suitable size but it didn't for me. The first image shows its first defaults. If I change to capsule and then back to box I get more reasonable values. Some kind of bug there I fear.

[slow dune]

does anyone know why this is happening?

the ground is a flat plane

i instantiated a plane to show the problem

the agent moves just fine if i click far from it but if i click close its floating?

[vestal mica]

I'm guessing you are hitting a different collider. Perhaps try the Physics debugger view to see if there are any colliders the wrong size or are getting hits when they shouldn't

[slow dune]

alright ill try, thanks

[slow dune]

that was the problem, thanks!

[haughty fossil]

Hello!
I have a question which might be or might not be quite complicated. I can't find any online documentation or tutorials (generally a bad sign...) on how to use nav mesh modifier volumes at runtime? I would like an object i spawn at runtime to mark the area around it a specific area type.
But from my testing I gather that NMVs only update the area type when baking before play

[uncut tide]

how can I check if my object is not on navmesh? Like, if any part of it is not on the navmesh, then something should happen.

bool isValidPos = NavMesh.SamplePosition(position, out hit, 1f, NavMesh.AllAreas);

I tried something like this, but it always return false. If I make the radius larger, then it works sometimes, and sometimes it does not.

[sacred dagger]

    protected bool SampleNavMesh(Vector3 position)
    {
        if (!stateMachine.CharacterController.isGrounded)
        {
            RaycastHit hit;
            bool hasHit = Physics.Raycast(position, Vector3.down, out hit, 2, 1<<LayerMask.NameToLayer("Terrain"));
            if (hasHit)
            {
                position = hit.point;
            }
        }

        NavMeshHit navMeshHit;
        bool hasCastToNavMesh = NavMesh.SamplePosition(position, out navMeshHit, 1f, NavMesh.AllAreas);

        if (!hasCastToNavMesh)
        {
            Debug.Log($"{stateMachine.gameObject.name} is attempting to move off the NavMesh");
            return false;
        }

        return true;
    }

this is how I check for my NPCs if they're on the NavMesh or not

Anyway, I also have a question. How do I get my NPCs to:

1. Get themselves out of areas where they can get stuck?
2. Find their way around obstacles?

I'll provide further details on request

[jaunty raft]

1. A*, 2) RVO

[sacred dagger]

what's RVO?

and I was told that Unity has built in Navigation, why do I need to use A*?

[sacred dagger]

I can't google something I don't even know the full form of...

[jaunty raft]

Those are algorithms that solve the two issues you asked for, respectively

if you have a stuck agent, you need to teleport it. A* is for avoiding getting stuck and solving mazes. RVO is for not running into dynamic walls and other actors.

[sacred dagger]

with all due respect, what's the full form of RVO? The first result I got was a Dutch company. I'm sure it's not what I'm seeking 🙂

[jaunty raft]

Reciprocal Velocity Obstacles

[sacred dagger]

thank you 🙂

[jaunty raft]

But nobody googles for that, just search ‚RVO algorithm’ like a regular person

[sacred dagger]

that's when you get better, not as a beginner or reading it the first time 🙂

[jaunty raft]

Might be a bit advanced for a beginner 🫢

[sacred dagger]

we all start somewhere

[jaunty raft]

Fair.

[river walrus]

is there a way to bake a 2D navmesh at runtime? I am using the Navmeshplus package?

[quasi fulcrum]

I have set up a simple enemy follow script and my enemy successfully follows and rotates towards me. Only issue is that the rotation is always exactly 90 degrees off

using System.Collections;
using System.Collections.Generic;
using UnityEngine;
using UnityEngine.AI;

public class EnemyFollow : MonoBehaviour
{

    public NavMeshAgent enemy;
    public Transform Player;

    // Update is called once per frame
    void Update()
    {
        enemy.SetDestination(Player.position);
    }
}

Is that an issue with the code or something I have to fix in the inspector?

[radiant palm]

Something in the inspector

Your model could be off by 90•

[quasi fulcrum]

where about would that be?

[radiant palm]

Probably in the model import settings

[sleek current]

experimenting with the navmeshagent in u6 for the first time, and having the popular problem that the navmesh agents are pushing my character controller around, are there builtin ways to make the agents consider the player and other agents as obstacles?

[noble anvil]

does anyone know what the problem is with creating an agent?

[warped fable]

Let's say I have a navmesh with a wall. I'd rather the character walk around this wall by following the grid more than its traditional plan of cutting the corner. Any ideas?

[quasi fulcrum]

maybe work with invisible walls?

[warped fable]

I came up with a plan to hybridize navmesh movement with custom grids. Idea currently is when I need a path, I'll check with the a* pathfinder for grids. If the pathfinder sees an obstacle the player shouldn't hug around, it sets some grid nodes around the object. Should lead to a cleaner, more xcom like movement

Keep in mind it isn't setting a node for each grid tile, only for around corners

[light saffron]

SamplePosition hit.normal =0,0,0 always, according to documentation. No idea why but i need normal, what's the cheapest way? I Would rather not loop through the entire navmesh triangles or do 3 samples to rebuild that. any better method?

[warped fable]

I take it you want the normal for the gameobject? You can try using a raycast that only cares about the gameobject. Either fire a new one in the exact same way as the SamplePosition or potentially reuse the one you're firing.

[outer bone]

I am thinking about how to go about cover systems. Would it be possible to somehow have something similar to the navmesh? The difference here would be that it would only generate points next to walls and other objects so you basically have a line along edges where the player can walk in a cover state.

This idea would be for contained levels so generated in the editor because i think manually doing it would take much more time

[sleek current]

I think this would have to sit on top of the navmesh, you need it for navigation around the levels, but your chosen navigation points are separate. unless your game is like a one way corridoor they still need to be dynamicly picked because if you have a wall where the cover is on it depends on where the enemy/player is relative to it.

[alpine glacier]

how to tell navmesh agents to give up if destination is too hard to reach?

[radiant palm]

what do you mean "too hard to reach"?

[alpine glacier]

basically, just stop searching at some threshold

[radiant palm]

what threshold.

[alpine glacier]

condition/

[radiant palm]

you are the one creating your game

a threshold could be distance

time

[shadow cove]

Hi guys. I've a small issue with NavMesh surface. (I m using unity 6) It looks like the navmesh surface float above my cube, but when I run the game, my character float above the navmesh surface. Is this a normal behavior or i have done something wrong ? thanks
(Its almost correct if I set on my agent the Base offset to -0.1)

[blazing pasture]

hi guys i have this weird issue where sometimes my navmesh works, then suddenly its not baked, cant even see it in the scene view

hmm seems related to it being in a prefab, works if i do it in hierarchy then apply

[shadow cove]

Ok i found something. It looks like my Animator controller doing this behavior.. but I ve no idea what to test / change to fix that 😅 any ideas ?

[odd kettle]

yo in what channel I could ask for help in general stuff?

[raw flint]

Not in AI channel 😁. #💻┃unity-talk

[kind spire]

I have been using NavMeshLinks as shortcuts for my nav surface but my agents seem to occasionally be getting stuck on them.
I checked around and found Off Mesh Links mentioned, am I suppose to use those for shortcuts across gaps and stuff or is there something else wrong?

[fading hedge]

hey i made an enemy which follows the player and is supposed to stop a distance away from the player but it ends up just sliding towards them. I set a drag value and tried putting physics material on the enemy and floor but nothing worked. Does anyone know how to stop the enemy from sliding?

[wicked badge]

are you using navmesh

@fading hedge

[fading hedge]

yeah i have a navsurface on the floor and a navmeshagent on the enemy

[wicked badge]

send a screenshot of the navmeshagent on the ai

of the inspector

[wicked badge]

stopping distance

just turn it up

also maybe up acceleration

when I used it it seemed like it was on ice almost and kinda slide around

when I turned acceleration up it seemed more responsive and made sharper turns

and make sure the humanoid settings are correct

[fading hedge]

yeah turning up acceleration worked ty

[fading hedge]

does anyone know why the enemy's movements are so jittery?

this is it's nav mesh agent and ai

using System.Collections.Generic;
using UnityEngine;
using UnityEngine.AI;

public class Basic_enemyAI : MonoBehaviour
{

    public NavMeshAgent agent;
    GameObject player;
    [SerializeField] LayerMask groundLayer, playerLayer;

    private void Start()
    {
        agent = GetComponent<NavMeshAgent>();
        player = GameObject.Find("Player");
    }
    void Update()
    {
        agent.SetDestination(player.transform.position);
    }
}

[loud granite]

so im currently using unity's navmesh system, and I pretty much want to make a moving obstacle walkable and non walkable at certain times during runtime, problem is it wont even be walkable when i start the game, i already used stuff like navmesh modifiers but it just wont work

[wicked badge]

turn the rigid body to interpolate

[fading hedge]

that worked ty

[radiant palm]

try navmesh surface component

[loud granite]

On a moving obstacle?

[radiant palm]

yup

look up how it works

but any navmesh-runtime stuff usually comes down to the navmesh surface

[hexed radish]

how can I fix the enemy rotation?

[radiant palm]

it looks like its at a 90 degree offset

so it will most likely be an issue with the exported model

so id look at those settings

[loud granite]

Ill give it a shot, btw is it also possible to make it only apply for some layers or navmesh agents

[radiant palm]

Yes very easily possible to only apply to certain layers

Look into it

[loud granite]

Ty

[hexed radish]

I think there's a solution for this

instead of export again

[loud granite]

do i put the navmesh surface component on the obstacle, if so then what do i do since nothing works

this just seems impossible to solve at this point, Ive looked it up everywhere

[radiant palm]

send a screenshot of the obstacle and its surroundings so I can know what you’re working with

What doesn’t work? There isnt a navmesh? The ai don’t use the navmesh?

[loud granite]

Btw the obstacle is the player

And im also working in 2d

[radiant palm]

oh both of these would have been very useful to know

i dont have experience in 2d sorry

[loud granite]

Its cool

[dim plover]

Hey, a question. In my 2d game, I flip my characters when they move left, I mean the component x of movement direction is negative. The problem is that sometimes they are fluctuated, they are flipped rapidly especially when they move up/down, the component x of movement direction is around zero
I can add some conditions to mitigate flipping consecutively but is there any better way, does nav mesh ai have something for it?

[lost plume]

Pulling this down again..

[vast sonnet]

I have set stopping distance to my nav mesh agent
but even so the agent comes staright to me and stops

[valid mirage]

Still Debug.Log turns false? How it can be false even SamplePosition is true? It find a place on a NavMesh but dont set the destination.

[molten sequoia]

Probably didn't calculate the path yet.

[trail lagoon]

I am embarassed to ask this, I have been using Unity for six years and haven't messed with the Navigation stuff in unity that much. But what is that gray box outline thing on the floor and how do expand how large it is?

The agent won't move past it, it's like it's stuck in a box.

Oh also, it works in another scene with the exact same floor and the exact same agent and the exact same navmesh. It's only in this scene that it decides to not work.

[tardy junco]

If you're referring to the rectangle area that looks darker, it's probably not related to navmesh. At least not directly. Did you try clicking it? Perhaps it's caused by a separate object? If the navmesh is baked for that object only, that would explain why the agent can only move inside of it.

[trail lagoon]

The Grey box only appeared when entering play mode. The way I managed to fix it was just moving the ai to a different point on the navmesh and it functioned normally. However, the weird gray box still appeared, but it was able to exit the box. When it did exit the box, the previous box would disappear and the new one would be in the area that the ai was in. It must've just been a weird glitch.

[molten sequoia]

Probably just visualizing the part of a navmesh the agent is currently on.

[alpine glacier]

    public void SetDestination(Vector3 position)
    {
        agent.SetDestination(position);
    }```What validator do you usually put into your setdestination logic? I have a problem... when you use this once inside an unwalkable area, the agent will try to settle to the nearest possible point it can. But if you keep using this method inside an unwalkable area per frame, the agent just keep standing still since it keeps solving paths. What should I check to avoid that behaviour?

Also, can we limit the maximum amount of turns the agent will do to pathfind? Or limit the length of path?

[tardy junco]

Maybe cache the last destination and compare against it before setting it again. maybe also consider not calling it every frame. Can also check the status of the current path.

[tardy junco]

You can't really affect that.

[alpine glacier]

lemme play with it more, thanks!

[ocean cobalt]

I'm having an issue with some code to make an object "jump" across a NavMeshLink.

This works fine the first time the agent crosses the link, but if the agent tries to cross a second time, it freezes entirely

This is where the coroutine is called

[alpine glacier]

Does unity default allow for 3D free flying agents? Or can we set them up to handle 3d free movements? Or at least possible?

[molten sequoia]

Unity's navigation does not support 3D navigation in that form. You can always roll your own or get a third-party navigation solution.

[noble fog]

@fiery pendant Off-topic to the channel.

[fiery pendant]

? this is the AI channel not? where should I ask this?

[noble fog]

Please read the channel name. And it's not related in any way to Unity either.

[alpine glacier]

Is there a more proper way to make agent's speed vary depending on what they are stepping on? Or raycasting where they at the best approach?

[radiant palm]

Send raycast from feet down to floor -> assign the floor a tag or key identifier -> compare tag to see if should slow down or not

[noble fog]

https://docs.unity3d.com/Packages/com.unity.ai.navigation@2.0/manual/AreasAndCosts.html

[alpine glacier]

Didnt knew that actually slows them tho

Yea, they're not. This just controls the preference

[noble fog]

It's up to you to use that information and apply to movement speed then

[vital bison]

Hey! Is it possible to get the travel distance between the agent and transform?
I only find "remainingDistance" but that compares it with the current target. I want to compare the travel distance and then set the destination

[timid raven]

One solution is to call CalculatePath() https://docs.unity3d.com/ScriptReference/AI.NavMeshAgent.CalculatePath.html, calculate distances between corners of created path and add them together.

[covert creek]

busy struggling with how to go about this but how could I have an NPC with a navmeshAgent be hit by a vehicle with a rigidbody

Trying to get this working with ragdolls

Tried adding a rigidbody with isKinematic but that brings the vehicle to a halt when the collision takes place while setting isKinematic to false

[tardy junco]

A ragdoll and a navmesh agent are 2 separate systems that would fight for control over the object. Which is why you shouldn't have both active at the same time, but switch between them at the right timing.

[iron rain]

Hi, is there a way to set a destination to an agent but not moving it without disabling the agent component? other than just setting the speed to 0

[timid raven]

Temporarily set updatePosition and updateRotation of agent to false?

[iron rain]

Just tried it, it doesn't move the object but it still updates the position where the agent is supposed to be, so it doesn't work. Nice try though

[tulip gazelle]

UserWarning: torch.set_default_tensor_type() is deprecated as of PyTorch 2.1, please use torch.set_default_dtype() and torch.set_default_device() as alternatives. (Triggered internally at ..\torch\csrc\tensor\python_tensor.cpp:453.)
_C._set_default_tensor_type(t) what to do about it?

[timid raven]

Oh, yeah, it doesn't disable the agents internal simulation. My bad :0

[fathom arch]

Did you read it? It tells you what to use instead

[keen sparrow]

i tried to make A* pathfinding algorithm but I'm using 3D and its half working, it will go up the ramps about half way then it will just stop, I thought that including Y axis height would fix it but it didn't.
here's the code https://gdl.space/feyunazace.bash
and this is what is happening

[fathom arch]

Not enough context/code to tell what is going on

Also it would help if you draw a debug line from the character to the current waypoint

And even use Handles.Label in OnDrawGizmos to draw some info (such as current costs) on the waypoints. Make sure to wrap it in #if UNITY_EDITOR though

[outer bone]

I have a question, how does the navmesh work internally? i would like to know how they make the navmesh conform to all the surfaces. is every point in the grid raycasted downwards? and every hit that is detected is a node?

I found a tutorial for a vehicle navigation system so that a vehicle can drive autonomously but it requires a more complex navigation system than i think could be achieved with the build in navmesh so that's mostly why i would like to know.

[tardy junco]

We don't know for sure. This is part of the internal implementation that unity doesn't reveal. It could be raycasts or combining the appropriate geometry and simplifying it in some way. You'll need access to the source code to tell for sure.

[tardy junco]

The navmesh is pretty complex. You can pretty much do anything with it.
But there are alternatives, like A* navigation assets.

[outer bone]

Yeah, i mostly want to follow the tutorial which was made for unreal. They made their own A* i think. For vehicles a custom approach will be better i think since a car has to turn and reverse sometimes compared to a character that can just always rotate around it's own axis

[tardy junco]

I don't see why you couldn't implement that with a navmesh. You don't have to use the navmesh agent, you can navigate the navmesh however you want.

[outer bone]

I am not too certain if it really is completely adaptable to vehicles since they behave different than cars. like the turns. i can try though but i would just like to follow the tutorial i found for most of the logic.

[tardy junco]

Sure, you can do whatever you want. Just saying that it's probably possible with the navmesh as well.

[fathom arch]

IIRC it is based on Recast https://recastnav.com/ but I could be wrong

[keen sparrow]

Oh yeah sorry, I didnt see your message my power went out 🥲

[outer bone]

Ah thanks, i will see if i can learn something from it

[stone bobcat]

so idk if my issue is with the navmesh but the slime enemy in my game isn't moving but still tracks the player an plays the animation accordingly. Been trying to fix this since yesterday but to no avail, can someone please let me know the problem. I'll past the enemy inspector and the code below so you guys can check it out

Any help would be appreciated

[tardy junco]

That's not the correct channel. This one is about navigation package, not machine learning.

[tardy junco]

You'll need to record a video of what's happening, as we can't see the issue in the screenshot.
However, it's likely caused by you mixing the navmesh agent and character controller. These would fight over control of the object. Don't use both. At least not at the same time.

[crude moat]

can you have the channel description updated, then?

ask questions and discuss anything related to AI

[tardy junco]

I'm not a moderator, but the description is indeed confusing.

[crude moat]

<@&502884371011731486> ^

[rustic obsidian]

Moderators can't change it. I'll ask the admins to. But it's also right next to the #1202574086115557446 channel, and has navigation in the name

[crude moat]

are there any channels for ML discussion?

[tardy junco]

#1202574086115557446 as vertx pointed out.

[crude moat]

so no chat 😦

[rustic obsidian]

Nothing stopping you from making a thread and chatting there as long as it's on-topic

[viral owl]

Question, I have a NavMeshLink with Auto Traverse Off Mesh Link = false (and updatePosition = false), but sometimes (not always) the navmeshagent position automatically teleports to the other extreme of the link.
Why that happens? How can I fix that?

[sharp charm]

Trying out the new unity 6 behavior graphs. It works well aside from one issue I'm having where my agents speed gets set to 0 at start

no idea why, it seems to be related to the graph since disabling it keeps the agent's speed correct

Aaaand that's another case of solving my own problem by asking about it. The graph is setting the speed to 0 because I didn't change a default value in the node

[fallow drift]

does anyone know how to fix ml agents lag spikes, im doing the rollerball tutoral, and their is a ton of lag spikes

[fallow drift]

woooooooooow yall are so talkive and helpfull /s

[tardy junco]

Everyone was sleeping. What do you want?

[fallow drift]

???? its was 12pm when i sent the first message

[tardy junco]

You can use the profiler to troubleshoot performance issues.

[fallow drift]

where is that?

[tardy junco]

It was 2 am here.

[tardy junco]

It's a tool for investigating performance.
Google it.

[alpine glacier]

omg..... agent is timescale dependent... speeding up the time causes the agent to overshoot turning

[fallow drift]

wait did you mean learning?

im confused

[molten sequoia]

#1202574086115557446 for ML stuff. I recall a big spike from communication with the Python backend. I don't think there's much you can do about this.

[fallow drift]

bandwagoner

your a bit late

that issue was solved 8 hours ago

[fallow drift]

lol the mod doesn't have a big vocabulary, and doesnt know what bandwagoner means

[keen sparrow]

i think he knows what bandwagoner means, he's probably wondering and confused why you called him that in the first place since it makes no sense, given the context

chatgpt is an idiot

[fallow drift]

inded

[molten sequoia]

It was primarily to direct you and future discussions to the appropriate channel. I left the pointer in since I didn't see a message indicating that the issue was solved.
Don't post here if you can't behave and keep messages on topic.

[fallow drift]

ok

[tardy junco]

Lol, I didn't even notice it was about ML and not navigation. But yeah, the answer is still the same: use the profiler.

[fringe raft]

Hi Folks, sorry for a possibly stupid question, but I've upgraded unity and I can't find the NavMeshSurface in the scene, but I do have a NavMesh being applied. I searched for t:NavMeshSurface, but it's not there. I'd like to rebake it, but I can't. Any idea what I could be doing wrong?

[tardy junco]

Check the navmesh window. Perhaps you have it baked with the old API.

[fringe raft]

Thanks - I think I've got it - rubber ducking always works :D. I deleted a navmesh in the assets then added a new NavMeshSurface to my scene

[vocal fiber]

does anyone know why my navmesh on my terrain is doing this?

[dense sparrow]

It's a little complicated, gotta check a tutorial and set up your objects accordingly. Houses should be marked as unwalkable and the terrain as walkable

[tardy junco]

Can you take a screenshot of the whole object(with the navmesh surface) inspector? Including it's transform.

[vocal fiber]

i kinda fixed it because i added the navmeshsurface on my monster and not the terrain but i think its not the good way because it still doesnt work

[tardy junco]

That's definitely not a great idea. It should be on the root of the objects you want to compose the navmesh.

[rocky quest]

Just saw this channel was a thing and was hoping to get some help on my rollaball tutorial.

#💻┃code-beginner message

[vital bison]

Hey! I have a enemy that I want to reposition whenever the player get caught, I reposition the enemy by simpley changing its transform.position to a random vector.
However the actual " navmesh agent" doesnt follow correctly if there is a wall inbetween the last and current position.
https://i.gyazo.com/584236737f9d669a2e6aa5b6bf9d6e6c.png as you can see the navmesh got stuck on the other side of the wall, how do I update/change agent position correctly?

SOLVED: The correct way to teleport a navmesh agent was to use NavMeshAgent.Warp

[spring magnet]

Hey so for my game I have randomly generated terrain divided into chunks with nav mesh surfaces on each chunk of the terrain. The problem is that my AI agents can't go across each chunk since there is a small gap for each of the chunks, I tried adding a small local nav mesh surface to each enemy that bakes a small area around them as I thought this would allow for them to us that to traverse the gap but they still get stuck. I can't really add nav mesh links because of of the random terrain and each chunk would require 100s of links for each vertices. I'm really stuck and have no idea how to fix this. Any help is appreciated.

[pallid hazel]

it's not possible to have a navmeshsurface scale along with the object it's attached to during runtime without rebaking each time is it? It already translates and rotates which is why I'm not sure if I'm doing something wrong

[granite spire]

I have a map with a bunch of stairs and I'm using links but the ai isn't tracking them. How can I put it in his path?

[alpine glacier]

    private IEnumerator SolvePathAndAssign(Vector3 targposition)
    {
        CurrentDestination = targposition;
        
        NavMeshPath path = new NavMeshPath();
        agent.CalculatePath(targposition, path);

        SolvingPath = true;

        // how to check if the navemesh path is done?
        while (path.status != NavMeshPathStatus.PathComplete)
        {
            yield return null;
            // agent.CalculatePath(targposition, path);
        }
        SolvingPath = false;
        agent.SetPath(path);
        agent.isStopped = false;
    }```

I want it to only assign the path to agent when it got a path already solved

So that it doesnt have to stop and calculate the path

Actually I need an asyncronous method, what should I use?

hmm.. there's no async navmesh calculate path?

[tardy junco]

Does this not work?

[alpine glacier]

It can fail since CalculatePath is sync

or rather it isnt doing what it's intended

the while is only evaluated once since the path.status will never change at that point

[tardy junco]

ah, indeed. Setting the agent destination should be async though

[alpine glacier]

yea, but it doesnt do what I want, wehre it will try to keep walking while solving another path

so it doesnt abruptly stop while thinking a new path

[tardy junco]

I see.

[alpine glacier]

making an invisible agent solve the path in async, and then passing the path to the original agent? That seems doable, right?

and maybe chop the corners the the original agent doesnt need (the corners it already passed)

CalculatePathAsync when

[alpine glacier]

        if(Brain.unitController.AgentStatus == AgentStatus.Idle)
        {
            // Walk random
            if(  Brain.unitController.agent.velocity.sqrMagnitude < 0.1f  && !Brain.unitController.agent.pathPending )// check if the agent is not pathfinding
            {
                Debug.Log("xxxx");
                Brain.unitController.WalkTowards(Brain.transform.position + Random.insideUnitSphere * 4); // todo: add roam data
            }
        }``` another q ![cat_shy](https://cdn.discordapp.com/emojis/1153690654615027722.webp?size=128 "cat_shy") 
When time scale is fast, the agents barely move, no clue why tho

[muted vale]

any idea why this navmeshagent walks on the terrain but clips through the bridge?

[molten galleon]

does the bridge have collision

[muted vale]

Has a Box collider on the parent I just tried, and a Mesh collider for the child

My player can stand on it

[molten galleon]

do you have a charactercontroller or something on your nav mesh agent? can you show

[molten galleon]

oh it has a rigidbody. nav mesh agents shouldnt have that

[muted vale]

But if i but a long cube in the water next to the bridge he will walk over that fine, he has a kinematic rigidbody to push me out of the way when walking and it works fine everywhere else

[molten galleon]

if you remove the rigidbody he wont push you anymore?

[muted vale]

Yes, and if i use an invisible cube under the bridge he will navigate fine, so I guess I am just using that

[atomic socket]

[SOLVED] Hi,

I am using NavMesh in Unity 2022 LTS in my 2d game and am getting a strange problem when trying to spawn NavMeshAgents on separated NavMesh areas.

In the below screenshot, I try to spawn agents in areas B and C but they will always just spawn in NavMesh area A. It’s as if the game does not realise the areas B and C exist. As soon as I build navmesh area “bridges” between the areas they spawn as expected in the correct location.

What could be the problem here? I use BuildNavMesh() to dynamically update my navmeshes when I change levels in game but the Scene view seems to show the meshes have been built correctly during runtime? Anyone else run into a similar issue?

[atomic socket]

Turns out if you're using Netcode for GameObjects and your agent is a NetworkObject you need to delay the NetworkObject.Spawn() for one frame after Instantiating your prefab.

[plain pivot]

I'm having issues generating jump links on my nav mesh, you can see drop links generate fine, but jump links don't seem to be generated properly

I'm not sure what I'm doing wrong

[plain pivot]

Sounds like a bug to me.

Have you filed a bug report?

[covert creek]

Hi everyone, does anyone have tips for making performant crowds, how could we optimise what's happening with the Navmesh Agents or might it be better to make a custom solution?

[tardy junco]

The main bottleneck with navmesh agents in this scenario would be from dynamic path/collision solving with other agents. If you disable them from colliding with each other, it should be quite a bit faster, but they would stack up on each other.
You can write your own solution for dynamic collision and make it as performant as you need it to be(potentially even running the calculation on a background thread or jobs).

[covert creek]

That's another thing I haven't found anything on running navmeshagents on jobs/burst do you have any resources or documentation that could lead me down this path? We're using a sort of AI LOD system by enabling/disabling components based on distance and occlusion and once they're close enough I want to switch them to rigidbody movement following calculated paths

[tardy junco]

You can't run navmesh agents in jobs. You can some custom math though.

[covert creek]

What about calculating paths or is all Navmesh agent logic unable to run in jobs

[tardy junco]

I think it's all synchonous and runs on the main thread.
https://docs.unity3d.com/ScriptReference/AI.NavMeshAgent.CalculatePath.html

But as I mentioned earlier, the path calculation iself is probably not the bottleneck.

[covert creek]

Possibly, will need to learn how to properly read the profiler for that then

[plain pivot]

Is there a workaround to do this nowadays?

(Also I have no idea why they can't do vertical jumps, the technology exists for falls...)

[viral owl]

Any idea why could I get this exception?

Error Message: EXCEPTION_ACCESS_VIOLATION_READ

[ 00 ] block_remove
[ 01 ] tlsf_free
[ 02 ] public: virtual DynamicHeapAllocator::Deallocate(void *)
[ 03 ] private: DelayedPointerDeletionManager::CleanupPendingMainThreadPointersInternal(void)
[ 04 ] public: virtual DualThreadAllocator<class DynamicHeapAllocator>::Allocate(unsigned __int64,int)
[ 05 ] public: MemoryManager::Allocate(unsigned __int64,unsigned __int64,struct MemLabelId,enum AllocateOptions,char const *,int)
[ 06 ] private: NavMesh::ConnectIntLinks(struct NavMeshTile *)
[ 07 ] public: NavMesh::AddTile(unsigned char const *,int,enum NavMeshTileFlags,int,unsigned __int64 *)
[ 08 ] public: NavMeshCarving::ApplyCarveResults(void)
[ 09 ] public: NavMeshManager::UnloadData(int)
[ 10 ] public: CallbackArray1<int const>::Invoke(int)
[ 11 ] protected: RuntimeSceneManager::UnloadSceneInternal(class UnityScene *,enum UnloadSceneOptions)
[ 12 ] public: virtual UnloadSceneOperation::IntegrateMainThread(void)
[ 13 ] private: PreloadManager::UpdatePreloadingSingleStep(enum PreloadManager::UpdatePreloadingFlags,int)
[ 14 ] public: PreloadManager::UpdatePreloading(void)
[ 15 ] ExecutePlayerLoop(struct NativePlayerLoopSystem *)
[ 16 ] ExecutePlayerLoop(struct NativePlayerLoopSystem *)
[ 17 ] PlayerLoop(void)
[ 18 ] PerformMainLoop
[ 19 ] MainMessageLoop
[ 20 ] UnityMainImpl(struct HINSTANCE__ *,struct HINSTANCE__ *,wchar_t *,int)
[ 21 ] UnityMain
[ 22 ] __scrt_common_main_seh
[ 23 ] BaseThreadInitThunk
[ 24 ] RtlUserThreadStart
[ 25 ] UnhandledExceptionFilter

[tardy junco]

Seems like possibly an engine bug. Does that happen when you unload a scene with some navmesh related objects?

[normal sonnet]

Idk if it's a common issue but, whenever my NavMesh agent moves toward an obstacle, he slows and jitter as if he was hesitating or something, If it isn't a common issue I can try sending some videos

[alpine glacier]

does anyone know how to fix it?

[viral owl]

Sadly I couldn't contact with the person which have the error. But our game do use NavMesh so it's quite likely the scene contained it. The problem is that it doesn't happens "always" so it's super difficult to identify the source of the error.

[hearty granite]

What would be the best approach for this? I have a building with certain areas that can't be traversed by agents that don't have a matching area mask, but if the mask does match I want it to act like the surface is just a single surface (so no offmeshlinks, as it creates unnecessary delays). I tried stacking multiple surfaces on top of each other, but that seems to break offmeshlinks. I made sure that every offmeshlink is only connected to a single surface, but agents don't seem to be able to swap between different surfaces. Does anyone have a solution?

[hearty granite]

Found it: Had to use navmeshmodifiers

[deep socket]

Hello, im making a zombie AI tha tcould climb other zombies and walls to get to the player, are there any better AI solutions to do that or is it douable with the present Default navigation solution?

[keen sparrow]

you could use the built in navmesh with some logic to make it happen or you could make your own navmesh system using A* or Grid system using A* with some logic to make it happen

[tired viper]

why is the navmesh so high up?

[cyan stump]

Anyone have any tips on how to do A* pathfinding with NPC's so that they don't run into each other? I tried making the NPC's increase node cost but that messes up their own pathfinding. Should I try some steering or replusion or something for them instead? Anything else I can do? Maybe switch to navmesh?

[tardy junco]

Make the nodes that other npcs are standing on unwalkable

[cyan stump]

So I was using a layermask and putting all the NPC's on the NPC layer and using that to adjust costs, is it not correct?

[tardy junco]

Does it work?

[cyan stump]

no cause the NPC's own nodes are also getting increased costs leading to weird behavior

[tardy junco]

Then no

At least not the way you implemented it.

[cyan stump]

I guess I do not understand how to put in logic that makes them ignore their own nodes while not ignoring other NPC's nodes

[tardy junco]

Well, you'll need to look at the documentation.

[cyan stump]

Do you think Navmesh could be used instead?

[tardy junco]

Sure. Navmesh would work out of the box.

[cyan stump]

What documentation do you suggest looking at?

[tardy junco]

The documentation of whatever asset/plugin you're using.

[cyan stump]

I did it myself lol

[tardy junco]

Aah. Well, then there's no one to ask other than yourself.

[cyan stump]

How would you do it?

[tardy junco]

How are you making walls unwalkable?

[cyan stump]

I set them to cost 100

and all the other nodes are cost 2

and NPC's and Player are cost 50

[tardy junco]

I see. So, it seems like you have a state shared among all agents.

[cyan stump]

Yes

[tardy junco]

You need to provide separate state for agents.

For example, each agent could hold an additive mask that you apply to the global map.

[cyan stump]

Would this need multiple node maps?

[tardy junco]

Or something like a list of other agent positions. When you calculate the path, check if an agent is in there and add extra cost.

[tardy junco]

Perhaps. Or a dictionary with a node position/Id as a key.

Many ways to go about it. I don't know your implementation, so it's hard to say what would be a good way.

[cyan stump]

I am using colliders to increase node costs, is this incorrect?

[tardy junco]

Depends. I'm not sure how they fit in the whole implementation, so hard to answer that.

[jovial wharf]

Guys and girls. I dont know anymore what to do. I am extremely frustrated. Me and my friend are trying to create an agent with ML-Agents which plays Tablesoccer/foosbal. We already set up an environment a long time ago.
But he have the problem, that after some time it keeps converging to the values -1 and 1. After some time it just stops doing anything and moves to the extremes.

The Input looks something like this then:
Translational:
1, 1, 1, -1
Rotational:
-1, 1, -1, -1
(4 Rods -> Translation + Rotation)

It makes no sense, that it learn this. It gets heavily punished for that exact behaviour. Thats also why our agent start with Rewards between -3 and 3 because its acting randomly, but in the end when it stops moving the rewards drop to -200 and below.
There is so insentive for the agent to do that. But yet it does that. Is there someone who has ideas on what it going on it would be hugely helpful. We already put 100s of hours into this project because it is for university and are on the edge of going insane.

Please, does someone have an idea on how to fix it?

Observations:
Positions, Rotations, Ballposition, Ballspeed, Rotation/Translationspeed. Everything is normed between -1 and 1 and correctly mirrored for each team.

Rewards:
Small Reward for shooting, even smaller penalte for shooting into the wrong direction, Reward for Goal, Penalty for Goal from Enemy, Small existential penalte, High Penalty for going into extremes (-1 and 1).

Actions:
Translation and Rotation between -120 and 120 degress for the rods

[spring bridge]

How do you detect Cover points for AI in a shooter game?
Dynamically - not baked in.

From my understanding there's only two ways - #1 using nav mesh and every point the nav mesh ends becomes a cover point.
#2 using objects that are part of a particular layer, and using the intersection of that object with the nav mesh to create a cover point

However - this would mean any wall/surface becomes a cover point which is inaccurate/unrealistic. A cover point is an edge from which you can conceal/cover and fire back from if needed.

So I made a script that detects the edges of objects of 'Cover Layer' which includes walls and obstacles, and it detects the height to determine if it is tall enough to conceal the enemy and also short enough to be fired from above - leaning over the cover.

If the edge of the object is an open edge, meaning it does not intersect or touch another object of the cover layer nearby, then its a proper cover point. If it does touch or intersect with another object, such as the corner of a room being made up of two walls intersecting/touching, then it becomes a 'corner' point instead of a cover point.

But my script is just not optimized still and makes mistakes, its about 80% effective. Has anyone solved this already?

[eager verge]

8 years ago, i would have just raycasted at points along the mesh.
https://discussions.unity.com/t/check-if-a-collider-is-visible-from-a-position/635767
But these days I'd probably use the gpu to see if it's occluded (suggested in the final example)

[uncut pagoda]

How do I make navmesh avoid trees and objects place in terrain. Currently, it just bakes through them and my ai are walking through trees. Using Unity 6 URP

[pastel epoch]

anyone knows how to stop the navmeshagent to stop repathing once it reached its destination?
auto repath is off i just set the path with agent.SetPath(path);

https://i.imgur.com/ehXJr7j.gif

once it reached the destionation it shouldnt repath to it once i move it

i tried using NavMeshPathStatus.PathComplete but that is always true

the only solution i found is

if (agent.hasPath && agent.remainingDistance <= agent.stoppingDistance)
{
    agent.ResetPath();
    Debug.Log("ResetPath");
}

[tulip gull]

Anyone know how to make an object carve into one nav mesh and not carve into another nav mesh?

[fallow drift]

can someone help me i dont know why ml agents if doing this

[tardy junco]

That could mean that the agent never reached it's destination. What settings do you have it configured with?

[tardy junco]

Ml agents questions go to #1202574086115557446
This channel is only for Navmesh and everything related to it.

[pastel epoch]

https://i.imgur.com/PNlcJsT.png

idk why NavMeshPathStatus.PathComplete would be always true tho

[tardy junco]

Path status is in regards to calculating the path, not following it.

[tardy junco]

Stopping distance is set to 1, so I guess that's not what I guessed originally.

[fallow drift]

No its for all, oh sh!t they changed it this used to be called ai-chat, CURSE YOU MODS!

[pastel epoch]

that was my first guess aswell

[tardy junco]

Does this happen if you set destination instead of a path?

[pastel epoch]

Yep both have the same behavior

[tardy junco]

Weird. I don't remember such behavior. Is that unity 6?

[pastel epoch]

yep unity 6 but i didnt see anything in docs related to this or any change in notes

might have to dig deeper

[half pagoda]

Satisfaction. Not perfect, but some good progress at least.

[dark river]

    [SerializeField] private NavMeshSurface navigationSurface;

    public void UpdateNavigationMesh()
    {
        try
        {
            navigationSurface.BuildNavMesh();
        }
        catch (Exception ex)
        {
            Debug.Log(ex.Message);
        }
    }``` when i do something like this the editor just freeze and sets the cpu usage to %100 but when i click bake on the editor it takes less than a second to complete what do i do wrong?

also im sure that i call UpdateNavigationMesh once so not like every frame

[tardy junco]

Try breaking with a debugger when it freezes.

[fluid leaf]

What is the right way to use ai navigation system? Is it through attaching monobehavior scripts in navmesh agents, or using animators state machine or should i use the behaviour tree package that unity provides?

Which is the best (easy and robust) way for creating ai NPCs in unity

[long wave]

Not sure if I should ask here, but I'm trying to use A* pathfinding https://arongranberg.com/astar/

Can I somehow use it to let AI move on like a small city with exterior area and also buildings that you can enter and have multiple floors?

[molten sequoia]

Both Unity Navigation and A* Project should be able to generate navmeshes and navigate them out of the box for that use case. Are you facing a specific problem?

[neon swift]

Hello, In order to make ml with ml agent on this rover, I imported the urdf model using URDF importer plugin but I have a problem with a joint which as you can see on the screen, it goes through the place it's supposed to rotate, I don't have enough unity skills to fix it handily, if someone has an idea, thank you 🙂

[crude moat]

I'm working with NavMesh for the first time, and I'm supplying the fields for my Navmesh Agent. Some of the properties pertain to the speed, acceleration, etc. of my agent, but these values aren't constant for my game's agents. For example, the agent has a slower top speed when backpedaling (running away from looking direction) than when walking forwards, as one example. What are the implications of these values? I'm thinking to just supply some approximate constant values for now - documentation ref

---

Ah, okay, I'm actually seeing that NavMeshAgent handles way more of the Navigation proces than I want because it has bearing on the physics of the GameObject it governs, which is not what I want. I found that, when the NavMeshAgent was active on a GameObject and I tried to punch it off of the map, the... well honestly, I don't really know what happened, because the object's model stayed on the stage, but the HPBar above the player fell off the stage. I don't know why the coordinates differed, but it isn't what I wanted. So, I think I'll just have the GameObject handle its path by calling methods from the NavMesh directly.

[crude moat]

Is there any way that I can make a NavMesh surface's cost a property of the surface itself rather than the agent? As related to my previous message, I can't use NavMeshAgents for my implementation because it impacts the GameObject's physics, but I do want to take advantage of cost decisions in the NavMesh.CalculatePath operation

[heavy vessel]

working on ML-Agents Hummingbirds tutorial. These errors came, but I don't think they should exist

the reason I don't think they should exist is cause i was really frustrated by following the tutorial and it still giving errors so I copy and pasted the source code

and it still gave errors, the ones I put above

only happens when it says "public override void" like here

so I tried removing the "override" from it, and then there were no more errors. But then when I saw my scene there was nothing in it and I do not know why

Help would be extremely appreciated. I don't know how much people care for ML-Agents these days but this is for school so it's rather urgent

[molten sequoia]

@heavy vessel Move it to #1202574086115557446

[cinder current]

Hey, folks. So right now we are trying to implement melee enemies and I’m struggling to figure out how to move the agent with root motion while attacking. What would be the go to approach for stopping the root motion putting the agent in a wall for example

[high ginkgo]

!cs

[pulsar wasp]

Join the C# Discord server, a programming server aimed at coders discussing everything related to C# (CSharp) and .NET. https://discord.com/invite/csharp

[worn torrent]

Hello everyone

I need to know how to calculate a random point on my baked navmesh surface area

I need to send my enemy to a random point ON the baked navmesh area, I tried layers didn't work. Can anyone please help me with this

Unity Version 2022 LTS

[tardy junco]

Calculate a random point in space in your desired way, then sample the navmesh for a closest point.

[thick vine]

Is it possible to make a* not look like the object is following a grid? Like not move through the centers of each grid space?

[late pulsar]

How can I connect this? Can't find any solutions (Not that experienced).

[worn torrent]

I calculate a random point using random.range method and putting some maxRange flaot value.
But what would be the code for checking the closest point in navmesh ?

[tardy junco]

You should check the docs. I don't remember the exact name. Something in the Navmesh class.

[tacit dawn]

when i bake the navmeshsurface, i get this area that's not covered. I have 3 meshes here....number2 the ramp, number 3 and number 1 horizontal boxes....Any ideas?

[neon patio]

Any solutions to NavMesh for 2D I found one called NavMeshPlus. Is that one that people recommend/use or does Unity 6 have a secret Unity Package I don't know about)

[cold juniper]

how can i do the navigation bake thing on a more complex object like this?

[cold juniper]

nevermind i messed with some settings and fixed it

[south bough]

why navmesh bake is not visible? can someone help

[radiant palm]

you need to enable gizmos via the little wire circle icon

use the internet for additional guidance

[heavy vessel]

i hear ML Agents is kinda dead

so what would I use now if i wanted to develop an AI thing in unity

[severe fable]

What's dead about it
Both the package an SDK are being actively updated

[heavy vessel]

ok

however there arent many new tutorials

anyone know where i could find those or change them to work with the new versions of unity

[crisp jackal]

Apologies for the long video, if you don't want to watch it here's a brief explanation

I've extracted the tree collides from my terrain and gave them all their own gameobjects temporarily. And now i'm trying to bake the NavMesh data so it accounts for the trees as collidables so I can then just delete the gameobjects and potentially save on resources, instead of having not only all of the trees actual colliders but ALSO objects for every single one of those trees that have NavMeshCollider attachments.
I'm not sure WHY, but when I bake the terrain, it sees my player and my monster as a collidable and draw the available walkable space around them, so I just gave them their own layers and excluded those layers from the baking process.
But no one what I try, I can't get the bake to account for the trees game objects I've created.

Again attaching NavMeshObstacles WORKS, like the monster that uses a NavMeshAgent will avoid the gameobjects, but baking the data doesn't account for the obstacles so I HAVE to keep the like 500 tree game objects.

[molten sequoia]

#1202574086115557446 for ML questions

[gilded comet]

hello, soo, idk what happen but my NavMesh agent sometimes shaking in random occasion

[atomic temple]

hello i need help im using astar algorithm to path find the shortest path and on that path i have to move a car from start to end and back to start while returning to start i face a problem of the car stopping just before the start can any one help me

    {
        // Ensure there is a path and the agent is set to move
        if (ConnectionArray.Count == 0 || !agentMove) return;

        // Get the current target position
        currentTargetPos = ConnectionArray[currentTarget].ToNode.transform.position;

        LocatioOfNextNode = currentTargetPos.ToString();

        // Move the car towards the current target position
        Vector3 direction = (currentTargetPos - transform.position).normalized;
        transform.position += direction * currentSpeed * Time.deltaTime;

        // Rotate the car smoothly to face the direction of movement
        Quaternion targetRotation = Quaternion.LookRotation(direction);
        transform.rotation = Quaternion.Slerp(transform.rotation, targetRotation, rotationSpeed * Time.deltaTime);

        // Debug.Log("currentTargetPos : "+currentTargetPos);
        // Debug.Log("Direction :"+ direction);
        // Debug.Log("currentTrget : "+currentTarget);
        if (Vector3.Distance(transform.position, currentTargetPos) < 0.5f)
        {

            currentTarget += moveDirection;
            Debug.Log("CurrentTarget : " + currentTarget);
            Debug.Log("ConnectionCount : " + ConnectionArray.Count);
            if (currentTarget == ConnectionArray.Count)
            {
                Reached = "Location reached ! Returning To start";
            }


            // if (ConnectionArray.Count <= currentTarget)
            // {
            //     moveDirection = -1; // Reverse
            //     currentTarget = ConnectionArray.Count; // Set to last index
            //     Debug.Log("CurrentTarget : "+currentTarget);
            //     Debug.Log("ConnectionCount : "+ConnectionArray.Count);
            //     if (currentTarget <=-1)
            //     {
            //         Reached = "Reached start!";
            //         agentMove = false;

            //     }
            // }


        }


    }```

[raven rapids]

I wonder if their destination is being rapidly changed

Select one of the objects with a NavMeshAgent on it. You should be able to see a gizmo in the scene view that shows where it's trying to go -- it's the circled crosshair in this screenshot (not the other blue lines)

[fluid leaf]

Are we supposed to use animation controller and nav mesh agent together?

[tardy junco]

Depends. Generally you want to avoid using 2 systems that both try to move an object unless you control when each of them does that. Animations(and animation controller by extent) don't normally move the object they belong too, they usually move the bones in the rig that is usually a child of the object with the animator. However, if you have animations with root motion and the animator has root motion enabled, it would try to move the root object(the one that the animator is attached to), at which point it might fight for control with other systems, like navigation.

[fluid leaf]

Isnt animation system based on state machine which is also used to create ai s?

[tardy junco]

finite state machine is just a design pattern in software engineering. It has a many uses across different fields.

[fluid leaf]

Oh ok

[tardy junco]

Which has nothing to do with what you can and can not use together

[fluid leaf]

What was the intended use of add behaviour in animation states of unity

[tardy junco]

It's to add a custom behavior(code) to the states.

[worldly lance]

is it ok to add multiple navmesh surfaces to a terrain? it feels really cost effective.

doing this in order to have 2 agents on 1 surface.

does it bake the surface twice or only once but for both agents?

[raven rapids]

Nothing wrong with it.

You only need one surface per kind of agent, though

you don't need 30 surfaces to support 30 identical NPCs

[worldly lance]

you mean so if I have 30 NPCs I will have 1 surface, and 30 navmesh surface components?

[worldly lance]

that's awesome, thanks

[raven rapids]

No. You will have one NavMeshSurface.

a NavMeshSurface produces a navigation mesh, and then many NavMeshAgents can use that mesh

[worldly lance]

so how do you set a surface to allow multiple agents

on my terrain, the navmeshsurface can only allow one agent type.
so I got humanoid or monster, but it won't let me do both.

[raven rapids]

If you have two kinds of agents, then you need two surfaces

[raven rapids]

hence this

[worldly lance]

oh identical

no, they are not identical.

[raven rapids]

so yeah -- one surface per agent type!

I have three, since I use three kinds of agents

[alpine glacier]

Hi. If I ran out of memory exception during baking navmeshes for many large terrains, Is it because their navmeshes take this much memory, or is it the baking process thats specifically memory expensive?

[raven rapids]

What settings do you have on the NavMeshSurface components?

I wonder if the "Render Meshes" option is more expensive than "Physics Colliders"

[tardy junco]

There would definitely be more memory consumed during the baking compared to the runtime, so the latter is definitely true, but the former is potentially true as well.

[deep socket]

void Destination(Vector3 destination)
{
    NavMeshPath path = new NavMeshPath();
    nm.CalculatePath(destination, path);
    nm.path = path;
} ```this doesnt calculate off mesh links?

[radiant palm]

try adding a little pause inbetween creating the new navmeshpash and the calculating the path, as it may be your calling on the path before it’s been created

[radiant palm]

a little pause i.e yield return new waitforseconds(0.1f) or something

[raven rapids]

I had a problem once where an agent refused to move at all. I was telling it to go somewhere unreachable every frame, and there wasn't enough time for it to actually finish calculating the "best effort" path

Note that this was different from what you're doing -- you're calling CalculatePath directly and then giving that path to the agent

I'm pretty sure CalculatePath is synchronous.

[smoky pivot]

Hey guys , so I worked on this ar indoor navigation project , where you can have a single starting point and multiple target points . Thing is I want those target points also to be the starting point. For example , lets say I navigate to the LAB , now with LAB as the starting point , I need to navigate to all other targets. How do I do this?

[tardy junco]

Code it. Have a list/queue of target positions and iterate to the next target position when you reach the current destination. This has nothing to do with navigation. Just simple logic code.

[smoky pivot]

Thanks . Btw I worked on another project too , and initially it worked fine with the line rendering part but then when I changed the starting position , I could not render the line , why is that?

[tardy junco]

No clue. Hard to say anything without seeing the code and potentially debugging it.

[smoky pivot]

using System.Collections;
using System.Collections.Generic;
using UnityEngine;
using UnityEngine.AI;

public class setNavigationTarget : MonoBehaviour
{
[SerializeField]
private Camera topDownCamera;
[SerializeField]
private GameObject navTargetObject;

private NavMeshPath path; // current calculated path
private LineRenderer line; // line renderer to display path
private bool lineToggle = false;

private void Start()
{
    path = new NavMeshPath();
    line = transform.GetComponent<LineRenderer>();
}

private void Update()
{
    if ((Input.touchCount > 0) && (Input.GetTouch(0).phase == TouchPhase.Began))
    {
        lineToggle = !lineToggle;
    }

    if (lineToggle)
    {
        NavMesh.CalculatePath(transform.position, navTargetObject.transform.position, NavMesh.AllAreas, path);
        line.positionCount = path.corners.Length;
        line.SetPositions(path.corners);
        line.enabled = true;
    }
}

}

[tardy junco]

!code

[pulsar wasp]

Posting code

📃 Large Code Blocks

Use links to services like:
https://gdl.space/, https://paste.ofcode.org/, https://hatebin.com/, https://paste.myst.rs/, https://hastebin.com/

📃 Inline Code

Surround code with three backquotes. Not quotation marks.
To format as C#, add cs to the first line:
```cs
// Your code here
```
Add a comment with a line number if there is an error message.

[smoky pivot]

using System.Collections;
using System.Collections.Generic;
using UnityEngine;
using UnityEngine.AI;

public class setNavigationTarget : MonoBehaviour
{
    [SerializeField]
    private Camera topDownCamera;
    [SerializeField]
    private GameObject navTargetObject;

    private NavMeshPath path; // current calculated path
    private LineRenderer line; // line renderer to display path
    private bool lineToggle = false;

    private void Start()
    {
        path = new NavMeshPath();
        line = transform.GetComponent<LineRenderer>();
    }

    private void Update()
    {
        if ((Input.touchCount > 0) && (Input.GetTouch(0).phase == TouchPhase.Began))
        {
            lineToggle = !lineToggle;
        }

        if (lineToggle)
        {
            NavMesh.CalculatePath(transform.position, navTargetObject.transform.position, NavMesh.AllAreas, path);
            line.positionCount = path.corners.Length;
            line.SetPositions(path.corners);
            line.enabled = true;
        }
    }
}

[smoky pivot]

Better

Is it got to something with the line renderer?

[tardy junco]

A bit busy now. Will have a look later

[smoky pivot]

Oh sure

[tardy junco]

Assuming there is a path between these points(transform.position and navTargetObject.transform.position), the line renderer should render it. Make sure you check if there is a path and that it's complete. Might need to actually sample the navmesh for a closest point. The navmesh can only find path between points directly on the navmesh. And transform.position of arbitrary objects are not guaranteed to be on it.

[fallow drift]

should i feed the ai euler angles or rotation? this is for ml agents btw

also i dont care that this is called ai-navigation, this channel used to be called ai-chat, and i will jump of a bridge before i use the forum ai-tools channel

[fallow drift]

nvm chatgpt helped me

unlike yall

(na the mods gona kill me, getting banned speedrun be like)

[tardy junco]

It was missnamed before. This channel was always about the navmesh navigation system.

[fallow drift]

we NEED a ml-agents channel, ai-tools sucks, because its a forum channel

[jaunty raft]

There’s the hint: don’t talk about it 🙃

[fallow drift]

ya but this is important to me.

[tardy junco]

What's the problem with a forum channel?

I think it's actually pretty good, as the issues stay in the field of visibility for longer and don't interfere with each other.

The lack of answers is due to the nature of questions being very niche or not providing enough info. And not many people in this community dealing with the topic. Even if you were to ask here, the results would be the same.

[fallow drift]

...fair

[lost quarry]

Hello i am trying to implement steering logic for a vehicle and i am running into a problem (pun not intended)

Basically its this, it detects an obstacle, tries to u turn, but because of how constrained steering is, it ends up running into it anyway, when in reality it would have to back up

Any ideas?

[jaunty raft]

Can only be solved by detecting the situation and executing a pre planned 3-point turn. You could maybe just do the initial reverse then retry the forward solution, rinse and repeat.

This will lead to edge cases where you have to accept a solution with the agent clipping or potentially leaving your agent stuck. In general, agents with a steering constraint and forward/backward motion capabilities are an unsolved navigation planning problem. It can be solved locally but not in combination with pathfinding. Your only real option is to avoid the problem. Which means only allowing the agent to be in states that can be solved either by forward steering or with preplanned 2/3-point turns. This is easy when there is always enough room, in the immediate vicinity of the obstacle, to execute the maneuver.

[lost quarry]

i ended up doing the reverse rinse and repeat, but curious about other ideas still, its hard because i also want the agent to avoid other agents so i dont really think a pre planned 3 point would work in this situation

i wish there was more info on obstacle avoidance like that online but i guess it kinda is an unsolved problem :/

[jaunty raft]

yeah, thats why, and you've already reached the best you can do with that reverse and retry, especially if you want RVO avoidance in the mix

you can do some gymnastics with a local A* flood search bounded by distance to find a path for a reversing agent, and evaluate those paths for a locally optimal solution, this can produce relatively neat behaviour but can't solve the navigation problem entirely, as its basically a brute force approach that is untenable performance wise. And A* doesn't understand steering constraints anyway, so you also have to design the environment in a very constrained way.

[raven rapids]

i would wait for the player to look away and then let the car pivot in place 😛

[lost quarry]

If(collideWall)
Transform.position = (random)

[jaunty raft]

you are too AAA

[raven rapids]

A little lie here, a little lie there

That’s all that games are, really!

[dusk stump]

howdy peeps, cranky old dev getting back into unity for the fun of it; ive been spiking on capabilities to constrain my design ideas (cant do it easily? PUNT IT). ran into an issue with ai-navigation and wondering if im actually stuck and should punt it, or whether im just Doing A Dumb and can sidestep the issue easily. Here's the sitch --i got a start on a nice little procedural hex grid happening, with code-driven details like trees and grass and elevation and water level:

hexes themselves are "real" meshes with just variety in materials. But the outlines are a procedural mesh that uses fancy math from a yt video to get the outline seen.

trying to bake tnhe navmesh at runtime leads to curious errors not seen in editor:

my instinct is saying to sidestep ai-nav and just do hex-center to hex-center. only thing i was trying to get out of ai-nav was some more realistic-looking behavior of the little fighting creatures when they moved and attacked from hex to hex

just want to make sure theres not something stupid like a checkbox that says "make procedural meshes work with runtime baking" or something silly like that XD

i guess the other sidestep would be to replace the procedural hex outline (which currently raises a small lip around the painted terrain) with a basic full mesh shape, and have the painted terrain jut slightly above the grey "outline"

but even then based on the navmesh baking im seeing in editor, i wonder if its going to be a lot of lift to accomodate for what seems to be the navmesh's desire NOT to have figures walk up the side of the edge:

my (limited understanding) interpreted of that navmesh rendering is that the figure will gracefully walk up an invisible slope to get to the raised hex to the NNW of it

[radiant palm]

youre not gonna believe this…

for that error: go to the mesh import settings (so the hexagon) and there should be a box that says “Read/Write enabled”. Tick that box.

[alpine glacier]

https://docs.unity3d.com/ScriptReference/AI.NavMesh.SamplePosition.html why doesnt this work on unity6

[molten sequoia]

What's the error?

[alpine glacier]

the AI package is uptodate (cant upgrade in package manager)

[molten sequoia]

Is that the correct class? Do you have a conflicting class in the project?

[alpine glacier]

I cant seem to f12 it

find definition

but it's green

[molten sequoia]

Does it compile in Unity?

[alpine glacier]

oh.. it's conflicting with something else

but I dont have any class with that name

[molten sequoia]

IDE could also just be confused if Unity compiles it without errors.

[alpine glacier]

likely a library problem?

I'd reimport just in case

[molten sequoia]

Just regenerating solution might also fix it

[raven rapids]

I'd consider using physics colliders instead of the actual mesh when baking. It would coincidentally avoid this issue -- but it'd also be a lot more accurate.

[dusk stump]

hot damn that did it! tyty

[radiant palm]

no worries

[eternal mauve]

I would like to inquire whether the ONNX file in the unity ML agent contains all types of learning or if it only retains the best results, Additionally, how can we filter it out if needed?

[tardy junco]

It contains the model weights. No more, no less.

I'm not sure what you want to filter out of it.

[eternal mauve]

Example : My Agent has 50 Episode From That if 10 Episode have some miss - training or say some issue during training agent learns that and performs that next Time When I Start New training From That Last Trained Model So How To Solve This Issue ?

[tardy junco]

Episodes is the amount of training that it went through. The more you train it, the better it would be at certain task. There's really no point in comparing the number of fails during training, especially disregarding the training stage at which these failures happened.

[tardy junco]

And I'm not sure what issue you're talking about.

If after completing the training, the agent still fails at a task, you'll need to train it more, potentially changing the reward/punishment or even the input data that it receives.

[eternal mauve]

ok

[livid marsh]

When I Inspect a script in the Unity Behavior Demo 's Behavior Graph, visual studio code opens to a blank script... hmm any ideas

[thick raptor]

How would I go about implementing a flying navmesh system?

[tardy junco]

You don't. You can only move a long the surface of the mesh, basically only 2 degrees of freedom. It doesn't have volume. You can fake it by modifying the visuals height, but wouldn't work for avoiding obstacles in 3d space.
If you need actual navigation in 3d space, you'll need to implement your own system. Some kind of A* in 3d space.

[thick raptor]

I mean, I guess I could do a if raycast hit, move along y until it doesn't etc

Like those robot projects for children

Some simple stuff and see how it turns out

[raven rapids]

3D navigation is exciting.

https://www.gdcvault.com/play/1022016/Getting-off-the-NavMesh-Navigating

I'm going to try and implement this, eventually

[haughty bridge]

hey guys! how can i fix this error of the navmesh not going below the hips of the character?

solution: i just re-added the navmesh agent component to the asset and it did the hitbox for me

[raven rapids]

It looks like you had the agent attached to the hip bone

[toxic osprey]

I have my navmesh surface properly set up and also the surface looks flat

however when I try to run

The zombie seem to get stuck randomly

running on its on spot

before manage to move again

without showing any error in the console

[tardy junco]

How did you debug so far and what did you discover?

[toxic osprey]

I have completely no idea

I tried debugging

and the zombie is targetting the player

and when he tries to walk

he just keep walking on the same spot

for no reason

[tardy junco]

Assuming you're judging whether it's walking or not by the animation - this is really not related to actual movement. Animations are handled by the animator and don't move the object. The navmesh agent does.

You should use debug logs, rays and gizmos to output and visualize important information.

For example logging the timing where you set the destination of the agent, drawing a ray towards the destination in the scene.

Logging the state of the agent and the found path.

[toxic osprey]

I drew the ray towards destination and it is pointing towards the player

[toxic osprey]

How do I do that?

[tardy junco]

Debug.Log.
And you can find the different properties of the agent and path in the docs.

Or just look through the suggestions in your ide.

[pure flare]

try also changing the values of the radius and the height

i have the same issue but the difference is that it was by dealing with heights

maybe can work

[granite spire]

I have a basic navmesh system placed on a creature for my game but he picks the shortest route which at times does not look the best like climbing stairs from the side. Is there a way to incentivise normalness into him? I heard that there is some cost things I can do. I'm using 2022.3.47f1 btw

[granite spire]

Also he sometimes clips underground how can I get him to relock to the ground

[raven rapids]

If you don't want an agent to climb up the side of a staircase, then you need to make that area unwalkable

[granite spire]

So that's just it no way to add cost incentives

[jaunty raft]

Depends on the implementation whether that’s possible, you can do whatever you want, that said, nav mesh (as a concept) is terrible at representing node weights in a useful way. Typically stairs and climbs would be represented as links between mesh islands with authored weights.

[granite spire]

I'm using the normal navmrsh and he can go throughout the place but as he chooses the shortest path he usually gets stuck or acts weird. I want a way to make him avoid certain areas but still make them accessible for him

[twin hound]

is someone familiar with the A* pathfinding in 2D space?

I have a problem where my AI seems to be flying off to space, I know this is due to the layer mask. Is there some way I can "Invert" the layer mask so that blue areas turn red and vice versa

[raw flint]

Well, inverting a bit mask in code is just a bitwise comlement in front: ~mask

[twin hound]

sounds reasonable...now I can read through 500 lines of code xD

[high ginkgo]

<@&502880774467354641> can we have a seperate channel dedicated only for ML Agents?

[noble fog]

@high ginkgo Don't ping staff with suggestions, use #1161868835423526933

[high ginkgo]

oh ok, sorry about that

[placid vector]

good day everyone so i want to make a nav mesh for the ai however as you can see in the picture the bake is rather not clean. what are the available solutions i can do

[raw flint]

I can't see, where exactly is it not clean? Navmesh is always a simplification of the original mesh (resolution can be chosen btw). I just don't see how that wouldn't be sufficient

[placid vector]

there are someparts that are not connected like this picture for examplem can my agent still be able to walk over that?

[placid vector]

i also kept getting this error over and over, i have a 2 floor level room shown in the picture

[placid vector]

also i seem to be having trouble trying to bake it around the sphere however the navmesh seems to ignore it completly

[radiant palm]

if there are gaps inbetween paths, you’ll need to generate navmesh links too

[radiant palm]

the ai are not correctly spawning on the navmesh, there are multiple solutions to this online

[radiant palm]

not too sure for this one, try ensuring it’s on the correct layer / is included in your baked surfaces

[pale notch]

Hey I need help on how to build a navmesh at runtime
I have the latest navigation ai package (1.1.5) yet i dont have an assembly reference to NavMeshSurface ?

I am on unity 2022.3.16f1

Even with the navmeshcomponents package on github it doesnt work

Like I cant put a navmeshsurface as a navmeshsurface variable

[placid vector]

the ai seem to spawn normally and be able to walk up until the sphere which is its waypoint

[pale notch]

Bump

[raven rapids]

You certainly shouldn't be trying to use both that library and the proper AI Navigation package

Are you using assembly definitions in your project?

[left stream]

Is there any interest in this project I have been working on. I was inspired by the old Autodesk Kynapse AI toolset (used in Medal of Honor, Watchdogs, and various other games). But I am getting burnt out, and thinking of open sourcing the project. It has a nice modular architecture and has pretty cool features like Shooter Avoidance, cover point generation, precomputed visibility between nodes, and node Cost Modifier Volumes. It is integrated with NavMesh as its base. It utilizes an octree for quick positional node lookups, and a simple A* implementation for pathfinding.

Its goal is to provide data for developers to create realistic AI decisions and environmental interactions.

The green orb is a starting location. The 2 pink orbs to the left are "shooters". Notice how the generated path is not the "shortest" path, but the shortest path with the most cover from the shooters. A user of the toolset could identify that some of these points have 50% visibility to shooters, and play a "crouch run" animation, making the movement along the path realistic.

Additional data generation can easily be added by implementing a custom "GenerationPhase" and adding it to the Generation Pipeline.

And each node has a map of visibility to other nodes. So the user can easily ask if one point has visibility to another without needing to perform a Raycast. Visibility is a percentage value, calculated by multiple raycast.

[sharp gulch]

how come my agent completely stops when it reaches a destination? even with acceleration at like 5, it just abruptly stops. this is a big issue when chasing players in my multiplayer game because the player can just walk slowly and the agent just keeps stopping

[granite spire]

what;s the deceleration

[sharp gulch]

like 5

[raven rapids]

What's your stopping distance set to?

[sullen mango]

Heya, been looking into AI assets like Emerald AI. Has anyone used it and can recommend it? Perhaps alternatives? #💻┃unity-talk message

Actually maybe better suited to ask in #💻┃unity-talk

[abstract nacelle]

This is right up my alley. Would love to follow the project if there is a place to.

[alpine glacier]

Hey guys, I got some problems with the navigation

[raw flint]

Please do provide all the information required for someone to help

[alpine glacier]

So i can't fully make the Nav-Mesh, even if it has the Nav-Mesh Modifier it doesn't work and I want NPCs to go on Objects But it just can't bake it..

[alpine glacier]

So does anyone have the answer for that problem?

[tardy junco]

Maybe if you provide a better explanation.

[alpine glacier]

Like some parts are baked and some are not

[tardy junco]

You'll need to provide some more context. Like a screenshot or something as well as show your navmesh setup.

[tardy junco]

So where exactly do you expect to have navmesh?

[alpine glacier]

On the tables and other stuff

[tardy junco]

Did you try reducing the minimum region area? Or modifying the agent type settings?

[alpine glacier]

No

I did not try that

[tardy junco]

Well, then try it. These areas are too small to generate a navmesh with these settings

[alpine glacier]

I swear it did work before but idk what happened to it

[alpine glacier]

And why not with small areas?

[tardy junco]

Because of the settings that I mentioned.

If it worked before, you either baked with different settings or something else changed. Are these object even included in the bake?

[alpine glacier]

Yes they are

[tardy junco]

Then it's the settings.

[alpine glacier]

Okay that's a bit odd..

[raven rapids]

Notice how narrow the navmesh gets as it goes through the door

and how narrow it is in the hallway, really

[alpine glacier]

So the characters can navigate?

[alpine glacier]

Is there a other way to do it?

[raven rapids]

Your agent radius is 0.7

that means that the agent is a cylinder with a diameter of 1.4 meters

that's huge!

they won't fit through a gap smaller than that

[alpine glacier]

Well what should I set it to? 0.2? Or 0.1?

[raven rapids]

A smaller number!

You can try different values and see how it looks

[alpine glacier]

Alright!

Thanks

I will see if that works

[crimson oxide]

Hi, I'm trying to figure out why my navmesh is being built too high above the level geometry. I read that this can be fixed by enabling Build Height Mesh, but for some reason it is appearing greyed out and inaccessible in my Inspector. Using Unity 2021.3.45f1

[alpine glacier]

Honestly i want the old system of the navigation to come back

The window < AI < and then navigation window and that's all

[woven crown]

Sorry for the late reply, but I'd be really interested in checking that out. I've always been disappointed by the limitiations of the built in navigation system for considering stuff like exposure maps or calcualting flanking routes.

[zinc heart]

👍

[zinc heart]

anyone know why I can't generate a nav mesh here

[raven rapids]

What are your NavMeshSurface settings?

[zinc heart]

fixed it by making the map larger

[raven rapids]

Consider using physics colliders instead of render meshes

[meager bison]

Hi, during the training of an ML-Agent the "game" is running at a faster rate, is there a way to do that for inference aswell?

[frail folio]

Ey, I have a few questions, beggining for it's true that the AI navigation package doesn't work by default with 2D? Cause I tried to use it with 2D and couldn't manage for the navmesh grid to show, but I am using a extension of that and it is not showing either, and I am wondering if it might be cause I don't have the toggle for it show turned on, but pretty sure I do so...

I don't know what I am doing wrong

[frail folio]

I followed this and I am not sure how right it is https://www.youtube.com/watch?v=HRX0pUSucW4&t=101s

[restive obsidian]

Hey everyone,

I have an issue with my unity ML-Agents env while I am trying to use the gym wrapper. I
have compiled my env with the agent's behaviour type set to Default. When I am trying to execute my python code to train my custom model the code stucks on "UnityToGymWrapper(unity_env, allow_multiple_obs=False)". It opens the env and doesnt move to the following lines of code. Any suggestions? I import the relevant packages as follows :

from mlagents_envs.environment import UnityEnvironment
from gym_unity.envs import UnityToGymWrapper

The rest of the test code looks like this:

unity_env_path = "./UnityEnv/Test.exe"
unity_env = UnityEnvironment(file_name=unity_env_path, no_graphics=False)
gym_env = UnityToGymWrapper(unity_env, allow_multiple_obs=False)

obs = gym_env.reset()

for _ in range(1000):
action = gym_env.action_space.sample()
obs, reward, done, info = gym_env.step(action)
if done:
obs = gym_env.reset()

gym_env.close()

[raven rapids]

You'll want to look at #1202574086115557446 .

[restive obsidian]

Thank you, I'll take a look there!

[celest wing]

Hey, I'm having some issues where my navmesh agent won't track the player once they get past a certain distance. The monster is navigating a large maze of trees so I don't know if this is a limitation of the navmesh system or if there's something I can do to fix it. I've attached an image of the map and my navmesh agent settings. Any help would be appreciated, thanks!

[raven rapids]

It may be taking too long to calculate a path.

I had a problem once where I was setting an unreachable destination every frame.

It wound up taking longer than that to calculate the partial path, so the calculation never completed.

Try only setting the destination occasionally.

#💻┃unity-talk message

@cursive knoll continuing here:

The obstacle avoidance radius should have no effect on normal navigation.

It's unrelated to the actual agent radius used when baking a mesh

[cursive knoll]

so I set it higher

[raven rapids]

Interesting, so they simulate getting pushed around

[cursive knoll]

it completely ignores every attribute it has: speed, turning velocity, acceleration....

[raven rapids]

Darn

If obstacles aren't giving you good behavior, then I'd consider something more manual

[cursive knoll]

and no, this isnt carve

[raven rapids]

You mentioned using raycasts

[fossil orbit]

Why not just temporary overide the destination when a bullet/arrow is withim certain radius?

[cursive knoll]

yes, but I dont exactly know how I could go with that

[raven rapids]

Something like that, yeah

[cursive knoll]

thats what im trying this time yes

[raven rapids]

I'd consider creating a "danger function"

[cursive knoll]

problem is when it's trying to chase the player

[raven rapids]

It would measure how bad a position is based on how long it'll take for a bullet to get there

[cursive knoll]

because I cant make it dodge and chase the player at the same time

[raven rapids]

You could then randomly sample positions and move to the safest one

in fact, this could be a fun application of gradient descent...

[cursive knoll]

oh speaking of sample positions

[raven rapids]

I wrote a Burst+Jobs enabled system to find the lowest (or highest) value point in a 2D space

where the value is created by summing up gaussian functions

so, imagine finding the lowest point in hilly terrain

That would probably be overkill for reasoning about dodigng a few bullets

[cursive knoll]

    Vector3 RandomWanderPoint()
    {
        Vector2 randomPoint = Random.insideUnitCircle * maxRadius;
        Vector3 newWanderPoint = new Vector3(randomPoint.x, 0f, randomPoint.y) + wanderCenter;
        NavMeshPath navMeshPath = new();
        if (!agent.CalculatePath(newWanderPoint, navMeshPath))
        {
            //Debug.Log("Wander point out of bounds. Recalculating...");
            return RandomWanderPoint();
        }
        if (Vector3.Distance(newWanderPoint, transform.position) <= minRadius)
        {
            //Debug.Log("Wander point too close. Recalculating...");
            return RandomWanderPoint();
        }
        else return newWanderPoint;
    }```

I wrote this whole thing for calculating random wander points

[raven rapids]

I'd just use SamplePosition to snap the point

[cursive knoll]

and needed a lot of edge cases

like the point being outside the map

etc

[cursive knoll]

huh

[cursive knoll]

yeah....

[raven rapids]

I used it to decide where to place buildings in an RTS

resource nodes are good, existing buildings are bad

find the global maximum and slap a building there

[cursive knoll]

I mean, my idea right now would be to use a spherecast in front of every single projectile

[raven rapids]

I would do something like this:

[cursive knoll]

and if it collides with an enemy tank, tell that enemy tank to move somewhere

[raven rapids]

• Look for all bullets that are within 3 seconds of us
• Sample points along their trajectories for the next 3 seconds
• Pick random points and reject them if they're too close to any of those bullet-points

hm, but that could get really expensive

it feels a bit clunky

[cursive knoll]

why not just use spherecasts from projectiles?

I don't need it to be perfect AI

I just want it to kinda try

[raven rapids]

Good point

[fossil orbit]

You could do something simpler, like

• using sphere trigger to detect incoming bullet,
• store and sort bullet that enter the trigger
• remove bullet leaving trigger
• check bullet direction compare to position, and decide dodge direction from that.

[cursive knoll]

like if I manage to hit it with a ricochet projectile, thats fine

[raven rapids]

I'm wondering if you could stop them from driving into a projectile

since they'll strictly be avoiding getting hit while standing still

[cursive knoll]

I just want it to make an attempt at dodging projectiles coming straight towards them in a linear way

[raven rapids]

Actually, here's an idea

Every time the tank decides it needs to dodge, it puts its position in a list

these are now repulsors: when picking a point to escape to, it tries to move away from all of the repulsors

I like these "gravity" models. They're pretty simple to implement and can give good results

I remember using this for an RTS AI way back in high school

[cursive knoll]

wait... I only have enemy tanks on a layer, and I'm only spherecasting that layer.... why am I getting nulls?

[cursive knoll]

I don't really get what you mean

[raven rapids]

[Serializable]
public struct Influence {
  public Vector3 position;
  public float score;
}

public Vector3 GetDesire(Vector3 position, List<Influence> influences) {
  Vector3 result = Vector3.zero;

  foreach (var influence in influences) {
    Vector3 delta = (influence.position - position);
    float factor = delta.sqrMagnitude;
    result += influence.score * delta.normalized / factor;
  }

  return result.normalized;
}

Something shaped like this!

It's just like gravity: the closer you are to something, the more it affects you

In this case, you would add a strong positive influence at the player's location, then add negative influences for the bullets

This will naturally steer the tank away from bullets

[cursive knoll]

how....exactly?

I get the part

on how it calculates positions

but how would I make it move to that position?

wouldnt I need to update the agent's destination every frame?

[raven rapids]

You would no longer be using the agent to directly control your position. It'd just tell you how it wants to move.

You would then mix this with the influence direction

Imagine the agent says "go right" and the influence says "go up"

maybe the tank should go right and up

[cursive knoll]

yes, but how

[raven rapids]

by...adding the directions?

the tank would then drive in that direction

[cursive knoll]

but i need positions to move to, not directions to move in

[raven rapids]

an agent can tell you the velocity it desires

even if it is not updating your position and rotation

[cursive knoll]

im sorry, im having issues understanding this

[raven rapids]

Currently, you are letting the NavMeshAgent set your position.

[cursive knoll]

well, I let it change the velocity, yes

but its completely up to me where it's destination is

[raven rapids]

That's a given. You always tell it what the destination is.

I'm saying that the agent should not be directly controlling where the tank is

https://docs.unity3d.com/6000.0/Documentation/ScriptReference/AI.NavMeshAgent-updatePosition.html

Setting updatePosition to false can be used to enable explicit control of the transform position via script. This allows you to use the agent's simulated position to drive another component, which in turn sets the transform position (eg. animation with root motion or physics).

You can set this to false, then read https://docs.unity3d.com/6000.0/Documentation/ScriptReference/AI.NavMeshAgent-desiredVelocity.html

[cursive knoll]

I was thinking of something like this:
Draw spherecasts from a projectile
If the spherecast hits an enemy tank, make it call a function (maybe tell the function the direction of the projectile) which makes it move out of the way temporarily

[raven rapids]

You could also just calculate a new destination, yes

[cursive knoll]

this would work for when they are in "wandering" mode

but problems occour when it needs to chase the player

[raven rapids]

You still might want to use the "influence" system to decide where to flee to

[cursive knoll]

I can't just make it go elsewhere

[cursive knoll]

personally id just generate a random point

id just need to make sure somehow that the random point isnt inside the trajectory of the projectile

[cursive knoll]

this does look interesting

i just dont know how id use it

i think itd be extremely noticable when the tank switches from chasing to dodging

[raven rapids]

in my game, I have updatePosition set to false on my agents

I just use their desired velocity to tell the entity's locomotion system what to do

[cursive knoll]

I dont really get it honestly

im trying to but its weird

[raven rapids]

When you turn off updatePosition, the agent's simulated position can move away from its transform's position

I just warp it back every frame

[cursive knoll]

wouldnt that make for jittery movement?

also I have an idea

[fossil orbit]

I dont think it would, infact if you dont move the tank manually, it wont move at all. You just using the agent to get the velocity, which you can use to move the transform anyway you like

[raven rapids]

No, because the agent doesn't control anything

It just tells you where it wants to go

[raven rapids]

Precisely!

[celest wing]

that fixed the issue, thanks!

[cursive knoll]

ohh the velocity, I get it

[cursive knoll]

i was thinking of something like this

where, whenever the tank gets hit by the spherecast the bullet is projecting, it tells the tank to move in a perpendicular way from it

it doesn't matter if the tank doesn't manage to move out of the way because, idk, a wall is in the way

call that the player outsmarting the AI

I just need it to make minimal effort at trying to dodge it

and I was thinking that maybe if its chasing the player, make it go in a 45 degree angle instead of a 90 degree one

so it still tries to move towards the player, and because it's chasing, its more likely to get hit

oh hey it kinda works! @raven rapids @fossil orbit

I can set the amount of time it should dodge for as well

at least it tries a little bit

i dont think i want more than that

[cursive knoll]

although I will certainly look into this on my main project

I want to use navmeshes there so I can make enemies move towards you

but I won't want to use the movement of the navmesh

I didn't know that you can turn it off while keeping the velocity

[magic stirrup]

Anyone know a good way to add a cooldown after a sequence has been executed in Unitys new Behavior Graph. I have a ability that my enemy performs and want to add a cooldown of X seconds to it. It is apart of a Random node so I don't want it to run if it is selected until the cooldown has resetted.

[hardy cave]

Hello guys, how do people work with Behavior pakage, Blackboards?
I cannot find any way to have a variable modified or added in a blackboard and in another script to recieve it, or to see the same blackboard modified. Isn't a blackboard suppose to be a block of data with global access?

[radiant palm]

#1202574086115557446 is probably the more knowledgeable channel

[jaunty raft]

A blackboard (as a concept) is supposed to be the limited view of the world that a given ai-behavior can ‘see’ and access. Whether it’s global or shared depends on the implementation/configuration of the blackboard. In unity, if it’s a SO you should be able to reference it in other behaviors, if it’s a component it’s probably local to the behavior on the same gameobject. Potentially there is an option for having truly global blackboard that is automatically available. I would look for ways to have these three different scopes and that should then illuminate your issues. This stuff is conceptually the same for all the different behavior tree packages/assets/libraries.

[hardy cave]

This put a little light on my problem, Thanks!

[gleaming briar]

morning people, actually afternoon... does anybody knows why can't i bake-Add this shape into my scene, as all my characters are ignoring it like ghosts? (( what i tried: baking based of mesh render, baking based of collider, make collider convex (althou the resulting shape wouldnt work) )) but it won't take it... ALSO i baked a simple plane to add to the scene which baked correctly but the cilinder guys ignored even that one. so what do you think?

[gleaming briar]

i haven't found any tutorials on the new system

[gleaming briar]

here, i'm not kidding, it's a complete breakage of physics

that nav link didn't help btw

[gleaming briar]

nope, i thought i just had to adjust those cilinder guys before baking specifically for thei sizes but like the first image, the navmesh won't accept that

im gonna scale everything everything in the scene x10

[gleaming briar]

there's no option then, i'm gonna make in blender the plane and the hill as one mesh, and that will be the only mesh

[raven rapids]

I don't know what you mean by "cylinder guys"

[gleaming briar]

ah those are the navmesh agents, the "walkers".

[gleaming briar]

to answer the question, first put the plane and the path inside a parent empty object, give the navmesh surface component to the parent and to bake it, change collect objects to "current object hierarchy"....
PLUS very important if you were working on the previous version of navmesh and then you updated, there might be an extra surface that is impossible to access and so to remove, so your agents will still stick to that floor and go thru anything even if the new baking looks proper. IF that is the case, the only way to solve it: get everything into a whole new scene.

[long moat]

Some part of my Navmesh agents are buried down to the floor

Does anyone know how to fix this? FYI the agents have Rigidbody and their own box colliders, separate from the Navmesh Agent component.
How can I remove the Navmesh Agent collider and use my own?

[heavy vessel]

but why isnt the latest version available

[night star]

Is there supposed to be bake button or has it changed to somewhere else?

[night star]

It was just debug view 🙂

[ashen pine]

Is it possible to use root motion animation in conjunction with a Nav Mesh Agent? I haven't seen much on that. I found a couple of YouTube videos, but it seems that it's not too straight forward. Thanks.

[raven rapids]

You can set the agent to not control the position and rotation of the object it's attached to.

Annoyingly, it will still set the X and Z rotation to zero

So I wound up attaching the agent to a child object

You can then reset its nextPosition to its current position constantly

You'll also need to use .Warp() if its simulated position gets too far away (I believe that setting nextPosition doesn't work if there's an unwalkable area in the way)

I then read its desired velocity and use that to control my animations

which then use root motion to move the character around

[idle flax]

How do I make the navmesh agent correctly follow a target that's in the air? whenever the target ascends too high or lands on a ledge the agent can't get on, the AI either effectively stops or follows a completely wrong target. The agent is grounded at all times, and I just want it to get as close to the destination as possible, on ground

[raven rapids]

You can use https://docs.unity3d.com/6000.0/Documentation/ScriptReference/AI.NavMesh.SamplePosition.html to find a point on the mesh

That might help. I'd expect the NavMesh Agent to be doing that already, though

[idle flax]

i think it might be connected to the fact that there is a navmesh on the higher surface

but no way to get to it

[idle flax]

Kinda losing my mind.

        Vector3 destination = player.position;
        destination.y = 0;
        NavMeshPath path = new();
        if (!agent.CalculatePath(destination, path))
        {
            Debug.Log("hi");
            NavMeshHit hit;
            bool closestEdgeFound = NavMesh.FindClosestEdge(destination, out hit, 1);
            Debug.Log("Can find closest edge? " + closestEdgeFound);
            if (closestEdgeFound)
            {
                Debug.Log("hi2");
                agent.CalculatePath(hit.position, path);
            }
        }
        Debug.Log("Can find path? " + agent.SetPath(path));

if can find path, it's good
if can't find path, for some reason you can't find closest edge either (this is when the player is above nothingness)

[idle flax]

this is the behavior with this script

and if i just do this

        Vector3 destination = player.position;
        agent.SetDestination(destination);```
this is the behavior

[idle flax]

Is this supposed to happen?

[raven rapids]

I ran into a related issue a while ago

If I set the destination every frame, agents could fail to move at all

[idle flax]

changing FindClosestEdge to SamplePosition with a high maxDistance fixed it

[raven rapids]

AFAIK agents can spend several frames calculating a path, and I was never giving them enough time to actually find it

[raven rapids]

You can do this once per frame and then reuse the result many times

Even an expensive query would be tolerable at that point

[idle flax]

yup, i'm gonna do that, just glad it works at all lmao

[heavy vessel]

but why does it say a warning "Heuristic method called but not implemented"

[woven crown]

I'm having an issue where when additively loading the next level, it appears like enemies from the new level are using the navmesh for the previous one at first (and get snapped into different starting positions then they should). Not sure if that's the exact issue since I would have assumed agents would get their navmesh from the surface in their scene but I might be missing something. Has anyone run into this issue before? It works fine with a syncronous load or async reloading the scene, it's just the first async that's messing up their start points.

[grizzled bone]

Can anyone tell me, how to end patrol after the last waypoint has been reached? Currently it loops indefinitely and never exits patrol mode.

[zinc heart]

what menu is this I wanna try it out

[grizzled bone]

Its Unitys Behavior package. has been recently dropped by Unity and the team behind this has been layed off so I stopped using it and switched to Behavior Designer instead

[crimson oxide]

Does the new Navmesh Component system support auto offmesh link generation like the previous version did?

[raven rapids]

for jumps and drops, yeah

[sick shoal]

What might be the reason NavMesh.SamplePosition fails? I've increased the search distance to 10 units (Npc's are 1.7 units tall) and yet it still returns "false" every now and then, even if the NPC is standing right in the smack middle of a navmesh.

        private void GeneratePath(Vector3 to)
        {
            destinationLocation = to;
            NavMeshPath path = new();
            if(NavMesh.SamplePosition(transform.position, out NavMeshHit hit, 10, NavMesh.AllAreas))
            {
                if (NavMesh.CalculatePath(hit.position, destinationLocation, NavMesh.AllAreas, path))
                {
                    pathPoints = path.corners;
                    curPoint = 0;
                    return;
                }
                DevLog.Log("failed path",this);
            }
            else
            {
                DevLog.Log("failed to find my position",this);
            }
            curPoint = -1;
        }

[alpine glacier]

Hey can anyone teach me about the NavMeshLink?

Aka the NPC jumping Platform

[alpine glacier]

I'm using unity 6 btw

[shy grove]

https://www.youtube.com/results?search_query=unity+6+navmesh

[thorn acorn]

Hello,
I want to create my own AI behavior, and I want it to be more complex than just detecting a player and following them. So, I found the Behavior package in Unity 6.
Is it a good idea to use this package and NavMesh for a 2D game, or do you recommend using another package?

[restive delta]

Hmm... I'm going to do some research myself, but I have a package that gives a player the ability to use a grappling hook to the edges of elevated surfaces. How hard would it be to adapt the NavMesh logic to add that in for enemies? The facile solution involves setting up OffMeshLinks (yes, I know those are deprecated with the latest version) with the animation, but that feels like it requires mapping every possible point as a Link.

[desert elk]

im trying to create a survival horror game set in a house, but my navmeshagent (the enemy) refuses to interact with any stairs or ramps. how do I fix this? here is my code: using UnityEngine;
using UnityEngine.AI;

public class enemyaipatrol : MonoBehaviour
{
GameObject player;

NavMeshAgent agent;

[SerializeField] LayerMask groundLayer, playerLayer;

//Patrol
Vector3 destPoint;
bool walkpointSet;
[SerializeField] float range;

//state change
[SerializeField] float sightRange, attackRange;
bool playerInSight, playerInAttackRange;

// Start is called once before the first execution of Update after the MonoBehaviour is created
void Start()
{
    agent = GetComponent<NavMeshAgent>();
    player = GameObject.Find("Player");
}

// Update is called once per frame
void Update()
{
    playerInSight = Physics.CheckSphere(transform.position, sightRange, playerLayer);

    if (!playerInSight && !playerInAttackRange) Patrol();
    if (playerInSight && !playerInAttackRange) Chase();
    if (playerInSight && playerInAttackRange) Attack();
}

void Chase()
{
    agent.SetDestination(player.transform.position);
}

void Patrol()
{
    if (!walkpointSet) SearchForDest();
    if (walkpointSet) agent.SetDestination(destPoint);
    if (Vector3.Distance(transform.position, destPoint) < 10) walkpointSet = false;
}

void Attack()
{

}
void SearchForDest()
{
    float z = Random.Range(-range, range);
    float x = Random.Range(-range, range);
    destPoint = new Vector3(transform.position.x + x, transform.position.y, transform.position.z + z);

    if (Physics.Raycast(destPoint, Vector3.down, groundLayer))
    {
        walkpointSet = true;
    }
}

}

[humble pine]

@desert elk did you find the answer? 🙂

[rapid pawn]

[Behavior package bug]
I compiled a Linux dedicated server program, but when I started it, the issue shown below. Is it because I used Behavior Graph, which depends on App UI, causing App UI to make the server attempt to enable display functionality?
I tried checking the “Editor Only” option under Project Settings → App UI, but the error persists, and the Linux server still fails to start. However, After I removed this package, the server started working properly.
Does anyone know how to solve this problem?
unity engine version is 6.0.30f1
The error is:

[App UI] Unable to find an AppUISettings instance, creating a default one…
Unable to init server: Could not connect: Connection refused
(demo.x86_64:1810695): Gtk-WARNING **: 11:48:54.295: cannot open display:

[drifting compass]

I'm trying to add a custom type to my Behavior Graph node like so:

[Serializable, GeneratePropertyBag]
[NodeDescription(name: "Publish GameEvent", story: "Publish [GameEvent]", category: "Events", id: "c26072e26a1392180f36c6e3693e8353")]
public partial class PublishGameEventAction : Action
{
    [SerializeReference] public BlackboardVariable<GameEvent> GameEvent;
...
}```

The class:

[Serializable]
public class GameEvent
{
...
}

However, this gives me a ton of Reflection Errors:
NullReferenceException: Object reference not set to an instance of an object
Rethrow as TargetInvocationException: Exception has been thrown by the target of an invocation.

Ah, I have to remove the fieldName reference from the Story.

However, GameEvent does not show up in the Node Inspector

[deft cobalt]

Never had this problem before, but why dose it look like i have two or even three (!) layers of navmesh on my floor, and why dose it look like it is tearing? Im using simple planes for the floor

[storm orbit]

What's the best method to ensure a car using the navmesh system adjusts its rotation and position in releation to the terrian its moving over?

At the moment i can either have it move over the terrain but never adjust to the elevation or using some child objects, wheels and a rigidbody have it bounce around and spin about

VehicleRoot (GameObject) ← Rigidbody
├── NavAgent (Empty GameObject) ← NavMeshAgent
├── VehicleBody (Car model) ← Just the visual model
├── Wheels (Empty GameObject) ← Container for WheelColliders
│ ├── FrontLeftWheel (WheelCollider)
│ ├── FrontRightWheel (WheelCollider)
│ ├── RearLeftWheel (WheelCollider)
│ ├── RearRightWheel (WheelCollider)

Anyone successfully implemented this?

i found this neat trick but cant figure out how to make it work - https://www.reddit.com/r/Unity3D/comments/1gxghsn/flat_navmesh_trick_is_anyone_else_doing_this/?share_id=5bsoad67D5dplaC7NpwP1&utm_medium=android_app&utm_name=androidcss&utm_source=share&utm_term=1

[tardy junco]

If you want a mix between physics and navigation, you should move the car along the generated path manually(by applying torque to the wheels) instead of relying on a navmesh agent.

Navmesh agent would compete for control over object position, just like physics or animations(with root motion)

[storm orbit]

not with my object structure, the model and the navmesh agent are separate child objects VehicleBodys position will only change if root does

[tardy junco]

Obviously, you'll need to change the structure. As it is now, it's not gonna work even if we assume that nav agent works nicely with physics

In fact, if you follow my suggestion, you wouldn't need a navmesh agent at all.

[storm orbit]

Not helpful. Your resolution to the issue it not to use it?

the linked video shows its possible to use the navmesh

[tardy junco]

Yep. Use navmesh, but not the agent.

I'm not telling you not to use navmesh. Go back and read my messages again.

I'm only telling you not to use navmesh agent.

[storm orbit]

Not Helpful.

[tardy junco]

Why not?

I mean, you can technically keep the navmesh agent. Just prevent it from controlling the position/rotation and use the generated path manually. Which is basically what I suggested, just via a proxy

[storm orbit]

Why would i create my own nav agent controller to use the navmesh instead of just using A*. The Navmesh agent is for easy of use.

" all agents are moving on a flat plane and don't have any concept of elevation.

However, the unit bodies are vertically displaced, depending on the terrain height.

This is like doing a heightmap lookup for each agent position and elevating the unit bodies accordingly."

In this example the agent is on the parent and the child is just the model.

my issue is the model flips about and rotates wildly

[tardy junco]

1. You are still using the navmesh(A*) to generate the path. Navmesh agent includes steering motion implementation that you have 0 control over.
2. If you REALLY want to use a navmesh agent(which doesn't make sense to me), you can pull it out of the vehicle object and then make the vehicle move towards that navmesh agent. This is a super dirty and silly solution that would be hard to control.

[tardy junco]

Of course it will. Navigation is not meant to work with physics by default.

[storm orbit]

VehicleRoot (GameObject) ← NavMeshAgent + Script
├── VehicleBody (Car model) ← Just the visual model

Scene
flat plane 0,0,0 - custom layer
terrain on top at 0,0,0

Baked the flat plane only

+++++++++++++++++++++
using UnityEngine;
using UnityEngine.AI;

public class TankTerrainAdjuster : MonoBehaviour
{
    public Transform tankModel;  // Assign the tank model child in the inspector
    public Terrain terrain;
    public float heightOffset = 0.5f; // Adjust for tank's pivot point

    void Update()
    {
        AdjustToTerrain();
    }

    void AdjustToTerrain()
    {
        if (!terrain || !tankModel) return;

        Vector3 agentPosition = transform.position;
        Vector3 terrainPos = terrain.transform.position;

        // Sample height from the terrain
        float height = terrain.SampleHeight(agentPosition) + terrainPos.y;

        // Sample normal from the terrain
        Vector3 normal = GetTerrainNormal(agentPosition);

        // Apply vertical displacement
        Vector3 adjustedPosition = new Vector3(agentPosition.x, height + heightOffset, agentPosition.z);
        tankModel.position = adjustedPosition;

        // Apply rotation
        Quaternion targetRotation = Quaternion.FromToRotation(Vector3.up, normal) * transform.rotation;
        tankModel.rotation = targetRotation;
    }

    Vector3 GetTerrainNormal(Vector3 worldPosition)
    {
        TerrainData terrainData = terrain.terrainData;
        Vector3 terrainLocalPos = worldPosition - terrain.transform.position;

        float xNorm = terrainLocalPos.x / terrainData.size.x;
        float zNorm = terrainLocalPos.z / terrainData.size.z;

        return terrainData.GetInterpolatedNormal(xNorm, zNorm);
    }
}

I was making it more complicated with the wheels

removing all that and just treating it as a rectangle works

[tardy junco]

!code

[pulsar wasp]

Posting code

📃 Large Code Blocks

Use links to services like:
https://paste.mod.gg/, https://hastebin.skyra.pw/, https://paste.ofcode.org/, https://paste.myst.rs/, https://scriptbin.xyz/

📃 Inline Code

Surround code with three backquotes. Not quotation marks.
To format as C#, add cs to the first line:
```cs
// Your code here
```
Add a comment with a line number if there is an error message.

[tardy junco]

Yeah, that would work, but it doesn't make any use of physics.

[bitter sparrow]

any advice on how to fix agents getting stuck on these slopes?

my first thought was to increase max slope but thats already maxed out

[molten sequoia]

Collider interfering?

[bitter sparrow]

Ah yeah it could be that

We do need colliders though, any advice on that?

[molten sequoia]

Does the collider need to collide with ground?

[bitter sparrow]

Maybe disable collisions between terrain and viking?

[unkempt tusk]

    // IEnumerator Perma()
    // {
    //     isAlive = true;
    //     while(isAlive)
    //     {
    //         RandomMove();
    //         yield return new WaitForSeconds( Random.Range(waitTime.x, waitTime.y) );
    //     }
    // }

    private float nextMoveTime = 0f;
    void Update()
    {
        // if (!isAlive) return;

        if (Time.time >= nextMoveTime)
        {
            RandomMove();
            nextMoveTime = Time.time + Random.Range(waitTime.x, waitTime.y);
        }

        _Update();
    }``` Why the coroutine triggers `"SetDestination" can only be called on an active agent that has been placed on a NavMesh.`
While the Update() version don't? Anyone can explain?

I prefer using the coroutine, but that causes a navmeshagent error

"SetDestination" can only be called on an active agent that has been placed on a NavMesh. happens when gameObject.SetActive(false) is called

[torn pelican]

you can't call setdestination on an inactive navmeshagent

[south cypress]

Does Unity have an official 2D NavMesh solution?

[copper hedge]

Don't think so. I was just looking for that yesterday

I'm using navmeshplus (trying to figure out isue with it currently)

Anyone know why the area aroundd the object would be so large? (for non-walkable area)

Bottom left circle btw

[copper hedge]

ofc

figured it out after I posted

Had tilemap collider on. Forgot to take it off

[vapid whale]

Is there any way at all to use a custom shape for a navmesh obstacle? being stuck to either a capsule or a triangle is a bit of a pain...

[vapid whale]

wait im an idiot theres a really easy way to do this

i just gotta add a mesh where i dont want her to go, bake the navmesh, then just delete the mesh after its baked lol

[unborn sandal]

I have a problem with my multi-training. As you can see, when a ML-Agent touch the wall or the ball, the episode ends, giving a positive or a negative Reward. The only problem is that the if one AI ends his episode, ends the episode of all others. Please help me!

[torn pelican]

Noticed NavMesh has a couple various static functions relating to stuff like sampling the navmesh. Is the AI Pathfinding package designed to only really handle 1 active navmeshsurface at a time? I was interested in managing two for the same level and I want to know if that's going to cause any problems for me

[torn pelican]

🥺 👉 👈

[fathom arch]

There's no limit to navmesh surface count, not sure where you got that from

But you do need to add navmeshlinks between surfaces or you can't travel from one surface to another

It can probably be automated but I haven't tried yet

[torn pelican]

Sorry i realise i missed a crucial piece of information here

I was curious about a workflow using two bakes of the same place

not sure where functions like this actually sample from data wise https://docs.unity3d.com/6000.0/Documentation/ScriptReference/AI.NavMesh.SamplePosition.html

[fathom arch]

What do you mean with this exactly?

You can have different surfaces for different agent types if you want

In the same place

[fathom arch]

That just samples all of the navmesh surfaces that use that int areaMask

Navmesh data isn't really exposed, other than CalculateTriangulation, which from my experience needs some post processing

[torn pelican]

oh?

[torn pelican]

does disabling the surface component remove it from those kinda results?

[fathom arch]

Probably, but you'd have to try. I don't remember

[untold pendant]

don't delete that mesh, but rather hide it and make sure it has no collision. If you delete it you will have to re-add it every time you want to bake, which... doesn't make any sense

[vapid whale]

Ahhh good idea thank you

[idle flax]

Any idea why they might not be walking? I'm setting the destination and all, and it works on other maps, but not on this one? Could it have something to do with it being very, very big?

[fervent fern]

hey does anybody have any idea how i would make a navmesh agent that works in the local space of its parent ? I have some npcs roaming around a little level section that is moving/rotating constantly and while ive found a way to keep their target destinations at a certain point in local space but the npcs themselves always end up sliding about or clipping through the floor or ending up in completely stupid situations and im not really sure how to go about fixing it

and when the level moves fast they sort of lag behind with but not completely if it moves kinda slowly its fine sometimes kinda

[fervent fern]

okay i managed to get it working in a stupid way by just adding the difference between the final position of the navigating object last update and the position of the object before it has been moved to the agent this frame to the agents position

still weird with big fast rotations but idc

if anyone knows a better solution it would be appreciated

[tardy junco]

Is the path even calculated? Can you get the path and draw some debug lines along it?

[brisk pumice]

trying to bake a navmesh but get this error

NavMeshPlus.Components.NavMeshSurface.CalculateWorldBounds (System.Collections.Generic.List`1[T] sources) (at Assets/Libraries-APIS/NavMeshPlus-master/NavMeshComponents/Scripts/NavMeshSurface.cs:438)
NavMeshPlus.Components.NavMeshSurface.UpdateNavMesh (UnityEngine.AI.NavMeshData data) (at Assets/Libraries-APIS/NavMeshPlus-master/NavMeshComponents/Scripts/NavMeshSurface.cs:228)
NavMeshPlus.Editors.Components.NavMeshAssetManager.StartBakingSurfaces (UnityEngine.Object[] surfaces) (at Assets/Libraries-APIS/NavMeshPlus-master/NavMeshComponents/Editor/NavMeshAssetManager.cs:127)
NavMeshPlus.Editors.Components.NavMeshSurfaceEditor.OnInspectorGUI () (at Assets/Libraries-APIS/NavMeshPlus-master/NavMeshComponents/Editor/NavMeshSurfaceEditor.cs:264)
UnityEditor.UIElements.InspectorElement+<>c__DisplayClass76_0.<CreateInspectorElementUsingIMGUI>b__0 () (at <043e10ec4a0f4999a2729072194b9cfe>:0)
UnityEngine.GUIUtility:ProcessEvent(Int32, IntPtr, Boolean&)```

yet when I set "use geometry" to "physics colliders" it works but not with "render meshes" and I need render meshes. the weirder thing is this only happens on SOME scenes when I try to bake the tile but all other scenes work fine

but all the scenes are copies of one another just with variated assets

I narrowed down the issue to one of my tilemaps in my scene but I just don't understand what about that would cause this error since those objects are the exact same ones the other scenes have

[fathom arch]

Are you setting the destination every frame? If yes, the agent probably gets stuck in the calculation phase and never gets to actually move

[idle flax]

it works on other maps though? why would that happen on this one?

[fathom arch]

My guess is that the calculation takes longer on a bigger map so it doesn't begin moving immediately

[idle flax]

oh true, i'll try it out then :D

[fathom arch]

You can use this to check if it's still calculating:
https://docs.unity3d.com/6000.0/Documentation/ScriptReference/AI.NavMeshAgent-pathPending.html

Can also give it a cooldown timer because you rarely need the path to update every couple of frames

[idle flax]

i added a cooldown and it seems like calculating the path isn't the problem. maybe the destination itself is?

[fathom arch]

Start by debugging if the SamplePosition is failing or not

(FYI, NavMesh.Raycast also exists)