🤖┃ai-navigation

[fallow terrace]

What is the simplest finite state machine, you use. Like I have a simple enemy ai patrol, detect player, move to player, attack. I used a list of strings and swap out the current state string to tell it which state it's in. All in one script. Is there a simpler way to do it? I just want the easiest and fastest way possible for my less important enemies

[real sonnet]

A bunch of if-else statements :p

[alpine glacier]

Enums to

If you want

[flint jackal]

I made a chatbot builder in Unity

[stable saddle]

@upper mulch Physics.IgnoreCollision() between all colliders that need to ignore each other

[sturdy summit]

hey guys, for my navmesh i am using the Navmesh Components to bake the mesh at runtime. But everyttime when i add it, it adds an error in my log saying

RuntimeNavMeshBuilder: Source mesh Item_WarriorSholderArmor does not allow read access. This will work in playmode in the editor but not in player

thats an item on the player, but im not even wanting that to 'update' or w/e, why is this happening?

it crashes my unity aswel

[spare light]

Yo How do i Make a Horror AI And Jumpscare In Unity vr?????

[stable saddle]

@sturdy summit you maybe just need to find the original 3D model in your assets and turn read/write on

[light saffron]

Anyone likes risk and uses the Unity planner in production? Seems to scale well, how is it?

[molten sequoia]

I have no problem with early access stuff as long as it will continue to improve and get maintenance, but this is not the case with AI Planner.
AI Planner is not on the roadmap of the AI team: https://forum.unity.com/threads/ai-planner-version-0-2-4-released.1000133/page-2#post-7972680

[bitter steeple]

I'm experiencing a weird bug where if a navmesh agent is pressed against a wall by another agent of higher priority (whose velocity is controlled by the player), it gets violently pushed in the other direction. How can I prevent this from happening?

[hybrid pelican]

Hi hi. So I'm having issues with navmesh agents getting stuck running at slightly steeper slope edges instead of getting over them normally. Would anyone happen to know what's going on or how to fix it? Note: the character is controlled with the starter third person asset and has a capsule collider so the issue isn't a box or something issue to my knowledge.

[prime parrot]

Can you PM me more detail. I am using AIML for chatbots but would love to talk a bit

[lunar quarry]

Is there a way to change the center values in a* using code?

I tried using Pathfinder.Center.y += 10; Pathfinder is the name of my AstarPath variable.

but it doesn't work

I don't think center is the correct definition.

[empty condor]

I noticed there is a small gap between the generated navmesh and the terrain. Is there a way to reduce that or offset the agents?

nevermind, it was changing voxel size

[light saffron]

UGUI has not been developed for 5 years but is stable, is AI planner broken? i haven't run into anything yet

[worthy plover]

Is that a unity terrain? (generated with the terrain component)

If so, it might be the case you'd need to generate a mesh yourself from the terrain data, and then use that mesh data to generate the navmesh

[empty condor]

it was unity terrain, but decreasing voxel size fixed the problem

[hybrid pelican]

It seems totally usable though it infinitely loads when building on run for me and in general it's crazy hard to find any information on how to use it, somehow there's less help from tutorials than the documentation but even that requires a lot of parsing and testing since that's seems to just be how experimental/preview packages are: little to no documentation. At least the 2018 talk helps a bit and shows off one of it's uses. I'd like to delve into it further once I can get it working but if you can get it to stop the infinite building and some reward/cost issues, you can do lots with it!

[light saffron]

oh that's not usable then, have you tried to build in other versions of unity?

[hybrid pelican]

more like I haven't used it enough; I got the endless building to fix by going through all the semantic and ai files and making the settings have complete ends and starts, so you just need to set them up properly. getting it to work or trying the planner and getting actions to do what you want however, I have not gotten to yet because I'm focusing on other elements first but I'm sure you can get them to work just fine :P. it's supposed to be super good performance wise and isn't made specifically for Unity since it's pure c# it can be ported to any other engine that uses c# according to what I know, so the Unity version shouldn't change much. just the documentation that's jank :P

[light saffron]

oh really? i thought it used jobs and burst to get the performances on string

i'm surprised build is affected by plan validity, does it hard convert action semantic to c# or a hard coded graph on build?

[glossy sundial]

Hey! So I am really stuck and I have no idea how to move forward. I have 2 sets of checkpoints (1-4) & (5-8) and I have 2 NPC's. I want NPC 1 to patrol checkpoint 1 to 4 and NPC2 the Patrol Checkpoint 5 to 8.

I am currently using a FSM to get them to patrol and I have assigned their own patrol points to the AI class however they are still moving across to each others patrol area

https://gdl.space/anuxegiyeq.cs

A link or a point in the direction to read up on stuff would be amazing

[hybrid pelican]

I'm really not 100% sure but the two times it did the infinite load before it was before I redid the database stuffs. but yeah; you have to build the semantics, problems, ai, whatnot all into c# scripts before it can run it proper-like. and it might use jobs, during the talk one of the creators talked about how the system is fully c# and can theoretically can be ported to anything c# but at the same time I don't remember importing burst but it's in my project so the unity package itself could very-well use burst and jobs, it'd make sense for performance seeing as it's a search algorithm and everyone knows how bad those are on a large scale on one thread :P

[light saffron]

thanks for your knowledge. did you end up using something else for your AI?

a shame they abandon projects all the time

[hybrid pelican]

currently I'm getting inventory working so that ai can interact with that but the ai itself is just a navagent wandering with animations and will look at you if you're in it's sight frustum, it quite literally does nothing because I haven't fully delved into it yet. I'm thinking I'll still use the unity ai planner package for alife sim stuff if I can figure out how to configure it and get it to work how I want, that's just turning out to be more annoying than I'd think because so much isn't explained fully and no one online has futzed with it enough to make a tutorial of any sort (if I figure out how I might do so myself even :P). to my knowledge the ai planner is complete for the most part and it's only missing unity engine stability and documentation. from what I know some experimental packages are effectively complete just without the testing needed to release them to the average unity user, makes sense if you think about it, they have so many employees and sometimes have some make passion projects into unity to allow others to use it but don't see it worth while to get a while bunch of testing into it since it'd be unlikely to collect many users from the average unity user since it's not the plug'n'play sorta thing that the asset store provides. technically dots started that way, just that the person who's passion project it was was one of the unity co-founders so it was never not gonna get abandoned. either way, I definitely agree, it'd be nice if they had better testing/stability and documentation/examples for their many sub-projects. ai planner in particulary seems to have pretty open devs and if you report problems or bugs they'll jump on it; I haven't tested so myself but I've seen them working with forum members as soon as 2021 (and that's only within the stuff I've seen :P). anywho, sorry for rambling ;-v-

[light saffron]

@hybrid pelican you don't need planning for that, utility AI is plenty enough

[hybrid pelican]

ALife or combat AI? ALife would definitely gain a lot from ai planning since it’s basically a more wide GOAP in my case. The combat ai, while not implemented at the moment and will likely get scaled back will likely have agents coordinating to do stuff like flanking and cover fire which of course can be hard coded or done through other methods, would be both easier and more deep with a GOAP adjacent system. I’ll be implementing something hard coded as a launch point but will be getting a GOAP-ish system either way. The Unity AI planner just makes it easier and faster since I’m likely going to have upwards or two thousand agents (though not all planning at the same time, and most certainly not every frame).

I’m going the goal oriented action planing route because that’s what I’m most experienced in :P I’ve only ever done GOAP, rudimentary state machines and desire systems so while goap is a more difficult to implement and hard to perfect system, it’s what I know and have done a dozen times :P

[molten sequoia]

I haven't dove into it beyond consuming content about it due to the state of it. I would be much more comfortable with it if it had been in feature freeze and had years of stabilization and use after that. I also imagine UGUI is maintained, which brings a lot of confidence.

[light saffron]

now you gotta show some stuff you did a dozen time

[hybrid pelican]

XD GOAP isn't rocket science, once you wrap your head around how the architecture works you can implement it pretty easy-like. getting it to run and work really well is the hard part; I may have made a dozen GOAP systems but none were super deep or fast since I didn't spend enough time with them and didn't have much of a testing space. I mean; action planning's literally just a search algorithm that's overly complicated at times, if you know how to implement A* you can certainly get a simple action planning algorithm with some learning.

[light saffron]

you're misinterpreting what i said, show us what you did with GOAP

[hybrid pelican]

Ah, my bad, my bad. Sorry about that (-v- ;)y.
Here's a video I quickly compiled of two projects that have goap-esque systems in them. Both were structured around the ai so there's literally no game and henceforth the ai is pretty much useless. The first one is made in Unity, second is made in something else. I have about two others (made in Unity) with goal oriented action planning in them but they are only ai and only the architecture at that. So basically I didn't do anything with goap, but I made some ai with goap :P the video doesn't illustrate the system's functionality but that's the state I left them in.
https://youtu.be/UVyhsjCqreE

[wild swan]

Hello
when baking Terrain for AI it not selecting all the terrain
there are spots all over the terrain not being selected, how do i fix that?

[jaunty raft]

you need to adjust your max slope angle if you want them to be walkable

[wild swan]

the max slope? make it lower or higher?

worked, thanks!

[vestal widget]

Does anyone know where i can dowload the unity ml agent assets

[placid ginkgo]

HOW MAKE AI FOLLOW PATH IN 3D

[lone ore]

whats a good pathfinding option? my game is (mostly) not grid based and is 2d, the tilemap tiles and the non grid based objects will be created/destroyed

oh and also the game is topdown so there is no gravity

[lapis hamlet]

The easiest way is to make your game run in 3D but look 2D with NavMesh

otherwise making your own A* pathfinding that automatically updates it's chunks when objects are created/destroyed

[lone ore]

alright thanks

ill try a* since i have already gotten pretty far into making the game

[lapis hamlet]

yes, as a tip, if you know you will need a complex system that you don't know how to do , do it first.
I lost countless projects because I hit a brickwall in the middle

[molten sequoia]

https://github.com/h8man/NavMeshPlus for extended 2D support

[lone ore]

i didnt know that existed, thanks ill try it

[turbid panther]

hello guys
I have a weird nm agent behavior where it kinda starts to "grind" against a wallkable area edge
it looks very janky and very noticeable
has anyone had similar issue ?

here it occurs in a narrow passage, but it can happen for any edge, even if there's a very big open space right next to an edge

[hybrid pelican]

I'm not sure I understand the issue... navmesh agents create paths that are the most fast way of getting from A to B, this means that they will beeline to the edge of navmesh which goes around an object or obstacle and then follow the edge until it is able to find a place to path to that's not on the edge. Unless that's not what you're referring to... it's hard to tell with the video flitting by so fast.

[turbid panther]

that's what supposed to happen, and happens in 95% of times (figuratively)
but sometimes an agent constructs a path that runs right next to an edge of a wallkable area a bit too close
which causes the agent to move at about 1/3 to 1/2 the speed. It looks like if it tries to stop to not step out of the wallkable area, correct the path, start moving again, and repeats that in a loop at a very high speed

in video you can see how it moves much slower when passing an obstacle

[hybrid pelican]

I'm not 100% on what's going on because I've never encountered that myself, aside from when I'm using Nav Mesh Obstacle.

Can the agent walk up slopes? The agent can freak out if it's not set up correctly.

[turbid panther]

the area is 100% flat

obstacles also use simple volumes instead of meshes

[hybrid pelican]

Eh? Volumes? I've never heard that terminology in relation to navmeshes. Stuff like obstacle avoidance (and even the carve feature on the Nav Mesh Obstacle) make the pathing and agent navigation bug out. If volumes work in that way it may be producing similar issues. I usually only use the static baking unless the object is dynamic at which point I ignore it or use the navmesh generation unity-script.

Ah, I just researched a bit. Why are you using volumes exactly? Seems like the wrong use-case unless I'm misunderstanding.

[turbid panther]

Not only volumes
it's on per-object basis, simple objects that can be represented as a box or a cylinder use volumes, complex objects use meshes
I assume volumes must be faster, will be very surprised if it's the opposite
I generate navmesh (and world geometry) at runtime, so it's easy to switch between volumes and meshes

the statue that causes navmesh to glitch is just a cylinder

[hybrid pelican]

That's... odd. I've personally not delved into volumes and haven't heard of any performance increases by using them. I'm likely wrong though. You can--if you want--generate navmeshes at runtime using https://github.com/Brackeys/NavMesh-Tutorial 's script (the brackeys video shows it off). It's apparently unity official and I've had it work with 2021.1.22f1 so it'll likely still work... maybe.

Elsewise I'm really not sure; sorry m8.

Perhaps skirting volumes fiddles with walkable speed?

[turbid panther]

navmesh generation is completely fine, I can inspect the navmesh and validate that everything looks like it should. and I'm generally happy with how it works, one of the most fleshed-out parts of the project 😃
it's definitely an agent issue, thanks for looking into the question 👍

there are no actual volumes after the navmesh is constructed

they are used in the construction process

to make unpathable zones

[hybrid pelican]

Oh? Aren't volumes supposed to be for marking areas of the navmesh to be either impassible or cost different. Is there native navmesh baking?

[turbid panther]

yeah, you can use them to assign different costs too
if you pass the area mask of the impassible area it will just cut a hole in the wallkable area

but you can pass any other area mask

that's what I do for water surface, so it's navigable, but requires a corresponding mask to be present on an agent

[hybrid pelican]

Hmm... I have obstacle's with carve on and they build the navmesh correctly but still have agents move slowly around them, maybe the volumes work similarly?

[turbid panther]

they affect the output of the navmesh generation job, they aren't really present and don't continuously modify the navmesh

maybe it's the opposite, agent just has issues with pathing near the edges of the impassible areas
that's why it does that in your case, as well as mine

[hybrid pelican]

Darn, I'm really not sure there then. Sorry.

[turbid panther]

no worries, ty for concern 👍

I have a solution in my head in case it's just a "Unity thing"

just wanted to know if it's a known issue with known workarounds

[hybrid pelican]

;~v~ it really could be. Using a non-navmeshagent and just baking paths then use the path nodes to move with your own script or just use a c# library like Recast but yeah... unity \_( -v-)_/

[turbid panther]

pretty much, will just use the agent to make paths and then move the units in my own code 👍

[hybrid pelican]

I wish you luck then. Be it in fixing the issue or custom navmeshagent :P

[turbid panther]

ty 😄

[summer mica]

How do i do "ai following the party leader" similar to something like chrono trigger or xenoblade but in 3D

[pale violet]

The advance server says that its too advance

I wanted to make a battlefield-like game but I wanna make an AI soldier enter the vehicle like tanks or jeep

[zealous mantle]

The advanced "server" is correct. Your question is vague.

[hybrid pelican]

You'd do it like any other thing. Have the ai switch to a mode where they path find and move with a car instead of themself and just lock them into the car or make them invisible. With so little information it's hard to give a remotely helpful answer. How does the ai work, how do vehicles work, how does driving work, how does entering a vehicle work?

[winter hound]

how do i bake navmesh areas in script?

[summer mica]

I don't think you can without a navmesh surface component

And if so iirc it's:

gameObject.GetComponent<NavmeshSurface.BuildNavMesh();

[tepid bear]

Hey, does someone know how to program AI movement in unity visual scripting / unity bolt

[wary ledge]

The one thing I've been putting off in my project has been AI... Mostly because every time I code it, I recode it... and recode it.. and rinse and repeat...

[red rampart]

can one bake a navmesh surface while the game is running?, or will the nav mesh agent know where to and not to go if there isnt a baked navmesh surface?

[floral sleet]

See the info 3 posts up by Linkychu. You can rebuild the navmesh at runtime, but when I used it (a long time ago) it takes a fraction of second and the game freezes for a bit. It might be faster/smarter now, so give it a try.

If your AI uses a navagent, then you need a navmesh.

[alpine glacier]

noob moment but anybody know why increasing step height isn't changing this navmesh on these stairs? It's not working

I tried going as high as 0.5 and the navmesh looks exactly the same after baking

[tardy junco]

Are you sure that it's not working? It looks like there's a mesh polygon going up the stairs.

[alpine glacier]

well I got it to start changing now, I just cant' get it to where I can walk over all the steps. But I think I'm giving up and just going to teleport the player to the second floor.

[tardy junco]

Make sure that your agent radius is not bigger than the agent radius in the baking settings.

[alpine glacier]

it's not 😦

[tardy junco]

Well, without any further details, there's not much I can help with.

[alpine glacier]

well, that's all the information really. I have stairs, and my agent cannot walk on those stairs. I have tried making individual box colliders for each step and then tried adjusting dozens of values for step height and none of them worked. Then I have tried making one giant sloped box collider and again doing dozens of different step height values. Neither approach has worked. Is there another approach I can try?

[tardy junco]

Did you try debugging the path?

[alpine glacier]

can you explain with more details?

[tardy junco]

You can get a path from the agent and debug it with debug rays/lines or something to see visually if the path is calculated properly

[floral sleet]

Does your navmesh agent have a Rigidbody component? I'm just thinking that even though the agent can calculate a path to upstairs, it might be hitting that step poking out and not going any further. If it doesn't have a rigidbody, ignore me 🙂

[molten sequoia]

Step height is just the threshold for whether the navmesh should be generated over a step in the first place, it doesn't affect the offset between geometry and navmesh.
IIRC navmesh by default generates the navmesh based on meshes, not colliders, if you want to create your own navmesh correction geometry.
At the end of the day you are probably going to face more issues if you use physics to restrict the navmesh agent. You can relatively easily query paths and handle the movement yourself if you want to keep the physics behaviour without the conflicting input from navmesh agent.

[red rampart]

im sorry to botber, but i cant get this to work p.p, says 'the type or namespace 'NavMeshSurface' could not be found ...'

[real sonnet]

Did you add the navmesh components package to your project ?

Did you add the using UnityEngine.AI import in your script ? Or whatever it is called, I don't remember 100%

[summer mica]

You need to install it from github from the package manager

https://docs.unity3d.com/Packages/com.unity.ai.navigation@1.1/manual/index.html

[red rampart]

aha! works tysm

[wary ledge]

So I had a similar issue recently.

The way that most folks used to solve it, was to use a separate mesh for the collider/navmesh that is usually just a slope, then hide the mesh.

[alpine glacier]

I will try that. I need to increase max slope too I think

[turbid panther]

it supports async baking, so even if takes relatively long you can make it seamless
except the step of gathering navmesh sources, which is synchronous. But you can implement your own async source gathering routine and feed it to the navmesh components api

for me it takes about 2-3 frames to bake, but my work station CPU is quite powerful

[alpine glacier]

hi 😊 ! how do you guys make ai fun? i have a fsm currently in my project, but my ai is still generally speaking, just running and shooting at the player. im hoping to make combat a bit more interesting.

[olive orbit]

https://www.gameaipro.com/GameAIPro2/GameAIPro2_Chapter02_Combat_Dialogue_in_FEAR_The_Illusion_of_Communication.pdf

The universal answer with game AI is: cheat. 🙂

[alpine glacier]

yeah i read up on fear a while ago. however they use goal oriented action planning, while i have more of a fsm.

[olive orbit]

Yes, but that article isn't about GOAP.

It's about dialogue

One idea in game AI is that a feature that's not advertised doesn't exist.

I.e., if the player doesn't experience some form of feedback related to the system it doesn't exist.

What Jeff Orkin does is turn it around. I.e., by having the AI comment on its immediate context, it will seem to make smart decisions.

So you can have your AIs run around and shoot, but also say they're flanking, or flushing someone out, or retreating, or whatever, based on where they end up dynamically.

[alpine glacier]

Anyone know how to get the warp function to work on a server build?

[autumn depot]

pls help me

so basically im adding ai script to my enemy

so he could follow me

but instead i get bunch of errors and the enemy is on the floor

[summer mica]

You haven't properly defined "myTarget"

You probably made "myTarget" a private variable or forgot to assign it in the inspector

[autumn depot]

what about the enemy on the floor

[summer mica]

Rigidbody physics are probably enabled on a navmesh agent, so probably set the rb to kinematic if you haven't done so

[autumn depot]

didn't change anything

[summer mica]

Can you send your script

[autumn depot]

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

public class speedfollow : MonoBehaviour
{
public GameObject myTarget;
public NavMeshAgent myAgent;

// Start is called before the first frame update
void Start()
{
    
}

// Update is called once per frame
void Update()
{
    float dist = Vector3.Distance(this.transform.position, myTarget.transform.position);

    myAgent.destination = myTarget.transform.position;
}

}

[summer mica]

No reason to check the distance in the script because you're not really using it right now,

However still unsure on what is causing your player to fall like that. My last guess is that it's something to do with your agent settings 😔

You can also lock rotations with a rigidbody

With rigidbody constraints

[autumn depot]

i did nothing to the settings

[autumn depot]

help me

[summer mica]

what does your speed object look like

[autumn depot]

Like on the video

The streamer ishowspeed

[autumn depot]

are you talking about the scripts?

[summer mica]

yeah in the inspector

[summer mica]

is there an animation on the armature?

[autumn depot]

no

[summer mica]

and it only happened after the navmesh agent was put on?

[autumn depot]

if i disable nav mesh agent it will get a bunch of errors and it wont work

[summer mica]

turn off the script in the editor and hit play

[summer mica]

while it's flipped, click on the object in the scene without exiting play mode

and check if the z arrow is down/up

of the speed object (not the armature

[summer mica]

bruh -90 degrees here , even adding a rigidbody on both the speed object and armature with rotation frozen and both doesn't help?

[autumn depot]

he noclips

[summer mica]

are u sure that you don't have any scripts that look at an object or rotates an object anywhere in the scene?

[autumn depot]

you can check me in vc if u want

[summer mica]

can't rn 😔

[autumn depot]

muted

[autumn depot]

please i rlly need help

[summer mica]

okok i'll try just give me a sec

[devout jackal]

hey doe anyone know a lot about a* 2d path finding?
I've been able to set it up for an enemy to path find to the player but I'm trying to figure out how to make it update on run time since in my game obstacles can be moved/placed.

[summer mica]

https://www.redblobgames.com/pathfinding/

[exotic turret]

I have developed a* 3d path finding with field of vision of enemy before. But never have developed it in a changing environment. This is hard considering the layout of the map can change which beats out the purpose of the algorithm. However, this is not certainly not possible to do at all. In my cases, the player can move (the target can be in different place at a time). What I do is after the enemy run its path-finding a* algo and move the finishing position. If the player is not detected by collision, it will run again its path-finding algo to follow the points. It will not stop until it is collided with the player.

Suggestion #1:
In your cases, I think rerun the a* algorithm when the path it takes has a blocked obstacle. The current position will the new starting position and just re run the a*algorithm.

Suggestion #2:
If the obstacles has a limit movement, just marked the area of its possible movement as 'Wall'. So the A* algorithm can just ignore it.

[coarse lance]

hey, can somebody tell me why my white boy won't jump down the edge?

[red rampart]

why does my nav mesh agent freak out whenever it reaches its destination?
when it reaches its destination, it keeps shifting about around the destinatiin instead of staying on it

[coarse lance]

so what can i do ? 🙂

[floral sleet]

It sounds like it keeps trying to reach the same destination, hence the twisting and turning. You'll need to check/debug your script to see happens when the agent gets to the destination.

[light saffron]

has anyone tried feeding the unity navmesh.AddNavMeshData with a sphere? I'm wondering if nav of planet can be done by bypassing recast

[stark stirrup]

When using navmeshagent I check if distance < 0.1f to stop it

Anyone has tips on how to make AI fair? Or articles about this topic?
My AI are usually unbeatable or when I try to make them easier (like making aiming slower or add noise to their aiming) they become too easy all of a sudden

My game is symmetric (both player and aí have the same capabilities)

[jaunty raft]

Unsolved problem of all game AI

the whole discipline is concerned with only this problem. How to make AI appear to play fair and be fun to play with/against

So if you want info on that, you can pretty much find tips in any resource that deals with game AI

[stark stirrup]

Any example of those resources you could recommend? Like books or something like that

[stark stirrup]

Ok found an amazing one http://www.gameaipro.com/GameAIPro3/GameAIPro3_Chapter33_Using_Your_Combat_AI_Accuracy_to_Balance_Difficulty.pdf

[jaunty raft]

That’s a good book series in general

[red rampart]

i did that and it didnt workkk

[stark stirrup]

have you tried a bigger distance? I don't remember the exact number I used

[coarse lance]

hey guys. i have a problem. I imported some mixamo animations (a pistol-bundle, fbx for unity, without character, T-Pose) and want to get them to work. No matter what character/avatar i Use, it always says that some bone (mostly hips) can not be found.... I even tried characters from mixamo and from asset store. When i tried to set up the rig based on any other model than "this model", it won't work. when i take "this model", the animations are more than clunky..... everything is set to humanoid

(its a pistol package and for example the run animation leads to a creepy hovering, whilst sliding the legs over the ground)

okay .... i fixed it... loop time..... i hate myself....... i needed like 4h for that ...

[outer bone]

Hello were in Unity can i find the navmeshsurface component to use in my project?
I see it being used in the FPS Microgame but it does not seem like a standard included component

[summer mica]

https://docs.unity3d.com/Packages/com.unity.ai.navigation@1.1/manual/index.html

does anyone know the best way to do a boss battle?

[jaunty raft]

behaviour trees https://www.gamedeveloper.com/programming/behavior-trees-for-ai-how-they-work

[lapis hamlet]

Anyone know/has ressources on how they do vehicle AI in the GTA series

[jaunty raft]

its entirely fake, there is essentially no AI beyond the immediate vicinity of the player, everything is just made to be a plausible façade with cars essentially spawning in just off camera. Also the AI is allowed to cause a lot of mayhem (which you notice if you stay at an intersection for a while) which also simplifies the AI's steering & avoidance requirements (if it can't solve a problem, just create a crash to make the traffic jam plausible)

[lapis hamlet]

they do still go from point a to point b successfully, go around obstacles and such. you are giving them less credit than they deserve.

and even if it is limited i'd like to know what's the ai behind the wheel

[dark hamlet]

could just be A* path finding?

[jaunty raft]

thats like saying math involves variables

[dark hamlet]

;/

[tribal agate]

I think this is the kind of stuff you're asking about: https://gamedevelopment.tutsplus.com/series/understanding-steering-behaviors--gamedev-12732

[lapis hamlet]

Yeah that's what i'm doing rn, works pretty well

I just don't want to connect all the nodes, but I think I found something that works for me

[wary ledge]

How do people make AI fun to fight? I've been currently struggling at this, and have been for a long time. Specifically top down shooters/twin Sticks etc.

I've taken a look at games like Enter the Gungeon, Hades, Issac, etc. The list goes on. But I'm just not sure as to what makes it fun, and now I'm struggling to make my game fun. :^)

[jaunty raft]

the effects/juice, a feeling of tight control over your avatar (snappiness/responsiveness), a clear sense of what is going on, enemies clearly telegraphing they actions and an overall sense of fairness, i.e. a clear indication that the challenge can be overcome with skill or better items. Absence of bugs. Generally success should not feel random.

[wary ledge]

Thats more game design. I'm mostly referring to the AI. Currently I have AI either:

• Running towards you
• Running away from you.
• Staying within X range.

However, they just seem flat and stationary.

[jaunty raft]

there is no difference between game design and fun

• Running towards you
• Running away from you.
• Staying within X range.
  this is completely meaningless without the context of the game design

a stationary enemy can be loads of fun

your ai's only goal is to achieve the results mentioned above. There is nothing in hack&slash game AI that has any other purpose than making stuff look and feel fun in a superficial way.

you can just reverse-engineer the action vocabulary of any enemy in a game you like and play around with the behaviour in a prototype of your own making to develop some intuition how to make stuff fun and to see how "they" did it, but you wont find a recipe that works universally for all games.

[tawdry fog]

hey, i'm using the navmeshsurface to dynamically bake a navmesh at runtime for a procedural level, but everytime i load it the game freezes, and i have to close unity with task manager. this only started happening today, it has been working perfectly before. any fixes would be great.

[timid cedar]

IndexOutOfRangeException: Index was outside the bounds of the array.
Unity.Barracuda.Tensor.set_Item (System.Int32 b, System.Int32 ch, System.Single value) (at Library/PackageCache/com.unity.barracuda@2.0.0/Barracuda/Runtime/Core/Tensor.cs:2091)
Unity.MLAgents.Sensors.ObservationWriter.AddList (System.Collections.Generic.IList1[T] data, System.Int32 writeOffset) (at Library/PackageCache/com.unity.ml-agents@2.0.1/Runtime/Sensors/ObservationWriter.cs:151) Unity.MLAgents.Sensors.StackingSensor.Write (Unity.MLAgents.Sensors.ObservationWriter writer) (at Library/PackageCache/com.unity.ml-agents@2.0.1/Runtime/Sensors/StackingSensor.cs:112) Unity.MLAgents.Inference.ObservationGenerator.Generate (Unity.MLAgents.Inference.TensorProxy tensorProxy, System.Int32 batchSize, System.Collections.Generic.IList1[T] infos) (at Library/PackageCache/com.unity.ml-agents@2.0.1/Runtime/Inference/GeneratorImpl.cs:276)
Unity.MLAgents.Inference.TensorGenerator.GenerateTensors (System.Collections.Generic.IReadOnlyList1[T] tensors, System.Int32 currentBatchSize, System.Collections.Generic.IList1[T] infos) (at Library/PackageCache/com.unity.ml-agents@2.0.1/Runtime/Inference/TensorGenerator.cs:173)
Unity.MLAgents.Inference.ModelRunner.DecideBatch () (at Library/PackageCache/com.unity.ml-agents@2.0.1/Runtime/Inference/ModelRunner.cs:204)
Unity.MLAgents.Policies.BarracudaPolicy.DecideAction () (at Library/PackageCache/com.unity.ml-agents@2.0.1/Runtime/Policies/BarracudaPolicy.cs:125)
Unity.MLAgents.Agent.DecideAction () (at Library/PackageCache/com.unity.ml-agents@2.0.1/Runtime/Agent.cs:1360)
Unity.MLAgents.Academy.EnvironmentStep () (at Library/PackageCache/com.unity.ml-agents@2.0.1/Runtime/Academy.cs:578)
Unity.MLAgents.AcademyFixedUpdateStepper.FixedUpdate () (at Library/PackageCache/com.unity.ml-agents@2.0.1/Runtime/Academy.cs:43)

delightful.

My working code suddenly started producing an error that appears to come from an area I can't access (something inside tensorflow?) UGH.

[lapis hamlet]

#archived-machine-learning ?

[timid cedar]

doh!

[frozen ibex]

Hey everyone i again need help. I am currently programming a game with an endless space and have enemies, which are circles. how do i make these enemies spawn at random position, but not to far away from each other? they also dont move.

im not really sure but it would be just like in this game made by dani: https://danidev.itch.io/balls

[real sonnet]

Spawn only the first one at a random position. For subsequent spawns, use the previous one as the starting position + a Random.insideUnitCircle

[frozen ibex]

can you tell me more about Random.InsideUnitCircle?

[real sonnet]

https://docs.unity3d.com/ScriptReference/Random-insideUnitCircle.html

[frozen ibex]

thx

oh so one circle replicates it self?

and then the process repeats?

[real sonnet]

yup

[frozen ibex]

Wow nice idea

thanks

how long have you been using unity?

[real sonnet]

you're welcome, have fun coding

years, but that's not my main job , just a hobby

[frozen ibex]

i started like a week ago but i have some experience with godot

[real sonnet]

oh nice, welcome to Unity

[frozen ibex]

i must say, unity feels much better to use

and c# is very powerful

may i ask how old you are?

[real sonnet]

34, but you'll find more talent among the young 😛 I'm not 100% expert

[frozen ibex]

well youre sure better than i am with my 14 years haha

[real sonnet]

don't worry, I learnt programming at 18 only, you guys this generations begin way sooner, you'll get an edge and still have a lot of time to build great things 😉

got to go dinner, have fun coding !

[frozen ibex]

Thank you!

[opal furnace]

Hi mates,

I am looking for a method for AI Nav Mesh for flying agents. I have any experience with bipeds or quadrupeds npcs, agents on the ground using FSM or Behavior Trees, but no idea of Npc behavior in the air. Anyone can enlighten me?

[light saffron]

I was just looking into that, on Unreal you use an A* that navigates in a voxel or octree version of the level

[stark stirrup]

http://www.gameaipro.com/GameAIPro3/GameAIPro3_Chapter21_3D_Flight_Navigation_Using_Sparse_Voxel_Octrees.pdf

btw, I was very successful in making a flying enemy for an FPS without using any of this logic (it's levels were kind of open though)

basically, I implemented the following rules:

• If player is not in line of sight (view is obstructed), start testing points around the agent (with increased distance and predictable) until you find a point where the view wouldn't be obstructed... move there.
• Move forward/backward to "optimal distance"

worked like a charm

[floral sleet]

My flying a.i is quite simple, but works pretty well.

I have a set of destination points that an npc can fly to, but i use the SensorToolkit from the asset store to dynamically avoid obstacles that get in the way.

I also use PandaBT to determine which destination point to go to based and which is closest/easiest to reach.

[stark stirrup]

What does PandaBT add? How does it determine the destination point?

[floral sleet]

That's just a behavior tree that decides what the npc should do next; flee, patrol, attack etc.

If the npc is patrolling it will pick up a a list of destinations based on large sphere collider. I'll the iterate over those destinations to see if anything is blocking its path, if it is then I skip to the next destination.

My game is set in space and the npcs move quite slowly, so I can spend a few frames fir the a.i to think about what to do next.

[alpine glacier]

hi! right now my ai is always looking at the player and shooting. however, the projectiles (not hitscan) almost never hit the player, ever.....

what would be the best way to fix this?

[stark stirrup]

Thanks Brianer

[round ruin]

I am using NavMeshBuilder to generate the mesh at runtime as the Components sample shows, but what I can't find is info on generating the off-mesh links or setting some of those variable via code, such as jump distance or drop height. Can it be done?

[opal furnace]

@stark stirrup @floral sleet @light saffronvery useful these infos, I am gonna iterate through these ideas and check the result I get. Thanks!

[light saffron]

@stark stirrup method works well for a few agents, plus you can easily dial in the clumsiness of the agent which gives it personality. I wonder how HL2 did its combine funship, it had a very interesting personality

[tepid dirge]

Hey is there any videos that can help me with AI? Been trying to create AI padestrians like in Gta. But trying to find resources for it. But essentially there seems to be nothing that helps.

[floral sleet]

There are loads on Youtube
https://www.youtube.com/results?search_query=unity+pedestrian+ai

[tepid dirge]

These waypoints systems i find on youtube are really flawed. Mostly because They leave alot out, And there not referenced in future videos or in the past videos they have. (Ive looked on each creators channel on that search. Ive tried finding it). Can you recommend me a good channel?

[floral sleet]

I struggled with the idea of A.I for a while, even simple stuff confused me. Then I watched this Udemy course by a wonderful tutor who explained basic A.I in a simple, easy to understand way.

If you can wait until the course is on sale, then I recommend this.
https://www.udemy.com/course/artificial-intelligence-in-unity/

[tardy junco]

What exactly do you think they leave out? What is it that you have trouble understanding?

[tepid dirge]

Its my first time doing AI waypoints so. And spawners for the AI. + intersections. Find it especially hard to do. They leave out certain scripts to make the other scripts work for the actual thing I want to do. They provide no context most of the time and just say "I have a basic controller for my AI" What is that supposed to mean? They provide no context to what the AI does. And there is this one creator that describes everything except for some essential parts that need to e in it for me to understand it. Like from what I said before. How do the scripts work. They give you tiny context but never go into full detail.

They don't even leave in the scripts on github for me to learn what they do. And when they do they leave alot of things out so without me completely understanding what they are doing. I don't know what I'm doing wrong if I get it wrong.

[tardy junco]

Maybe provide a link to a tutorial that you tried recently and specify what part was it that you didn't understand?

a basic controller for AI can mean many things, yeah. But it should be somewhat clear from the tutorial context.

[tepid dirge]

Yeah but for example https://youtu.be/MXCZ-n5VyJc

That's where the "a basic character controller for AI" comes from

[tardy junco]

Ah, that channel is pretty good, although he does target more of an intermediate-advanced devs imho.

[tardy junco]

Do you remember the timestamp?

[tepid dirge]

Its at the beginning of the video

It's near 1:30

[tardy junco]

He does provide some explanation of the controller though

Th reason that he doesn't go into details is because it's not relevant to AI specifically. It might as well be a player character controller. The controller implementation doesn't matter as long as the AI can control it.

His controller specifically is simply moved via the transform towards some point.

[tepid dirge]

still though the AI itself its pretty essential for the traffic system to actually work though

[tardy junco]

You can simplify the whole thing as:

transform.position += (destination - transform.position).normalized * speed * Time.deltaTime;

And it will roughly be the same.

[tardy junco]

As I said, the controller is not relevant to the ai.

If you need a character controller tutorial there are plenty of tutorials that get into details.

[tepid dirge]

I dont need a player character controller im just struggling to understand this because its kinda my first time doing something like traffic.

And its kinda frustrating Bc i feel like im missing alot of knowledge.

[tardy junco]

Like, what exactly about that script do you have problems understanding?

[tardy junco]

If you can write a player character controller, you should be able to understand the script he's using for the AI characters.

The underlying logic is the same

Namely: provide some input into the controller and it will move the character.

[tepid dirge]

Yeah i just cant visualize it. Thats why im struggling with it. I can make character controllers fine. But it feels like im missing something from it and its making me feel frustrated bc of it.

[tardy junco]

Well, go through his script line by line and make sure you understand it. If you have problems understanding a certain line or the whole concept(after understanding each line), feel free to ask specific questions here or in the coding channels.

[delicate drift]

what type of AI should i use to make a nextbot-type system? i'm working on a small project for a couple friends but just cant seem to get the nextbot feel like in garry's mod to work.

i've worked with AI a couple times before but there aren't good tutorials online for what im trying to do.

[light saffron]

nextbotAgent.SetDestination(player.transform.position) i think 🤣

[karmic bobcat]

What is the avg that i should let a MLAgent learn before getting conclustions ?

[delicate drift]

does SetDetection allow for wall/obstacle collision? if not, how would i implement that

[light saffron]

yes with an agent component

[cobalt schooner]

Hey! New here!
Taking my shot at some game development and ran into some issues...
I have a Sprite that is rendering behind my tilemap, I think I've pinned the issue down to Nav Mesh Agent -> Area Mask, when flagged as Everything my GameObject is behind all of my other assets, but if I uncheck everything it appears no problem, but now my pathfinding doesn't work.
Running Unity 2020.3.26f1 and using NavMeshPlus

[cedar swan]

Hey, I'm new at Unity. I'm trying to install and use ml-agents in ubuntu 22.04. But I can't make it work I would like to use the most up to date version. However, the one that's not generating compilation errors (I newer versions I get a lot of conflicts with namespaces not being found) is Unity 19.4.40f1 and ml-agents 1.1.0 preview. Eventhough, I don't get compilation errors when I press "play" a pop-up window shows up "Unity has stopped working... no fix available"

Can some one help me out configuring Unity and ml-agents for Ubuntu 22.04??

[harsh flicker]

Hello. I want to make patrolling AI. He has to patrol house. But when he patrols he look at patrolTarget. And it does not look good as u see. How i can fix it. I want him ti look forward while he is moving by path.

[split sandal]

Hi! im not the best at code and was wondering if someone can help me out with AI

[alpine glacier]

Ask the question

[split sandal]

wdym?

im just trying to get Ai to chase the player

[alpine glacier]

Just don’t ask to ask next time https://docs.unity3d.com/ScriptReference/AI.NavMeshAgent.SetDestination.html

What about that A* pathfinding ?

[split sandal]

wha?

[alpine glacier]

That’s an asset if you’re talking about that

This is the normal solution

[split sandal]

cant i use this?

A * Pathfinding?

[alpine glacier]

Navmesh Agents and functions like SetDestination are the default solution for AI in Unity

[split sandal]

ah

ok

ima just try and find a tutorial

[wild swan]

is it possible to check if the AI cant reach the player? obstacles on the way or player up a cliff?

[real sonnet]

https://docs.unity3d.com/ScriptReference/AI.NavMeshAgent.CalculatePath.html

[wild swan]

thanks

[sudden scarab]

Just upgraded a project to 2022.2.0b6 and it appears that NavMeshSurfaces can no longer be serialized

Is this a known issue or intended behaviour?

[fringe crag]

anyone else using the experimental component based navmesh surfaces for runtime building? I'm noticing that NavMesh.SamplePosition and NavMesh.CalculatePath do not work

i'd love for someone to tell me i'm crazy and they do work 🙂

[plucky dragon]

Hi, anyone know how to override NavMeshModifier with NavMeshModifierVolume? it always goes through Volumes first and then NavMEshModifier leaving me with wrong NavMesh. Is there a way to reorder NavMeshModifiers?

[plucky dragon]

I have checked how the NavMeshComponents orders the list of modifiers but it seems volumes are after the base modifiers, so its not its fault it seems, but rather something to do with how NavMeshBuilder works, but I cannot look into it as Unity does not wont you to see how it goes through the list provided, any idea how I could make sure to use Volume after base modifier

[real meadow]

How could i make enemies react to physics? For instance falling or being pushed by the player

[light saffron]

compare rigidbody.acceleration to the force applied by the agent

[orchid solstice]

hey, i think this is the right channel, but anyone here knowledgeable abt mlagents ?

Python\Python310\lib\functools.py", line 856, in register
    raise TypeError(
TypeError: Invalid first argument to `register()`. typing.Dict[mlagents.trainers.settings.RewardSignalType, mlagents.trainers.settings.RewardSignalSettings] is not a class```

I come across this error after running "pip install mlagents"

[real sonnet]

there's the #archived-machine-learning channel

But to try to answer your question, double-check all versions you're using are compatible together

mlagents package, Python, and so on...

the mlagents doc on github should tell you which versions you should have

Looks like you're using Python 3.10 ?

The mlagents documentation recommends staying with 3.6 or 3.7

[brittle galleon]

Hey channel. Trying to put some of the final touches on some AI I'm writing, and right now I'm dealing with combat. I've got my AI successfully finding targets, and only shooting at them when they're within a certain viewing angle in front of the AI. And if not, rotating until they are in that angle...and finally rotating until they're directly facing them. I'm now working on not shooting just because they're in range and within that angle. Meaning, not shooting if there's something between them and the AI (i.e. a large structure). Achieved that by raycasting from the barrel of the weapon to the target and checking to make sure it's a clear path.

What I need help with now, is figuring out a good way to tell the AI to move if there's not a clear shot. How can I find a good point on the NavMesh, where the AI will have a clear line-of-sight to the desired target?

[orchid solstice]

Oh yes, but I got confused, as therr aren't any download files

Rather zip folders

[real sonnet]

A naive first implementation you could try:

• sample X positions in a grid around the player.
• for each position, raycast towards the player,
• save the ones that hit the player in a shared collection (each AI should be able to read from it)
• each AI then finds the closest valid position from itself
• navmesh.samplePosition() or whatever it is called to get the closest point on the navmesh, then setDestination

if "sample X positions" becomes too heavyn then you'll look for ways to optimize: delay refresh of casting only once every some amount of frames/seconds, use the batch command version of the raycasts, delayed reading and sorting distances for the AI too, and so on...

also play with position intervals in the grid, min max distance, and so on....

Depends a lot on your game and especially on the environment for these spatial queries stuff

[brittle galleon]

Got it. I can't use a simple raycast from the sampled positions, because the weapon's have some pretty big offsets from the AI's center, but I can always factor in that offset. Thanks a lot! Gives me a good place to start from.

[real sonnet]

are you aware of shape casts variants instead of rays ?

[brittle galleon]

If you mean ones like SphereCast...yes.

[real sonnet]

👌

[wary ledge]

Do I need a NavmeshSurface for each agent type?

[glossy yarrow]

nope

only one and bake it it will work for all nav mesh agent

[wary ledge]

How would that work then? As NavmeshSurface 1 would allow for a agent that is 1 high to walk under something. While I could have an agent that is 2 high and I dont want him to be able to walk under it.

[glossy yarrow]

then yes u need too nav mesh surface

[stone owl]

In case it helps, what I have done in the past is have specified areas like cover for example that is sure to have vision over certain areas.. we work these locations into the level design, then query each position in a radius if they are in angle/view etc of target, and send the agent there.

This way you can avoid lots of calls to a potentially large grid and gives you more control the final position an agent is firing from. whereas if you leave it to chance, maybe he has vision and is in angle in the middle of the open which is not a very strategic firing point etc.. totally up to your game just my two cents, a little of both usually works great in my experience anyway.

[brittle galleon]

GREAT insight for robust AI. Thank you!

[real sonnet]

Nice trick but maybe not for ALL agents, as it will make them 100% predictable 🙂

[hollow egret]

hey all. I'm trying to make ai for this game im making which has a procedurally generated map with floating asteroids. The map is also extends indefinitely and generates as you go out further. I'm using A* project for my pathfinding. Currently, my solution for pathfinding is having a graph that is scanned every 0.5 seconds that is centered on the player and follows it around, but obviously this solution is really bad and causes terrible performance. How would i go about implementing good pathfinding that can react to obstacles moving and being destroyed? I'm not very familiar with a* so I'm unsure how I would go about doing this right, since redrawing the graph when objects usually are just moving slightly seems very redundant but i'm unsure how i would do it otherwise

[tall idol]

how in the world do i use navmesh

[floral sleet]

There are lots of tutorials on YouTube. Have you tried those?

[hollow egret]

lmao

[plush hollow]

Does anyone know the easiest way to fix this? I have a wall under this (second floor) which I believe is causing it.

[tribal agate]

Is this an image showing the floor of a 2nd story, with a gap in the navmesh cause by the wall on the floor below?

If so, if you mark the wall as a walkable surface and rebake I think it'll fix it

[plush hollow]

I did manage to fix it by changing the radius of the agent, not sure why that mattered but hey. Marking the wall underneath would have possibly worked but it would look strange when the AI walks through it from downstairs.

[visual surge]

Greetings, if there's anyone who's willing to do some paid coaching on the topic of Behaviour Trees -- DM me. In any case, I'll explain the problem below.

Basically, I'm trying to use Behaviour Trees to implement an AI in the game. But it feels like I can't grasp the paradigm of operating sequencers and selectors.
I have a Unit.

1. I'd like Unit to go Idle when no enemies are present on the map.
2. I'd like Unit to go to defensive position if enemies are present on the map and he can't see any of them to attack.
   If he detects the enemey...
3. If he has his skill recharged -- approach the enemy -> use skill.
4. If skill is not available or on cooldown, approach the enemy -> attack.
5. If both skill and attack are on cooldown -- walk towards the defensive position, waiting for either skill or attack to recharge.

Here is what I have (https://imgur.com/a/DXqWngY), but it looks bad since I'm basically ignoring the return value of casting the skill, and my Unit gets stuck on endlessly running towards chosen enemy. I'd probably need to re-evaulate the closest enemy every frame, but that too sounds like bad design...

So: is there anything I'm missing, and what's the correct way of handling nested IFs in a behaviour tree? Like 1) if enemies are present globally branch, 2) if enemies are seen branch, 3) if enemies can be reached branch, 4) if skill is off cooldown... etc.

Perhaps Behaviour Trees is not what I'm after?

[jaunty raft]

i think you are asking the wrong question, if you are thinking in if-then-else a BT will always stay obtuse and you best use if-then-else and not a BT. In a BT you describe behaviour as "try if all these things work (sequence)" and "try if any of these things work (select)", inside these queues you put checks before actions and you get the "execute/succeed when all conditions are met" or "execute/succeed unless condition is met"

[light saffron]

Avoid BT until you can no longer do your AI with FSM, when you reach that point, then skip BT and do GOAP instead 🤣

or utility, or influence map, etc...

[civic ore]

Whats the best choice for ai for a 2d platformer?

[jaunty raft]

it depends

[civic ore]

On what

[shell isle]

which is better to use in unity car game AI perceptrons or sigmoid neurons?

[shell isle]

The type of game you want to create! or simply the purpose

[civic ore]

The purpose is an enemy ai which has states like patrol idle etc has pathfinding so can chance the player and will attack with melee if in range

[real sonnet]

You described a Finite State Machine (FSM)

[shell isle]

taking into account of pathfinding which mainly needs ai use ML agents by unity. and for the the purpose of attack use an invisible cylinder parent to the enemy. it should have a radius equal to range of attack. If something triggers it then play animation of attack.

[civic ore]

Thank you

[errant turtle]

I'm very comfortable with ai, but often I run into issues with my design patterns.

One I find myself using often is a switch case in update (or my own 0.5 second pulse for performance).

Its switch case is based on an enum that just holds the state (idle, patrol, chase, etc). And the action logic is contained within each case.

Is there a better way to set up that sort of logic? It begins to struggle when using things like nsvmeshagent pathfinding, or setting up sequences of actions (i.e. go here, open this, then go there)

[tribal agate]

what you described is a common (and totally fine) way of setting up a simple finite state machine. However, it doesn't lend itself all that well to sequences of actions. For that, you'd probably want to research a behavior tree or goal oriented action planning.

One thing to keep in mind is that none of these AI patterns are mutually exclusive. You could have a finite state machine with a state that runs a behavior tree inside of it, for example.

also, a general principle that I've found very helpful for wrapping my head around writing AI is that you should separate the thinking from the doing. It sounds like you're trying to mix together the logic of "am I at location X" with "navigate to location X" and it is muddying up your code. I find it very helpful to separate out my code such that I have...

Controller -- a class (or collection of classes) that actually take actions in the game world. This class cannot think. It has methods like MoveTo, FireAt, etc
AI -- a class (or collection of classes) that takes in information about the state of the game world and makes decisions, then gives instructions to the controller based on those decisions. It cannot directly affect anything in the game world.

[errant turtle]

Many thanks for your extremely thorough and well thought out answer. Extremely helpful! You're exactly right, separation of logic is the way to go, and makes things wayyy less spaghetti, and was my biggest issue.

🍻

[wild swan]

my AI getting stuck on Nav Mesh Obstacle, any idea how can i avoid that?

[pallid jungle]

why is my nav mesh ai flying

what on earth is happening

[wary tide]

bro went sicko mode

[limpid plinth]

Hi, playing around with navmesh agents here today...

haven't used them much before... anyone know if there is a way to drive the agents without requesting a path? I just want the agent avoidance really, don't need the pathfinding

not sure if those can even be separated

[real sonnet]

yup, https://docs.unity3d.com/ScriptReference/AI.NavMeshAgent.SetDestination.html

[limpid plinth]

I saw that... won't that request a path?

[limpid plinth]

well, I got it working using navAgent.destination... not entirely chuffed about the performance... getting a pretty noticeable hit to frames

this might be overkill for what I need... the agents are pretty sparse, they really just need to be able to sidestep around each other

setting up a basic avoidance sidestep behaviour here (using a scheduled raycast for detection) to compare

[real sonnet]

Nope, if you click on the link and actually read the documentation, it takes a vector3 as only param. That's it, wherever it comes from.

[limpid plinth]

well, yes, I did read the docs... what it doesn't mention is whether internally, setting a destination doesn't also request a path

it's not requesting a path explicitly, it doesn't give you a handle to the path either... but as far as I can tell, it could be doing expensive pathfinding internally and just not showing it

I'm trying to figure out what avoidance solution works best in my case... my options are, built in nav agents, simple raycast-based sidestepping, or running my own RVO simulation

I was hoping the built-in one would work, because well, it's built in, it's already got navmesh baking as well for obstacles, it's pretty complete

but the downside is that it's a total black box, with limited tinkering possibilities

the simple raycasting is good because it's well, simple... but also pretty limited and being based on a single raycast, it's difficult to make it not jerky without a lot of fudging

[real sonnet]

Yes it requests a path

Thought you wanted to know if you needed to do it yourself, not that if you could avoid the computation cost

[limpid plinth]

ahh ok, yeah, I'm trying to avoid the computations

I'm currently fuddling with the sitestep behaviour... if that is too much headache for too little result, I'm going to try the RVO

[real sonnet]

By the way, it clearly stated in the doc, so you might have read it to quickly 😉

Sets or updates the destination thus triggering the calculation for a new path.
😄

[tribal agate]

I think what you'd have to do is be able to set steeringTarget manually, but it is read-only, so I'm pretty sure you can't use it as steering only

[limpid plinth]

yeah

I saw there is a navAgent.desiredVelocity, which is literally the input parameter for RVO... but that's also read-only for some reason

the big con with rolling my own RVO is that it's reinventing a very big wheel... unity already has it built in for navagents

[tribal agate]

I wouldn't be surprised if they're all read-only as a performance optimization. It means you can skip any validation of those values if you know they're only being set by the system internally. No need to worry about steering targets that are off the mesh, etc

[limpid plinth]

yeah

[tribal agate]

but who knows

[limpid plinth]

the big black unity box

I mean it is their logo... can't say it's out of character 😛

ok, the raycasting is a PITA and at best it's going to be janky...will try a custom RVO sim next

[real sonnet]

do you know boids ?

and especially the separation part

but yeah that's close to what you would try when you talked about raycasting and sidestepping 🤔

I assume

[limpid plinth]

yeah, RVO is more or less like boids, in terms of results

it still allows each agent to have its own idea of where it wants to go... boids is more free-form

[real sonnet]

I see

good luck with RVO then, careful trying not to go overkill

[limpid plinth]

thanks! 👍

the trouble with custom RVO is that it doesn't speak unity out of the box... need to write up data conversions and stuff

[limpid plinth]

ok, hooked up an RVO simulation... now to test it out

[limpid plinth]

well, no smoke came out, so we got at least that far

the RVO agents aren't moving yet... I need to work out what units this is using

[limpid plinth]

aha! movement

the time scale values are not in the units I thought

[limpid plinth]

hah! it works

best solution of the three

it runs on a bg thread, so it's pretty much free for performance, and the avoidance is pretty good

[real sonnet]

oh nice

yes was interested in performance

[winter hound]

hey im using a package called a* for my enemy ai anyone know how i make the enemies avoid eachother?

[real sonnet]

can you maybe talk a little bit more of the specifics of your game, just to figure the context of complexity we're talking about ? (how challenging is local avoidance for your agents in your game)

@limpid plinth

[wild swan]

Hello
this is patrol function for my AI, my issue is when patrolling AI can get stuck in navmesh obstacles, because it cant reach the destination point behind them, such as houses or resources etc, is there is a way to make AI detect if there is no path to destination and just generate new walkpoint?

[limpid plinth]

I would think it's not very challenging for them, and it isn't very mission critical either. The agents are mainly just pedestrians walking around the game world. Their interactions with the player are going to be very limited (they have a duck animation to get out of the way), so their job is mostly to just be there for visuals. The game is a vehicle builder/simulator.

[tribal agate]

you can check the agent's pathStatus and if it is partial, that means the path they found cannot reach the destination

you'd probably also want to check that they don't have a pending path first

[real sonnet]

I see. Thanks for feedback 🙂

[wild swan]

never used that before, do u have example code?

[wild swan]

whats the pending path?

[tribal agate]

https://docs.unity3d.com/ScriptReference/AI.NavMeshAgent-pathPending.html

[tribal agate]

try to guess. It'd all be one pretty simple if statement

pseudocode:

if ((not path pending) and (path status is partial))
  no path found to the desired destination

[wild swan]

tried print(agent.pathPending); , it false all the times, both normal and when he get stuck

[tribal agate]

path pending will only be true when you've asked for a new path but the pathfinding system hasn't finished computing it. It essentially means "all my path data is currently invalid, wait until next frame when I've actually finished my work"

[wild swan]

print(agent.pathPending + " " + agent.pathStatus);

it stuck with same result

that house is player placed with navmash obstacle

[tribal agate]

can you screenshot your navmesh?

looks like the house is not carving the mesh

[wild swan]

ah so i should enable carve?

[tribal agate]

yep

though if the house never moves just baking it into the navmesh is probably better

[wild swan]

as i said it placed by player

[tribal agate]

ah, I missed that

[wild swan]

like rust / ark building

no worries

[tribal agate]

okay, yeah, then just enable carve on the obstacle

and it should work

[wild swan]

yeah , just waiting on AI to get close

again, weird

the red line is their destination

[tribal agate]

so the one has a destination inside the house?

[wild swan]

yes

[tribal agate]

hmmm... path status might not account for obstacles? I haven't tested that...

[wild swan]

yeah all places to be stuck in are navmesh obstacles

i guess i could make a timer, if walkpoint is not reached return to walkpoint = spawnpoint

[tribal agate]

I guess you could use NavMesh.SamplePoint to find the closest point on the mesh to your target destination

then you know your always aiming for somewhere that at least on the mesh, if not reachable

[tribal agate]

yep, a timeout like that could work too

[wild swan]

i will look it up

thanks for the help friend

[wild swan]

worked
code incase someone else run across same issue

[alpine glacier]

Hey, so I'm trying to make enemies shoot back at me and deal dmg to me, but I'm having a hard time all day I am new to programming, so I do all my coding with the help of yt tutorials and so far I've made the enemy ai to follow me, but I can't get the shooting to work I don't know how to do it all... Any advice/help?

[tribal agate]

don't overcomplicate it. Just make a bunch of if/else statements that describe what you want the enemy to do.
pseudocode example:

if (is in shooting range)
{
  if (shoot is off cooldown)
  {
    shoot
  }
  else
  {  
    do nothing
  }
}
else
{
  move toward player
} 

[alpine glacier]

Ty I just don't know where to type all that stuff, I use visual studio I'm really new to coding I literally know nothing btw so sorry if I'm asking for too much, I started last week...

[tribal agate]

well, first off, you can't just copy/paste what I wrote. It is not real C# code. You'll have to figure out how to translate that into real code. But it should probably go in the same place that you made your ai follow the player. Likely in Update

[alpine glacier]

Oh alr I'll just have to learn how to translate it to c# code.

[limpid plinth]

uhm, speaking of AI, I had a bit of an argument with one today... aren't the robots supposed to be our loyal servants?

well, at least it wasn't the least productive dev discussion I've ever had... he's ahead of quite a few humans 😄

I'm not sure I should save that file 😨

[alpine glacier]

How do i add bots in topdown 2d shooter (a*?)

With tilemaps etc

[real sonnet]

Lol it mirrored your mindset perfectly

[alpine glacier]

Im having a hard time again, i cant translate that and every tutorial i follow just doesnt work for me for some reason, i did actually fix something and i made them look at me and follow me, but i cant make them shoot/damage me...

[flat granite]

anyone knows how to tell a nav agent to stop following an OffMeshLink?

somehow clearing currentOffMeshLinkData

i can tell the agent to stop, and then it would stop at the start position, but afterwards when continuing it would snap to the end position of the OffMeshLink

ResetPath() does not help

apparently only Warp() can do anything. wtf

[vernal hearth]

I have a tricky issue to solve. I got a bunch of bots that have to try and avoid obstacles coming towards them.
They can move in two axis, while the obstacle comes at them from the third axis.

What I need to figure out is how to best evaluate where to go if I detect that I am on a collision course with an obstacle coming at me. Because I could pick a random direction and just try and go in that direction to see how far I'd need to move in order to not get hit by this oncoming obstacle. But what if where I am going has another obstacle coming, but faster/slower?

Is it a matter of simply choosing how far into the future I want to look and then just go blind and once I get to my new destination figure out if I wanna move again? Or is there another smarter way to possibly do it?
These bots do not have to be overly intelligent. Just enough to create a bit of chaos in a game of 100 people.

[jaunty raft]

Simple rules (reflexes responding directly to sensor input) with little or no foresight often produce the most plausible and interesting behavior in these kinds of situations

[vernal hearth]

So flying blind, so to speak other than what is immediately in front of you

[jaunty raft]

Mostly yes

[vernal hearth]

I made this sort of mental layout of how to run through the options

Can we move?
|.false:
|..bot is inactive
|.true:
|..Are we moving?
|...true:
|....did we reach target destination?
|.....false:
|......keep moving
|.....true:
|......stop moving
|...false:
|....Obstacles coming?
|.....true:
|......pick random direction in unit circle
|......add x distance to direction
|......clamp target destination to playable area
|......start moving
|.....false:
|......wait for next evaluation loop

[jaunty raft]

The combination of a reflex avoiding a slow opponent just to run in a fast one and having to dodge that one aggressively is plausible and fun

[vernal hearth]

Basically every bot has the same evaluation above but each difficulty is decided by how long it takes between each evaluation loop

Reaction time, basically.

So an easy bot will take 1 second between each evaluation, a medium bot half as long and a hard bot will only take 0.05 seconds

[limpid plinth]

I'd suggest looking into state machines, when you have that sort of complex if-else birds nest situation

the simplest implementation of a state machine you can do is nothing more than three switch statements in a couple of methods:

enum State
{
  Idle,
  DoingStuff
}

State state;

void Update()
{
  switch (state)
  {
    case State.Idle:
      // update idle behaviour
      break;
    case State.DoingStuff:
      // update doing stuff behaviour
      break;
  }
}

void SetState(State st)
{
  switch (state)
  {
    case State.Idle:
      // leaving idle behaviour
      break;
    case State.DoingStuff:
      // leaving doing stuff behaviour
      break;
  }
  state = st;
  switch (state)
  {
    case State.Idle:
      // entering idle behaviour
      break;
    case State.DoingStuff:
      // entering doing stuff behaviour
      break;
  }
}

it's not the more organized way to set up a state machine, but it will get you a long way already

Also, you can check the value of st on that first switch of the SetState method to see which state you are transitioning into, and potentially set up rules there for which state transitions you allow or not (maybe return true or false from SetState based on whether the state transition was allowed)

but yeah, that is the most bare-bones implementation of an FSM I can think of

for really complex stuff (main character logic, complex NPC AIs and so on), then you'd probably want to set up actual types for states and transitions, but this will get you through most simple jobs

[light saffron]

was copilot helpful when you were programming?

go on the asset store and get the free version of BTPanda, it has all that

[limpid plinth]

oh yes, when it gets it right, it's briliant. For an adhd brain like me, it even helps you stay focused, because it can get you out of rabbit holes where you overthink stuff

[light saffron]

if you get a chance to take a video of you working with it, i'm interested.

to see the ratio of right and wrong suggestion

[limpid plinth]

it really depends a lot on how you work with it

it's not like intellicode, where you can ask it to code templates, like code blocks and stuff

it's more like a smart finish-my-sentence type thing

where you guide it by commenting ahead of you as you go

that's how I've been using it at least

[indigo field]

So im following a tutorial on how to make an ai enemy and I need help making it melee attack instead of using projectiles, if anyone could help I would appreciate it : )

[light saffron]

i'd download pandaBT-free from the asset store, it has such examples if i recall

[untold sphinx]

Hey guys, how do I store the neuro network data from ML Agents Training in Unity?

[dusty drift]

The type or namespace name 'Interactable' could not be found (are you missing a using directive or an assembly reference?), i tried to change MonoBehaviour to Interactable and it gave me this i was watching a brackeys tutorial btw is it changed

[real sonnet]

Looks like it's not really AI related.... The only interactables I know are for canvas groups and selectables(buttons and such) so trying to replace a MB with this makes no sense

Unless you created your own Interactable class, deriving from MB

[wild swan]

my AI in scene is placed perfectly , but during runtime it floats, it only does that with nav agent active, what causing the issue?

[wild swan]

solved, had to turn this on, why isnt it enabled by default -.-

[light saffron]

because it's been buggy forever until recent fix and adds cpu cost

[patent pilot]

How would one approach Galcon style AI? In other words, given a set of nodes with various amount of units in them, what would be the process to find an optimal source and target node to transfer units between?

[light saffron]

astar point graph using multipath

[light saffron]

and give each line a penalty that's the inverse of resources available * time of travel i guess... or just code your own A*, these turn based things don't need to be that performant

[modern dragon]

Can we make Nav Mesh Agent a Sphere? Currently i can only see it having Cylindrical Body. In my game i want sphere ball agents.
How can I achieve this?

[oak trout]

Hello. I want to learn more about pathfinding. In my game i want a layered pathfinding system. i could only find 2d pathfinding tutorials so right now i have a basic a* pathfinding on a 3d grid. I downloaded Aron a* project but it is very advanced and hard to learn for me. There is a layered pathfinding system there but it's in the pro version which i'm not going to buy yet. Also i would like to make my own code to learn better. I'm looking for any sources to help me.

[molten sequoia]

You probably don't need to solve for some optimal formula to have a really difficult AI in a game like that. I would probably start with some random attack/defence planning with target nodes being semi randomly picked with weights based on unit count and distance.

[patent pilot]

thank you, I will check it out

[patent pilot]

I see, the ratio of distance vs units is the main heuristic I was after, I am thinking of making the AI send reinforcements to nodes that player gathers his units around in expectation of an offensive.

Also thank you for the tip about splitting the planning for an attack and a defense, originally I tried to evaluate all possible commands in one big list and select the best one in a given situation, whilst it could pretty much be two processes running simultaniously 👍

[red grove]

I'm currently using nav.SetDestination(playerPos) every frame to make the ai follow the player. But when the player is in an enclosed room (here closed by the fence with nav mesh obsacles) the nav mesh agent won't move. Any ideas how I can fix this and make the ai approach onto the fence?

[languid ibex]

heyho. I am using the experimental AI navigation. (AI Navigation 1.0.0-exp.4). I have a script on the NavmeshAgent object, that bakes a volumetric navmesh surface at runtime around the agent. The script works fine as long there are not too many agents at one place. Has somebody tried that and had similar problems? Here is a video of running the runtime navmesh builds at large terrains (4km).
https://www.youtube.com/watch?v=ZO3TgoBxfds

[worthy peak]

anyone with Citizens Pro

[floral sleet]

Probably best to exclude the door from being baked as an obstacle, then the ai can calculate a path to the inside of the room.

If the door is open, the a.i can enter but if the door is closed then the a.i will be blocked by the doors collider.

[languid ibex]

yeah lots of people 😉 but is this navmesh built on runtime?

[worthy peak]

wdym ? it said it's support export build

[languid ibex]

ok, sorry got that wrong, thought you reply on my question.

[ashen smelt]

Anyone here got any good alternatives (except Aarons A* project) to the NavMesh navigation system that is built into Unity? I'm building procedural levels. And the built in system locking the main thread to build the navmesh is a big problem. I'm not even sure I need navmesh specifically, but at least something that can create a decent set of path nodes to follow in order to reach a desired goal.

Regular steering works ok for open areas, but as soon as there is a bunch of obstacles, pure steering breaks down and get stuck quite often.

[limber forge]

I had something similar, and ran into the same problem, my solution was to store all the agents in an array, find the center of their group and make a dynamic navmesh from there, with its size varying in real time using a bounding box so the agents would always be in the navmesh.
This solution however could pose a size problem this time, so you might wanna find yet another solution, like grouping your agents depending on their distance, and having a single navmesh per groups or something, depend on your needs

[languid ibex]

Cool solutions, I will look into that. Thanks a lot 🙂

[sturdy lark]

Honestly I don't know if this even belongs here or in code advanced orcode beginner, but

Can anyone help me i am trying to use mlagents, i used them before in an almost identical setup and only changed the code in slight bits, but for some reason the movement or general connection does not work

public override void OnActionReceived(ActionBuffers actions)
    {
        Debug.LogWarning(actions.DiscreteActions[0]);
        int action;
        if (actions.DiscreteActions[0] == 2)
        {
            action = -1;
        }
        else
        {
            action = actions.DiscreteActions[0];
        }

        Player.GetComponent<Rigidbody>().AddForce(new Vector3(action * Time.deltaTime * 25, 0f, 0f), ForceMode.VelocityChange);
        // Debug.LogWarning(actions.DiscreteActions[0]);
    }

    public override void CollectObservations(VectorSensor sensor)
    {
        sensor.AddObservation(transform.position); Debug.Log(transform.position.x +  " " + transform.position.y + " " + transform.position.z );
        sensor.AddObservation(WallsX[0]); Debug.Log(WallsX[0]); Debug.Log(WallsX[1]);
        sensor.AddObservation(WallsX[1]);

        for (int i = 0; i < 5; i++)
        {
            int pos = getPlayerPositionInBlocklibary() + i;
            float BlockID = blockLibary.getBlockID(CurrentBockList[pos]);
            float distanceToBlock = getPlayerDistancefromBlock(pos);
            Vector2 Oservation = new Vector2(BlockID, distanceToBlock); // Vector2 ( Typpe of block, distance to this block)
            sensor.AddObservation(Oservation);
            Debug.Log(BlockID+" "+distanceToBlock);
        }
        //sensor
    }```mlagents 2.0.1

what am i doing wrong

[sturdy lark]

i am now testing i could have forgotten that a decision requester exsists

[sturdy lark]

yes, I did

sorry if i botherd anyone

[mystic thistle]

All right im having big problems with instantiating navmeshagent. I just keep getting those errors.

Vector3 t_position = buildingAttributes.spawnPoint.position;
        NavMeshHit hit;
        if (NavMesh.SamplePosition(t_position, out hit, 100f, NavMesh.AllAreas))
        {
            GameObject spawnedUnit = Instantiate(GameManager.Instance.allUnitPrefabsList[unitGlobalID], hit.position, Quaternion.identity, unitParent);
            spawnedUnit.GetComponent<NavMeshAgent>().Warp(t_position);
            if (buildingAttributes.buildingOnThisNode.GetComponent<NodeAttributes>().nodeColor == NodeAttributes.NodeColor.blue) spawnedUnit.GetComponent<UnitAttributes>().unitTeam = UnitAttributes.UnitTeam.blue;
            else spawnedUnit.GetComponent<UnitAttributes>().unitTeam = UnitAttributes.UnitTeam.red;
            spawnedUnit.GetComponent<UnitAI>().enabled = true;
            spawnedUnit.GetComponent<NavMeshAgent>().enabled = true;
        }

I've tried using SamplePosition, tried Warping and enabling navmeshagent component after all of that (it is disabled by default) but the errors still appear. When I place the same prefab normally and enable navmeshagent its working perfectly.

Ping me if you want toanswer please

also tried this way, same problem

    private void SpawnUnit(int t_unitIndex)
    {
        Vector3 t_position = buildingAttributes.spawnPoint.position;
        NavMeshHit hit;
        NavMesh.SamplePosition(t_position, out hit, 1000f, NavMesh.AllAreas);

        GameObject spawnedUnit = Instantiate(GameManager.Instance.allUnitPrefabsList[unitGlobalID], transform.position, Quaternion.identity, unitParent);
        spawnedUnit.transform.position = hit.position;

        if (buildingAttributes.buildingOnThisNode.GetComponent<NodeAttributes>().nodeColor == NodeAttributes.NodeColor.blue) spawnedUnit.GetComponent<UnitAttributes>().unitTeam = UnitAttributes.UnitTeam.blue;
        else spawnedUnit.GetComponent<UnitAttributes>().unitTeam = UnitAttributes.UnitTeam.red;
        spawnedUnit.GetComponent<NavMeshAgent>().enabled = true;
        spawnedUnit.GetComponent<UnitAI>().enabled = true;
        
       

NavMesh.SamplePosition(t_position, out hit, 1000f, NavMesh.AllAreas); returns true, and it still doesnt place my unit on the navmesh.

[real sonnet]

Could you try to rebake your navmesh ?
Also try to warp it to hit.position, I don't understand why you would sample position to send it back to t_position

Actually I'm not even sure you would be able to warp it befire it is enabled

before*

Maybe try to :
Instantiate at sampled position
enable
(eventually warp now)

[real sonnet]

Don 't worry you don't bother anyone when asking for help if you're stuck 😉 glad you solved your problem, next time if you want mlagents help, you can go to #archived-machine-learning channel

[sturdy lark]

thank you

[dusty drift]

I have been trying to make a enemy and have been watching a tutorial series from this https://www.youtube.com/watch?v=Mp8yFUKxldk&list=PLGUw8UNswJEOv8c5ZcoHarbON6mIEUFBC&index=6 but the enemy isn't moving

i have everything exact i have checked many times i will also send the code

[wild swan]

is it possible to have structure prefabs that is placed on runtime and the AI and walk over them?

[dusty drift]

idk really know what that means but it is just one cube on the floor so nothing could interrupt it from moving

[wild swan]

can it go on top of the cube?
for example let say the player build a large stonewall with stairs in runtime, so it not baked, is it possible for the AI to climb the stairs and walk on top of the wall?

in 7d2d game i saw AI jump and walk on top of player placed blocks

[signal saddle]

hey guys im new into ai

someone can suggest me a youtube channel to learn?

[jaunty raft]

No, but there are books you could read: http://www.aiwisdom.com/ http://www.gameaipro.com/

[signal saddle]

Nais

[signal saddle]

thanks

[mystic thistle]

Hey, any idea on how to make agents not collide with agents on the ground? I'm making flying units and I've arleady set it that it can fly above anything except for other units. How can I do that? I'd just set their obstacle avoidance to none, but I still want them to avoid each other in the air.

[molten sequoia]

Separate navmesh for flying units perhaps?

[mystic thistle]

I actually made another navmeshsurface which is just placed higher and instantiate air units on that navmesh

it took me more time than it should

but thanks

[languid ibex]

Thanks a lot for the tips 🙂 Your ideas worked well. I implemented now a Manager for the Navmeshes, that dynamically groups and/or ungroups the meshes based on distance. https://www.youtube.com/watch?v=hFV5c9Ybbsc

[limber forge]

Glad I helped, good job ! 😄

[opal flume]

I have a flock system with bees, but I can't find an approach to make the flock move around looking for gameObjects with tag "run", any idea on how should I start solving this issue?

Flock: https://hatebin.com/vpgfzohpjq
FlockManager: https://hatebin.com/ppjstuqnwy

[tardy junco]

Just add another vector in your direction calculation.

[keen oak]

I was wondering if someone can help me with my ai a little, I just need help with board evaluations, and the numbers for them, this is a 24 by 24 grid chess game with fairy chess variant pieces. https://github.com/Mysticpasta1/ChessAIInUnity/blob/chess-ultima/Assets/Scripts/BoardPieceValuation.cs

[desert yoke]

guys is it possible to train bot to smart bastard?

[verbal grove]

My nav mesh mesh agent jumps only if two points of offmeshlink are in same X. If they're not, it just not jumping. How to fix that?

[light saffron]

there is an async version of navmesh build

[light saffron]

this is cool, i didn't know that the builder could create separate chunks of navmesh

[languid ibex]

Its the exp AI Package: com.unity.ai.navigation. With this you create Navmesh Surfaces to navigate which can be even vertical too. And with this you set it to volumetric. You can see the params here: https://nonon.zone/2022/10/11/charactercontroller-17-runtime-navmesh-surface-builder/

[light saffron]

Ah yeah I've used that nav package in a spherical world, i love unity's nav but the bits of navmesh couldn't be merged, zero write access to the navmesh so an agent on one side of the planet could not path to an object on the other side, or if an agent was following another one, there were movement artifacts when their distance was about 1/8th of a sphere.

[junior heron]

somebody can help me with an ai for chess game?

[fiery lion]

I'm working on a tower defense game, for debug reasons, whenever my ai reaches its goal, it'll snap back to spawn, however, due to the fact i can build blockades, it places my AI near the nearest edge instead of the spawn point (which is also on a navmesh)

does anyone know how to fix it, im using the new package

clarification, it sets it to the spawn, until it actually just snaps back to that edge.

(it should be a valid position since on start the location it spawns at is its own transform.position)

[fiery lion]

checked the docs and apparently using NavMeshAgent.Warp worked instead of changing the transform.

[verbal grove]

How to auto generate two-directional offmeshlinks? For some reason unity generate only one directional.

[languid ibex]

In this example they are bidirectional... i am not sure what you mean, but here is the link: https://learn.unity.com/tutorial/runtime-navmesh-generation#

[floral sleet]

I have an A* class on a single gameobject that will find the shortest path between two points in an Octree graph. The NPC's in the game reference that single gameObject/component whenever they need to move to a new destination.

But I've noticed that the game is dropping frames whenever an NPC uses the A* calculate a path and I was wondering of ways on how to fix this?

I assume the first step is to convert the pathfinding algorithm into a coroutine(?), but will that create a problem if two or more NPC's try to calculate a path at the same time? So I'm then thinking, is it better to let each NPC have it's own copy of the A* class, or should I have a manager class to manage the queue of NPC's wanting to use the A* algorithm? What do you think?

This isn't for a fast-paced shooter, so I don't mind if the NPC's have to wait around for a few seconds before a path is calculated for it.

[jaunty raft]

How big is the search space and average path length?

[floral sleet]

At the moment the search space isn't too large, but this might change. There are about 3,000 nodes and 30,000 edges.

Path lengths generally sit between 50 and 100 nodes, but this path is all over the place so I pass this it to another method which shortens it either 3 or 4 nodes. This process works, but it's a bit slow.

[ashen smelt]

Why dear Unity, does the NavMeshQuery system lack means to get the area of a polygon. 😢 Or does anyone know how to get that info given only a PolygonId?

[jaunty raft]

This should be solvable in milliseconds per path if you use an efficient backing data structure for your open/visited nodes, even without a*, using just dijkstra would be in the milliseconds

[floral sleet]

Hmm. Ok. I'll profile each method to see where the problem might be. I've just read about Async/Await too, so I might see if that can help out as well. Thanks.

[jaunty raft]

you can't really use c# threads effectively to parallelize work in unity, you'd need to use the job system

[floral sleet]

Really? I've seen a few videos and articles on how they are the preferred over coroutines in Unity, and they seem simpler to implement as well. The jobs system is a bit too complicated for me and could be a bit of rabbit hole.

[molten sequoia]

You might not care, but you are likely to lose support for platforms that don't support threading, which right now is just WebGL. UniTask and Jobs fallback gracefully.

[glacial anchor]

Noob question: How do i hide the blue Navmesh? I tried to turn off all the different Navmesh options in gizmos but it still shows. The only want i can hide it is to disable the NavMeshSurface component completely

[mighty charm]

My navmeshsurface isnt detecting anything. How come?

[floral sleet]

Did you set your ground as Navigation Static, and have you clicked the Bake button?

We'll need details of what you tried, before anyone can tell what step you might might have missed.

[floral sleet]

When you select the Navigation tab, you'll see a window in the bottom-right of the scene window. You can hide the navmesh from there.

[mighty charm]

I did click the bake button, wdym by navigation static?

[floral sleet]

Which guide/tutorial are you following? Anyway, here are the steps, in particular, step 2:

https://docs.unity3d.com/Manual/nav-BuildingNavMesh.html

[mighty charm]

This one: https://youtu.be/UjkSFoLxesw

[floral sleet]

It looks like a part of the process was missing from the tutorial, as per some comments:

[mighty charm]

Oh ok, ill test that later. Thanks

[floral sleet]

You can set floor to static either on the gameObject itself, or from the Navigation tab. They both do the same thing.

[glacial anchor]

How do I make a NavMeshObsticle that Agents can step up onto? For example I want to have a box that is dynamically instantiated. Once it spawns navmesh agents should be able to use it to walk on in their path finding.

[languid ibex]

You could use the new ai package and generate the surface at runtime with volumetric generation. https://youtu.be/JjQg2vGHK_8

[stoic tulip]

Hi I am trying to make use of A* pathfinding project running into some errors anyone that can help?

I tried navmesh but the bake button doesnt work in that

[stoic tulip]

Basically I am trying to make a gps as a fun project for my neighborhood area to play and mess around with ai the plan is when we click one point to another on the map the route is rendered as a line so i need pathfinding
tried navmesh bake option isnt working in A* project i put all obstacles in obstacle layer and hit bake but the bake result ignores all obstacles

[stoic tulip]

Anyone can help in this prob

[light saffron]

you could use thread ninja

[tardy junco]

Async await Tasks are running on the main thread in unity by default. You can make them run on different threads, but you won't be able to access unity API from them then. Also, jobs are way more efficient performance wise. But yeah, if you just need something to run in the background without much care for how fast it is and don't need to access unity API from it, then a threaded Task is a good option.

[muted pike]

I can't find any videos on ai that uses 0 animations and is moved with ik does anyone know where I could find a video or documentation on that?

[jaunty raft]

ai is typically moved by a kinematic character controller and animation is driven by whatever the actual movement is (vs root motion)... basically all AI movement works this way and you do not have to search for anything special. If you find some tutorial that uses root motion for AI, you should carefully consider if its approach is scalable to your particular use case.

[muted pike]

Well I wanna use physics sorry I didn't specify.

[real sonnet]

Keywords are active Ragdoll or a plain physics driven animation / physics driven ai could give good results

In any case don't expect to find a perfect tutorial that provides a result out of the box. You'll probably have to experiment a lot on yourself before reaching the result you hope

[muted pike]

Tysm I'll definitely try that!

[floral sleet]

I use the free version of PandaBT. That has some documentation and examples you can learn from. Otherwise, BTs mostly use a common terminology for describing nodes in the tree, so most YouTube guides on BTs can be useful.

I actually learnt BTs from a Udemy course teaching beginner a.i, and lots of practice.

[light saffron]

do you know why you need bt vs other form of ai?

[stone sequoia]

Been trying to bake a navmesh at runtime, but in script NavMeshSurface keeps coming up as missing namespace. Is there a different namespace than UnityEngine.AI for it? I already added the package the way the docs say via add package by name option, and I enabled preview packages in project settings. Beyond that I have no idea why I can't access NavMeshSurface. It seems to work fine in the Unity Learn Example

nvm wrong namespace

[plush hollow]

I'm having a weird issue with the NavMesh where it seems to disappear when I hit save after baking it and eventually that causes .SetDestination() to return with can only be called on an active agent that has been placed on a NavMesh.

[light saffron]

cool. the best guides i've used are panda bt's sample scenes. and zomboid blog

[crystal karma]

Hey! Can anyone in here give me the quick rundown for manually completing NavMeshLink versus OffMeshLinks? Right now I'm reading currentOffMeshLinkData, getting the start/end positions and manually animating between them. I understand you can get the current navmeshlink by casting navMeshOwner, but I'm not sure what type to cast it to (NavMeshLinkData? NavMeshLinkInstance?) and I'm not sure how to make use of the fact that navmeshlinks are now areas instead of lines

Just to be clear I'm trying to preserve existing offmeshlink functionality but also adding support for navmeshlinks so I can make use of different navmesh surfaces

[crystal karma]

So I fixed some assembly import stuff and now the NavMeshLink type is available. If I cast navMeshOwner to UnityEngine.AI.NavMeshLink I can get a reference to the original component but that just tells me the start and end positions for the link, it doesn't make use of the fact that the link is now an area. Is there a way to access the path calculated across the area by the actor?

[queen prairie]

does navmeshagent.move() still avoid obstacles? doesnt seem like it. how does it work?

[tacit fractal]

How can I solve that error? Failed to create agent because it is not close enough to the NavMesh?

[tacit fractal]

it's close

Ugh, I solved it, thx

[light saffron]

when you query this channel then find the solution, share the solution

[tacit fractal]

I only set the same Agent Type for NavMesh and Agent

[jovial creek]

I'm working on the AI system for an action game.
My question is about pathfinding.
I have a long range pathfinding a position a bit far from the player say I want it to be 5 units away for it to stop pathfinding and do an attack.
What should I do if the player gets too close?
how can I tell the enemy where to go if it's TOO CLOSE

keep in mind the levels are randomly generated so I can't make predetermined spots for it to go
should I just make it raycast in a direction away from the player and go to the hit point if there's nothing in the way?

what if the enemy is a corner?

or should I make it so the enemy has a short range attack and have the enemy attack until it dies?

how do games typically approach this problem?

[jovial creek]

I posted in the forum if anyone wants to answer there, since this channel doesn't seem to get much activity :(

https://forum.unity.com/threads/behavior-for-long-range-enemy-when-player-is-too-close.1354871/

[light saffron]

Several ways, The simple one is to path until distance is reached then path away from enemy. My favorite is to sample a set number of random position on the navmesh which are within that distance and chose the one with the shortest path - a bit costly but unity's navmesh can push a LOT of queries and you can always jobify those queries if performance is an issue - it rarely is an issue and if it is, then your game might be one of those with swarming enemies, in which case another strategy for pathfinding is required.

[jovial creek]

Ah thats a pretty clever strategy thanks

Im using sebastian lague's pathfinding solution since my game is 2d and nav mesh doesn't work in 2d(i know there are ways to hack it but id rather not) tho unfortunately so ill have to program that in

[thorny citrus]

Dose anyone have a simple public collision path I can have a look at and edit

For enemy AI

[thorny citrus]

yes

so if your behind a cube it will find a path around and go to you

think of it like minecraft ai

bruh

i need a enemy ai script i can edit that will move around the block with colision instead of getting caught on it and getting stuck

so it can move around and try to get to the player

should work thanks

[jovial creek]

what's the best way to solve this problem?

have a second collider on walls to increase the non walkable tiles around it?
remove collider from enemy?

keep collider on enemy and set it to ignore wall layer collisions?

or should I change the create grid function to use a variable that adds detection range for nodes?

public void CreateGrid()
 {
     grid = new Node[gridSizeX, gridSizeY];
     Vector3 worldBottomLeft = transform.position - Vector3.right * gridWorldSize.x / 2 - Vector3.up * gridWorldSize.y / 2;

     for (int x = 0; x < gridSizeX; x++)
     {
         for (int y = 0; y < gridSizeY; y++)
         {
             Vector3 worldPoint = worldBottomLeft + Vector3.right * (x * nodeDiameter + nodeRadius) + Vector3.up * (y * nodeDiameter + nodeRadius);
//would I add to node radius with a sperate variable here?
            Collider2D cast = Physics2D.OverlapArea(worldPoint, worldPoint + new Vector3(nodeRadius, -nodeRadius, 1), layerMask: unwalkableMask);
             
bool walkable = (cast == null ) ;
grid[x, y] = new CopyNode(walkable, worldPoint, x, y);
         }
     }
 }```

[coral gulch]

Hey guys, I have been thinking on how this would be. How would you make an AI similar to the cannibals in The Forest where they just observe you and get close but in a determined radius?

[coral gulch]

Thanks for the detailed response. Great ideas. Also do you purely make AI in code, or do you also integrate it into the Animator machine states, or use any plugin?
Thanks.

[coral gulch]

Awesome thanks for the help! I will let you know how it goes 😄

[queen prairie]

anyone familiar with behaviour trees?
i would like to run idle movement and check for player in range(2 checks apart) in parallel.

          .Selector()
                .Parallel()
                    .Sequence("Check Player In Aggro Range")
                        .Do("FirstAggroCheck", () =>
                        {
                            if (CheckPlayerInAggroRange())
                            {
                                print("First check");
                                return TaskStatus.Continue;
                            }
                            return TaskStatus.Failure;
                        })
                        .WaitTime(2f)
                        .Do("SecondAggroCheck", () =>
                        {
                            if (CheckPlayerInAggroRange())
                            {
                                print("second check");
                                aggrod = true;
                                return TaskStatus.Success;
                            }
                            return TaskStatus.Failure;
                        })
                        .WaitTime(2f)
                    .End()
                        .Sequence("Idle/Wandering")
                        .Do("Set Next Position", () =>
                        {
                            destination = AIUtility.BrownianMotionWalkPosition(transform.position, 40);
                            return TaskStatus.Success;
                        })
                        .Do("Move to a Position", () =>
                        {
                            controller.SetDestination(destination);

                            if (controller.DestinationReached())
                            {
                                print("reached my destination");
                                return TaskStatus.Success;
                            }
                            return TaskStatus.Continue;
                        })
     
                    .End()
                .End()
            .End()
            .Build();

[acoustic minnow]

How would I simulate this track in order to use a genetic algorithm?

[muted pike]

Does anyone know why my ai will stutter and keep going in both directions?

void AI()
    {
        GameObject plr = GameObject.Find("Player");
        float distToPlr = transform.position.x - plr.transform.position.x;
        Debug.Log(distToPlr);
        if(distToPlr < 1.25f)
        {
            rb.velocity = new Vector2(moveSpeed, rb.velocity.y);
        } 
        else if (distToPlr > -1.25f)
        {
            rb.velocity = new Vector2(-moveSpeed, rb.velocity.y);
        }
        else
        {
            rb.velocity = Vector2.zero;
        }

(sorry for the sloppy code :/)

[zealous mantle]

Your first if statement will always be true if you're close to the player until it's too far away then it'll switch to the second.

The third statement will never happen.

[muted pike]

so should i use a switch

[zealous mantle]

Your first statement needs to check if the distance is between a min and max range.

[muted pike]

alr

[zealous mantle]

If it's less than 1.25 but greater than 0.25 for example.

Also, using Find every frame like that is going to kill your performance.

[muted pike]

Alr tysm

[uneven current]

yall prefer Utility AI or GOAP or Machine learning, Behavior Trees or wat ?

I find myself spending hours on tinkering with each one , but they each present their own problems

[jaunty raft]

they all solve different problems, real world, 80% of the time a BT is what you need, Utility AI is just a fuzzy predicate in a BT, GOAP almost always leads no no perceivable entertainment value, ML is not AI, its a niche field of study in AI that concerns itself with function approximation and can be used as part of any symbolic AI (like Utility functions in BTs)

[crystal karma]

It's such a project specific question

I feel like you're almost asking "what kind of genre should I pick for my game"

[crystal karma]

I feel like the first question is the type of behaviour that you want your AI to have, and to answer that you have to answer what behaviour you want the game in general to have

All the techniques you described are useful for creating different types of experiences. A state machine is fine for creating an enemy in a wolfenstein type of game, but if you're making a deep RTS you would want an AI that's able to make plans and follow them

It's like asking what kind of brush you should paint with without knowing what you're painting or even the medium : p

They're techniques that you apply to create an effect

My own personal experience with AI in games is that it's best to be as simple as possible so I agree with the POV that state machines are good for most applications. I'm working on an imsim type of shooter right now and we started out with a very complex AI that used fuzzy logic and would grade it's choices based on the effectiveness against the player's tactics, so it could learn over time. Then it turned out that all that stuff made it unpredictable and not very fun, then eventually I chopped sections out of it gradually until we were left with basically a simple FSM that follows the player or takes cover depending on some very simple criteria. It's better to design with the experience you want in mind. Complex AI sounded cool superficially but the experience of playing against it did not create a feeling of being able to learn the enemies' behaviour and adapt.

but that means nothing in a different context. If I was making a game that had long term planning and strategy in it then I'd want the AI to be able to express that too

(turns out if you let your AI learn what works best eventually it figures out that sitting in doorways spamming grenades is effective)

[round bridge]

finally, somebody that agrees with my strats

[jaunty raft]

this is basically the story of every game AI programmer ever 😉 somehow nobody seems to believe it until they experienced it

[light saffron]

the end boss learned your pattern from observing from afar

another world style

[crystal karma]

Did that happen in another world??? I dont remember

So is there really nobody that's using the new navigation package that can give me some tips on how to utilize navmeshlinks?

I got it working but I have no idea how to leverage the fact that navmeshlinks are areas instead of lines

[light saffron]

tiger creature lurking in the background

[uneven current]

do you make custom trees or do you use plugin ? I see something from KIWICoder not sure if I should start with that

[acoustic minnow]

Is a policy gradient just a way to see how much certain inputs would affect certain outputs?

[kind laurel]

Hope this is the right place. Anyone know why after I bake the nav mesh, there’s random spots along my terrain that are not included in the bake? I’ve changed settings for the agent itself in terms of step height and max slope but nothing is changing tbis

[crystal karma]

Looks kinda like intersecting objects

Is it possible you have some objects there that have renderers with no materials in them? These get picked up by the navmesh baker as collision geometry

[kind laurel]

Could be

There’s no game objects around the affected areas

[crystal karma]

What happens if you remove the terrain and put a giant cube there instead to try to bake a flat surface?

Do you still get holes in those locations?

(make sure your stretched cube platform thing is at the same elevation as the terrain)

[stoic tulip]

Hi I am using the A* pathfinding project the algorithm is working for my game I want to use a line renderer to draw the path so its visible
can anyone guide on how to do that

[stoic tulip]

Top down 2d
u click on the destination and the player moves to it

[jovial creek]

Sebastian lague's video shows how

[stoic tulip]

could you please link the video

[haughty current]

https://www.youtube.com/watch?v=-L-WgKMFuhE

[hasty forge]

Did NavMeshModifierVolume get removed for 2021 unity and beyond?

Not finding any documentation on it anymore

[tawdry trout]

Poll on visual node based no code ai development (as seen in unreal and unity for games) https://www.reddit.com/r/artificial/comments/ypye9v/how_many_of_you_would_use_a_drag_and_drop_visual/

[crystal karma]

Is this chat like, both for AI in the context of game development and AI in the context of machine learning...? because I feel like these two things are not full of overlap

[ashen smelt]

Most of the time, they're not even close. I'm assuming both goes here though

[spice cobalt]

Anyone have any tips on why manually created Off Mesh Links might not be working? My agent seems to just bounce around them but never takes them.

[wind arrow]

I got a problem:

This is the error: "SetDestination" can only be called on an active agent that has been placed on a NavMesh. UnityEngine.AI.NavMeshAgent:SetDestination (UnityEngine.Vector3) AIScript:Update () (at Assets/AIScript.cs:24)

it's on same object

[hasty forge]

try disabling and reenabling the navmesh agent

[wind arrow]

ok

[hasty forge]

that error means your navmesh agent is not connected to the navmesh

have you generated a navmesh?

[wind arrow]

nope

[wind arrow]

one sec

[wind arrow]

let me google

bake?

now it kinda works

better

now it works, thanks

[lethal current]

Hey everyone!
I could use some help to point me in right direction when it comes to pathfinding.

I am making a 2D game and I'm looking for an efficient way to navigate player to one of blue points on path. So let's say I click on one of the blue dots, and I want the player to slide along the black path to finally reach the point and stop there. Player should always slide on the black path no matter which point I choose.

Which pathfinding method would be the best one to go with in this scenario? Could I possibly use one of Unity methods or should I code it my own way?

It actually looks too simple for many pathfinding methods that I have found so far.

[floral sleet]

You would still need something like A* to find the shortest path between two nodes. I don't know much about 2D, but Brackeys has this guide on setting up pathfinding on for 2D. It might still work:

https://www.youtube.com/watch?v=jvtFUfJ6CP8&ab_channel=Brackeys

[mystic thistle]

Hey, when I call agent.setDestination(), the agent smoothly turns towards it's new destination. Can I use that from agent's class but in other scenario? I want to use this smooth rotation when my agent spots an enemy and turns towards him.

[little nymph]

Anybody know how to make navmesh agents chase the player but prevent the player from being able to run circles around the agents?

[muted sand]

I have a Tree which gets ignored after baking and I can't figure out why. Anyone got any tips? Here is the Tree object in Inspector and in the Scene after baking.
There are cubes in my scene as well which don't have this problem. Does it have anything to do with Mesh stuff? The tree image comes from a Sprite Renderer.

[sterile vine]

You could try few things. Like puting the capsule half inside the ground. Removing the isKinematic from the rigidbody, check the tree as Static, and change the layer to something else than ignore raycast. One of them or a combination of those may be the right answer

So i'd like to implement my own text to text AI solution, like GPT-3 or chatbots.
The ultimate goal would be to be able to just chat vocally with an NPC in my game and the NPC to answer. Just like so : https://www.youtube.com/watch?v=WHrVRAIaA9o&ab_channel=ThomasSimonini
I have solid basics in IT and a master degree, so I could solves the various issues when they happens.
So there are few steps inbetween, like converting the mic input of the player in text, then feeding the text to the AI. And getting back the text from the AI, and then converting it back to vocal. And ofc another that would be to know how to feed the AI with base text to customize each NPC.
The part I want to focus on, right now, is a text to text (so all the vocal part i'll see later) with an AI that I could embark in my game to request offline. I don't want to use the openAI API to allow offline and to remove the payment issue.
Could anyone point me in the direction of an open source chatbot/ai that I could embark in my project ? Maybe a tutorial stuff, or a git repo ? I'm looking into GPT-J and GPT-Neo for now.

[jaunty raft]

You need to find (or make) a model that is trained specifically for NPC dialogue in your game or a similar one. There are various ones available for interviews and other real life situations, they may work for you. But your key issue will ofc be accuracy and parsing any ai generated text into actual gameplay commands. A chat ai is still very far removed from achieving this and major strides in connecting ML with symbolic ai have not been made yet

[sterile vine]

This "key issue" is actually quite simple for me to achieve, just some parsing and formating on the fly.
The real black spot that I have is to find a way to embark an AI in local.
I found that the open source GPT-J has a dedicated conversational feature, that looks just like I want
https://nlpcloud.com/how-to-build-chatbot-gpt-3-gpt-j.html

I must find if I can embark that

[jaunty raft]

wdym by embark anyway? Distribute?

[sterile vine]

To put the AI inside the unity sources of the project so I can resquest the AI and talk with my NPC even if i'm not connected to the internet

[jaunty raft]

the word is embedding

[sterile vine]

Oh thank you for the clarification ! 😄

[jaunty raft]

Why would that be any issue? You can just do it if it’s an offline model.

[sterile vine]

Yeah, i've never done that before, I've only played with APIs and open collab

So the problem that may not be one is that i've no idea how to proceed

I found the sources here https://github.com/kingoflolz/mesh-transformer-jax/ along with pretrained models

[jaunty raft]

i think if the only experience with ML you have is in playing around with premade repos and never made a model yourself this is going to be a long journey

The key issue to solve is probably running tensor flow from a unity app

[sterile vine]

Yeah, i've never done that. However it seems like it wont really be required to train a model myself, since this GPT-J is shipped with pretrained ones that works just fine for my needs from what I can see of the playground on their API.

[jaunty raft]

you still need to run the model and convert any python bits to something else or run the required py in a separate process

a model doesn’t run itself

[sterile vine]

I've had a tensorflow thingy working with MLAgent when I trained some models with the unity package it may help in this journey

[sterile vine]

Yeap, that is indeed the problem 😂

[jaunty raft]

understanding tensorflow and integration of it will help a lot

The py notebooks that data sci uses for this stuff are just convenient, but not required, you can use a model from anywhere that supports its infrastructure (usually tensor flow)

https://scisharp.github.io/tensorflow-net-docs/#/

Also don’t underestimate the difficulty of parsing the text you get from any ML classifier

this is something that is by far a less popular technology than simple fuzzy classification and using it for fuzzy output (images/music/chat/sales trends)

if you pipe the outputs simply into a behavior tree or fsm, like any voice recognition telephone hotline, you aren’t really giving the player a good UX

[sterile vine]

I see, well I'll dig a bit more around that thank you @jaunty raft for your support !

[marsh lintel]

Hi! I'm trying to generate a navmesh in a procedural world. I added a navMeshSurface to my terrain chunk prefab (it's a 3d square area that gets spawned in a grid) and when objects are finished spawning I call the .buildNavMesh function to generate the navMesh. However.. what I get is this:

I think that every terrain is baking a complete map of all the objects in the whole map. That's because, as the strong blue color suggests, those are many navmesh stacked into eachother. Some chunks only have one navmesh on it... and only some trees are getting taken out of the navmesh.... Can someone help me?

[opal flume]

where can I find behaviour bricks sample scenes?

And FSM too

[sterile vine]

😂

[mystic thistle]

Hey. I'm using navmeshcomponents.
I need to rebake navmeshsurface at runtime, I did it in like my 2 other games, but this time for some reason I can't get reference to NavMeshSurface in any way. I can't find it (there's one) in the scene in awake, I can't drag and drop it (even though it shows like I could drop it, after I do so it still shows none) and also it says that there are no nav mesh surfaces on the scene.

even when I set it that way
groundSurface = environment.GetComponent<NavMeshSurface>();
it still doesn't find it's reference

even though NavMeshSurface is the only component on environment transform

[marsh lintel]

It makes sense aha

btw if you want to set a default response, just answer it manually (write something like "What's a telephone?"), so the a.i. will know what to say when it encounters a modern item

at least, this works on gpt-3

[sterile vine]

I mean the player could ask anything and not only telephones

[marsh lintel]

I was also trying to use gpt-3 and j for university purposes, and if you haven't a very, very powerful computer you can't run gpt-j in realistic times. I generated 100 tokens in 4 and a half hours, using CPU... To use GPU, your GPU needs at least to have 16GB ram, and it will still take a lot of time. The simulation that I saw in a video took 6 seconds to generate 20 tokens in a 5k€ quadro RIG, meaning that there were more than one of that monster GPU costing 5k each, and it still took 6 seconds

[marsh lintel]

Sure, but it would know that he can answer "I don't know what a X is"

test if if you want, it might work, it might not

usually if you give example responses he will answer correctly the next time

[sterile vine]

Ah i see. Well that suck to need to be this heavy for that

[marsh lintel]

It's a 6B model, sadly it's normal

gpt-j davinci has a 175B model, it would take much more time to run in local

[sterile vine]

Yeah the problem with the use of a cloud system is that you pay for everybody that use your app. In the opposite of using the computer that is playing the game. If tomorrow 500 person play my game and run the AI non stop i go bankrupt lol. It's not even worth to use a cloud system for a game, since with a 8 hours story, the player that pay your game 20 buck will cost you more than that just for the AI in the game

[marsh lintel]

Yeah. Using those nlp ai's is still out of reach, sadly. I thought that too.

You can only do a demo for yourself, or for someone that has a very powerful pc. No way that an average computer can compute anything so big fast enough.

[sterile vine]

😿

[proper sparrow]

Hello! I have a question. When using ML Agents in unity, the ppo RL algorithm is considered active or passive RL?

[marsh lintel]

Is this the place to ask for navmesh problem? or should I move my question to code-general?

[muted sand]

Sadly enough, these suggestions (and combining them) didn't do the trick.
I've played around a bit with another object for which the baking was behaving as expected, and discovered that it was only baked correctly if it has these:

• Cube (Mesh Filter)
• Mesh Renderer
  If any are removed, it isn't seen by the Navmesh Agent. Interestingly enough, a Collider isn't necessary. I'll read up again on the Navmesh docs, see if I've missed something.

[muted sand]

Depends on the question I guess, but I agree there's quite some overlap

[muted sand]

Alright, I've added a base cube to the sprites. This way I can easily visualize which areas are walkable and which aren't in the editor, and since the camera ignores the cubes they won't actually show up in-game. It works and it might even be better for game design purposes.

[sterile vine]

Good to know thanks for the feedback

[left nymph]

try to be more specific about the AI:
Utility AI
Behaviour trees
GOAP
Hierarchical Task Networks

then you should find plenty

machine learning is pretty rare in game ai

[fading burrow]

Whats the best way to get a random NavMesh destination? Rn im taking the current sector, adding or subtracting random numbers to it, then checking if theres a path to it (I migth be messing this part up?) but itd be really nice to be able to say things like only choose a destination on this type of layer or NavMesh filter thingy

[brisk raven]

https://docs.unity3d.com/ScriptReference/AI.NavMesh.SamplePosition.html

[fading burrow]

So does this work by taking a destination- that I would generate- then finding the closest spot it can actually get to, and allows for choosing what masks it can place it?

[brisk raven]

did you read it

[fading burrow]

yes

i did misunderstand it it looks like

im a little confused. how does the nav mesh hit output work?

gdi. and now its crashing

[fading burrow]

Is there a way to check if theres any objects nearby that doesnt ignore line of sight? raycast would work if it didnt have a direction, i tried detect sphere but it doesnt seem to be working with 2d, and im not sure if that ignores line of sight

[brisk raven]

Physics.Linecast doesnt have direction

[rustic obsidian]

A linecast is just a raycast.
What's the actual problem?

[fading burrow]

I want to check if any objects are close by, without ignoring line of sight

[rustic obsidian]

So like, a spherical sort of outwards raycast that hits things

[fading burrow]

Yea

[rustic obsidian]

There's no such thing. Alternatives could include collecting nearby objects using physics operations or other logic, and then loop, casting at selective parts of them that you think are enough to ensure something is deemed visible

[fading burrow]

How would I collect nearby objects?

[rustic obsidian]

Maintaining a list of objects that hit a trigger is a common way

[fading burrow]

Sorry very new to this. To detect objects that hit a trigger- would I need to do something like make a second collider on the object that nly acts as a trigger? So if I want a detection radius of 4, have a circle collider with a radius of 4 that doesnt do physics collisions, that acts as a trigger to check if theres a direct path between them?

[brisk raven]

physics2d.Overlapcircle and then do ur path calc

[rustic obsidian]

The circle collider (marked as trigger) with radius 4 would only be acting to collect the certain objects you want to consider line of sight for.
When they enter the trigger you add them to a list, when they exit, you remove.
For each object in the list you have to do any number of raycasts you want from the thing you care about to points on the object (which could just be one, to its origin, no idea what you need), and if it hits it it's considered close and in LoS.

You can also use OverlapCircle if you don't want to maintain a list yourself, which could be totally fine too

[fading burrow]

oh overlap circle is probably the easier way to do it

[brisk raven]

you can loop inside the array and check each one for a valid path

[fading burrow]

now to figure out how arrays work in visual script...

[brisk raven]

#763499475641172029

[fading burrow]

I had the ai pathfinding working pretty well, messed with some stuff, now the ai doesnt know to go around this wall

its both marked not walkable and an obstacle

Please @ me if you respoond

[ashen smelt]

Are you baking using NavMeshSurface or the Navigation tab? Either way, it seems as if the trees are "collected" on the lowest LOD level (which probably is a flat impostor). Change the layers of the tree parts, or use colliders instead of renderers when baking the navmesh

[whole ocean]

Hi guys, anyone knows a why to make NavMeshAgents (nma) fall if the floor (mesh) beneath them disappears? Rn it tp back to the closest point existing in the rest of the navmesh. I know that if I disable the nma component and set kinematic/gravity to true, then destroy a portion of the mesh, it works, but it's just a hack that works on certain scenarios.

[tardy junco]

It's not really a hack. You need to disable the navmesh agent or stop it from controlling the object position, then let it fall in whatever way you're comfortable with(either physics or moving it manually).

[marsh lintel]

Hi, how can I change a navmeshagent active navmesh? I have many navmeshes in my game, they are completely dynamic (created around some entities with a navMeshSurface), but sometimes when an agent "pick" a navMesh it will not change that even if he steps inside a new navmesh

Like in that case, where the selected navmesh is the left one, and when I move the object to the right navmesh it will not change the active navmesh and it will simply get stuck there between the two never crossing it

[woven summit]

Hi - I've got a basic question about using Navmeshes. I've got a terrain with a navmesh surface on it; my navmeshagent dutifully navigates across it just fine - however, I can't seem to figure out how to make the "river" (a second 3d Plane object that's slightly above the minimum elevation of the terrain) impassable. The Plane/Water Level has a NavMeshSurface set the "Not walkable" by default, however, the NavmeshAgent continues to navigate along the riverbed.

[amber helm]

Figured I'd share since I think it's amazing 😄

If you're doing 2d and using tilemaps, you can easily create navMesh's that match your flooring exactly by using this code. All you need to do is make sure the floor object of the tilemap is tagged "NavMeshFloor" and you're good to go. The navMeshComponent is just a prefab with a NavMeshSurface2d on it set to -90 on the x rotation (Needed for the 2D navMeshPro add-on).

Screenshots showing the difference between normally building the navMesh with modifiers and building it at runtime with this function I created.

private void GenerateMesh()
    {
        TilemapCollider2D meshFilter;
        Mesh mesh;

        // Spawn a new Game Object to hold the Nav Mesh Surface and set it to Render Meshes.
        GameObject navMesh = Instantiate(navMeshComponent);
        navMesh.GetComponent<NavMeshSurface2d>().useGeometry = NavMeshCollectGeometry.RenderMeshes;

        // Finds the Tilemaps Floor Game Object, adds a Tilemap Collider to it and creates a mesh basedon that collider.
        GameObject[] floorTiles = GameObject.FindGameObjectsWithTag("NavMeshFloor");

        for (int i = 0; i < floorTiles.Length; i++)
        {
            // Add a tilemap collider to the floor area, creates a mesh with it and then removes the collider.
            meshFilter = floorTiles[i].AddComponent<TilemapCollider2D>();
            mesh = meshFilter.CreateMesh(false, false);
            Destroy(floorTiles[i].GetComponent<TilemapCollider2D>());

            // Creates a temp Game Object to hold the Mesh Filter and Renderer. Destroies the mesh after being applied to the mesh filter.
            GameObject navObject = new GameObject("navObject_" + i);
            navObject.AddComponent<MeshFilter>().mesh = mesh;
            Destroy(mesh);
            navObject.AddComponent<MeshRenderer>();
        }

        // Generates the Nav Mesh based on the created Mesh Filter then deletes the Mesh and temp Mesh Game Object.
        navMesh.GetComponent<NavMeshSurface2d>().BuildNavMesh();

        // loops back through and deletes the navObjects.
        for (int i = 0; i < floorTiles.Length; i++)
        {
            GameObject navObject = GameObject.Find("navObject_" + i);
            Destroy(navObject);
        }
    }

[bitter rune]

Like in that case where the selected

[stone owl]

I can't say neither my nav modifier/nav volume modifier components are working.. is there some trick to it?

I'm basically trying to mark a 1x1 cell as unwalkable during runtime

. Baked the navmesh (visible in scene)
. Created a "wall" tile with a modifier component set to unwalkable

but still, see no change to the surface or consequence to my pathfinding

[fringe edge]

Hey I'm struggling with this, how am I supposed to make a navmesh agent that flies?

Or tbh I don't even want it to 'fly' really, I just want it to aim to hover above the player. There doesn't seem to be a lot of good information about stuff like this.

[fringe edge]

Tried just attaching a model to an invisible navmeshagent of a new type just now and the agent won't even work

Why is it even possible to create another agent type if only one works???

[stone owl]

Offset the model from the navmesh agent bounds

so the agent will move using the navmesh, the model will appear to fly

or don't use navmesh but you'd be pathing yourself.

[fringe edge]

Yeah this is what I ending up doing

[bitter rune]

you need to bake a navmesh per agent

[devout arrow]

I have a python program (neural network) that needs to be fed inputs like position and rotation from the scene, and its outputs need to be fed into a C# behavior script to control movement. Any ideas how do I do this?

[brave compass]

https://github.com/pythonnet/pythonnet
Did you check that out? Not sure if that works with Unity, it's on my way too long todo list of cool stuff to program 😛

[worn oxide]

What's a good place to start with AI? I'm looking to start building AI for enemies and NPCs in a multiplayer game, the NPCs each have a daily routine, with different daily paths they'll take.

Would using behavior trees be a good foundation for everything or is that overkill? I'd like to keep it relatively simple and easy to iterate, but I don't know much about what different methods are out there for implementing AI.

[jaunty raft]

Behavior trees are neither over nor underkill, it’s how you do it properly. If you don’t want to use an asset you can easily DIY it or use one of the free code based ones.

[worn oxide]

@jaunty raft Thank you, are there any assets that you would recommend? I was looking into Behavior Designer: https://assetstore.unity.com/packages/tools/visual-scripting/behavior-designer-behavior-trees-for-everyone-15277

[jaunty raft]

I’d recommend node canvas personally, but BD is fine too

[worn oxide]

I'll check that out as well! Do you find that designing behavior trees visually is generally easier than coding them?

[jaunty raft]

It’s much easier to debug them visually

and allows a designer to build/tweak them safely without technical knowledge and on top of a vocabulary that your (as the programmer) provide

[worn oxide]

I see, I really appreciate your help, and I'll look into NodeCanvas too!

[jaunty raft]

There is also flow canvas which is the generic visual programming companion to it, in case you need it. It’s great for high level flowchart style code that designers may want to tweak

[young bloom]

Hey guys, Im working on a project and I have a fairly odd requirement. Im looking to change the plane that the navmesh operates on so that what would normally be height on the Y axis would instead be the Z

Is such a thing possible?

[devout arrow]

I did not, ill take a look. thanks

[alpine glacier]

how may i go about finding empty space between/beside opponent team players in a football game?

i mean of course i can figure out a random empty point, but I wanna make it look like realistic behaviours

[dawn flower]

anyone here use candace ai?

[storm violet]

I have problem with AIs crowding around one point and not moving to the next one. How can i make the position check by radius with center on this point and not by specific point

Like this

[cursive wadi]

how do i make my enemy ai to not take the same patth as other enemies around it i am useing A* pathfinding and a custom script

sorry it is a bit too long

for discord

also it is a 2d game

[cloud jetty]

I'm wanting to make a multi-stage, multi-ability 3d boss battle, I dont know anything about AI, when I google about it, I dont receive a solid answer on what or how to go about this? I'm being thrown multiple different "types" of AI i could use, although everyone doesnt seem to agree on which ones easier/better? I thought i'd ask here for some advice! What's the best AI Method for this? Best place to learn AI?

[jaunty raft]

Books http://www.gameaipro.com/, or download a behavior tree asset from the asset store and follow its tutorial/examples (Node Canvas or Behaviour Designer)

You can also look for yt videos or articles on the web that explain how behavior trees work. Just skip the tutorials on yt if you want to actually learn that stuff

[cloud jetty]

Alright thank you!

I did look into a unity asset "Eliot AI" which I was recommended, have you ever seen/used it before?

[jaunty raft]

Never heard of it

Use node canvas or behavior designer, they are the real deal used in many commercial projects

[light saffron]

pandabt v2 is much easier to use than BD and NC

performance: BD is about 3x faster than NC when i tested

[jaunty raft]

Apples to oranges

[light saffron]

all 3 are bt so not sure what you mean

[jaunty raft]

apples and oranges are both fruit, they still don't grow in the same climate, taste the same or are substituting each other

[jaunty raft]

since you didn't understand: behaviour designer and node canvas have a visual scripting frontend (for designers to use) that is supported by a scripting API to allow the code team to make and curate a list of custom nodes for designers to use as their vocabulary in designing/tweaking/understanding/observing/debugging behaviour, pandabt is scripting based and targets a different audience & workflow where everyone is ready to define/tweak/debug stuff in plain text, consequently these different approaches need to focus on different things in their implementation.

[pulsar galleon]

Hey guys (noob here) what's the best way of doing 3D AI movement? I'm trying to use navmeshes with the experimental navigation package, but is that the recommended way?

[jaunty raft]

what's best depends on what you want to do. NavMesh is the built-in way to do agent pathfinding/navigation, there are other options that may be worse/better depending on your needs. In particular, navmesh assumes you have a walkable plane, so you can't do navigation that is not locked to the surface of a mesh, like in a game where your player can fly.

[pulsar galleon]

I have a grid system which builds the "floor" at runtime with multiple tiles. I just want my Utility AI to be able to navigate this surface and avoid any walls or objects that are placed on the grid. NavMesh seems like the best way of doing this, but when I download the experimental package and add the nav mesh surface to my tile prefab, they don't connect to eachother.

Would it potentially work if I were to group all the tiles under a single game object and give that game object the nav mesh surface component?

The terrain will never change shape, and I don't expect the ai to fly, jump, change it's y level, etc

[light saffron]

there you go, much clearer than your fruits . i agree that in team it's very different target audience. if you do both design and code the difference blurs out.

[rugged palm]

has anybody tried building ai based on neural networks. does unity have a solution for this or is there something on the asset store I can use?

[south plank]

help!!! im trying to figure out why my navmesh agent is so weird with steering
it sticks to walls
it follows a path but it sticks to walls

[spring creek]

I really need help, i have been trying to install MLagents for hours now... every seems to work when i type in the CMD "mlagents-learn --run-id test1", but when it says "then it loads as it should and saya "press play in unity", but as soon as i press play in unity i get this error in the CMD "runtimeerror: could not infer dtype of numpy.float32"... As far as i can read it means i have some sort of compatibility error, which i do not understand

I use: python 1.10.0
PyTorch 1.11.0+cpu
ml-agents: 0.30.0,
ml-agents-envs: 0.30.0,
Communicator API: 1.5.0,

If you need more information from my side, pls let me know and i will get a hold of it

[cursive wadi]

would i be able to put a delay on A* pathfindg sscript so that it cans only after all rooms have spawned

[woven mason]

Hello, I am using Unity 2021.3.14F1 and I can't seem to figure out why the Custom NavMeshLink is not working. I've read Unity documentation and been on the forums but I still can't figure it out. Any help would be appreciated! In the video I showed you my setup.

[lapis hamlet]

It's a tough thing to trouble shoot, try setting it up following a tutorial very closely in another scene

if you can get that working then you can modify set scene to your requirements until it breaks, and you'll have found your problem or more likely it will just be fixed

[lapis hamlet]

cans ?
can ?
can what

[cursive wadi]

fuck sorry so thaat it can scan only when all the rooms are done spawning

[lapis hamlet]

Well A* just solves a path for a given input

you can call it whenever you like

[devout arrow]

Currently have a drone which is attempting to fly to a goal. As expected, its a complete mess at the start, but Im wondering if I should reward it for not crashing? I have it get a reward every time it touches the goal and gets punished every time it touches the sides of the cage.
Im using MLAgents

[frosty peak]

guys im using navmesh agents for my bots but they get stuck on eachother is there a way to make them go around eachother

i tried using nav mesh obstacle (and carve) but they freak out

obv doesnt work

is there a workaround?

[rugged palm]

Are the bots using physics?

[frosty peak]

Yea

Rb?

Or do u mean something else

@rugged palm

[rugged palm]

yeah that's what I meant

[rugged palm]

how do their colliders look like?

[frosty peak]

Just a capsule collider for now

[rugged palm]

What does it mean when they get stuck on each other? Do they eventually get unstuck or do they permanently get stuck? Just trying to visualize

[frosty peak]

No they just block eachothers paths sometimes

Cuz they collide

I made them not collide with eachother using the physics matrix

But it looks stupid lol

I want them to go around eachother instead

[rugged palm]

oh yeah I don't know of a native solution to solve that with navmesh outside of setting all the ai as dynamic navmesh obstacles themselves. You will probably have to code if that does not work. You can also give them a physics material that helps them seperate if that doesn't impact any other parts of your game.

[frosty peak]

Yea I tried using the nav obstacle component to make it not walkable around them

But that means they cant walk either

[rugged palm]

can't you set some sort of invisible object around them that is a child of the transform and set that as the navmesh obstacle.

oh nvm

i see why

[frosty peak]

Wouldn't that still effect its path

Yea

[rugged palm]

they will block themselves

[frosty peak]

Yup

[rugged palm]

What type of game is this? how complex is the ai

does the ai only need locomotion

[frosty peak]

It doesn't have to be complex I just wanted it to look nice

[rugged palm]

you could also make the obstacle only block behind them and to the side

[frosty peak]

If there's no fix I'll just keep the thing I did rn

I think the problem wouldn't be fully fixed if I did that

Good idea but there's no need I'd rather have it not buggy at all

Rather than a little bit buggy

[rugged palm]

ok. If you want to code. You could also use raycast or something and make it slow or something when someones ahead of them

[frosty peak]

Yea I didn't want to

But thanks for the idea

[sharp gulch]

trying to make a simple ai that follows a gameobject. what do i do?

[frosty peak]

Do u have a navmesh agent component on ur "AI" game object

[frosty jackal]

anyone else still using the marathon-envs project for thier ai training?

[flat sedge]

Hey everyone, I need to select one of the following exercises and I was wondering which one would you guys suggest me in terms of difficulty ( I want it to be easy) .

A. Distributed scheduling via auctions 

-Specification of the distributed scheduling problem using agents’ valuation functions
- Connection with competitive equilibria
- Specification and implementation of the ascending auction algorithm (convergence and termination)
- Experimental results

B. Computing solution concepts of Normal form Games 


- Computing Nash equilibria
- LCP formulation and Lemke-Howson algorithm
- Computing minmax and maxmin values
- Computing dominating strategies

C. Congestion games 

 (theory and implementation)
- Definitions and congestion games as potential games
- Non-atomic congestion games (NCG)
- Nash equilibria
-  Specification and implementation of an algorithm for computing Nash equilibria in NCG
- The price of anarchy 

D. Fictitious play and reinforcement learning for computing equilibria 

- Repeated & zero sum (stochastic) games
- Fictitious play (FP) (theory and implementation)
- Reinforcement Learning (RL) (theory and implementation) application to a game for the computation of equilibria
- Experimental results comparing FP and RL

E. Distributed Optimization

Formulate the matching problem as a repeated game
Solution concepts (Pareto optimality and Nash equilibria) for the matching problem
Matching problem formulation as an MDP
Reinforcement Learning (RL) (theory and implementation) to solve the matching problem
Comparison of results with the naïve auction algorithm and the solution concepts identified.

[jaunty raft]

they all seem to be scalable in terms of difficulty, i.e. once you understand the concepts/principle you can spend an arbitrarily large amount of time to make it all "nice". I'd pick the one that deals with the concept most interesting to me. None of them are particularly easy, since they benefit from (maybe require) intuitive understanding of game theory

one caveat: "distributed" always makes stuff hard 😄

[flat sedge]

Understandable and kinda makes sense. My professor said that the D has 70% implementation and 30% theory. And I consider myself to be more practical. Of course I will full understand the theory but you know what I mean.
Which platform/programming language would you suggest?
I know python,c#,java and I made my diploma thesis in ml-agents.

[jaunty raft]

tbh, i'd do them all, they all sound like a bunch of fun

[flat sedge]

I have plenty more projects that I need to finish before the semester ends sadly.
Im attending an ai master

[jaunty raft]

if you want to make it "nice", use C# and visualize in unity!

the language doesn't matter since these are algorithm problems

[flat sedge]

That's what I wanted to do, but with some ready to use libraries not out of scratch haha

[jaunty raft]

maybe don't use C++

[jaunty raft]

what kind of libraries?

[flat sedge]

I haven’t used c++ that much anyways

[flat sedge]

I mean something like ml-agents provided by unity or something

[jaunty raft]

that probably misses the point of the exercise

[flat sedge]

I mean its like using sklearn for training a classifier. You don’t write it out of scratch

[jaunty raft]

what kind of course is this?

is it about fundamentals or application?

[trail vale]

ChatGPT is awesome

[jaunty raft]

its an impostor with an ability to imitate the voice of an expert

[little nymph]

Howdy all! I'm trying to have my navmesh agents follow their leader in formation. Currently I have them hard-coded to go to an empty gameobject behind their leader unless attacked by player. But I have to hard code this each time. Is there a way to loop through the empty gameobjects to see which ones are not taken instead of hard coding each AI to have it's preset gameobject? Thanks!

[whole ocean]

(If I understood it correctly*) You could have a property to track if an agent is following a leader, maybe bool type (or more complex if you need more information), and another property to track if someone is following me, again bool type. Thus, if an agent is not following someone, then loops trough the collection of "followable agents" and check if any has the second property set to false, meaning it is not being followed and this agent can follow it. Hope it helps 👍

[little nymph]

this is sort of how I'm doing it at the moment. There's a bool set for is following vs leading. My issue is trying to select a position for the follower to go to and to check to make sure it's not already taken

[whole ocean]

Ahhh okey, I think this could be solved with a grid. A "Leader" has a grid behind, around or in front of it, depending of the Leader "type". Then every follower queries for an empty position (if any) to the grid, thus the agent can follow it or do something else if it is already full

I restricted the problem to a finite number of followers, but maybe a dynamic size grid can be implemented instead

Thus, a potential infinite number of followers could follow the leader

[whole ocean]

I'm guessing you need an organized looking formation right? Like the ones in Age of empires?

[little nymph]

Sure basic rectangle formation

[whole ocean]

Yeah, the grid sounds like your solution then. Let me know if you were able to solve it anyway 👌

[little nymph]

solved it by adding each nearby unit into a list and then assigning a leader unit. the units then each get assigned a location for the formation based on an array of gameobjects

[mystic stag]

is there a way to include trees on a terrain object when baking a navmesh? it doesn't carve any holes for them so my ai just goes straight through them

[lunar plover]

im using astar for ai its working pretty good apart from when i made stairs for a quick test he can climb all of them apart from one because its too tall so he decided to just go through the stair/clip through it and i dont want him to do that i just want him to stop walking if he runs into it

[rugged palm]

Make sure your tree has colliders

[mystic stag]

yeah they have colliders and enable tree colliders in the terrain collider component is checked too

[rugged palm]

hmm... that's weird. Does your ai have colliders too?