🤖┃ai-navigation

[tardy junco]

Hmm.

[late thunder]

Im using this btw https://github.com/Blueteak/Unity-Neural-Network/

[tardy junco]

Weird. It kinda works against unity logic. The moment you destroy a GameObject, its components are also destroyed and should return null reference if you try to access them...

[late thunder]

hmm yea im not sure what i did there either actually

[real sonnet]

destroyed at the end of the Update loop though ?

so still available this frame ?

[tardy junco]

Yeah, but unity should still throw an error if he tries to access a component marked for destruction. Or so I thought...

@late thunder what's the purpose of that Destroy there anyway?

[late thunder]

otherwise i spawn useless cars

i only want their brains

so I can just return a list of brains

ill do that actually

[tardy junco]

MonoBehaviours can't live without a GO. So it's probably destroyed at the end of the frame.

[real sonnet]

By the way it seems you give them a random brain "when spawned", to then give them the one from one of the parents ? Isn't it useless ?

[late thunder]

its for the first generation

[real sonnet]

oh okay

[late thunder]

they dont have a brain yet

[real sonnet]

fair :p

[tardy junco]

And GetBrain function probably returns something, right?

[late thunder]

yeah it returns this.brain, or if it is null then it makes a random brain

so im thinking the problem is lying there but i just cant find it

[real sonnet]

yeah try to clone instead of passing reference

[tardy junco]

So CloneCC is something that you don't need(want to dispose of) along with Clone object?

[late thunder]

yeah

so im writing a function now to only copy the brain

or clone

[tardy junco]

Then I don't get the whole point of instantiating a new object...😅

[late thunder]

it was temporary

but if it works it works

im changing it now though

List<NeuralNet> CloneSelection(CarController[] selection, int size, GameObject prefab)
{
    List<NeuralNet> clonedList = new List<NeuralNet>();

    for (int i = 0; i < size; i++)
    {

        NeuralNet nn = new NeuralNet(6, 8, 2);
        nn.SetBrain(selection[0].GetBrain());
        clonedList.Add(nn);

    }

    return clonedList;
}

i changed it to this

it still doesnt work tho

[tardy junco]

What's Brain? Is it a Monobehaviour?

[late thunder]

its a NeuralNet

https://github.com/Blueteak/Unity-Neural-Network/

im using this small custom library

[tardy junco]

Okay, it's not. So this part should be fine in my opinion.

Here you said that they all contain the same number. Is that unwanted behavior?

[late thunder]

No thats good

i want them to be cloned

so thats working just fine

[tardy junco]

Okay. Then what's the problem now?

[late thunder]

this

somehow, it clones incorrectly or something once, and then it clones correctly the rest of the time (forever)

i can make a quick video

@tardy junco @real sonnet (sry for @ )
https://streamable.com/t4hjax
this has been ruining my life for the past week

this is consistent behaviour btw: the first time it clones, its always mutated, or random. the generations after that are perfect clones of the one I clicked on

[real sonnet]

Yeah they all get c1 brain on L:143

[late thunder]

yup

[real sonnet]

and then you want to activate back mutation and have all except one behave differently

[late thunder]

yea

[real sonnet]

aaaand I dont remember what goes wrong

[late thunder]

so the good one doesnt get mutated and ruins the generation

[late thunder]

it doesnt keep one un-mutated. not applying mutation at all doesnt solve the problem either: the first clones are random (i think), and the clones after that are perfect clones of the random clones

[tardy junco]

How do you pick the good one?

[late thunder]

i click on it

[tardy junco]

Where's that code?

[late thunder]

wait heres the new code

https://hastebin.com/ahenikakoq.csharp

for genetic alg

[tardy junco]

I don't see any clicking related code...

[late thunder]

this is in carcontroller

[tardy junco]

Try debugging the gameobject and making sure it's the one you actually clicked.

[late thunder]

everything is Car(clone)

[real sonnet]

Might be just an opinion ? Or try to increase mutationRate ?

wait I'm confused xD

[tardy junco]

add gameobject as the second parameter. That will allow you to click the debug to highlight the object.

[late thunder]

but it should not matter, because if I do mutate the population, then it still wont keep the best from the prev gen

[late thunder]

oh ok thx cool

[late thunder]

i disabled it on purpose

to show that it clones the wrong one

or a random one

[real sonnet]

yeah I just have a hard time getting my head around it sorry

my bad

[late thunder]

nooo youre already a big help you dont have to use your time to help me

[real sonnet]

that's fun though

[late thunder]

its not fun after 10 hours of debugging 😦

[real sonnet]

So if I get it now, you expect the clones having the same brain as the fittest one from prev gen ?

(without mutation)

[late thunder]

Normally, yes. But now I am manually clicking on them to test if they are cloning the right one

[real sonnet]

so on second gen they should all get to the green car position

ok

[late thunder]

yeaaa

[tardy junco]

line 174:

newCarController.GetBrain("Spawn");

What's the purpose of this? Did you forget to assign it to something?

[real sonnet]

its actually the oposite of what i was thinking. I was looking for a reference vs value problem, but instead your clones dont get the fittest brain

[late thunder]

its to get a brain (in the constructor of CarController i dont get a brain)

GetBrain also makes this.brain = new brain if it doesnt have a brain already

public NeuralNet GetBrain(string i = "")
    {
        if(this.brain != null)
        {
            return this.brain;
        }
        else
        {
            Debug.Log("Nieuwe Brain: " + i);
            this.brain = new NeuralNet(6, hiddenLayerNodes, 2);
            return this.brain;
        }
    }```

the string is just for debugging purposes

[tardy junco]

Okay... Not the best descriptive code, but that explains it...

[late thunder]

yea im kinda new to coding

[late thunder]

yup

[tardy junco]

Is there any way to identify the brains?

do they have an ID or something?

[late thunder]

yea im using that specific weight as an ID

[late thunder]

the thing at the bottom

I am going to see if the selected brain and the cloned brains have the same ID (already did this, but checking again)

[tardy junco]

wait, your EnableMutations is unchecked.

in the video

[late thunder]

yea

so they should be the same

[tardy junco]

And they are, no?

[late thunder]

I am going to check now again if they have the same ID

[tardy junco]

I mean, they all go to the same point

Now I don't understand the problem anymore.

[real sonnet]

yeah same

[late thunder]

its supposed to go to the car im clicking

sorry for not explaining clearly.

[real sonnet]

i was looking again and noticed that haha, I'm so confused right now

[tardy junco]

did you debug as I suggested?

[late thunder]

im circling my mouse around the spot it should go

yea 1sec

it appears to have exactly the same ID, meaning the brains of the selected car and the clones are the same. It drives differently tho somehow

[tardy junco]

How do you debug it?

And did you try what I suggested?

[late thunder]

Huh, weird. its returning a different instance id

I log the instance ID when I click it, and then log the instance ID in th e foreach loop

and its different

GeneticAlg

CarController

[tardy junco]

Perhaps you have several of the chosen ones?

[late thunder]

I just checked

only 1

weird

[tardy junco]

How did you check?

[late thunder]

logged only once

[tardy junco]

Hmmm

[late thunder]

does it change the instance id or something when you put it in a list

nah that would be weird

im going to output every instance id

wut, the instance id shown when clicking a car isnt even in the list of sortedCars

what in the world

[late thunder]

the brains are the same tho

[late thunder]

I give up

[real sonnet]

:/

[lost verge]

How does unity make that graphic in the navigation editor window?

The graphic that shows the character's height and stuff

[steady merlin]

Hey i wanna archive some enemy avoidance with steering behaviour. But somehow it works not as i want - i walk towards the highest dot product value but thats kinda wrong i guess

cod >>> https://hatebin.com/fdrodqmncf
reference >>> https://www.red3d.com/cwr/steer/gdc99/

can someone pls help

[uneven monolith]

hey guys, i have an ai with nav mesh but the terrain asset is fucked so i cant use only nav mesh, how can i make im create random points within a certain range, this one that i have, creates a walking point in the range of the on that came before i want a general range

sry if it is a bit confusing

[tired void]

Hey! I can´t get my cooldown working, any ideas? The code snippets using it: https://hatebin.com/evmfvqsriq

[hardy anvil]

all the refs are wrong. google the keyword

also, use a coroutine for a timer

plenty of examples on google

[hasty pivot]

would wave spawning count as ai becuase im trying to use brackeys wave spawner and they do spawn but they spawn infinitely

ive been looking at his video and cant find what i did wrong can anyone help me?"

i just need help how to stop them from spawning infinite amount of spawns

[gritty cloak]

Hello 👋 Can any one recommend a good AI / Behaviour Tree framework? There are a few options out there,but is there like one that sticks out / battle-proven? I tried setting up my own framework but quickly hit the limits of my skills.

[full sierra]

@gritty cloak I used Opsive Behavior Trees from the asset store, I like it, allowed me to visually troubleshoot my state, just like blueprints

The only part that was always confusing to me, is that every action and every decision needs to be it's own script. So unlike having a state machine that is part of your Character class, the scripts have to find a reference to the Character they are responsible for each time they execute. But in the end it didn't end up being any major overhead to use a guid and a singleton that would return the correct character

[gritty cloak]

@full sierra You mentioned blueprints... do you mean Unreal Engine? I really liked their AI system. Is it comparable?

oh wow the trailer for that asset is an official unity video. i guess i found my solution then. seems as advanced as i'd need it to be

[full sierra]

Yea Unreal Engine, well Unreal Engine allows you to blueprint everything. I have not used Unreal, but the idea is the same. If you can see you code snippets running, it is easier to debug for most of us. AI is tricky to debug with a log, you need a visual reference (IE: Blueprint like system)

[gritty cloak]

yea, just bought the asset cause yolo but looks very familiar. just have to get used to how it works in unity. thanks for the recommendation!

[rancid frigate]

good day people. Im hoping someone can assist with navmeshagent's. Im setting up an enemy npc with navmeshagent to follow the character target, but no matter what I set the NMA speed to it's still keeping pace with the player, even setting speed to 0. Can anyone point out what I've obviously missed?

[tardy junco]

How are you moving it?

[rancid frigate]

@dilch I fixed that issue 🙂 I had forgotten to add a capsule collider lol. Just struggling with an issue on the anim right now

meh. I always type your name wrong too 😛

@tardy junco any chance you're familiar with importing animations from blender to unity (same model in both)?

[tardy junco]

A little bit. They should work out of the box if they use the same rig.

It's better to ask in #🏃┃animation

[rancid frigate]

should lol. and yeah I have put the post in there, its a dead zone currently though 🙂

whats working flawlessly in blender is dropping the character transform on the Y axis as soon as the animation starts playing in unity

http://203.86.195.138/AnimDropping2.mp4

there it is 🙂

got no idea why when the rig/model is the same and works on the blender end. Im only guessing but I'm assuming I missed a step at either export or in unity

thing is, as you'll see in the video, the actual GameObject transform isn't shifting when it happens

the entire rig position doesnt match the blender pose on any key either

[alpine glacier]

that happens cuz ur animation and model are not set to humanoid

[gritty cloak]

Hey, what's the purpose of setting an agent during baking a navmesh? I thought there are different types of agents you can setup already?

[tribal dew]

Hello, Just wanna spread attention for my "Help wanted" thread on Unity Forum in Navigation & AI, Seems like here is waaay more ppls than on forum...
Thread is about problem with navmesh agent local avoidance in RTS, if anyone willing to point me out maybe a good asset which resolve this i will be grateful.
https://forum.unity.com/threads/how-to-approach-this-problem-with-navmesh-agents-local-avoidance-asset-suggestion-maybe.1025272/

[robust mantle]

Can I pathfind only where there is a specific light level and how would I do this?

[full sierra]

You will need data on your light level, so you can include it in the map. Some ambiguity in the term light level though, light level could be something your performing on each tile, or maybe you mean "Can I determine the light level of a unity light shining on a specific tile at a specific location" in which case you got to ask the guys in the lighting channel.

[prisma prawn]

hey guys

how do i make a A.I like i am a total newb so i don't know what navmest means and stuff

so does anyone know any posts or videos that might guide me through?

[trail shell]

check brackeys channel he has a tutorial on that.
Note : you can make more complexed AI logic by using some alogorithms like A* or theta* and more without using navmesh

@prisma prawn

[prisma prawn]

Sup

@trail shell thx

[sonic rain]

@prisma prawn Hi, I just wanted to add that AI and Pathfinding are not the same thing. When you ask about NavMesh you are asking about finding a path between two points, which is related to pathfinding. AI is more about ways to have entities/game managers which can "intelligently" make decisions on their own depending on the game's world state. For example: on a FPS game, how should a Bot react when they see a grenade being launched towards it? Should the Bot run? Should it take cover? Maybe, if it has enough time, should it pick the grenade and throw it back to the enemies?

Well, pathfinding IS useful for AI because, on games, we usually needs our entities to navigate through the scenario, so AI and Pathfinding are two terms which usually are together, but they are not completely coupled to each other

So, I'm not sure if you just want to know about pathfinding, but if you're interested in understanding AI to create your own Bots, and you say that you're just beginning with the matter, I suggest you three AI models:

.
Finite State Machine (FSM): in games, commonly used for AI and animation. You create a graph of states and create transitions, actions and decisions in order to define how the behaviour of the FSM agent will be
A introduction to the concept by Robert Nystrom: https://gameprogrammingpatterns.com/state.html

Behaviour Tree (BHT): also very commonly used on AI. You define the behaviour of the entity by creating a tree (in terms of data structure). This model was used in Halo and Bioshock
Nice Gamasutra article explaining the concept: https://www.gamasutra.com/blogs/ChrisSimpson/20140717/221339/Behavior_trees_for_AI_How_they_work.php

Goal Oriented Action Planning (GOAP): used to create planners which can come up with plans to resolve problems in a very dynamic way. It's the AI model used in F.E.A.R
Very useful source on GOAP: http://alumni.media.mit.edu/~jorkin/goap.html

So, if you're looking into knowing more how AI works, I suggest you to read these posts.
As for implementing your first AI, I would try FSM first as I think it it a good start for beginners, and as it has a nice evolution: **Hierarchical **Finite State Machine (HFSM). Hit youtube with the search "Unity3d FSM" and see what better fits your current level

[prisma prawn]

@sonic rain thank you so much this article helped

but i want to create a AI enemy in a horror game like the one you see in granny or in metal

which one should i use if i want to create a AI similar to that

[sonic rain]

@prisma prawn
For that, both FSM and BHT seems like a nice fit. But I would use FSM in this case. Your collection of states would represent the possible states in which the Granny could be. Is she chasing you? Is she just wandering randomly in the house? Is she attacking you? Etc

Simplifying and summarising a lot, just to illustrate:

Every FSM has its starting point, so in that case, Granny would start by choosing a random place in the house to go

And the it would go from/to states depending on what happens

Like if you make some noise, she would go to where the noise happened and see if she finds you...if not, then she goes back to patroling again

[prisma prawn]

You type fast

[sonic rain]

Movement can be aimed by NavMesh system or not, that's your choice

[prisma prawn]

Yeah I understand thank you so much for your help

Like really thank you

[sonic rain]

No worries, just really consider diving deeply into this, because AI is an important subject on game development and if you make your own FSM system (or BHT, GOAP), you can try to make it re-usable between other projects

So you can have the same basis FSM for other types of enemies

Investing time on reading and making it correctly is a good thing for you as a gamedev

[prisma prawn]

Thx for the heads up

If I earn from my game I'll you give u a buck no joke there lmao

[deep cloak]

Hi all,

I've the following issue with navmesh

did posted it on reddit with screens as well : https://www.reddit.com/r/Unity3D/comments/kiqy1t/navmeshissue_on_multiple_floors/

Any idea what i'm doing wrong?

[thorn hemlock]

Is there an updated guide on how to bake navmesh at runtime? Everything i've seen online seems very outdated.

I'm randomly generating a world and I want to bake it after it's completely generated.

There is no such thing as a "NavMeshSurface" in the latest unity version, and literally every single guide uses it.

[sonic rain]

@thorn hemlock AFAIK NavMeshSurface is the indeed the way to go for baking on runtime, but it doesn't come by default on Unity, you have to add it as an external package which can currently be downloaded on Github: https://github.com/Unity-Technologies/NavMeshComponents

[sleek oak]

There seems to be a major problem on generating localized navmesh in realtime using unity terrains. With localized I mean generating navmesh only around objects that need it. Every time I request navmesh from the terrain it rebuilds it completely from terrainData.ExtractNavMesh, which is incredibly slow. I couldn't find any way to cache that either, since the function above and everything related to that is hidden in the engine.

[slim garnet]

@sleek oak it sounds like you should file a bug report, in theory there is nothing inherently slow about ExtractNavMesh so it is probably miswritten on Unity’s side

[sleek oak]

@slim garnet I'm guessing it's baking the navmesh from the heightmap though? So it creates a mesh from heightmap every call. At least that's what I figured since it's so incredibly slow

[slim garnet]

well it should clearly be cached

terrains are already subdivided so it should be able to cache most of the terrain when e.g. obstacles are placed on top of a specific spot

[sleek oak]

Yeah, I'll just create a minimal project and submit a bug report

[hazy oar]

This makes me so happy! I definitely did not spend 10 days on it!

[warped grove]

@hazy oar Cool! Did you create your own algorithm for this? Or did you use some pre-built program like the ML program from VS to train your model?

[hazy oar]

I did it myself.

You said that at almost 2 in the morning?

The lazers check if they should spawn every fixedUpdate. A float increases based on the velocity is used to shoot lazers, as passing a threshold resets it and calculates where to place it.

Based on the activation delay and the time it stays active it determines the closest point it estimates they are going to be at in the middle of it (shown with a trail) to start the lazer. It then sends a raycast in a random direction with a random length from 0 to a float variable, plus .001, then minus the direction by .001 as to prevent it from going inside what it hits, because then doing the other ray casts hits the same point. It then does 2 more raycasts in opposite random directions, and creates a lazer object from the points.

Took 100,000 times longer to code than describe.

It is almost impossible to avoid it if it places it at the end of you jump, especially placed flat because you can't change the jump height, so you kinda have to fall into it.

Mabey I will change that.

[strong karma]

why arent my ai going to other meshes, that are apart of my nav mesh surface? do i need to use navmesh links?

[lone tundra]

@strong karma yes.

[strong karma]

alright thanks

[lone tundra]

You need to connect discontinuities one way or another

[strong karma]

alright

[severe crescent]

is there a way to create a simple ai that picks between rock, paper, and scissors using enum?

[warm python]

not ai

[severe crescent]

oh okay my bad

[warm python]

but random function can be used

[severe crescent]

random.range?

[warm python]

idk

ask some expert

i never did it in c# before

[sick bobcat]

Don't say that. I was accused by Fogsight of trolling JUST for saying that.

[warm python]

[sick bobcat]

I am NOT kidding.

[warm python]

then me ded

[sick bobcat]

Okay, back to topic.

[warm python]

yes

in java you can use math.random between 1 and 3 and then write an if else function to pic rock for 1 sizzors for 2 and paper for 3

[stone owl]

UnityEngine.Random.Range(0, 4) @severe crescent

I believe with range the last number is never picked so you'd make it 4 instead of 3.. Could be wrong

If int = 3 enum state = scissors, so on

[severe crescent]

Ahh okay thank you

[halcyon latch]

"SetDestination" can only be called on an active agent that has been placed on a NavMesh

Im getting this error. Does anyone know how i can fix it

nevermind i fixed the problem

[inner lotus]

my ai guy when it hits me it has a siezure and flys across the world help

this is setting

this is code

I GOT IT I REMOVED RIGID BODY

[little nymph]

Hey all! I'm working on a stealth game with melee combat. I'm trying to figure out how the AI should act. Perhaps similar to the Assassin Creed games but much simpler.

So, as a player, should you have the option to sneak past a guard, or takedown from behind. Should the enemies hear you if you attack their comrades? Should AI attack you in a group or 1 on 1?

Here's a small clip with rough gameplay:

https://www.youtube.com/watch?v=xf7a1d711xg&ab_channel=PhantomProgramming

[ember pawn]

@little nymph Impressive thus far. I would go the route of saying the AI should be able to detect you if within X radius of you taking down another but only if the take-down wasn't completely stealthy. (If possible?)

If you are detected, all AI in X radius should come for you. perhaps there are places you can hide to remove the 'wanted level' ?

Hello all question for everyone. I have an AI NashMesh set up on my Terrain. How can I 'slice' the mesh up a bit so that I can mark areas that should be avoided and so on?

it just makes one big navmesh over everything when baked

[sleek jacinth]

Any ideas on how to make a simple fps game ai for enemies

[quick plinth]

You should check out Coroutines

[unkempt ocean]

Hi can someone tell me how to make a ai wander

pig me pls

[molten sequoia]

@unkempt ocean Create a component which takes some sort of movement input, like target position or direction. Control those inputs based on where the AI desires to be.

Unity ships with few of these type of components, like Character Controller and Navmesh Agent.

[noble granite]

Can someone help me with how to make a Navmesh not just follow you around but follow you around until a certain range, then start shooting you

cuz rn he just follows u and gets up in your face

ping me pls

[unkempt ocean]

thx

[twilit dock]

Anyone know the best way to have an AI move along a straight path?

I want one of my AI to charge in a straight line until it hits something

So for example, the cyan dot would be a player. Red is the enemy that's just spotted them. The player would run along the cyan line out of the way of the enemy. The enemy would continue along the green path until they hit the wall.

Currently thinking about crawling along in 0.5m increments and sampling downwards, then caching it and repeating until I have the furthest point

[molten sequoia]

I would probably just crawl from previous position to next position

I imagine the AI moves over time

[full sierra]

Moving in a straight path is straight (no pun) forward, is the problem with knowing how to detect hitting something. This depends on your model, you might do this with physical colliders, or a populated tile graph

[viral pumice]

Is there a way to constrain a navmesh agent to move only in two directions?

[full sierra]

put an invisible collider on both sides of the 3rd axis

might be a programmatic way to do it but I have never used navmesh, I like A*

I wouldn't even use path finding for 2 dimensions

I mean 1 dimension

[coral oasis]

how do i increase the slowdown speed of nav mesh agent ?

i don't like the way it (kinda ) drifts to side when i want it to change direction.

i am trying to make a moba like mechanic

[twilit dock]

@full sierra I ended up doing the crawl. It's for a single enemy done once every so often so the raycasts cost very little overall

[full sierra]

What does the raycast do?

[molten sequoia]

@stray swallow Seems to be up for me https://arongranberg.com/astar/

[mortal umbra]

Hello there, I want to work with obstacle avoidance with A*. What could be the solution for this?

[tardy junco]

Introduce obstacles into your A* algorithm.🤷‍♂️

[mortal umbra]

Any source for that please.

[tardy junco]

Should be implemented in your A* api. What are you using?

[turbid kelp]

does anyone know any good tutorials for 3d enemy ai that shoots at the player and wanders, all the ones on youtube ive found are outdated

[arctic lava]

Does anyone know why I can't bake my nav mesh? The target of the enemy AI is the player prefab. You have to press the play button in the main menu scene to go start the game. I don't know what I'm doing wrong.

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


    public class zombieAI : MonoBehaviour
    {
        NavMeshAgent NMA;
        public Transform target;

         void Start()
        {
            NMA = GetComponent<NavMeshAgent>();
            
        }

        void Update()
        {
            NMA.SetDestination(target.position);
        }
    }

[tardy junco]

How are they outdated?

[tardy junco]

Did you actually bake it?

[arctic lava]

i did. I pressed the button, turned on show path polygons. It didn't work.

[tardy junco]

Did you set walkable objects as navigation static?

[arctic lava]

um....

im going to check.

[tardy junco]

Did you actually read about navmesh in the manual / watch a tutorial explaining how to use it?

[arctic lava]

a tutorial

[tardy junco]

Okay, so it worked for them and not for you?

[arctic lava]

yeah.

[tardy junco]

Then you probably missed something in that tutorial. Or it's bad. Can you link it?

[arctic lava]

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

[tardy junco]

Assuming you followed the tutorial properly, it should work.

[stark stirrup]

hello, I want to do Enemy AI for my FPS Prototype. I have a lot of requirements and each one of them doesn't sound too hard to hard to make on it's own, but I'm afraid I'll run into some spaghetti code once I join them all together

I really like making reusable components for micro behaviors and use them to compose my stuff and would like to follow that pattern here

any recommended tutorials or reads on how to structure enemy AI?

[tardy junco]

Finite state machine perhaps?

[stark stirrup]

any good reads on FSM?

[tardy junco]

Plenty. Here's a good video for example https://youtu.be/V75hgcsCGOM

[turbid kelp]

idk but the tutorials i tried none of the code worked properly

[tardy junco]

You sure the problem was with the tutorials? Were there compile errors in the tutorial code?

Instead of discarding the tutorials, you should stick with one and if there are problems during the tutorial, ask fro help in #💻┃code-beginner.

[alpine glacier]

Worked a bit on this one today, the agent's "Brain" is continually provided classified visual info based on what he can see from his perspective. He picks his next action based on very simple motivation/goal definitions (here he wants to "Kick" all the "Physics objects" for "Fun") and can refer back to earlier visual input (Memory) if he doesn't have anything to do or can't see anything new to do. Will be fun to see how it turns out when interacting with other agents.

I wasn't expecting that at the end he decided to run up to my first person camera component because apparently he classified it as a "Physics object" since it had a rigidbody and really wanted to kick it 😄

[stark stirrup]

thanks! you mentioned the brain which I think I will need to centralize the information for the various scripts

you also mentioned the memory which I was thinking about (like last known location)

I'll also check Jason's tutorial, I really like the way he codes

[wary charm]

Ai in my game are pretty lame, I am new to video game ai, and I want to expand on my game ai with a behavior tree system. As a solo developer and full-time student, is it worthwhile to make my own implementation (I am not using navmesh agents) as opposed to an asset store asset?

[tardy junco]

Unless it's for educational purposes, probably no. There are some decent free implementations on the asset store. If you want some challenge you could find a generic open source implementation for c# and port it into your project.

[wary charm]

Ok, thanks!

[alpine glacier]

Indeed! I read a few books on neuroscience and psychology past few years and I'm trying to model and conceptualize my understanding of it in the implementation. Incidentally, the executive central decision component of the "Brain" is called "PrefrontalCortex", while the visual interpretation happens in the "VisualCortex" component 😄

I think a very qualitative entirely heuristic based AI would be kind of uninteresting to explore so I'm incorporating ideas from Steel, P., & König, C. J. (1992). Temporal motivation theory to quantify my agent's behavior. Goal is to be able to train the agent with reinforcement learning to maximize his motivation to stay alive and clear the area of any perceived threats.

You can see there the agent's stream of conciousness, the top log is the current visual input (filtering out the useless things), second is the kind of stream of consciousness and bottom is somewhat the same just a log of the agent's memory. He can do a bit more than just kick boxes but that's the scene I have at the moment 😋

[alpine glacier]

I'm kind of meh when it comes to those kind of state machines, on top of that I think rolling your own generic state machine framework just for the sake of it is also probably going to get very clunky too.

When it comes to state machine oriented tasks. I've encountered a few at work/hobby and I always see much more value writing a more language oriented interpreter approach which allows for way more flexibility/complicated behavior and eventually cleaner and less code overhead. Technically, that's still going to be a state machine I guess but even then when you conceptualize those using turing machines they're way easier to work with.

[full sierra]

It's 2021. Mankind invents AI. It likes to kick things.

[alpine glacier]

The AI realized the secret to life, the universe and the meaning of existence, its resolute decision was that every object needed to be kicked at least once.

[turbid nest]

is there any reason a navmesh agent wouldnt respond to movement events? for example chaning the transform.rotation to make him rotate in a direction while standing in place

or while using an attack animation being able to turn

or get a little boost of forward velocity to make the strike look more impactful

the same scripts i use for my players animation events are not working for the AI when it comes to anything movement related (turning, pushing, launching) it is just a basic NavMeshAgent

[turbid nest]

it seems like applying Velocity to a navmeshagent is useless

[tardy junco]

As long as the agent updates position and rotation, you won't be able to move it properly in other ways.

[stark stirrup]

I think I read somewhere you need to turnoff navmeshagent when you want to do that

[turbid nest]

yeah i disabled nav mesh but i still cant push him forwards even with the navmesh disabled

@tardy junco

[tardy junco]

How are you pushing him?

[turbid nest]

velocity.x = (character.transform.forward * _pow).x;
velocity.z = (character.transform.forward * _pow).z;

nevermind, i knew it was something stupid lol, i wasnt calling the move function on the character controller for the AI only the player

thats why my animation events werent working

[valid turtle]

Hi, I used polarith AI from the asset store but couldnt make it work. This was a month ago. What do you guys use for AI? Preferably free since im not making anything serious with it just now

There were a couple free AI tools that seemed to be OK. Here are the free ones https://assetstore.unity.com/tools/ai?category=tools%2Fai&free=true&orderBy=8

[stark stirrup]

Hello, first time using navMesh here. I have a simple follow script where the enemy will follow my character. The issue here is that if the character is jumping the enemy won't follow because the destination is not on the navmesh (it's above it)

The solution I see is casting a raycast down from the player to know the ground position of the player

My question is if it's possible to do a raycast that has that only hits navmesh?

I've seen the AI RayCast but it doesn't seem to do what I want

Or if there is a simpler solution 😉

[tardy junco]

There's a method that samples the closest navmesh position.

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

[stark stirrup]

Yes I'm already using it but it's not working as intended because of height differences.

Here's my code

How do I paste code? 😛

[tardy junco]

use back ticks `

```

or upload it to hatebin

[stark stirrup]

            do 
            {
                NavMeshHit hit;

                Vector3 interationPoint = player.transform.position + interationVector * currentDistance;

                hasDestination = NavMesh.SamplePosition(interationPoint, out hit, maxDestinationDistance, NavMesh.AllAreas);
                if (hasDestination)
                {
                    _agent.SetDestination(hit.position);
                }
                else
                {
                    currentDistance += iterationDistance;
                }
            }

I'm using this because some parts of the level aren't walkable but I want the enemy to move to the closest position that is walkable

[tardy junco]

you could use the filter argument to filter out areas.

Why is interationPoint so convoluted?

What is it supposed to be?

[stark stirrup]

oh I forgot the while

but basically what I'm doing is that I start to look to go to the player position. If I can't, I move closer to the enemy and try again

iterationPoint in the beginning of the loop is = to player position, then it starts moving towards the enemy

[tardy junco]

why not just increase the max distance?

[stark stirrup]

I was afraid it might move the enemy to the other side of walls. I have some thin walls on my level that the player can walk above it (but enemies can't)

[tardy junco]

then assign them to different areas.

Anyways, put some debug logs and drawrays in that loop to see what's going on.

[safe fiber]

Dumb question, but how do i use AStar for flying AI?

[alpine glacier]

Hey! I'm making a zombie apocalypse style game and I've just added navmeshagent onto my zombies but now they walk inside of each other. I looked online where it said to add navmesh obstacles onto them but now the zombies move at super-speed and just sort of sorry the player on their heads? Any help with be great. Thanks 🙂

[gleaming kernel]

probably going to have to use a 3d grid

[safe fiber]

I couldn't find an option for a 3d grid though, unless I'm just missing something

[molten sequoia]

@safe fiber You should specify if you are using an existing solution (and which one) since AStar is the name of a general path finding algorithm.

[safe fiber]

I meant as in, just imported A* into project and atleast on the graph part where you'd usually set up nodes and whatnot, i did not find any "aerial"-like settings

[molten sequoia]

https://en.wikipedia.org/wiki/A*_search_algorithm This is A*
https://assetstore.unity.com/packages/tools/ai/a-pathfinding-project-pro-87744 This is A* Pathfinding Project asset.

A* Pathfinding Project does not support 3D graphs afaik

[safe fiber]

Ah apologies, i didn't understand your previous question correctly. Yeah, i meant the unity asset, not the algorithm.
If it's as you say, i guess I'll have to try find some other means for aerial pathfinding since no 3D graphs

[molten sequoia]

A* Project's forum is a good resource https://forum.arongranberg.com/t/3d-grid-or-3d-point-graph/2554/2

[visual osprey]

aaron granberg pathfinding can do aerial pathfinding

using point graph

[spice chasm]

I'm working on a tower defense game where the path the enemies takes is made longer by placing towers. Using navmesh to path the enemies to the end. Problem is that I can't seem to figure out how to check if placing a new tower would completely block off the destination. Thought I could place the tower down, check if the pathStatus was pending or invalid, and remove the tower if it was. Seems that the new paths(and mesh carving I think) are calculated for the next frame, which won't do here. Am I missing something simple? Hoping that is the case.

[spice chasm]

I figured it out. I had to use NavMeshBuilder.BuildNavMeshData and NavMesh.AddNavMeshData and some other things to get required arguments.

[alpine glacier]

Just sharing for fun. Been having fun with this goal/motivation based AI. Uses a definition system that makes it very easy to adjust and extend behavior. The soldier's goals are to kill all zombies as primary and to reach a destination as secondary. They use temporal motivation theory to determine the next task to execute.

I think I forgot I was going to make a game at some point and just went overboard...

[alpine glacier]

Hello Guys
Can you plz tell me what should I do when the player will die in my game. Which type of GUI should I create when the player will die . Plz help me

[full sierra]

Probably best to be as clear as possible and put the words YOU DIED on the center of the screen

[flat shard]

I am looking to create an enemy that adapts to the player's behavior and tries to predict what the player is going to do. What resources and papers can I look to for creating such a machine-learning model?

[full sierra]

There is a dedicated channel to machine-learning and this tutorial is great https://www.youtube.com/watch?v=zPFU30tbyKs&t=2349s

Just keep in mind it is nearly impossible to train in real time right now

Machines do not learn like humans do. Humans link learning new processes to related already trained NNs so that we can learn a new thing in just a few attempts. A machine is still just throwing random numbers around and working from scratch on everything, never linking to previous NNs. So it must run many many iterations and be done offline.

[slow viper]

https://media.discordapp.net/attachments/763495187787677697/801197635821371412/unknown.png

how do I modify the width and depth via a script

this is the a star pathfinding system

[full sierra]

Pathfinder is a script right? Are you using someone elses script? Will have to see if they made a public function to set the variables

Code Monkey's Dots A* system is far superior, the problem with a function based A* system is you will have to run it on it's own threads and wait for it's return, it is just too slow and if you want to get many paths in one run of Update your going to get stuttering

[marsh scarab]

I already have a state machine setup, but I have questions about the actual design of the AI. I'm making a boss enemy, and I can't figure out when to make it do what. Like, how often should it attack and retreat? If I have multiple ranged attacks, how should it interchange between each other? Should the timing for the attacks be fixed, or based on stuff like the player's position? How should it know which attack to do? Thanks.

[full sierra]

@marsh scarab This is all what will make your AI unique. These are called Decisions. You have Decisions and Actions and it can be represented by a tree. You can even move your state machine into a visual decision tree and it makes it so much easier to troubleshoot when you create a bug in the logic. One branch of the tree might be when a player is distant, while one branch is when the player is close, etc

If your not sure what your enemy should do, put yourself in his shoes and ask yourself, what would I do?

[autumn river]

Help i dont understand why but sometime this doesnt work when i reset my variable CurrentWp to 1 my agent dont care and go to the if

      if (myAgent.remainingDistance <= 0)
      {
        myAgent.SetDestination(Waypoints[CurrentWP].position);
        CurrentWP = CurrentWP + 1;
      }
      if(CurrentWP >= MaxWp)
      {
        CurrentWP = 1;
      } 

[glad cape]

Not sure how useful it'll be , but here's a git repo showcasing procedural navigation

https://github.com/Mylab6/BasicUnityProceduralTileGen

[autumn river]

its ok i found a solution but my agent is dumb he fail a lot trying reach the Wp

[potent urchin]

Does anyone know why an agent would not take any actions?

[wheat nacelle]

idk

[alpine glacier]

That's given the assumption that you want completely unsupervised training on difficult NN structures. Which would probably be hard to translate into fun game AI.

@flat shard you could theoretically model your AI's future prediction with Bayesian networks, they can have a relatively limited number of parameters to adjust and might be something interesting to take a look at.

[alpine glacier]

Are you comparing the distance between the three dimensional axis or on a two dimensional plane? If you have three then you need to remember that the Y axis can interfere in how close your two agents think they are.

[stark stirrup]

I have a question. How are FPS enemies usually done in terms of AI in AAA titles?

Is it FSM? ML?

If you need an example: Doom Eternal or Hitman (since they have different styles)

[real sonnet]

I think Behaviour Trees are still something in indie games, but I guess AAA have moved to HTN, and the like

It's sometimes hard to qualify, a big studio will work on its own implementation and mixing BT, HTN, Utility scorers, some FSM, and so on, depending on how high or low level the decision making is

Recycling said AIs for next games and adding little variations, until someone finds a new interesting pattern and share/write about it

Dunno for the 2 specific games you mentioned though. I'd be curious to know for Doom eternal

No AAA studio uses ML yet (unless Blizzard with AlphaStar AFAIK)

It's a long story but it's not suited for studio workflows

The key point being, IMO, that a FPS bot can be VERY dumb, because more often than not it's expected to die a few seconds after you see it

Hope my answer sticks to the topic

[wispy minnow]

Can someone help me with my coroutine, Im trying to make a charging type enemy in a 2d game (the move faster every 5 seconds for 5 seconds)

https://pastebin.com/sQs68y33

[finite rock]

can u be specific??

[wispy minnow]

every 5 seconds the enemy charges at the player at a faster speed than their regular movement speed for 5 seconds

@finite rock

[finite rock]

float Speed = (1 + Mathf.Round((CurrentTime - StartTime) / 5.0f)) * SpeedMultiplier;

that's all you need

set the StartTime when the charge should star and keep tracking the current time every update/frame or maybe a timer then get the time difference and divide it with 5, you will get how many times the enemy has to move faster, then just multiply it with the speed

nothing fancy needed for this logic just simple math

https://pastebin.com/BN1xuKL0

@wispy minnow

[finite rock]

Serius question. Where do you pull these terms from like HTN etc? I have never heard of it 🤔 Want to build my own source of information stream so i can keep up to date. This is somethin new thought no?

[wispy minnow]

thanks!

[real sonnet]

Years of keeping an eye on the topic. Reading books, watching videos, looking assets on the store, and so on

Hierarchical Tasks Networks are pretty new and still a thing

Must read: http://www.gameaipro.com/

Must watch & subscribe: the YT channel of Thommy Thompson, "AI and Games" https://www.youtube.com/user/tthompso

[finite rock]

it's a $60 book :C

Knowledge is free imho, and the writer of the book also learned it from somewhere i guess, he didn't invent it, or did he?

[real sonnet]

ALL the talks of Dave Mark, for UtilityAI: http://intrinsicalgorithm.com/ search his GDC Vault videos

It's multiple authors from the industries, and artciles you can find probably online for free

I just find it nice to have it all compiled in a paperback book

@finite rock it's all free on PDF format if you like reading online

(gameaipro)

[finite rock]

added to my reading list

thank you

[real sonnet]

Then there's multiple papers if you can find them

[finite rock]

its a good starting point thank you

i'll dig deeper if something catches my interest here

[real sonnet]

But just the 4 or 5 references i mentionned are great ressources

yeah, enough to dig for a start

[vale egret]

Hey guys, I originally asked this in the tech channel but I think it's more appropriate here:

Anyone ever done pathfinding where you want the pathing units to stop at certain nodes along the way if they are close enough?
Like you want the units to go from a to b, but you'd also like them to stop at c if it is not prohibitively out of the way
I imagine it'd be A* with some tweaks to the heuristic but I'm wondering if there are established approaches before I take a crack at it

The use case is a unit that is kind of like one of the traders in civ games. His goal is to take things from A to B but if he can stop at an outpost along the way he should do so because they act like checkpoints in my game

I think I'm just gonna modify the heuristic to be more costly the more tile the units travels without passing through an outpost

[wooden stag]

I love navmesh

[tardy junco]

Why does it need to be implemented in the pathfinding system? It should be part of your particular ai behavior. If you were using a navmesh for example, you could calculate a path first, then see if any of the points along the path are somewhere where you want it to stop, then save the original target and set destination to the intermediate target.

[vale egret]

I have my own pathfinding solution. It would be great if I could do this without touching the pathfinding itself cause I don't want to complicate the system too much but I'm not sure how

If it's along the path then that's good but I wanted to program it so that the path can deviate a little to traverse an outpost

So long as the deviation isn't too great

I could do this by making my heuristic a delegate that I can pass as an argument but I'd also need to make some add some way of allowing the path to pass through these outposts (which it usually can't) so I'm open to any ideas that don't require me to reimplement the pathfinding

I'm probably doing a pretty poor job at explaining this so I get it if it's a weird question. That being said, if anyone knows what I'm going on about I'm up for talking about how to architect this solution

I feel like I should just add delegates for the calculating the cost of every move and the heuristic at every tile. Then I could control which tiles are accessible using the former.

[paper mango]

Hello, I'm working with "mlagents" with unity at the moment, but I ran into a problem.

Problem: When I call the function "AddReward", the Cumulative Reward does not go higher then the value i put in the "()"

for example:

AddReward(0.1f); //--- I add a reward
Debug.Log(GetCumulativeReward()); //--- Print Cumulative Reward

When i call this function it just jumps to 0.1 and stays there, even if I call it again, and the same if i put "-0.1" in (it jumps to -0.1 and stays there)

What i have treid:

1. Made sure I called "AddReward" and not "SetReward"
2. Used "SetReward" and just put "GetCumulativeReward() + 1" in the "()"

Both did not work

any fixes?

[sharp pollen]

Im using a navmesh agent on my enemy, but it faces the wrong way when chasing my character, why is this?
I have tried rotating the prefab, but this hasnt seemed to fix the issue

[halcyon latch]

hey so when my agent chases me theyre partly in the wall, does anyone know how to fix this. I made the walls static and the agent doesn't seem to be going through them, only touching them

[lapis jetty]

good day,
is there some way, to get the path between start and destination AND to get all area costs OR all areas that are passed?
because i only found stuff on google that returns true or false, but i need the all the costs or all the areas that are passed.
are there some functions for that?

edit:
i have a small "dungeon" with 3~ rooms which are connected with corridors.
the corridors and rooms have different Navigation Areas.
i want to tell my agents, they are only allowed to move from 1 room to the other room and not more (turn based game)
what is the best way?
maybe i just go in the wrong direction with my first idea

oh and would be great if u mark me, so i get notified when someone answers 😄

[full sierra]

@lapis jetty Trying to figure this out right now. I am trying to convert the pathfinding to a floodfill, but I don't know what to set my initial hCost to when setting up my path node array. Normally the hCost is based on an end position. I just want to keep going until all tiles have been removed from the open set

I got it!

All you have to do is take the A* code and set the initial hCost to int.MaxValue just like you set your initial gCost. Then make sure not to break your open list loop if the current node = the end node. In fact after taking away that reference, there is no need for an end node reference at all

Here is my floodfill code, which was just a modification of the A* code https://hatebin.com/azrzqgmqdb

There is still a problem, when I tested starting at 5,5 it worked, but if I start further out on the grid, it does not fill the entire grid

[twilit saddle]

If you just want to do flood fill, isn't a breadth-first-search sufficient? A* seems overkill. (sorry haven't peeked at your code to see what modification you did)

[full sierra]

Well I have this cool A* system already built to check for things like walking through touching corners, etc. So it is more of a pathfill I guess

Also it is the best performing A* routine I have ever used, so using at as a flood fill is nice to keep any frame drops

It is working as it is because I just figure if a tile is outside of the result then it is too far to walk to anyways, but that would not work for everyone.

[south mango]

I need help with a reward function for a car AI using mlagents

It can move forward and backward and it has a brake and it can turn left and right and I have checkpoints around the map but the ai just holds brake the whole time

Rarely moving forward

I punish it for braking and moving backward too much but still itching

[real sonnet]

And I assume you reward it enough when distance increase ? you might have better luck sharing code in #archived-machine-learning

[worldly trench]

I started learning GOAP as my first AI system to implement

But it seems dated to 2012

Does anyone have a recommendation of a 2018-2020 AI archicture to learn and build for a first AI system to implement?

[tardy junco]

Start from state machines or behaviour trees. 😄

[worldly trench]

I got state machines down pat

But id rather not have to rewrite a new agent every time I want to make a different type of agent

Im looking for an architecture that is more scalable and reusable

[tardy junco]

Didn't you say it's your "first AI system to implement"?

Let me tell you a story. around 2 years ago when I was starting with gamedev I had huge ambitions. I though I'll create the best rpg/strategy game of my dreams that will of course implement the most realistic AI. So I went with GOAP. There's an open source implementation called ReGOAP on the asset store. At first it worked great, but at some points I started encountering bugs that I couldn't explain or debug, because I couldn't understand the source code. It was just too advanced for me. I ended up ditching that implementation and tried another architecture: HTN(Hierarchical task network). But it ended up the same way as my GOAP attempts. Since then a lot of stuff has happened and I had to abandon that project, but recently I've been working on a real project that I actually getting paid for. And when implementing the ai I understood one thing: if you can get it working by simpler means - do it.. I ended up using a simple state machine implementation and it works wonders. It's true that I don't have a lot agents types, but if I needed I could make the solution more generic and make it easier to add different agent types.

Point is: don't chase for "super advanced new AI technique" that is out a month ago. It is not solid, you'll have to implement it yourself or pay money to other people, you'll have to learn how it works. Consider if your project really needs a complicated system that you'll have trouble debugging, or can it be implemented with a simpler solution. More importantly, it's not just the tools that you use, but HOW you use them. A good state machine implementation and proper use case could be better than a very advanced GOAP, HTN or any other advanced AI architecture that is not use properly.

P.s: GOAP is considered pretty fresh. It's not like we get new AI architecture every year and it doesn't mean they come to replace older solutions.

[worldly trench]

Ive been following along with this tutorial

https://learn.unity.com/project/goal-driven-behaviour?uv=2019.3&courseId=5dd851beedbc2a1bf7b72bed

It boiled down GOAP into really simple mechanics

Its a little frustrating to re-bake a new Planner when divergent levels or use cases come up

But its not exactly hard to understand or anything

P.s. Is a state machine really an AI? Its just been the thing I use to control states for any type of thing im making

P.p.s if you cant do something with an Enter, Update, and Exit, do you really understand what you are doing :P

[tardy junco]

AI is what makes your agents move in your game and do things. If you just have sequence of if statements in update that does it, it's already ai.

[tardy junco]

Oh, the concept itself is easy, I completely agree. But when you get down to details or try to use someone else's implementation that uses interfaces/inheritance and multithreading heavily, you can get lost easily. Especially if you've just started coding.

In my case it was probably a racing condition due to the implementation using threads, but it was so long ago, I can't remember now.

[tardy junco]

Then you make it generic enough, to fit all cases. But I agree, that there are situations, when a state machine might not be enough. GOAP is a good architecture, if you get it to work properly.

[worldly trench]

Im only really considered writing my own, I like having control over my pet implementation

Keep it simple, only have the essentials I need, use my standards and conventions, make it something I can really wield

GOAT is great so far, but its not scalable up to hundreds of divergant agents

But at least the re-use factor is like 6 out of 10

Compared to a state machine where the re-use factor is more like a 3.5/10

[tardy junco]

I'd say it all depends on your implementation. If you find that you can't reuse something, it's probably too specific and you can make it more generic.

[tardy junco]

I remember following that course myself. The reason why I decided to use an asset after all was that this course and many other learning resources didn't cover backwards planning which, if I remember correctly, is way faster.

[worldly trench]

I think the critical failure here is a GOAT planner, as you have to rework it often, I was hoping to learn something that could add a bit of architecture mix to make that part more reusable

[tardy junco]

Why would you need to rework the planner? It simply finds a path to the goal, where path nodes are actions.

It's as generic as it can get.

What you'd need to reimplement with GOAP are actions and goals.

[worldly trench]

Maybe im confusing two aspects of goat into one thing

Im talking about the process of creating graphs conceptually for your use cases that you then turn into actions and goals

Id think a decision tree would be more flexible then this part

[tardy junco]

Donno. Is seems pretty simple to me. Let's say you have a goal "satisfy hunger" that requires "food eaten". Then you have an action "eat food", that requires "food" and satisfies "food eaten", and actions like "Take food from storage", "Hunt for food", "gather food", "beg for food". All of them would give you "food". You add the desired actions and goals to your agent and it will plan the actions to get to the goal, for example: "gather food" - "eat food".

You can make as many different actions and goals as you want and put them in different configurations on your agents. That's as modular as it can get.

[worldly trench]

It was like that at the start

But now I have over 300 actions and goals, and tweaking and messaging them to get some realistic simulations is becomming unweildy

[tardy junco]

Then perhaps your actions or goals are too specific.

[worldly trench]

I think my game has too many stages

[tardy junco]

I don't know what actions and goals you have, but for example if you have goals like gatherFoodTypeA and gatherFoodTypeB, the amount of actions will explode very quickly.

[worldly trench]

Maybe I could use some sort of reinforcement learning to weight the acitons better

Just thinking out loud

Re-jiggering my design documents to figure out weighting better sure is a pain

[tardy junco]

Machine learning is a completely separate field. Might be more complicated to get into it, than fixing your GOAP implementation.

[worldly trench]

If I used a decision tree, then weighting actions is no longer such a barrior

But then its not really goap any more

[tardy junco]

You mean behavior trees? Well, as I said, if a simpler solution works, go for it.

GOAP is really a dynamically created behavior tree.

[worldly trench]

You know

I just realized when checking all the states

This might be a great spot for a lambda operation

Well

I will just figure out wieghting iteratively for now

Im back in code now lol

[alpine glacier]

is it possible to tell a navmeshagent it can't use an area no matter what, even if it can't get to its destination otherwise?

[sonic rain]

@tardy junco totally agree with you. I was fortunate to do my first AI with FSM, because even though it is harder to maintain due to spaghetti, it is still something that you have more control, which is of course intentional -- in FSM you specify exactly every possible decision/action that can be done, and in what order, as GOAP creates it in runtime, making it way more obscure to understand/debug it at some times

PS: I'm not complaining about neither FSM or GOAP, they just have their differences and that's it

By the way, I'm not sure if you already tackled this, but I'd suggest you to use HFSM (Hierarchical Finite State Machine), which is super useful on reducing the amount of spaghetti graph, because the whole package of "Fight Sombody" states can be encapsulated inside another state, so it cleans a lot the graph

Also something else that you might be interested in (or already knows about) is Utility Theory, which is a mathematical AI model where decisions are done based in math curves. It is super interesting, I suggest you to look at some GDC talks on the matter, if you didn't yet

[tardy junco]

Haven't heard about HFSM. I'll have a look at them, thanks! I remember utility theory from political studies back in university, and also remember I read somewhere about it in the context of AI, but I don't remember the contents. Gonna have a look at it too.

[real sonnet]

@tardy junco For UtilityAI, don't miss the work of Dave Mark (http://www.intrinsicalgorithm.com/) and Kevin Dill (AI engineer for the military at Lockheed Martin)
That's the GDC vault talks @sonic rain is talking about. Dave M is head at organising most of these AI talks

[sonic rain]

Yeah Mr. Dave is the big reference on Utility AI 😄

[real sonnet]

@worldly trench If i read well and I'm not off-topic of your concerns, don't be afraid to mix multiple patterns where it helps you 😉

[worldly trench]

I made the mistake of diving right into GOAP

But instead I should have focused on how wieldable the modules of the AI architecture are

A hierarchical task network planner is no good if tuning the weighting takes up more time then implementing a game

[real sonnet]

Design a game with balance is definitely not easy

[worldly trench]

And I am starting to realize

[real sonnet]

It's interesting and rewarding though ?

[worldly trench]

That when an AI clearly demonstrates a personality in its decision, its more fun then an AI that demonstrates "intelligence"

[real sonnet]

True that

[worldly trench]

Yeah, im learning a lot about architecture

[real sonnet]

and you can expose the settings your AI use to the player so he understand them, helps making your AI more believable in the decisions they make. Some games do that very well

[worldly trench]

Thats a interesting idea

[real sonnet]

https://www.google.com/imgres?imgurl=https%3A%2F%2Fforumcontent.paradoxplaza.com%2Fpublic%2F542698%2Fideology%25202.jpg&imgrefurl=https%3A%2F%2Fforum.paradoxplaza.com%2Fforum%2Fthreads%2Fis-impose-ideology-supposed-to-change-ai-personality.1358271%2F&tbnid=fh2pQWM8k40m7M&vet=12ahUKEwijjLjCjbruAhUbgHMKHU5MCg4QMygFegUIARCyAQ..i&docid=tyIaya1962_q-M&w=993&h=859&q=crusader kings AI personality&ved=2ahUKEwijjLjCjbruAhUbgHMKHU5MCg4QMygFegUIARCyAQ

[worldly trench]

I should probably replace my planner weights from world and local situations to something more like personality traits

Hmm

[worldly trench]

Thats very nice

[real sonnet]

Crusader kings III exposes AI personality very well too

[real sonnet]

🤔 not replace. You still want your AI to *sense the environment

personality adds a layer and mutates the perception

to add even more variation and unpredictable behaviour to your agents

Any AI framework in a nutshell: SENSE > THINK > ACT

[worldly trench]

Im thinking maybe a Spatial UI that when highlighting over a enemy you see things in the weighting

Could be nice

[worldly trench]

Very nice

[real sonnet]

Yep, if it helps the player understand how your AI behaves, so he can learn and understand the systems of your game, yeah go ahead 👍

[worldly trench]

It will make it easy to debug also

I could really use some data visualization of my states and decisions

[real sonnet]

that takes part in the fun of mastering the game, and eventually overcoming it

yeah debugging too, but remember you can also make editor tools for that 🙂

[rancid copper]

How I can use a class from an embedded package?

I have this class from Digital Human:

and when I try to use it like this:

using System.Collections.Generic;
using UnityEngine;
using Unity.DemoTeam.DigitalHuman;

public class NewBehaviourScript : MonoBehaviour
{
    // Start is called before the first frame update
    void Start()
    {
        GameObject snappersHead = GameObject.Find("SNAPPERS:Head");

        if (snappersHead != null) 
        {
            SkinAttachmentTarget sat = snappersHead.GetComponent(typeof(SkinAttachmentTarget));
            Debug.Log(sat);```

It gives the error: The type or namespace name 'SkinAttachmentTarget' could not be found (are you missing a using directive or an assembly reference?)

damn, wrong category..

[paper star]

I have a really basic enemy AI where they just move towards the player. How can I stop the enemies from clumping up? I am a beginner programmer. Can I maybe force them to stay a min distance of 50 pixels apart from each other?

[tardy junco]

I'm not sure what an embedded package is. Is it the same as a separate assembly? Is it in your project? Did you import it via unity? Does it have an assembly definition file?

[tardy junco]

Is it 2d? How are you moving them?

[paper star]

I am making a 2d game and setting the positions of each of the sprites using Vector2.MoveToward(current, target, speed).

I noticed that when dodging the enemies, they tend to clump into a single super powerful enemy. I read an answer on Stack Overflow but didn't understand it at all so I came here. Can I make the enemies stay a minimum distance apart?

[tardy junco]

You'll need some kind of algorithm that will move enemies apart. Like flocking algorithms.

For that you'll need to make sure that your enemies are aware of each other. Are the enemies controlled each from its own script or do you have some ai manager that updates them?

[paper star]

Each are controlled by their own script.

I just followed BlackThornProd's Follow AI tutorial. How would I make all of them controlled by one script?

[tardy junco]

You could have a reference to all of them in one script and call some functions on all of them in update or something. But you don't have to do that.

If it's separate scripts, you could use a static list of enemies in the script itself, for example. Or you could overlap circle in some radius around the enemy. Or, if they have a collider, you could use collisions(but that wouldn't work create with Vector move towards). Anyways, you need a way for your agents to know what's around them. When that's done just write some simple avoidance algorithm.

[tulip ravine]

anyone good with NavMesh could spare me 15 minutes?

[tardy junco]

Just ask your question and someone might answer.

[tulip ravine]

I need help with all of navmesh from setting up to bake to setting agents and stuff

so uh....

[tardy junco]

What's the problem?

[tulip ravine]

i dont know anything. just managed to setup the mesh, need help with cost and stuff

[tulip ravine]

How do I apply Costs and stuff

[tardy junco]

Costs? Never used them. I think it's when you use additional areas in your navmesh.

I'm sure some tutorials cover that.

[tulip ravine]

thing is i cant find any explaining them well

[tardy junco]

I can see at least a few with a quick google.

There's also a manual page specifically for that:
https://docs.unity3d.com/Manual/nav-AreasAndCosts.html

[worldly trench]

Im looking for notes/a review of this sweet AI architecture I have in mind

Would anyone be up? Its just two pages

[frozen token]

Hi. Maybe, somebody, knows, what algorithm is used by Unity for NavMesh generation?
Is it something like Mononen model or Multiple-Layered NavMesh?

[tardy junco]

For generation..?🤔

[worldly trench]

As far as I am aware, its a black box to everyone

[devout spear]

https://github.com/Unity-Technologies/NavMeshComponents

[alpine glacier]

is there a way to make the peds walk on an offset of the path they're on?

[normal mountain]

ok so i've finally come back to ask about a certain nav mesh agent issue that i have, when enemies spot the player and they're on a different platform where they can't jump, they seem to leave their designated area of navigation, as in they float in the air... i have had this issue for a bit now

[grim mirage]

Curious what types of databases you all use for development?

[rustic zealot]

man unity really needs to use colliders instead of meshes for navmesh baking

it makes absolutely 0 sense

like what if i have weird meshes that causes weird navmeshes? do i have to make all the meshrenders visible every single time i bake a navmesh?

it's a very painful workflow

[normal mountain]

so i keep getting this error on unity with my navmeshagent that says this:

"SetDestination" can only be called on an active agent that has been placed on a NavMesh.
UnityEngine.AI.NavMeshAgent:set_destination(Vector3)
i really cant seem to find a solution here
i've even tried putting the enemy slightly above the navigation area as someone stated but nothin worked

[tardy junco]

Assuming you've bakes your navmesh and the agent is really positioned on it and has that are in walkable mask, is the agent enabled?

Can you take screenshots of the agent so that it's gizmos are visible relative to the navmesh?

[normal mountain]

the agent is in fact enabled and you are correct positioning is fine, the navmesh is baked, everything is set

here are the two screenshots

cause either he floats in midair or sinks to the other navigation area that he should NOT be allowed to access otherwise he starts moving weirdly

usually this happens after he spots the player and there are no walls to stop him

so basically if the player leaves his area

anyone that is not on a floating platform do not have this problem

[tardy junco]

First, I don't see the navmesh properly I'd like to see it from the second screenshots perspective.
Second, you agent height is 0.01??? Why?? I'm sure that could be the source of the issue. It doesn't seem like the model is 0.01 units to me.

I mean in the bake settings.

[normal mountain]

well i changed the agent height back to 2 already

and yet it still doesnt work

and here

[tardy junco]

And in your agent it's set to 6?? These values should correapond imho.

[normal mountain]

so change radius to 0.2 and height to 2 like in the baking area?

[tardy junco]

Yes.

Then you can level up your debugging by pausing the game right after the error and checking the navmesh agent at runtime.

I think you can use a try catch on that Set destination line and Debug.Break if in the catch.

[normal mountain]

the catch... explain that part?

[tardy junco]

Try-catch block allows you to catch errors and respond to them in the code. Look it up. It's a feature you should be using from time to time as a coder.

[normal mountain]

kk

i will look at that after i try what i just did for the matching up the radius and height to the baked agent

[tardy junco]

Google c# try-catch.

[normal mountain]

okie

and also it does kind of look like he isnt going off the platform but lemme disable a certain moving platform animation next to him to make sure he doesnt

ok so i realized that as soon as the enemy goes off the platform when spotting me, this appears:

"SetDestination" can only be called on an active agent that has been placed on a NavMesh.
UnityEngine.AI.NavMeshAgent:SetDestination(Vector3)
EnemyAI:InFront() (at Assets/EnemyAI.cs:106)
EnemyAI:Update() (at Assets/EnemyAI.cs:50)

so clearly he is for some reason still disobeying the navmesh with his code

[tardy junco]

Does it have a dynamic rb?

[normal mountain]

i was told to remove the rigidbody from someone else as it would not work well with the navmesh

[tardy junco]

At the very least set it to kinematic.

[normal mountain]

i kinda forgot what that does since i have not used it in a while

[tardy junco]

Basically. disables physics calculations for that rb. You shouldn't let both physics and navigation control your object at the same time.

unless you're doing some custom blending.

[normal mountain]

ah

so it gets rid of gravity

[tardy junco]

Yeah gravity too and any other forces.

[normal mountain]

kk i shall try that

he still floats off of the platform

doesnt seem like it made a difference

@tardy junco i mean i could show you the EnemyAI script if you wantr

[tardy junco]

Did you pause at runtime as I suggested?

[normal mountain]

yeah

[tardy junco]

Sure upload it to hatebin or something.

[normal mountain]

ok

[normal mountain]

and like said, this showed up as soon as they left the platform

https://hatebin.com/zbnhfsymhc

[tardy junco]

Not completely sure, but agent.autoBraking = false; might be a problem? Never had any need to use it.

The fact that your agent goes out of mesh bounds probably means that it's not breaking.

[normal mountain]

i guess so

[tardy junco]

This could also be a problem: this.transform.Translate(0, 0, Time.deltaTime * speed2);

If you're using an agent you should be very careful about applying any kind of movement manually.

[normal mountain]

well actually speed2 is the one that is meant to stop the enemy

speed and speed1 are the moving ones

[tardy junco]

Doesn't matter. Stopping is also applying some movement. You should avoid any and all kind of manual movement to the agent, unless you're 200% understand how it will work out.

[tardy junco]

Another thing is to get rid of root motion. It can also apply unwanted movement to your agent via an animation.

[tardy junco]

make sure to enable gizmos when you debug. We can't see the actual position of the agent like that.

[normal mountain]

getting rid of speed2 just makes them dart in random directions... ---

[tardy junco]

So here's the summary:

• don't disable agent's autobraking.
• disable any source of movement aside from the agent itself. This includes:
• setting rb to kinematic.
• making sure your animator doesn't apply root motion.
• getting rid of all the transform.Translate or similar commands.

[normal mountain]

the rigidbody already is kinematic, the animator does NOT apply root motion, and without speed2 the enemies chase the player weirdly

should i turn off use gravity?

[tardy junco]

if it's kinematic it shouldn't be affected by gravity anyway.

[tardy junco]

And I said get rid of all transform.Translate. Not just the ones where you use speed2.

If you disable and reenable the agent, make sure it's on the navmesh at that point. You could sample the closest navmesh position and setting the character pos to that position every time you enable the agent.

[normal mountain]

the only transform is speed2 at the given time

[tardy junco]

I see quite a lot with speed and speed1 in your code too. I don't have time to look thoroughly through your code, so I can't be sure whether they get or don't get called at that time. There's also a possibility of a bug where it is actually called when you don't want it. That's why commenting them out and testing is a good thing to try.

[normal mountain]

speed and speed1 is when they are currently chasing the player or they are going to a waypoint, that's it

the only time they leave the platform is when they have speed1 active... so that could probably mean that speed1 does something to the navmesh

a.k.a. when they see the player

getting rid of speed1 seems to make them not rotate fast at all or even run fast

so i need to think of a way for speed1 to not lead the enemy off the platform, but also have the enemies still have their normal speed

[tardy junco]

try playing with acceleration it seems to have an effect on the rotation speed too.

[normal mountain]

okie

alright it is not working at all without speed1

they dont move when they see the player

[tardy junco]

are you actually setting the destination of the agent?

[normal mountain]

yeah, it sets to player when they see him but they do not chase after him due to the lack of speed1

[tardy junco]

Is that destination on the navmesh?

[normal mountain]

cause for going for waypoints, since speed is in there, they go there successfully

it is in the script

agent.SetDestination yadda yadda yadda

[tardy junco]

is the destination position on the navmesh or very close to it?

[normal mountain]

for what

[tardy junco]

For set destination

[normal mountain]

so you mean......... uh... im confused

[tardy junco]

SetDestination takes a vector3 position, right?

[normal mountain]

yeah

in this case it is player.position

[tardy junco]

Is this position somwhere on the navmesh?

IS player.position on the navmesh?

[normal mountain]

no it's just this in the script:
agent.SetDestination(player.position);

[tardy junco]

I'm asking you if this position is physically on a navmesh?

[normal mountain]

no it's anywhere

[tardy junco]

Well, that's the problem.

[normal mountain]

if they see the player in the raycast and in their line of sight they will try to reach them where they eventually break through the nav area

[tardy junco]

are they supposed to break from the nav area?

[normal mountain]

nope

clearly not, if they are floating in midair then hell no they aren't supposed to

[tardy junco]

Then follow the steps I mentioned a few messages ago, and they will stay confined to the navmesh.

as for not following, you could try sampling a point on the navmesh and setting that as the destination.

[normal mountain]

but that is where the problem is, if i dont have speed, speed1, or speed2 they dont move period

[tardy junco]

I already explained you why they don't move. moving them manually adds to the problem, not fixes it.

[normal mountain]

manually??

[tardy junco]

Manually. As you did with transform.Translate.

[normal mountain]

oh

and wait where did you say why they dont move?

[normal mountain]

unless you meant this

which i dont know how to do

[tardy junco]

Instead of setting player.position as the destination, sample the navmesh for the closest point with SamplePosition and then set the resulting point as the new destination. https://docs.unity3d.com/ScriptReference/AI.NavMesh.SamplePosition.html

[normal mountain]

the issue with me when it comes to the scriptreferences, for certain ones if they are not used directly in my script by someone else who knows how it works already, then i dont know how to properly put it in... better yet i dont know which one to use

like which one do i even use in this reference, where do i put it

[tardy junco]

There's an example on the page.

Try to understand what the function and each of it's parameters do, and you'll be able to use it without problem.

[normal mountain]

the closest one i can think of is maxDistance

but i dont even see it being used in the example script that is shown

[tardy junco]

Of course it is. NavMesh.SamplePosition(randomPoint, out hit, 1.0f, NavMesh.AllAreas)

1.0f is the third argument and the only float. I don't understand how can you not notice it. 😅

[normal mountain]

so i use that on a line in if(state = "pursuing") ?

[tardy junco]

you mean in that if block?

[normal mountain]

yeah

[tardy junco]

Yes, check if it returns true. If it does, you can use the hit position as the agent's destination. Otherwise it couldn't find any navmesh position in the designated radius.

[normal mountain]

so how does the if statement look for if true? and does it trigger the player position if so?

[tardy junco]

trigger? It will look exactly the same except you'll use the sampled position.

if(SamplePosition)
Set your destiantion.

[normal mountain]

so
agent.SetDestination(player.position); ?

but replace player with SamplePosition?

[tardy junco]

No, do it outside in an if statement to see if you got a position or not.

[normal mountain]

so what i just said... but outside of the pursuing if statement

[tardy junco]

If(sample position)
  set position as the hit result
else
  throw error or handle in a different way

[normal mountain]

i notice another issue

randomPoint

the example shows a whole new bool statement with {} and yet i cant do that

nevermind

i got the thingie

wait... i already have a hit section at the HaveLineOfSight statement bool where it checks for if the tag is Player

how the fuck am i supposed to do this

[tardy junco]

it can be any name you want.

[normal mountain]

what can??

the bool? the hit??

[tardy junco]

both.

[normal mountain]

so i can put the thing in the havelineofsight and have the same hit?

even if the hit is a RayCastHit?

[tardy junco]

No. They are different types. The navmesh one is NavMeshHit.

[normal mountain]

i renamed the navmeshhit to booper so that is solved

i dont get what the hell im doing but here is where i am at

[tardy junco]

You don't need random point. you want to sample the position of your player.

[normal mountain]

so replace randomPoint with player?

[tardy junco]

And it's not the place we were talking about.

how did it get to HaveLineOfSight method? lol

[normal mountain]

cause that is where the raycasthit also is

where am i supposed to put it???

[tardy junco]

Why would that be related?

I told you where. Where you set the destination of your agent. agent.SetDestination(player.position);

instead of player.position, you want to set a destination on the navmesh, right?

[normal mountain]

and i told you that is inside another bool

[tardy junco]

What is inside another bool? How can something be "in a bool"? 😅

[normal mountain]

the fuckin bool InFront() { }

[tardy junco]

This is a fuckin method or function(that returns a bool), but in no way you can call it a bool and expect other people to understand that.

[normal mountain]

then that i guess function

cause i do have return true or false in there

[tardy junco]

That's the part about what it returns.

[normal mountain]

but now it isnt working due to the result thing

[tardy junco]

Either way, why is it in there? I think I was pretty clear with that message:

  set position as the hit result
else
  throw error or handle in a different way```

[tardy junco]

what result?

[normal mountain]

oh my fuckin satan, the result that came with the example

[tardy junco]

result in the example is just and out parameter they define in the function. You don't even have to use it like that.

The only part you should really care about is this:

            if (NavMesh.SamplePosition(randomPoint, out hit, 1.0f, NavMesh.AllAreas))
            {
                result = hit.position;
                return true;
            }

you can ignore the return and setting the result variable

[normal mountain]

well it is causing an error just by existing

[tardy junco]

all you need it to use is that hit.position for your destination.

[tardy junco]

Then fix it.

[normal mountain]

i dont know what the hell this means though

[tardy junco]

Since you're using result as an out parameter it has to be assigned in the function.

[normal mountain]

what part of it

[rustic obsidian]

result

you need to assign it

[normal mountain]

WHERE

[tardy junco]

ANYWHERE before returning from the function.

[normal mountain]

well guess what, regardless i still have the error

result = Vector3.zero;

[tardy junco]

Because you don't listen to me. You don't even need to use result.

[rustic obsidian]

Not saying what an error is is really unhelpful too

[normal mountain]

i mean i showed it in the screenshot

it's literally right there

[rustic obsidian]

well then you haven't assigned result before returning

if it's the same error it's the same problem

[normal mountain]

i dont know how to assign it

[tardy junco]

It must be set in every independent scope. IF your code goes through a function, gets to return and sees that out parameter is not assigned, it will give your this error.

[normal mountain]

cause im a dumbass

[rustic obsidian]

result = anything;

is assigning it

[normal mountain]

so not Vector3.zero??

[rustic obsidian]

that would be fine, but you're clearly just not assigning it before a return

it's skipping the assignment logic for a possible execution

all execution paths need to have result assigned

[normal mountain]

so like this?

cause that got rid of the error

[tardy junco]

That will work. You could also assign it a default value once before the branching, instead of doing it before every return.

[normal mountain]

so at the very beginning of the function or...

[tardy junco]

yes. Before all the if statements begin.

[normal mountain]

ok

[rustic obsidian]

I like to assign it once before each return as it makes it clear when you've missed some logic

[tardy junco]

I agree, but it this case he doesn't even need it.

[normal mountain]

alright i did that... anything else?

[tardy junco]

Why is there a for loop, why is it after setting the agent destination. You should be using the result if the Sampling as the destination. I think I mentioned it like 5 times now...

[normal mountain]

it's because it was in the example... -_-

[tardy junco]

Example is there for you to understand how to use the function. Not to copy-paste the example.🤣

They use it for a completely different thing.

[normal mountain]

how do i type the if statement as the replacement then

if (SamplePosition...) what comes next

[tardy junco]

I already explained several times:

  set agent destination as the hit result
else
  throw error or handle in a different way

[normal mountain]

the if statement doesnt work at all.... it cant just be If(sample position)!!

[tardy junco]

That's all you need. 1 if statement, 1 sample, 1 set destination. 2 lines of code.

[tardy junco]

Of course. It's pseudo code. But I think it's pretty self explanatory.

[normal mountain]

not really

[tardy junco]

Why? What exactly is not clear?

[normal mountain]

what i should put in the if statement with it!

[tardy junco]

sample position is NavMesh.SamplePosition(blablabla).

I mean cmon man, let's not get down do #💻┃code-beginner level. If coding is a problem, I'd suggest doing some c# courses.

[normal mountain]

i should not do that because if i try to see it through courses that dont relate to my issue, then im probably going to get stressed to the point of wanting to shoot myself AGAIN

and plus outside of AI i can do literally everything else correctly

i haven't come here a single time with something other than AI damn it!

[tardy junco]

Yet the issues you're having are clearly unrelated to AI. They are related to missing some beginner-intermediate understanding of c# or coding in general.

[normal mountain]

and yet everything you try to show me in that sense is something i dont fuckin understand cause it doesn't relate at all to my problem

considering that just about all of it doesnt apply to the code

and my issue DOES relate to AI, it involves NavMesh

[tardy junco]

It doesn't have to relate to your problem. Is generic coding knowledge. You will need it at some point(like now), if you're planning on developing(games) seriously. And it's clearly related to solving your current issue.

[tardy junco]

That was your initial issue and I already gave you feedback on how to tackle it. Now it's just up to coding it.

[normal mountain]

how though?!

how do i code it???

cause what you told me for sampleposition is nothing but bullshit it didn't even change anything

THIS

[tardy junco]

Very simple. 2 lines of code. Both of them are in the example, you just need to change some parameters.

If(sample position)
  set agent destination as the hit result

SamplePosition is a method that takes some parameters. Check the docs page again to see what it does and what parameters it takes.

I'm not gonna write the code for you.

Did you read the description of the method?

[normal mountain]

yes

i read it multiple times!!

but i guess im such a dumbass that it goes right through my thick ass skull!

[tardy junco]

Did you read what parameters it takes and what they correspond to?

[normal mountain]

yes

i read every fucking thing on the page

[tardy junco]

Okay, so it's simple from there. call the function. Pass in the relevant parameters and you're done!

[normal mountain]

you act like i know how to do that...

i cant even do the fucking if statement how am i supposed to do that?!

[tardy junco]

Then I'd really suggest going through some beginner coding tutorials. Calling a function is one of the basics.

if statements too.

[normal mountain]

i can usually do if statements pretty easily, but every time i ask about it, you make it as bland as possible for me to figure it out

[tardy junco]

I just don't understand what the problem is.

[normal mountain]

like if(sample position) isnt gonna do it for me

[tardy junco]

Why not?

[normal mountain]

i need to at least know what to put in that and then maybe i can do the rest of this shit

cause if (NavMesh.SamplePosition) does not work

neither does if (NavMesh.SamplePosition(player, out booper, 1.0f, NavMesh.AllAreas))

[tardy junco]

If I tell you what to put in it it's basically the same as writing the code for you. I can copy paste the method from the example, but you'll have to replace the parameters to what you need.

[normal mountain]

fine do that then

[tardy junco]

Why?

[normal mountain]

cause it is in the thing right under it

[tardy junco]

Why do you need that line 114??? 😆

[normal mountain]

cause you told me to replace the for statement?

[tardy junco]

I asked what was it for. I meant that you don't need it(nor you need to replace it).

2 line of code:

If(sample position)
  set agent destination as the hit result

all you need.

[normal mountain]

well i got rid of it so...

[tardy junco]

instead of the 10 that you have from 113 to 122

[normal mountain]

so put the agent.SetDestination inside of if (NavMesh.SamplePosition(player, out booper, 1.0f, NavMesh.AllAreas)) ?

FINALLY stupid auto-delete

[tardy junco]

It's not auto-delete. It's bot moderating proper sharing of code.

[tardy junco]

Yes.

[normal mountain]

ok.

and of course the InFront that gets called in the update function decides to NOT WORK ALL OF A SUDDEN WHAT THE FUCK???

[tardy junco]

Define "not work".

[normal mountain]

red squiggle shit

(Vector3 center, float range, out Vector3 result) this is the cause of it

[tardy junco]

Yeah. I don't know why you changed your function definition...

[normal mountain]

cause it doesnt seem to work without center

[tardy junco]

What center?

Did you not listen to me? 2 lines, instead of 10.

[normal mountain]

that is part of Vector3 rando = center + Random.insideUnitSphere * range;

[tardy junco]

You're probably still doing something unnecessary.

[tardy junco]

this is unnecessary.

It wasn't in your function before, was it?

[normal mountain]

it was

after i searched the "example"

[tardy junco]

I mean before that.

[normal mountain]

no it was not

[tardy junco]

you only ever needed 2 lines in your original function.

[normal mountain]

what is not needed in here then?

[tardy junco]

Originally it was 2 lines instead of 1. So you only needed to add 1 line.

Was the function originally returning something?

[normal mountain]

it was originally part of the result thing before i scraped it

[tardy junco]

the first argument of SamplePosition is the position that you want to sample from. The player position.

[tardy junco]

then you don't need a return.

[normal mountain]

so like this? if (NavMesh.SamplePosition(player.position, out booper, 1.0f, NavMesh.AllAreas))

[tardy junco]

Yes. Now it's only the SetDestination that still uses the player position.

[normal mountain]

yes

[tardy junco]

Yes. An I said like 10 times already that it's supposed to use the hit result.

[normal mountain]

so booper

[tardy junco]

Yes.

[normal mountain]

so this is the final result?

[tardy junco]

Set destination still uses the player position.

[normal mountain]

yeah? and?

[tardy junco]

And it shouldn't. That's the whole freakin point of what we were doing.🤣

[normal mountain]

what do i replace it with? booper?

[tardy junco]

Yes. And I think you already asked that and I already replied to that...

[normal mountain]

i thought you meant somethin else

[tardy junco]

I don't see any other SetDestination in that screenshot.

[normal mountain]

booper.position or.... what is it

[tardy junco]

Refer to the docs.

https://docs.unity3d.com/ScriptReference/AI.NavMeshHit.html

But yes. it is.

[normal mountain]

so yes

and that is literally the last thing... RIGHT?!

[tardy junco]

Assuming you didn't mess up anything else in the process, yes. Although I'd add an else statement that throws a Log or an error for ease of debugging.

[normal mountain]

i skipped my shower tonight for this so i have to make sure

and ok

[tardy junco]

You might need to play with the distance parameter depending on how far the player is from the navmesh.

[normal mountain]

alright... i will aim for probably 2 to start

...

the enemy automatically goes off the platform when spotting the player now...

that did the exact opposite of what i wanted

[tardy junco]

"goes off"?

[normal mountain]

floats! doesnt stay on the platform he is on originally! acts like the lower ground is at its height!

[tardy junco]

You're still using transform.Translate in your latest screenshot btw.

comment all that shit out.

[normal mountain]

he glitch moves now

when spotting the player

[tardy junco]

glitch moves?

[normal mountain]

he doesnt move, then he moves slightly

[tardy junco]

Probably because you set new destination every frame. It stops and takes time to recalculate the path again.

[normal mountain]

InFront() has to be activated in the Update function

[tardy junco]

It doesn't mean that SetDestionation has to.

Might also be that else statement.

What is that supposed to be checking?

where do you set the direction?

[normal mountain]

to see how close the enemy is to the player in order to start attacking

so that is needed PERIOD

[tardy junco]

Put some debugs in the if and the else to see whether they execute where you don't expect them to.

[normal mountain]

well apparently the useless shit that i put in that was the sampleposition failed to work 189 times

[tardy junco]

Then your player position is way to far away from the navmesh. That's also why the agent wasn't moving before.

[normal mountain]

what is

the navmesh is probably being interfered by the lower level navmesh...

WHICH I CANNOT ALTER IN ANY WAY

[tardy junco]

??

why do you have 2 navmeshes near each other?

[normal mountain]

because the floating platforms are the same navmesh just higher!!!

[tardy junco]

So they're not that near..?

And how does it interfere?

[normal mountain]

what now you're gonna tell me that there's multiple bake able navmesh areas?!

it interferes because what isn't altered is the height of the navmesh in the air from the one on the ground

[tardy junco]

There is, but I'm not sure that's what you need..?

[normal mountain]

i mean if anything that could work BETTER

[tardy junco]

When you sample your position, your player is probably directly above the lower navmesh, so yeah it will be sampled.

[normal mountain]

indeed

[tardy junco]

I'm not completely sure, but your game seems to be a platformer?

[normal mountain]

but yeah let's try the idea of multiple navmeshes that is my only hope and yes it is

a 3D platformer

[tardy junco]

I'm not sure a navmesh is the most suitable solution for navigation in your scenario.

Do you need your npcs to move towards the player, when it's in the air?

or on a different plarform?

[normal mountain]

yes

the lowest level of ground as a navmesh is perfectly fine though

[tardy junco]

What you could do is check if the destination is reachable first before setting it.

[normal mountain]

of the player?

[tardy junco]

I don't understand what's the current problem then?

[tardy junco]

Yes. That might stop the agent from moving if the player is not on the same platform though.

[normal mountain]

the problem is that the enemy will chase the player but eventually they will glitch into the lower level navmesh but since they dont do gravity, they float

and that is why if the gravity is on then the enemy will fall to the lowest level and start moving weirdly

[tardy junco]

Do you want them to chase the player to the lower levels?

[normal mountain]

no

if i wanted that then i would have made their jumping function by now

buuuut i will save that for a different type of enemy

[tardy junco]

Then they shouldn't by default. The platform area mesh is not connected to the ground is it?

[normal mountain]

no it isnt

[tardy junco]

Then it's probably one of your Translate calls.