🤖┃ai-navigation

[golden kettle]

@sonic rain Hey quick query again on UAI, so he mentions on one of this slides that you can keep multiplying these utility/axis values together to get an overall result for the related action.

However he highlights that by default you will end up losing value with the more in the axis as you keep multiplying you just keep getting lower and lower i.e:

0.9 x .0.9 x 0.9 =0.729

So someone else came up with a fix for this where they applied an offset fix which allowed you to multiply with the offset and it stops the reducing of the average, which seems a bit long winded and complex.

Now my question here is really why bother doing all that when really if you want an average could you not just do:

(0.9 + 0.9 + 0.9) / 3 = 0.9

(sorry for so many edits, pressed enter by accident without shift lol)

The only logical reason I can see for doing it with the multiplication with offset addition is because division is more time consuming on CPU than multiply, but I wasnt sure if there was any other reason

from a quick check it seems on average CPUs an FP division is 5 times more expensive than a multiply or add, but even so I would imagine you need to do far more math to calculate the offsets and factor them in than the cost of a single divide (adds/subs seem to be about same cost as multiplication on fps)

posted in reddit gameai, Dave is a mod there so hopefully he will jump on it and tell me

[sonic rain]

Hey @golden kettle ! Yeah, I was also puzzled by that. My conclusion was: the multiplication is really needed as it has the power to scale up/down some value. See, a multiplication of a big number with a small number will cut it down a lot, in comparison to summing it up and doing the average. Something like this:

Considering two scores:
A = 0,9
B = 0,1

Sum: 0,9 + 0,1 = 1
Multiplication: 0,9 * 0,1 = 0,09

So, as a comparison of the results, the Sum will increase the final value, and multiplication will decrease it "a lot". When we get the sum and make the average, it still results in a "high" value, which would be 0,5

Considering the two same score values as above, and applying the average method VS the method presented by Dave Mark:

Sum and Average: (0,9 + 0,1) / 2 = 0,5
Dave's Method:

modificationFactor = 1 - (1 / 2) = 0,5
makeUpValue = (1 - 0,09) * 0,5 = 0,455
finalConsiderationScore = 0,09 + (0,455 * 0,09) ~= 0,131

Results in ~0,131

Comparing the results again, they are very different from each other and they express different design choices. With multiplications, we are always scaling one value with the others, which happens in a different proportion in comparison to sum and divide

Another thing that I noticed is that, if you have a value 0 as score, the results will ofc differ a lot:
Sum and Average: (0,9 + 0) / 2 = 0,45
Multiplication: (0,9 x 0) = 0

But even when performing sum and average, in case any zero, we could perhaps just say "ok, cancel the sums, result zero immediately"...but that's a workaround which might break the methodology

Why am I speaking about zeroes? Because, when designing my own curves these days, there is some usability value when you want to return zero in one of the curves as a way to cancel the other ones. Like: I'm combining 3 curves to know if I should use a Skill. One of the curves will return zero if the skill is in cooldown, aborting the whole multiplication

Showing a little bit about this Skill thing that I mentioned...these are my curves, on my tool (this is the minimised view, it can be expanded to see/define the curves better)

Their results are combined in order to know if the Heal Spell should be used

As you can see, the first thing that is tested is: is the skill in cooldown? (i.e it's cooldown value is > 0)

If it is in cooldown, zero it. This, multiplied by anything, will cancel it. So no, in this case, don't use the spell

If the cooldown is zero, then well, the result will depend on the others. But yeah, multiplication by very high or very low factors, while normalized, allows you scale the importance of doing stuff. I'm not sure if the same is achievable with sums and averages, perhaps it is! But it would be bit different, but it might be not impossible

[left turtle]

I have an AI that generate its own sequence of action based on parameters and avalaible resources. It's not a state machine for sure, how would you name that type of AI?

[golden kettle]

@sonic rain Thanks for the info buddy I have been in a meeting most of the day but yeah as you say the 0.0f results indicate STOP, whereas with a simple average it would just drag the numbers down a bit, not short circuit

[real sonnet]

In short, "action planning". So GOAP ?

Or perhaps if it's not goal-oriented and you made up a new kind of AI, give it a new name: Context Action Planning haha

Reactive Action Planning ? 🤔

[trail lagoon]

Hey my NavMesh base offset variable keeps resetting to 0 in Play Mode, any fix?

[golden kettle]

@sonic rain I also came up with a semi novel way to handle multiple things per utility, so I already had the notion of a sort of blackboard style object which was used by our existing framework. So by default it's keyed on ints (utility/consideration type) and contains whatever data you want but as its a <K, T> situation under the hood I have just added a slightly customised implementation with a composite key where its made up of utility Id and another possible Id, so if you just provide the utility type it only has one entry for that utility, if let's say we want to track all targets it's all under same utility Id but the foreign key (being the enemy Id) basically let's be have multiple entries under same utility Id. This not only let's me keep things a big simpler in terms of usage but it also let's me use the foreign key to lookup individual targets/tiles/abilities for that variable AND I can basically average or score all entries under that utility Id to get a feel for things, I. E get average party hp scores, or min enemy hp score etc, not sure if I will find some downsides more I get into it, but going ok so far.

@trail lagoon not a clue but this is more ai discussion. Would ask in general

[left turtle]

It has 2 state
Survival and Normal
In normal mode he has a "safe/agressivity meter" based on how much he is advantaged. They don't wanna get stalled but don't wanna be suicidal nether. It build his sequence o action based on risk/reward taking account his resource ex (mobility, weapons) and his target resource as well.
In survival it's as the name say, they are trying to drag you out until reinforcment or escape if they're aint getting any

i built it long ago without knowing anything about IA and was wondering if it has a name

[golden kettle]

Fuzzy logic?

[trail lagoon]

I thought Nav Mesh was AI?

[golden kettle]

There is only really names for conventions or architectures, you have some fuzzy logic with a transition to states or something so I wouldn't say it's a specific convention just a mix of a couple of approaches.

[golden kettle]

At a stretch you could say pathfinding is, but I imagine you would get more success asking about odd editor behaviour elsewhere.

You may get lucky but I would ask in general or something too as I assume this is a lot smaller an audience than there and the issue isn't the ai or whatever its just the editor/mb ignoring your value for some reason

[trail lagoon]

ah, ok thank you.

[sonic rain]

I see. It makes sense to me...I didn't think about it as a solution on how to get averages or min/max, but your explanation makes sense. I currently can't point out any major disadvantage. I hope it keeps going fine!

[left turtle]

thats what i though, some people make it seem there's only a few way and everything fall into one of the category. For context it was a tactical rpg game

[errant quail]

i made my ai yaay

@golden kettle thanks mate

[golden kettle]

Np buddy

[golden kettle]

It's like with any pattern or convention, everyone knows what an fsm or bt is when you mention it, so we can all be on the same page conceptually. If you make an fsm but you didn't know it was an fsm and tried explaining it everyone would have puzzle faces wondering what you were on about initially, even if it's the same thing. In your case it just sounds like a bit of a few approaches

But at some point someone has to name a convention before it can become standard, look at Utility AI/Infinite Axis AI, it was an evolution of fuzzy logic and Dave gave it a name and now we all kinda agree on what it is

Like if I said to you, I have an object that listens to value changes from somewhere and contains a read only copy of the most recent value but you can be notified on changes to the value.

What is it?

(It's a computed property/value, but if I just said computed value up front people would be ons ame page from get go)

[left turtle]

i made a couple of IA in the past without knowing anything (competitive programming, where my AI had to beat other AI), i had a pratical diploma so I could code but not a university one so i barely knew of any concept I was actually using or found out myself xd. The only time i used an external source was when I used genetic AI cuz i read about it and though it was cool.

[left turtle]

Imo I call it the variable you put at the top with a getset

[golden kettle]

Lol well it's called a computed by most people 😋

But as you say until you know the common terms it's all nonsense, like when I first learnt factory pattern had a name, fresh out of uni I didn't have a clue about all these conventions and terms. It's only really when you need to work with others that there is some value in having common terms that everyone can understand

If you tend to work alone more doesn't matter what they are called until you come to discuss it with someone else

[left turtle]

Nah i dont work alone, but prefer to use easy wording that everyone can understand

including non thecnical people

[golden kettle]

That's a different thing then, like at work if I say to someone "make sure you plug in the repository via the di module and add an aop transaction handler" I'm only gonna say thst to a technical person

[left turtle]

and if i take shorcut ill only talk in a pratical manner, no weird word

[golden kettle]

But these words are not weird, they may be unknown to you, but assuming it's a common industry/technical standard term there is more merit following same vocab as everyone else

Simplifying technical jargon for muggle business people is a separate skill, but again before you can translate what the architect is spewing you need to know what his terms mean

If he's speaking his own flavour of terms rather than common ones everyone's gonna get confused

[left turtle]

my uni teacher didnt like it, but i quit cuz it felt like a waste of time, job secured, wasnt getting more money, was losing years of salary/experience

in the job i just learn term on the fly when i need them

by talking to people

[golden kettle]

You don't need degrees or anything these days, plenty of materials to learn and improve online, but regardless of this stuff the quicker you can all speak same terms and all understand each other the better, same in anything

[left turtle]

it's a bit like linguistic immersion travel vs learning everything from home then move

i prefer the first approach

[golden kettle]

Sure sure, but as long as both arrive at same destination I wouldn't care, its jsut if you purposely didn't want to adhere to common terms for whatever reason that I would be worried

[left turtle]

oh don't worry i'm fine with term, im just not fine with how teacher hard punish you for not using the exact wording they want

and again, i'd say it depend on teacher

[golden kettle]

I had a tutor refuse to accept my assignment because he said we had to make a website about anything, so I just made a website about cheese with horrible js effects and midis lol

[left turtle]

but when i was young and didn't know how thing work i though i absolutely needed a university degree

[golden kettle]

(this was like 20 years ago, not modern webby swishness)

[left turtle]

haha lmao

[golden kettle]

It definitely helps, but you don't NEED one if you are motivated and learn right stuff

I think that's one of the key things, knowing what to learn

I had this debate with a chap who was streaming his game dev, that I feel as time goes on there is less and less merit in paying for a degree in tech

The only major benefits are potentially the road map of what you learn and possible help in getting some work experience

If you got a university module road map and youtube each bit you could potentially still be as competent

[left turtle]

one time my teacher was trashtalking video games so in a argumentative essay i picked the subject "literrature shouldn't be in educational system", i failed but and I was able to prove she wasn't impartial in my evaluation haha it felt so good

my parent just though i was making trouble for nothing ¯_ツ_/¯

[left turtle]

Yeah i was defintively motivated to learn, just school wasnt for me. Even my current diploma is partially earned remotely from home

[fallen nest]

Not necessarily, the issue isn't with complex collisions cause it would be used with path finding. But the issue is that I can't figure a good way of getting around an obstacle overall. Someone had managed to do it without making the literal obstacle avoidance by implementing a system that would avoid certain vectors. The issue is I can't figure out how to make it avoid those vectors and also go the right way.

[fresh jetty]

Hey does anyone know how to fix that? I'm getting it only on the second "enemy" object, on the first with the same mesh agent settings works perfect but on the second "enemy" I'm getting this error (before bake map fixed it, but now it doesn't change anything)

oh I see this problem exists only on custom agent type :/

[fresh jetty]

nvm I fixed that, if you want to have more than one custom agent you have to import NavMesh components https://docs.unity3d.com/Manual/NavMesh-BuildingComponents.html

[alpine glacier]

1 for unity navmesh if you have a stationary object that does not move at all, why do people still bother to set it up as a navmesh obstacle in tutorials when you can literally just tick 'height mesh' then bake works exactly the same?

this way it excludes it from the navmesh entirely

which in my opinion is not only less work but will work way better has your game becomes more complex

pls @ me when replying

[alpine glacier]

2 I have this issue with the following code:

Basically when I run it on my game there is a significant delay in the enemy actually starting to move, I am not entirely sure what is causing this or how to fix it? I am on quite a powerful PC so it can't be the processing power? Additionally I thought it was because the enemy was note entirely on the navmesh or falling off it, but its not that either?

I think the most likely issue is, its recalculating the movements too often

but I am unsure how to fix that?

[alpine glacier]

https://i.gyazo.com/e7a2b7519fcf0095586302979b30181f.mp4

here is a video

of the issue

the video actually makes it look like they are moving fast, but sometimes they won't move until I go near them, drag them in the editor screen or even move at all

pls @ me when replying

[sharp mulch]

hello how can i make that the navmesh is on the ground and not floating like in this picture

[fresh jetty]

Hi does anyone know how can I make detection with camera for AI something like on the image? I mean if player is behind of the wall on camera look then don't detect, otherwise detect. I saw some ways with GeometryUtility, raycast or viewportpoint from world and nothing from them are works as I want, does anyone has better idea?

[solar oriole]

Hi i was wondering if anyone knows what could be causing this issue i'm having with my navmesh agent.

On scene play my navmesh agent goes from a normal rotation to being stuck at a -180 x rotation

[fresh jetty]

I had the same problem, it was caused by invalid exported fbx model from blender

[solar oriole]

In your navmesh agent can you pass in coordinates directly?

For the destination

instead of a mouse position etc

[alpine glacier]

https://www.youtube.com/watch?v=j1-OyLo77ss&t=24s and https://github.com/Comp3interactive/FieldOfView

[vast viper]

So i want my navmesh agents to NOT avoid each other, meaning it's ok if they're clumped. Their collider is set to trigger, radius to 0f, but whenever they walk "through" another agent, they still make way / push each other when passing

[copper current]

oof i hit a lag spike there

[wind crypt]

Hello anyone know how to check if agent has complete path to reach destination or not?

[brave sorrel]

Anybody knows what i can do against this?

[slender wing]

change your ml agent version to the supported version. version 2.

[tight bison]

my question is the same as akka's, is there an efficient way to tell if the agent has reached the destination?

[brave sorrel]

what kind of destination does it have?

[tight bison]

I'm tagging a bunch of objects having them with a tag "LightObjects" and "HeavyObjects" and just checking the furthest distance from the agent and setting the destination to it

[brave sorrel]

i mean you could do couple of try runs, where you try different things and you can compare them in tensorflow

[river fern]

So I made an ai but when it goes to the waypoint it just turns circles can some 1 help me?

[real sonnet]

Do you try to reach exactly the position ? If so, you should instead just check if distance < someThreshold where the threshold is something small but not zero, like 0.05f for example

And when you reach last waypoint, you could just stop updating the position (stop moving)

[river fern]

I'll look my code in a minute

[wispy bison]

Hello, world! I am currently working on an arcade racing game and I was wondering if anyone here knows any resources that could help me get started learning how to develop AI opponents for it.

I did some searching on my own but, besides some tutorials on YouTube, I didn't find much.

[dim needle]

i think some keywords here are:
navmesh, pathfinding, astar, finite state machine

[winged relic]

a good resource for in general a.i. concepts to get you started which has a chapter on cars, movement and path-finding http://www.gameaipro.com/

[old kiln]

nav mesh time

how to solve this issue

[old kiln]

my nav mesh agent is falling through ground

[alpine glacier]

hey guys, im a bit new to programming ai and stuff. currently trying to get follow target working after click to move is complete.
heres my current code: https://hastebin.com/alimarujog.csharp

(obviously wrong since it doesnt work xD)

[river fern]

Try the tutorial of pretty fly games on YouTube

[low stone]

hi am starting out in untiy, i want to create an AI for my enemy (2D platformer game) melee attacks or range attacks what is the best way to do so, any video on that topic would be helpful :)

[alpine glacier]

Erm, which one? 😅

[river fern]

The ai one first one

[tight bison]

I'm having a bit of a problem here, I'm making my characters go to specific objects. When they do they'd just send an output. Now I'm trying to check if the distance to target is smaller to stopping distance but sometimes there'd be characters that are just a above the stopping distance by a couple of float points

note the distance to target in the inspector tab

how do I fix this

[old kiln]

anyone knows any fix to this

[old kiln]

nvm I solved

fix is just to bake it from navigation

[alpine glacier]

ok i tried to follow the tutorial but im still not making much progress. its starting to be a pain in the ass lol 😆

[winged relic]

to be fair it looks like a simple logic problem, gone wrong 😉
Since your conditional is in the update function which runs all the time you are resetting the destination.

basically try commenting out else { m_minion.SetDestination(m_master.position); } or removing it.

it's overriding what your mouse click does in the next frame.

however I'm making some simple assumptions just looking at your code, I don't navmesh 😛

you also appear to have 2 references to navmesh agents, which could be confusing... do you know if you are triggering the correct one

[alpine glacier]

at this point im at such a stress trying to fix this im not even certain at this point haha

[winged relic]

rather than trying to fix it, perhaps try starting clean and following a set of related tutorials from start to finish to help understand how it should be used 😉
It may take a few tutorials, but eventually I'm sure things will start to click into place

[golden kettle]

Related to a query I had a while back to do with Utility AI, so I have a basic Utility AI framework and am trying to build up a suite of considerations and actions, however im struggling to wrap my head around the best way to collate the available information.

So for example I have a consideration for if something needs healing, and allies who can heal can monitor that utility on others, and same on enemies to know which one the weakest is. So when factoring in the best enemy to go for I ideally need to know considerations for:

• Distance
• How Low Their HP Is
• How Low Their Defenses Are
• How High Their Damage is

as if something is close but with high HP but something else is slightly further away with lower HP I should be going for that, but the thing is scoring alone (utilities multiplied) isnt enough alone, so im wondering about adding weights to things to make them more important, like:

Enemy 1

• Dist: 0.5
• HP: 0.8
• Def: 0.2
• Dmg: 0.2

Enemy 2

• Dist: 0.4
• HP: 0.3
• Def: 0.8
• Dmg: 0.8

So at a high level Enemy 1 is close-ish, very low hp but has high def/dmg, and Enemy 2 is further away, more HP but far less def/dmg. Ideally I would want to go for Enemy 1 as the low HP being higher is more important than other stats, so is it as simple as just adding a weight modifier and multiplying the HP by that?

I know its always mentioned in the GDC talks about adding weights but its discussed at actoin level not consideration never delved into deeply so was wondering if anyone else had any experience/thoughts on it

[winged relic]

@golden kettle isn't this where Curves come into play, it allows you to artificially inflate important stats or down others.

[golden kettle]

that gives us the 0-1 score for a consideration

so for example HP is using a quick rising quadratic curve to take the 40/200 hp for example and run it through the evaluator/curve to yield 0.8 (i.e you want to heal)

if it were linear or a more relaxed curve I may get 0.6 instead or something, so yeah they do influence the consideration value, but only in isolation

so in this case when I am formulating composite considerations or even an action I am wanting to imply that certain utilities are more important than others

as every unit has the notion of how close to death they are (needs healing), how strong they are (rough damage output), how defended they are (rough defenses) and other bits

so scoring them and knowing about them in isolation is one thing, but when you want to bring them together in some cases, low HP utility may be far more important than how high a damage output someone has

but as there is little REAL world info on this approach, the main gist is you just multiply them all together and at action level maybe multiply it based on a priority

but I want to add a weight to existing utility values when factoring them together with others

[winged relic]

I was reading a chapter which maybe relevant, it suggesting having a consideration to asses "survival chance". I.e. don't run to repair if it puts you at more risk, which could be flipped around to don't engage if chance of survival is low. Basically it maybe alter your list of considerations to include some more expressive values.
Hope that makes sense 😛

[golden kettle]

but even so, if I have 3 utilities

lets say health, risk, distance

risk may be far more important than distance

I guess risk is a bad scenario as thats contextual

but in some cases you may want to scale utilities somewhat

but again this raises another question, around composite utilities, currently I am building some utilities off others, like computed value

but you could say "dont do that, just make it an action"

[winged relic]

I know what you mean and can't help but think as I often have been this past week "Am I overthinking this", you maybe seeing a problem that does not exist. If the information in is good, the decision should be too

[golden kettle]

yeah, I feel that this is a valid issue, as right now im working through targetting logic and its tough getting it to know who to target as early on all the values are pretty similar as even if HP starts to fall on one, its kinda ignored somewhat by the other values dragging it down

[winged relic]

I'm in the slow process of building my UAI (just in spare time, so going slower than I'd like), I like the idea of Utilities to support other utilities and have indeed built this into my code. How it works in long run is yet to be seen 🙂 But I approve of the idea!
Targetting is a tough one, I was considering handling this logic elsewhere but that feels defeatist. So much as you are probably doing I've now added some inputs to my top level decision, so it can be assessed in UAI. It still feels like catching air though 😄

For me though I'm going to pull in some values from influence maps which should be the deciding factor / extra weight

[golden kettle]

at the moment I am quite happy with the overall implementation, as I have composite keys and also contextual advice so rather than it just being "Go Attack" it also contains a contextual accessor which lets you knwo what you should be attacking

due to this I kinda went with Considerations/Advice notion

as this doesnt outright DRIVE actions, it just provides high level weighted advice with context, i.e Attack X, Heal X, Move To Y, Take Cover At X and then when that unit takes its turn it looks at top rated advice then decides to go with it or not

[winged relic]

I see, interesting approach

[golden kettle]

as if all advice is pretty low or around same amount its more likely that a random outcome wont matter, but if one is far ahead of the others it makes more sense to follow it, but the decoupling of the actions and having it more a steering sort of system makes it easier to just consume anywhere

[winged relic]

do you have influence maps at the moment or how do you handle "if two enemies are close together and a third separated" type situation or does that not exist for you?

[golden kettle]

its just basic distance checks

like the distance utility is literally a Vector2.Distance style lookup

[winged relic]

if you have those scenarios, it maybe worth making up a simple influence map to help gauge the number of enemies in an area. that way you can spot targets alone, which might be more distant but are more vulnerable.

[golden kettle]

(run through a clamper and curve)

[winged relic]

I'm going to have to run, eating into work time. perhaps someone else with more experience will chime in 😉

[golden kettle]

possibly, if I wanted to make it more complex like that I could always query the enemies distance utilities to other allies/enemies and just find an enemy with distances to other allies being low/high etc

[winged relic]

the idea of influence maps is that it can save those calculations by simplifying the problem, but until you build one it can seem hard to picture how they help / can be useful. I'll leave it with you, best of luck!

[golden kettle]

thanks, they mention them in the GDC arenanet thing, but I cant see me getting much benefit out of them atm, all values are cached and only refreshed when they change (bubble up) or a predefined period passes in which they are deemed stale (i.e after a turn) so little overhead in them

[low stone]

Hi am trying to make a enemy AI, am facing a problem with my game crashing whenever the player attacks in the trigger collider of the enemy (this collider should make the enemy move toward the player if he enters the range), i really don't know where did i mess up. but i have a feeling something is wrong with my combat script or the enemy damage script

public class combat : MonoBehaviour
{
    // Start is called before the first frame update
    public Animator animator;
    public Transform attackPoint;
    public float attacking = 0.5f;
    public LayerMask enemyLayers;
    public int attackDamage = 20;
    
 

    // Update is called once per frame
    void Update()
    {
        if (Input.GetKeyDown(KeyCode.E))
        {
            Attack();
        }
    }

    void Attack()
    {
        //Animation
        animator.SetTrigger("Attack");

        //detect enemy
        Collider2D[] hitEnemies = Physics2D.OverlapCircleAll(attackPoint.position, attacking, enemyLayers);

        //Damege them

        foreach (Collider2D enemy in hitEnemies)
        {
            enemy.GetComponent<enemy>().takeDamage(attackDamage);
        }
    }

    void OnDrawGizmosSelected()
    {
        if (attackPoint == null)
        {
            return;
        }
        Gizmos.DrawWireSphere(attackPoint.position, attacking);  
    }

}```

[zealous mantle]

What does the enemy script look like, specifically the take damage function and the one that handles the range.

[real sonnet]

If I remember well, The Sims tackled this with "buckets"

Basically yes you can group Utilities by priority

and assign weight to it

Or don't even consider the next bucket if current is above some threshold

"Eat, sleep and such will always go first before watching TV and socializing"

You can come up with way more reasoners/selectors than "Pick the highest score"

you can have full-random, weighted-random, best or random within the n-first results, first to score above a threshold, and so on

lot of combinations

Hope that gives you ideas

[low stone]

public class enemy : MonoBehaviour
{
    public int maxHealth = 100;
    public int currentHealth;

    public Animator animator;
    // Start is called before the first frame update
    void Start()
    {
        currentHealth = maxHealth;
    }

   
    public void takeDamage(int damage)
    {
        currentHealth -= damage;

        animator.SetTrigger("hurt");

        if (currentHealth < 0 || currentHealth == 0)
        {
            Die();
        }
    }

    void Die()
    {
        animator.SetBool("isDead", true);

        GetComponent<Collider2D>().enabled = false;
         this.enabled = false;
         Destroy(this.gameObject, 5f);
        //GetComponent<Animator>().enabled = false;
    }

    

}

tbh if u have a good tutorial on enemy AI that would work too.

[molten sequoia]

Is the editor freezing?

[low stone]

no the game view is freezing

[molten sequoia]

Game view only?

Can you move things around through scene view?

[low stone]

yes i can

everything else works apart form the game view

as soon as the player does the attack in side the trigger collider (child object) the game pauses

also a weird thing i noticed, i added a new collider on the enemy object (total 2 colliders now) also marked it as a trigger but if the player attacks inside of the trigger collider the enemy still receives damage

@molten sequoia

[zealous mantle]

If the game is literally pausing (the pause button at the top of the editor is pressed) , it means you have a runtime error. Is there no red errors in your console when this happens?

[low stone]

yes that seems to be the case

here's the error message

[real sonnet]

SHould have begun with that 😛

What is line 40 in the combat script

[real sonnet]

Uhm did you add/remove code between the screenshot and now ?

THis line is suspicious and often leads to mistakes

enemy.GetComponent<enemy>().takeDamage(attackDamage);

[low stone]

ah yes i removed the unnecessary start function

[real sonnet]

the error is basically saying that something you expected to be set is in fact NOT. So when you try to access it, you get some NullReferenceError

Alright

So, you think you'll get only enemies with that, but in fact it seems you catch objects that don't have the <enemy> component (script)

(That's not really an AI question)

Multiple ways to fix this, do you know where to begin ?

[low stone]

ofc i don't 🌚

[real sonnet]

My first guess would be you have more things than enemies in your enemy layer or layers messed up. You can begin with logging what's colliding

In the first line of the foreach, try to debug log enemy.name

[low stone]

should i comment everything else?

[real sonnet]

nope

[low stone]

ok

u need the console view right?

[real sonnet]

in the end you would probably cache the script and check if not null, something like this ```cs
var enemyScript = enemy.GetComponent<enemy>();
if (enemyScript) {
enemyScript.takeDamage(attackDamage);
}

So the collider and the enemy script aren't on the same gameobject, right ? Can you show how the gameobjects are set up in the hierarchy view ?

[low stone]

sure

[low stone]

yh the enemy script is on enemy object and the collider is on a different gameobject

[real sonnet]

alright, can you try the code i posted

[low stone]

should i write it in my attack function?

[real sonnet]

Replace this line enemy.GetComponent<enemy>().takeDamage(attackDamage);

[low stone]

alr

[real sonnet]

I think you have the enemyTrigger on the enemy layer too. Up to you to decide if it should be or not ^^

the null check should be enough though

[low stone]

i think that seems to be working but still have an error message in console

and yes i do have the enemyTrigger on enemy layer

[real sonnet]

The first is from the Collab package, I can't do anything about it :p

[low stone]

ah well

TY ♥️

[real sonnet]

So it will work as is, but you could still put it to another layer, to save twice as much collision check per enemy ^^

You're welcome, happy coding

[uncut mortar]

im confused at why i get this error

( the graphics )

( ping me pls )

i have this like this

[surreal flax]

have you baked your navmesh ?

[uncut mortar]

i think so

[surreal flax]

... check

[low stone]

why isn't the pathfinder tool working :( the background layer is set correctly

is there a way to create a navigation grid manually?

nvm i fixed it, silly mistake

:P

ok so real problem, every time i try to make the player jump when the AI path finder is running the player animation is stuck on jump (the player do not jump but can be moved around on the x axis)

i really don't know where the problem is at, probably at the animator

am using charterer controller 2d script for movement

[final sleet]

Does anyone have opinions about locomotion/full body attack animations/rootmotion?
I've got my units using a nav controller and a blend tree for walking, but when they do an attack that's animated with their feet standing still they slide around as the nav controller moves them around.
I can do any of the obvious things like turning off nav and using rootmotion for the attack, using avatar masks to keep the lower body doing locomotion and playing the attack on the upper body, etc. - but would really like to talk it over with someone who's played around enough to have opinions and knowledge about what works well.

[uncut mortar]

When I click bake nothing happened 😭

[torpid stump]

Make sure that the Objects you're baking the NavMesh for are marked as "Navigation Static"

(top right corner in the Inspector there's a "Static" checkbox with a dropdown next to it)
(also, used bad wording there, i don't mean marking your actual NavMeshAgent as "Navigation Static", but instead the Objects where your NavMeshAgent should be able to move on)

[fallen nest]

So I'm having trouble figuring out how to make a good context based steering system. I noticed from a tutorial through Godot that they use a danger value that instantly turns off the interest if there's a collidable object, but the issue with that is if the target is against a wall, it's not going to go towards the target. There was a system someone managed to make where they some how don't use any collision detection, and just try to avoid specific vectors.

Is there anyone who's managed to do that, or have a successful basic context based steering, or local avoidance ai?

I guess the main issue is I can't figure out to get proper danger values

[final sleet]

@fallen nest there's lots of ways to do it, but the devil is in the details.
What's your project, how can the AI move, what sets the targets, have is the ai pathfinding, etc.

Was the tutorial that 2d game with the skeletons?

[fallen nest]

Not sure which you're referring to there, but it was a Godot tutorial talking about how to implement context based steering which contains a similar way to how I want to do it. Basically, the agent is suppose to be able to go through different directions (no rotations), but to figure out which direction to go, it's meant to act on a dot product that favors going towards a target position. However, you need a danger value to counteract the value to not run into a wall, and figure out a way to assign that danger value so that it only applies to actual danger and not when say it's hitting a wall that you need to go towards

And that's primarily my issue, figuring out how to make it so the danger value works properly enough to where we know it's okay to go in that direction despite it hitting an obstacle

[final sleet]

I think the way he got it working was to calculate a value for every direction

If it gets you closer to a desirable goal it gets a positive value. If it's pointing a bit towards the goal it gets a smaller positive value. So that allows multiple desirable goals.

And then if there's a wall or something a little undesirable there's either a penalty, so it'll prefer to avoid running towards walls...

Or if there's something to avoid like an enemy he was just clearing that direction entirely so it couldn't be chosen.

Which means if the enemy is between the AI and the goal it can still move towards the goal at an angle to avoid the enemy.

I don't personally use that style of avoidance, I believe there's some local Maxima, like if you move towards the goal at an angle then you might get to a point where the enemy is near the goal and you can't walk closer and get stuck. Usually you build in a strategy for that, like circle to the right or some randomness to add in motion that hopefully breaks the deadlock. It also struggles at handling walls and multiple instances.
But those might not be issues depending on your game style.

[fallen nest]

Yeah I mean the way I've been really wanting to do it has been to scale the different directions using the dot product minus the any undesired direction and going based off of the most unobstructed direction to get to the target. However same issue applies, how do you figure out how to use the undesired positions at

Because the main issue I'm having is using A* but not being able to figure out how to avoid other agents so they don't run head on, or go in the same direction for that matter

[final sleet]

Hmm, OK. So you know how the dot product will give you the component of a vector perpendicular to another vector?

Actually, what I suggest you do is forget about dot product for now, and figure out how your logic should work and how you can show it on the screen.

If you're trying to get to a goal while avoiding an enemy, can you do something like drawing a green circle around the goal and red circles around the enemies.

Then a line in each direction you can move. Something visual like that so you can see how it's making decisions and spot if something is going wrong.

Then you can get the angle between the possible direction and the goal to get the desirability of going that direction. Smaller angles are more desirable.

When that's working start layering on your other requirements. A small angle between a direction and an enemy makes that direction less desirable. Closer enemies have a bigger impact. Walls make a direction less desirable.

Do you need to pathfind around walls? If so you can astar a path and each turn in that path is a potential goal. Can't go direct but can go to that turn in the path? Great!

After the logic is working properly you can worry about making the calculations more efficient and using dot product.

But if you're using A* already I'd probably just use even more A* as my solution rather than something like this.

[fallen nest]

I'm not sure what you mean by angle though, why use that instead of dot product? Because that's what the dot product is basically

Also that's the other issue I have in figuring this out

[final sleet]

Up to you. I find it easier to work with angles when figuring it out, but yes the dot product is a more efficient way to get similar info.

[fallen nest]

Figuring out how to track walls and enemies

[final sleet]

That'll depend on how you built your game.

[fallen nest]

Because I've tried raycasts, however their very inefficient, they can get the agent stuck on walls from clipping the corner

[final sleet]

How are you building your nodes for the astar tree?

[fallen nest]

I use Aron granburg's a* so decision making is done through that

[final sleet]

Aha

That's a great asset, but throws a spanner in the works

[fallen nest]

Yeah it works perfectly for one ai, but I'm trying to figure out a way where multiple enemies don't bump and run into each other. Aron has this in their $100 pro package with RVO/Local avoidance ai

[final sleet]

I was about to mention that

[fallen nest]

But I'm just mostly trying to do something simpler and don't have the money for that lol

[final sleet]

It's good for traffic jams but won't make the air avoid going near enemies

So you probably need your own solution anyway

[fallen nest]

The one method I was wanting to go towards is explained pretty well in this video https://www.youtube.com/watch?v=6BrZryMz-ac

[final sleet]

Is your level a grid - or could it be a grid? If you know how to program it might be worth writing your own A*

[fallen nest]

It is, it's using tilemaps

[final sleet]

It's a fairly useful skill to have, because you can then pack a lot of information into the pathfinding graph.

[fallen nest]

Probably could, I'll have to think about it

[final sleet]

I've got to go to bed, we can pick this up tomorrow. But if you're not scared of reinventing the wheel what I'm thinking is:

• make your own A* graph from your tiles. Then you'll know where the walls are by seeing which tiles aren't connected to their neighbours.
• make your A* pathfinding work with your graph
• Add weighting to the level data (moving over grass is more desirable than moving over mud)
• then add weighting for things like danger. Moving near an enemy is less desirable than moving away.
• since weighting in A* is essentially the same as the path being longer - i.e. it has a higher cost, your AI can then circle around an enemy to get to the goal, but can still pass close if there's no other choice. And they can choose a further away goal with less enemies.

That tutorial has fun movement but no real pathfinding. Note how they all move in existing open areas. The obstacles just stop them moving, but they have no smarts about going around obstacles.

You could add something like that on top of A* if the pathfinding tiles are large enough, to decide on movement within the tile. But I probably wouldn't. I'd just put a pathfinding cost for moving over other units or something and that will already cause them to spread out.

And recalculate periodically, since everything is moving. Don't just calculate a path and assume it will stay clear while you walk it all the way.

[alpine glacier]

Ogod I wish you didn't post this, because now I'm rethinking my AI...

[final sleet]

It's cool isn't it? Made me rethink my AI too!

[alpine glacier]

https://www.youtube.com/watch?v=gZUDzlKdtf4

I was using this, which works in a similar fashion. But I love how they strafe around the player.

I came here to try and figure out weighted attacking... and now I'm like, I want this for my movement...

[final sleet]

You can bias movement like that with other methods too

[alpine glacier]

Yeah, I just need to look into it now.

My current problem with the above method is that when i started to implement weighted attacks, I found that they would attack, that attack would be weighted less, then it'd get to the point where they're waiting a long time between attacks as that weighted value is now so low.

[final sleet]

I'm currently working on combat and I'm planning to have them circle like that, inspired by that video... cavemen hunting a mammoth...

But with different techniques

Gnight!

[fallen nest]

Lol, I know, it's such a simple but really good ai system. They explain how they did it pretty well, but I still really get how they actually did it.

[alpine glacier]

Yeah Im rewatching the video now to try and figure out how to code that now...

At the very least, it'll replace my movement system.

[fallen nest]

If you can replicate it, I'd be shocked cause I've been spending weeks trying to replicate it lol

But I just can't seem to figure out how they did it, unless they did what I think they did

[alpine glacier]

Realistically, it shouldn't be too hard. It's VERY close to what I already posted.

But they're using some different/smarter things.

[fallen nest]

It shouldn't be, but my main issue is figuring out how to know where undesired positions are at

cause they said they don't even use any form of obstacle avoidance, it's purely by coincidence that it works that way.

[alpine glacier]

If you refer to the "Forget Real AI" video I posted, its probably similar to that.

[fallen nest]

Yeah I'll have to check that one out

[alpine glacier]

Scan the scene for boxes, apply negative pressure to them to push them away.

Though the smarter solution wouldn't be "Scan the Scene" and probably just a Physics2D.OverlapCircleAll and go from there.

EG like this.

Player being the blue circle top right. Green Circle being the OverlapCircleAll

[fallen nest]

Doesn't a overlap circle only work if there's multiple objects, not tilemaps?

[alpine glacier]

We talking about Unity Tile Maps? In which case, if it has a Collider it should hit it.

[fallen nest]

Yes, and it hits, but tilemaps are only one object

[real sonnet]

You can sample positions around your agent and draw raycasts from there

[alpine glacier]

^ Also works.

[warped idol]

I've been looking into implementing this as well lol

[alpine glacier]

I'd actually assume Raycasting might be the better option here.

[fallen nest]

That's the way I've been doing

Like basically I would use raycasts from the position of the agent to the direction vector, though boxcast is the only one that really works because raycast doesn't keep track of corners

[warped idol]

https://kidscancode.org/godot_recipes/ai/context_map/
Even though this is in Godot I used it to help implement it Unity

[alpine glacier]

Nice find.

[fallen nest]

Yeah that's the article was looking at just a few days ago and been trying to think how that would work in Unity, only issues with that system is that it's seemingly using rotations, as well as causing issues if you run into anything by negating that direction

[real sonnet]

Article looks good

[alpine glacier]

Would you say use CircleCast or just individual raycasts?

[fallen nest]

If you're doing tilemap, it basically has to be raycast if you're doing collision detection

[alpine glacier]

Ah yeah, I misread how CircleCast works.

[real sonnet]

Raycast is enough. Maybe shape cast for a big boss eventually ?

[warped idol]

I setup direction vectors and use them for raycasts

[fallen nest]

That's nearly exactly what I have, though I add the agents position to the Cosine an Sine

[warped idol]

I do
position + _contextDirection[i] in my raycast checks

[fallen nest]

Ah I see

[warped idol]

but yea same thing

[fallen nest]

But yeah no, my problem isn't necessarily figuring that out, it's figuring out how to properly add danger to the mix. Can't use dot product with the collider, can't use distance between the collider and the agent, and can't do anything that would instantly negate a direction.

Though unless the distance is used like in the spring method video, perhaps. I'll have to definitely try that out

[warped idol]

Right now for me the danger values are just 0 or 1 if the raycast hits

I'm using this for 3D so that why I create a Vector3 for the direction

Could use distance to scale between 0 - 1

[alpine glacier]

Sure wish I paid attention in math class now. :^)

[wary ledge]

How do you set your interest?

I'd assume it would be something like, get direction of player, get the closest angle assign it at as 1 and then slowly reduce around it?

[warped idol]

This is just for following a target atm

[wary ledge]

Yeah that's fair. I was curious. I guess that's where the dot comes into play

Thanks for the reply though!

[warped idol]

Right now I'm zeroing out a direction with a danger value but I'm thinking about changing it

[wary ledge]

I assume as well that if I add a "Danger" to the player as well, while still keeping them interested, that might make them strafe around the player as well.

[warped idol]

potentially. I haven't figured out cool ways the set the interest yet lol

[wary ledge]

Lol

I'ma play around with it tonight, I might pop back in.

[warped idol]

Cool

[wary ledge]

Cheers for the code sample though, was struggling at the setting the interest.

[wary ledge]

Just out of curiosity sake, have you done any debugging with your code? Like drawing rays?

[warped idol]

I'm doing that now

[wary ledge]

I feel like some of it is off from me mucking around with it.

[warped idol]

I did add something to the GetContextDirection

[wary ledge]

Seems to be working as expected now.

Though I am having issues where its still getting stuck on some things.

Or it can't choose between going left or right.

[warped idol]

That's where you'd play with setting the interest values

[wary ledge]

Yeah, even just drawing it out still seems wrong though. Like somethings missing.

https://hastebin.com/ewufiyogur.csharp

This is what I got currently.

Things that I noticed:

If the Vector2.dot is a negative it seems to fuck with this, as such, I fixed this by still having some very low interest to go backwards wherein the rare occasion, you'd want to. I need to smooth it out though.
Also, I feel with how I calculate my direction (EG: Just getting the best) is the better option? I could be wrong here though, but I feel thats what he was doing.

[warped idol]

I change the range of the dot product values to be 0 to 1 instead of -1 to 1

[wary ledge]

Yeah I noticed that.

The current problem I'm attempting to fix is the jittering. As once it reaches the target, it just... jitters back and forth.

[warped idol]

Yea that's why I blended the direction interest vectors by averaging them

[wary ledge]

My idea was weighting the direction to be higher in whatever velocity youre already traveling.

[warped idol]

could do that as well

[wary ledge]

I feel that doing that might add for some interesting physics to it as well. Via if they get knocked back.

I just dont know the best way of doing that would be.

Oh... Well...

That was easy, and turns out exactly how I'd expect... Wow...

https://hastebin.com/ajaloricin.csharp

Give this a shot @warped idol

I have my Distance set to 1 and Howmany set to 18 in this case.

[wary ledge]

Current problem I need to solve now is stopping them from getting too close.

[barren latch]

What is the best way to handle AI in other scenes?

For example I move from scene 1 to scene 2, wait for 10 seconds and then move back to scene 1. I want the enemies in scene 1 to have moved as if 10 seconds in that scene had passed

[wary ledge]

Store the time you were in Scene 2, do the calculations to teleport them when you go back to Scene 1.

[barren latch]

That could work

[wary ledge]

And of course store their Pos.

[barren latch]

It may be a bit more complicated if i still want them to have fought/ damaged things but it wouldn't be too difficult to come up with an approximation

[wary ledge]

At the end of the day, just store all that information.

I know Unity has the option now to load multiple Scenes. That data might be stored if loaded/unloaded Async? But i'm not sure.

[barren latch]

Ill have to look into that. I'd like to maybe have it so that my main hub is always loaded to process npcs actions etc

[wary ledge]

@warped idol , Any more luck/progress yourself?

[warped idol]

Nothing new yet

[wary ledge]

if(hit){
                danger[i] = distanceFromTarget / (1f + hit.distance * hit.distance * hit.distance);
            }

and

interest[i] -= danger[i];

Produces some interesting results.

for(int i = 0; i < interest.Length; i++){
            if(interest[i] != 0){   
                Debug.DrawRay(transform.position, rayDirections[i] * interest[i], Color.green);
            }
        }

        for(int i = 0; i < interest.Length; i++){
            if(interest[i] != 0){   
                Debug.DrawRay(transform.position, rayDirections[i] * danger[i], Color.red);
            }
        }

Inverse cube on the distance from the danger target makes it so the closer you are to said target, the more it'll push you away.

[warped idol]

Nice! I was going to look into that as well

[wary ledge]

The other thing I noticed was that the https://kidscancode.org/godot_recipes/ai/context_map/ uses a path. While this will help it move out of the way of some things, if you give it a full wall, it gets stuck. While I (Or we) could add A* to build a path to the target and use that as the target vector, that seems sort of moot at that point.

It looks like in Game Endeavor's video, he takes the the opposite of the negative value and adds it to the other side. (EG: If there's a 0.5 negative value at 90 degrees, he subtracts 0.5 from the intent at 90 degrees and adds it to 270 degrees. See 3:50)

Or something like that, because at 2:59 in his video it seems slightly different.

But I am 90% sure he selects the best line, vs the calculated normalized vector.

[wary ledge]

(Mine Right, his left. Size and everything matched up via screenshot)

Well I'm done for the night, this is the best I got. For some reason my values are a little bit further to the right when they shouldn't be.

I cant figure out the exact reason tonight, but, its at least progress.

[fallen nest]

That's awesome, that's literally what I've been trying to go for

Did you have two sets of lines, one being danger, and the other being interest? Cause that's what it's looking like from your side.

[wary ledge]

I'm half asleep, but yes. Two lines overlapping with the red taking priority

Blue being the chosen direction

[fallen nest]

So it's two sets of arrays like the Godot tutorial

2 questions though, are you using the dot product of the danger area, and how are you going to detect the danger in an efficient way

Unless you place red marks around the level as obstacles

Which has been in the back of my mind to do, but it feel like it wouldn't work effectively or even properly considering you'd have to add enemies avoidance mark

[fallen nest]

Hey btw, do you know how they did the GetDirection function? I thought it was to maybe normalize the two vectors, but that doesn't seem to work right

[wary ledge]

Right now I'm just doing a circle overlapping function for testing. Raycasting out works just as well.

The only problem with raycasting is it can miss smaller targets or targets that are further away.

[wary ledge]

Simple fix to that though is Raycast your objects and circle collide with anything else.

¯_(ツ)_/¯

Orincrease the amount of rays

[wary ledge]

Figured it out. Smiles.

He seems to make his dot value tighter though for things he dislikes.

But yeah.

[wary ledge]

After playing around with it a bit.

I feel that there might need to be A* still involved.

[wary ledge]

(Confirmed, I just popped into Game Endeavor's Discord and he replied to my questions)
He uses a Node Based A* for path finding to move around town/stay on the main path etc, while keeping things simple to avoid concaves etc.

[fallen nest]

Yeah I seemingly can't figure out how this person managed to do their movement system either. Since it's a normalized vector, I can't figure out if I'm suppose to use that with the method I've been doing where I get the dot product of where the agent wants to go towards the desired position

Or if it's suppose to be applied directly on to the agents position

[wary ledge]

You take the normalized vector, make that your velocity, and multiply it by your speed.

[fallen nest]

Just tried that, it seems to get stuck on only two vectors

[wary ledge]

Show code?In that video specificly, he shows you all of his code.

[fallen nest]

    {
        Vector3 intention = Vector3.zero;
        Vector3 tp = transform.position;
       
        Vector3 direction = GetDirection(tp,targetPos);
        float distance = Vector3.Distance(tp,targetPos);      

        float targetDistance = 0.1f;
        float targetSringStrength = (distance - targetDistance);
        intention += direction * targetSringStrength;

        foreach (Vector3 obstacle in obstacles)
        {
            // GetDirection gets the normalized direction from the hit.point from tp
            Vector3 direct = GetDirection(tp, obstacle);
            float dist = Vector3.Distance(tp, obstacle);

            float springStrength = 1f / (1f + dist * dist * dist);
            intention -= direct * springStrength;
        }

        if (intention.magnitude < 0.5f)
        {
            return Vector3.zero;
        }
        Debug.Log(intention.normalized);
        transform.position += intention.normalized * speed;

        // Ignore this for now
        return intention.normalized;
    }```

[warped idol]

@wary ledge So what is your logic looking like at this point?

I figured you made changes after talking to Game Endeavor

[wary ledge]

Nothing as of yet. I've been fucking around with Shaping. But uhh give me a few here.

[warped idol]

Cool. Yea I was messing around with shaping but nothing really worked yet lol

[wary ledge]

Play with the float springStrength = 1f / (1f + dist * dist * dist);

Here @fallen nest - https://hastebin.com/godotidiqe.csharp

[fallen nest]

Now it's seemingly pushing away from the target even though I have a low target distance

[warped idol]

looks like your direction vector might be backwards

try switching the tp & targetPos in your GetDirection call

[wary ledge]

^ what he said. Lol

[wary ledge]

Ignore the shitty quality gif.

[wary ledge]

https://cdn.discordapp.com/attachments/852925569095041066/852991166081138718/unknown.png

Last one, negative vectors working with it.

At this point in time, it's going to be all about tweaking some values here and there but I'm happy with the final product.

Next up for me is a weighted based action system. Afterwards who knows what. But hopefully I'll be able to come up with some ideas.

[fast sail]

How would I modify/make the navmeshagent to only move 8directionally?

[wary ledge]

After spending all day/two on the above, part of me is thinking I might want to go with the other option I had...

FML lol

[gleaming imp]

hey I'm bran new to unity and just getting into C# and well I have a small game I'm working on and I cant find a decent tutorial on a 3D ai can somebody here maybe pls help me?

[narrow nexus]

How do I make an object follow the player

[gleaming imp]

Same question here

[real sonnet]

There's A LOT of ways to achieve AI. And I mean it, each and every game on earth could have its own implementation. You could watch dozen of tutorials and pick what applies to your game. As a beginner I would recommend to start with some "beginner scripting AI" course on learn[dot]unity, it's free

[civic silo]

Hello, I'm looking for help with my AI that I'm currently working on.... I'm pretty much a beginner with coding and no idea why its not working, is anyone able to help?

[winged relic]

Rule#1 Don't ask to ask
Layout your question / problem as best you can, provide screenshots or code snippets where appropriate. If someone can help or provide insight they will 😉

[civic silo]

sorry! thank you for letting me know

[hollow path]

Hey, I have a question. I'm tying to make a werewolf ai with navmesh etc, which for now will just chase player. But it teleoprt on random place and just stand among trees and rotate sometimes. Did somebody know what's a probably problem?

[final sleet]

@hollow path are you setting a destination for the werewolf?

And have you set the terrain as static and baked the navmesh?

[hollow path]

Yes

[hollow path]

Yes

[final sleet]

Is it walking at all or just staying still?

Is it going to where the player was or some specific spot like 0,0?
Are you using rootmotion movement or letting the nav controller move the unit?

Are you setting the target destination every update or just once?
Is the stopping distance....

@hollow path oh! Your radius is WAY too large!

That's the radius of the werewolf movement agent. It should be something like 0.5

It probably thinks the werewolf is so large that it's already standing on the player.

[hollow path]

Yeah, I know I've fixed that, but it's still acting like a tree and just rotating to see a player

[final sleet]

So it's rotating to face the player, but not moving?

Consistently? If you move the player past it turns to follow?

Is the terrain using the walkable layer mask? You might want to just set the mask to "everything" until it's working.

[hollow path]

So it should just follow player, which dor start is on the middle, but player is moving, I'm telling him to go to player on every frame, I'm letting navmesh and nav agnet to control it, just saying where it should go by script

[hollow path]

Yes I'm using it

[hollow path]

Kinda, it's rotating something like 100-200 degres

Can I just call to you on dc, it will be easier, if u can do that

[final sleet]

@hollow path yeah, this seems specific enough you can message me. I'm happy to video call when I'm at my computer, but I'm in Oz timezone.

[hollow path]

That's a little problem. I'm in Europe time zone (west europe), so you are 11+ hours to my time, in which hours are you online (I mean your hours)?

[final sleet]

I'm around most days from about 11am to midnight. But I check discord between other tasks so I'm unpredictable.
I rarely know when I'm going to get a block of time in front of the computer.

Best bet is to DM me, share the project or upload or whatever works best for you so we can do as much as possible without being online together, and if you spot me online poke me and ask if I'm able to take a look :)

[drifting holly]

if i were to hypothetically make a very large collection of NPCs, and i do mean very large, would i be able to use my GPU to process them?

[winged relic]

you could run a compute shader to do certain calculations etc. but I'd probably be inclined to look at multi-threading on the CPU with unities job system. Depends on target platform I guess though, nothing to stop you from doing either

[hollow path]

ok, I'll mesage you private when I'll be ready

[final sleet]

If you reused the same objects for the NPCs you can use GPU instancing. If you want them to be skinned meshes it gets trickier and you need to use special shaders to store the animation data as a texture.

[crimson oxide]

My navmesh agents keep stopping and jitter in place if the target is on an area that's unreachable. The behavior didn't show up until today, and no changes were made to navmesh agents. Any ideas?

[alpine glacier]

Hello everyone, can i get some pointers on AI coding? i've been using statemachine for it, but it seems unnecery in some placeses and completely essential on others. Is there projects and/or videos that you can recommend?

[ruby owl]

how do I make AI cars?

[zealous mantle]

Make a player controlled card, then instead of using player input, use AI input.

[fallen nest]

You'll want to look at AI steering for that, many tutorial videos go over the basics of local avoidance with that.

[ruby owl]

ok thx

[fallen nest]

Hey by the way, did the script you sent me work for you? Doing what you have in your script basically, but still getting weird results

Such as the agent going into the obstacle position, which was really weird

Instead of the target

[fallen nest]

Never mind I got another way (sort of) working using this method

[raw tiger]

Hello i do a zombie game where on the CIRCLE zombie SPAWN and have to go on the CAMP for kill the player.
So can you suggest to me the best way to do this with MLAgents toolkit ?
I have add invisible wall for create the way to follow to the camp.
but i need to priority to attack the player when he are in good distance of him.

[mighty glacier]

Hi, I am building a turn based 3D shooter game and wanted to design an AI where enemy soldiers can move to a position where they have line of sight to one of the player's soldiers so that they can shoot them.

I set up unity's navigation mesh and can get the enemy soldiers to move close to the the player's soldier(s) during their turn but I was wondering what the best way to implement getting them to a position where they are able to shoot the player's unit. The nodes in the navmesh are pretty sparse, I was thinking of creating additional nodes inbetween the unity navmesh nodes and testing line of sight at each node before getting the unit to move to a node that is a) close to their position and b) able to see the player's unit from that node.

My questions are: 1) Is there a way to place additional nodes between generated nodes within the unity navmesh so I can reuse the built in pathfinding for shortest path? 2) Am I approaching this the wrong way, is there a more flexible way of achieving the above without the sparse granularity of using nodes to determine positions where the enemy units can move to?

[sharp mulch]

does someone know how to fix the navmeshagent sliding at the end

[crimson oxide]

You can change the agent's acceleration speed. This affects both acceleration and deceleration. So if you have a high accel speed (like 50f or something), he'll reach his max and min speeds pretty much instantly, causing an instant stop when he reaches his destination. Also turn on auto braking.

[polar compass]

Any idea what would cause this to happen in the baking?

(The building also seem to be entirely ignored)

[untold helm]

Cool, I like it!

It's not easy for training mlagent, it takes me about 2 months.
And some glitch are not sure if it is relate with checkpoint?

And AI car is flashing...
https://youtu.be/o0csiVGEuyQ

[alpine glacier]

So i have this enemy ai

And i want it to point towards the player

But every time i try it off by alot of degrees

Ive used lookto rotate and various ovr methods

[analog goblet]

I have a bit of advice I'd like to share with you guys about Unity 2D's composite colliders and the A Star pathfinding system:

I have recently discovered that when using the A Star pathfinding system in the Unity game engine along with any 2d colliders combined into a composite collider, the navmesh resulting from generating a scan from A Star has many errors and missed/mistaken areas that should or should not be walkable. To fix this problem, I have found that changing the composite collider's geometry type to polygons instead of outline patches up these mistakes reliably.

Of course, this will not do anything if your main A Star script has its circle collider diameter set too big to generate any meaningful pathways throughout your tilemap but I have provided the solution for any people who have done this correctly and are stuck. Hope this helps!

[half sigil]

Whenever my enemy gets hit by a bullet and it pushed back, it's pathfinding breaks. Any help appreciated.

[torpid stump]

How do you handle pathfinding? NavMesh?

[half sigil]

I have a 2D Top down game and I use A* pathfinding

[twilit saddle]

You'll need to calculate a new path after the knockback of course

[wild shell]

why is the navmesh bake button cant be clicked at all

[severe fable]

Have you looked in the Navigation window under Window>AI

[wild shell]

somehow I just fixed it, I make the objects static

idk I did it before and it wont work

[severe fable]

It's deffo important that all navigable surfaces are Navigation Static

[wild shell]

yeah I did it before but not working but now working. idk lol

[slow plinth]

Love this channel

Only questions

[raw tiger]

😢

[hazy oar]

This is not a neural network or anything, its a minimax algorithm.
I am trying to make a AI that plays checkers with different "strategies" which right now is just best and worst.
The problem is, Player 2 always seems to loose. They sometimes seemingly randomly make a terrible move, letting Player 1 capture pieces for free for seemingly no reason. I always run it at depth 6 or more so the issue is not that it doesn't see it. Perhaps it could be an issue with the Alpha-Beta pruning, or maybe I use the wrong variable somewehere.
https://pastebin.com/XugNKNGt
The issue is in the Search() function. The ApplyStrategyPart modifies numbers so if I want it to pick the worst strategy, I can invert the score the AI gets, and other stuff.

Huh, if I remove just one piece Player 1 suddenly looses quickly. Now, that would suggest that it is working fine, and it is just an advantage from starting first, but when it was balanced the mistakes it made were easily avoidable, and it even recognised that it was loosing right after the move as once the oponent moves the evaluation goes down a lot (for them).

[sinful belfry]

Could I pick everyone's brain with a Unity programming task I'm trying to accomplish? It's not related to a dev test or any kind of employment thing, so no rush.
I'm trying to get vehicle AI to follow targets around a track.
Once it collides with a target, I want the script to go to the next target.
Should I put my target objects into an Array, and just have a function get the next object from the array? My C# is a little rusty, but if the function gets to the end of an array, it should circle back around to the first item, right?
My end goal is to have the vehicle AI go from target to target until it does a lap, and then do it all again until told to stop (IE the race ends).
Any suggestions on how to approach this would be much appreciated. I see like a dozen different ways to do vehicle AI, but I felt like a target-to-target system would be flexible for any track ideas I had, especially if I want a more open world approach.

[stone owl]

sounds like you want a simple waypoint system.
https://www.youtube.com/watch?v=GIDz0DjhA4E

_
__
Detecting if agent's current path to destination is in a straight line, how would I do that exactly?

[real sonnet]

I'd try with the cross product

[hybrid hamlet]

Is there a guide to writing a good AI? All I have is a navmesh agent, types of environment props (hide, climb, throw) and aiming while shooting. It feels to basic

[ruby current]

Make it do a 360 no scope!

[hybrid hamlet]

Lol

There are no scopes

So I guess that'll be easy

[modern orbit]

Hello! I wanted to make an HP Bar for AI but my problem is that when I either get behind or beside the AI the HP Bar (Slider) isn't facing me. I want it to be facing towards the camera so I'd know how big their HP is without having to go to their front.

Please feel free to ping me if someone's going to answer me.

[tribal agate]

https://wiki.unity3d.com/index.php/CameraFacingBillboard Here's some ideas to get you started

[stone owl]

How does it deal with a player's presence? You say it can hide, when does that happen? Have you tested it, what did you find fun and not so fun?

an AI that can deal with (or pretend to in most cases) a player's dynamic decisions, all the unpredictable choices in movement they make is a lot of fun

eg we can hide if we see a player at a distance and we have no ammo, but what if the player runs full speed at the AI? hiding would seem silly since the player is running with it, instead maybe do a shove, roll to the side etc

[tawdry cairn]

Is it possible to implement 2d pathfinding system with no grid?

[noble fog]

A* pathfinding for example node based. Nodes could be anything, they can be grid, voronoi, etc.

[hybrid hamlet]

Right now I have several behaviors and states defined and I'm just combining them as I go along to mix and match until I find something I like. I have a rare and random behavior where the enemy will just wave at the player when it detects it can't win.

It's one of three rare and random behaviors. The other two are cowaring away and dancing. Just to spice it up. 😄

[tawdry cairn]

That is, for A* implementation, I have to divide the map into certain areas, like nodes?

[noble fog]

yep

[tawdry cairn]

Many thanks

[stone owl]

Haven't seen that yet personally that's pretty cool!

always nice to have surprises

[hybrid hamlet]

Speaking of A*, what does the F cost stand for? I can't find a clear answer.

I get that the H stands for Heuristic. The G.. Actually I also want to know what the G stands for lol

[real sonnet]

F is a math standard notation to express a result (here the cost) as a Function of smthg else (here a node)

then G is used because well, F is already taken, so you just increment in alphabetical order :p

Then H is used for the same reason G is already taken.

Nothing to do with A*

[stuck scaffold]

hi i am making a tps game i have a ai that shoots when the player is in range and also moves to different points but my game is tps so i need it to hide take cover run behind run ahead etc can someone help me with a basic code or any youtube tutorial

[rustic vortex]

Try looking up "AI State Machine" in google and/or youtube @stuck scaffold

[crimson oxide]

When using the Unity Navmesh, if a target is above the pursuer, the pursuer goes to his xz location, but doesn't actually try to use nearby offmeshlinks to try and get up to the higher area. Is there a way I can fix this?

[stuck scaffold]

ty @rustic vortex

[hybrid hamlet]

I think the H being heuristic, starting with an H, is just a coincidence then

Or the H was there first and F and G got added after :p

[sinful belfry]

So several days ago, I asked about a vehicle AI system using way-points. I had a few people graciously help me, and I actually got a very rudimentary vehicle AI system working. However, this presented some flaws. The point-to-point method combined with my physics based vehicle controller turned into a sloppy mess. My boat will launch past points due to its speed, over compensate in turns, and jerk violently towards the next transform point.
I see a ton of tutorials that implement a bezier curve for objects, but I'm stumped on how to make an object follow the curve using physics forces, rather than transforms. Example: Have my boat object apply a negative or positive torque force to turn onto the curved bezier path. Secondly, have it accelerate or brake to stay on the path at a good pace.
I know the information exists somewhere on how to do it, I just need to find the right tutorial haha.

[maiden dust]

is there like a base enemy ai for fps games that unity provides?

[sinful belfry]

Check the asset store, I am 100 percent certain they have a very basic Doom style enemy AI package.

[maiden dust]

ok thx

[winged relic]

I'd suggest increasing the distance from the waypoint which triggers next waypoint selection, you also need to manage speed if making a turn. Simplest is to either slow down based on distance to waypoint or have it as a parameter of your waypoint which the boat can use. I did this for a very simple demo I was making, but if you want something more inteligent then may need to do some research.
You could for example, where you have bends the waypoints are closer together look a few waypoints ahead and use some decision making based on distance between points. If they are closer together slow down, just spit-balling here as I didn't have to go deeper for my demo.
there are a few articles on handling vehicles in the following Link which should help give some more ideas

http://www.gameaipro.com/

[mellow rampart]

Howdy, I want to create a basic AI enemy for a VR fps game and am completely ignorant on coding. Are there any Unity assets you would suggest I download?

[alpine glacier]

NodeCanvas/Behaviour Designer.

[dusk meadow]

Does anyone here use the a* package? I am struggling with paths being correctly calculated, but entities aren't following the path the way I would expect them to. They usually don't follow it at all and just end up doing nothing.. Example here : https://www.youtube.com/watch?v=yZt6ZFIzHw8 . Why does the entity not follow the calculated, optimal path ?

[dusk meadow]

I found my problem! I had frozen the z axis, but enabled rotation in the A* script. Disabling rotation in the a* script fixed it 🙂

[weak gull]

i stumbled across this forum post. Is turning nav mesh agent off n using created path more performant?

The post is from 2017... so I'm not sure if anything has changed with the api since then

[molten sequoia]

@weak gull If you don't need the agent, you might be able to just do navmesh queries without it

https://docs.unity3d.com/2021.2/Documentation/ScriptReference/Experimental.AI.NavMeshQuery.html

[golden kettle]

Hello again, so ive been doing bits on and off with my utility AI style system and its been going ok, but im struggling to find a nice way to represent the damage or useful application of a skill/ability.

So for example I have an ability attack which does weaponDamage to 1 enemy, however I also have cleave which does weaponDamage*0.6f to 1 enemy AND any enemies on either side.

So currently my utility scorer looks at the potential damage output, which attack would always score higher, but if there was 2-3 enemies in a line cleave would score higher, but its a contextual utility, so does anyone have any advise on how to best express this, as it seems like its too complicated to just expose as a single utility value.

[golden kettle]

Hmm think I will add some sort of runtime modifiers where it has a predicate and a bonus and if predicate is met it applies a bonus to the consideration or something

[gloomy storm]

Does anyone know a good way to implement an inference engine in my project? Not for machine learning like the Unity Inference Engine, but just a forward-chaining rule-based system?

[golden kettle]

sounds a bit like GOAP

where its just actions linking to other actions and if you are at A and you need to go to D you may have A->B->C->D or A->C->D (latter would win)

[fleet crown]

I've made a basic AI, however, it is incredibly buggy and I have no idea how to fix it

code: https://pastebin.com/10BX5Vxp

the buggyness has mostly to do with handling slopes

[verbal gust]

Hello! I'm not sure if this is the best place to ask about a NavMesh issue, but here goes--

I'm using the NavMeshSurface component on an object rotated 90° to make a wall. I've placed a NavMeshAgent on the wall with a basic wandering AI to move it to random points on the wall.

After a short time, the NavMeshAgent will begin to stutter, eventually stopping. If I manually force it to move by dragging it in the Scene view, it "resets" and continues moving normally for a bit longer.

I tested it on my ground, and it worked perfectly, so there's something different about the wall. I don't have any Rigidbodies, so physics shouldn't be what's causing the issue. It's not a framerate issue--another effect is working perfectly in the Game preview (not visible in the attached video).

I've tried adjusting the Voxel Size and Tile Size, with no noticeable difference. Turning the Agent's collider on and off made no difference.

Any help would be appreciated, and please let me know if you need any additional information. (In addition, if anyone does respond to this, please ping me, I might not see your reply otherwise).

Object components, for reference: https://imgur.com/a/8lBNHFS

[zinc zenith]

try increasing mass

if it has a rigidbody

[fleet crown]

that doesnt seem to work

[zinc zenith]

there is a parameter in rigidbody which you need to increase so the game object can climb

check in brackeys fps tutorial

@fleet crown

[fleet crown]

Ill try that later once sourcetree is done its 20 hour commit sequence

Its taking its time

[mellow rampart]

Is this what you were referring to?

https://assetstore.unity.com/packages/tools/visual-scripting/behavior-bricks-74816#description

[alpine glacier]

@mellow rampart no.

If you type either name into the asset store you get the exact assets I talked about...

Ad·https://assetstore.unity.com/shop-behavior/designer
Behavior Designer - A Unity Asset Store Exclusive

https://assetstore.unity.com/packages/tools/visual-scripting/nodecanvas-14914

[brazen stump]

how would I go about pathfinding on a NON grid based 2D game

side scrolling, not top down

as far as i'm aware the navmesh system only works in 3D

and my levels aren't (and can't be) grid based

[noble fog]

Just basic A* implementation would work, with trigger events on the way to jump up or down.

[brazen stump]

I thought A* only worked on grid based levels

how could you make A* work without a grid

[noble fog]

Any nodes. Platforms can be presented as nodes

NavMesh also uses A* btw. https://docs.unity3d.com/Manual/nav-InnerWorkings.html

[light saffron]

do you guys know a way to batch queries to the navmesh? I need to verify that targets are reachable, in great number

[dense sparrow]

so like [Way1, Way2, Way3] and if Way2 is out of reahc, go for [Way1, Way3] in respectively?

by Way i mean a Point or P if u will

[versed sand]

A*, floor is marked red.
it has a tile collider but otherwise the player and enemies will fall to through the ground.
how do i solve this?

[light saffron]

exactly

[dense sparrow]

Should be failed simple if u take them into a List<P> and just use the remove function there

You could access them all at once with a forEach function on the List.

So , ignoring my syntax mistakes it would be like this :

UpdateList()
{
  forEach ( var point in Points) 
    if (is_distanced(point)) Points.remove(point);
}```

Or if u want a one liner you would do
Points.foreach((p) => { if(is_distanced(p) Points.remove(p); } );

And u may call the UpdateList every N seconds depending on how cpu intensive it would be. If you have a dozen of points I suggest N be 0.2 , if hundreds, N be 1.3 or something. If much more than that, you can go for an asynchronous method which yields after each M checks . M could be big

[wanton silo]

If I want to create an AI with some abilities based on player distance, time or health left. Wich is the easiest way? Ienumerator, stateMachine or i'm missing any easier way?

[light saffron]

how about: 1 agent has 20 potential targets and I want to calculate the weighted path distance of each one?

[light saffron]

give us a tangible scenario

[wanton silo]

@light saffron My bad! i'm doing a 2d Boss

i'm trying with this:

[light saffron]

tangible: tell us what you want the things to do as the various agents react

[wanton silo]

oh, I dont know if you mean this: I want an Idle animation, if the "player" its too far away, walks towards him and attack him, if he's near a corner of the room, change the normal attack to a special corner attack. and 2 or 3 "special attacks" choosen by time (for example if the player is out of range for 3s or more and the health is behind 75%) and if that conditions are met, choose from one special or another by random.

[light saffron]

exactly 🙂

see how you describe your scenario with simple linear "IF"? and then you get into a state which limits the number of options? that's State Machine

special attack with 75% condition to another state

so yeah you got this

[wanton silo]

so using animation attached to StateMachine is the way to go?

[light saffron]

I wouldn't use anything mecanim

maybe try the Unity integration of bolt, its FSM might be easier to use and performant enough, although I read that on 2021.x using bolt incurs massive slowdown on playback and domain reload

I made my own FSM, it's so easy to make

[wanton silo]

I'll check what's a FSM and what's a bolt and what can I do with that, thank you @light saffron

[light saffron]

you're welcome

[dense sparrow]

Oh sorry I just realized for some reason I typed "Failed" instead of "Done" in the message you replied to ... Probably my Android keyboard being corrective...

[dense sparrow]

You know you can just code this exact thing you said here :D just write down the psuedocode and you'll see it's pretty easy from there

[zealous mantle]

Start by defining what an enemy and what it should do.

Without knowing how you want your enemy to act, you're not going to be able to build anything.

[real sonnet]

try to google things like "patrol AI" or smthg like that, will give you ideas

there's a lot of different ways to do it

depending on your game

or "guard AI" maybe

[versed sand]

i have a problem with a*

i wish to have a tileset dedicated to the ground with a collider, problem is that would make it unwalkable in the red and blue squares. any solution?

[ripe gorge]

Im was trying to build a scene from scratch in the FPS shooter template but when i place the enemy and bake the ground in the navigatie tab the ai will follow me when he "sees" me but he doesnt walk around if he hasnt seen me

than he just stands there

[ripe gorge]

not even moving a it

[woven copper]

Hi, I'm making a kind of 2.5D game with 8-directional movement, and I'm trying to figure out how to restrict a NavMesh AI to those 8 degrees. Any advice?

[real light]

Anybody use NPBehave before?

[plucky dragon]

Is there a way to ignore painted "trees" on terrain when baking NavMesh?

[alpine glacier]

Err pal u could maybe bake navmesh before you paint

Delete grass,trees etc and bake it

then begin painting

make sure to back up tho

[brazen stump]

in the AStarPathfindingProject, is there a way to get the pathfinding to account for enemy height in 2D

currently this flying enemy will try to reach the player but get caught on platforms because it doesn't account for the enemy height

[keen crater]

anyone using Aron Granberg's A* solution for creating navmesh at runtime? wondering how it compares against Unity's Navmesh Components

[stone owl]

I'm willing to bet it's a lot better

Aron's take that is

[keen crater]

yeah same

just wish I could test it properly before dropping $100 on it

[stone owl]

Is there not a demo? Feel like there was maybe I'm confusing it

[keen crater]

yeah there's a free version and it contains a lot of stuff but not the runtime navmesh generation that I'm after

[stone owl]

Anyway you'll not find a better one I'll tell you that much. Hes been praised for a while for his work

Well worth the price if you need it

Investment you can take to any project really if it has ai involved, very flexible

..more performant, etc

[keen crater]

yeah for sure

[pseudo crow]

Hi all, I'm having some issues with Emerald AI and it telling me my gameobject is not an active agent, despite my navmesh being baked and a navmesh agent component being attached to the gameobject.
Would anybody be able to help me diagnose the issue?

[unborn olive]

Hi !
I have 2 problems concerning my FSM for my AI and I don't know if it's an issue inherent of the pattern itself. It's about transitions, and the fact that they are hardcoded in the states themselves. That mean that if I make a Patrol state, I usually add a transition that goes to a Chase state if my AI see my player, pretty basic for a melee enemy. Now if I want another AI that shoots when it sees the player during a Patrol without going through a Chase... well I'm kinda stuck. And it's just one example. I could make a quick fix and dirty fix but that won't work for all my cases.
I basically have to make a PatrolChase and a PatrolRanged, which seems very... not okay. I can get the running code out of the state itself and create an Action class and put my patrol code in a PatrolAction, but I'll still need 2 Patrol state that both use the PatrolAction but have a different transition.

My second problem is concerning attributes. Where should I put attributes that only interests some AI but that I need to be able to change from the Inspector ? I have a "brain" component that runs my FSM and serves as a "blackboard", an interface to the scene for my AI, but it's generic, and if I inherit from it, my States wont be generic anymore (since they need a generic brain). In my example, my chaser needs a ChaseDistance and my ranged AI needs a cooldown between attacks (it goes back to patrol while cooling down). If I put both in my brain or my "stat" class that my brain also has access too, I'll end up with a lot of useless attributes on both AI.

Maybe I am overthinking this ? Or is the FSM not adapted ?

[final sleet]

Your first problem is real. If you want an advanced AI it will get worse and worse as you keep adding combinations until it's too hard to add anything new without a huge risk of it breaking.

For the second problem the answer will depend on your design and how you like to write code. You might even have multiple places to store the information.

I'd suggest using scriptable objects to help with both issues.

Imagine having a scriptable object AIData with some method like GetValue("ChaseDistance"), then you can have a PatrolData class that inherits from it and let's you enter the right fields.
Or if you don't want multiple classes just have a dictionary of key, value. Then you can store anything for any AI.
You might need a separate method and dictionary for each variable type.

And throw an error if you ask for a variable that isn't set yet to catch any typos.

To solve the first problem, split your FSM into different tasks. You have the state, you have the state's it can transition to, and you have the trigger to transition.
A state will have one or more states to transition to, and each of those will have one or more trigger to make it do the transition.

So now you can build an interface for a state, trigger and transition. You can have a scriptable object for each, and the AI can have a state as a public variable.

So you can have a PatrolState class with all the specialised code. You can then make multiple objects with different settings, for example if you have waypoints it would follow, otherwise it would walk to random points. You could have speed, how alert they are etc.
That lets you have multiple behaviours. A guard with a patrol route. An enemy that spotted the player and is searching. A character that panics while looking for their child. A grazing animal. Just drag and drop the right one onto your characters.

Then you need a way to set different states to transition to. So guard might transition to chase or attack or flee.
You could store the list in that units AIData maybe. So the guard transitions to chase. The guard with the rifle transitions to attack. The deer transitions to flee.

For the triggers, same thing. Code up scriptable objects and drag them to the transition. So "spot player" or "hears distraction event" or "value is less than" and so on.
So you can then add "spots player" trigger to the "attack" transition and now you have an "attack the player if you see them" behaviour.

Put that as a transition from the patrol state and you have guards.
Put it as a transition from the idle state on the guy at the tavern that the player owes money to and you have a loan shark and a challenge for the player to get through the crowded room without getting spotted or drawing attention.

I'd suggest not including combat in the AI.

Just have an "attack" state, which can trigger your combat class. Let the AI know it's attacking but nothing else. Combat class can then decide where to move or if it should shoot. Splits the complexity into smaller pieces.

And allows different attacks. A horse attacking might kick the player and knock them over rather than using the normal combat system.

And... That's about it. There's more you can do, but you'd tailor it based on your specific game.

[unborn olive]

Oh thanks for the detailled answer ! I think I understand most of it, at least concerning Transitions and Decisions as scriptable objects. If I'm not mistaken, is it close to that kind of implementation : https://learn.unity.com/tutorial/pluggable-ai-with-scriptable-objects# ? And is the PatrolState a scriptable object also ?

[final sleet]

Yeah, my suggestion is using similar techniques to that.
I started by copying that technique to make sure I understood it well, then designed my own solution that can expand better and give me more ways to mix and match to get more behaviours from the same amount of coding work.

But if you start with what they're doing you can update and change the AI if you need more flexibility.
There's no need to split it into as many parts as I did if something simpler will work!

In fact, I don't! I always start simple. Each state might be a scriptable object and might even hard code when it changes to another state.
Then I add the more complex stuff only when I need it - keeps things simple for as long as possible, and gets it working really fast so you can move on to the next job, and come back to keep improving it later.

And yes, each state should also be a scriptable object. Makes it easier to see what state the AI is in and store data about it.

[unborn olive]

Ok ok I see, thank you for your time ! Also, as a side question, how do handle combat, and especially combat animations ? I tend to use animation events a lot to spawn projectiles and activate collider for melee weapons on the exact frame of the animation, but I wonder how other people handle it. Using hardcoded timer to know how far you are in your animation clip ? Another solution ?

[unborn olive]

This question extends to any state where I want to wait that the animation is completed. For example a monster spawning that would go into a SpawnState, play the animation and start its behaviour once the animation is completed. Again, I handle it by waiting an animation event that tells my FSM "ok I finished you can resume your logic". But it seems prone to errors...

[crimson oxide]

Has anyone run into the problem where a NavMeshHit normal always returns as a vector 3 zero?

[slim crest]

@unborn olive there is a course on unity learn about «goal oriented action plans» that teaches how to set up a flexible ai. maybe you can find something useful in threr. Search for goap

[unborn olive]

Yes i'm following the whole course

It covers FSM, goap and behaviour trees. I find FSM very easy to understand but I might have a wrong idea about goap and trees as "over complicated" patterns. Definitly gonna look into those

[fast sail]

How could i make the navmeshagent only move 8-directionally and not freeform like it does by default?

[light saffron]

it's not meant for that so you have to write your own controller and use unity navmesh to feed your controller with a raw path

people usually write their own A* pathing or maybe aaron's astar project has modifiers that will give you this behavior

[earnest forge]

calculate the path to where you want to go and split the path into 2 @fast sail

[real light]

Is there any free waypoint editors available and if so where can i find one?

[turbid panther]

hey guys
I have a question on Unity's path finding:
When I set a destination for an agent that cannot be reached, it is able to calculate a path to the closest reachable point
But when I call NavMeshAgent.CalculatePath to the same destination it always fails with InvalidPath status and never returns a partial path
Is there some gotcha I'm not aware of ? Can you make CalculatePath find paths like SetDestination does ?

[molten sequoia]

@turbid panther There might be some heuristic when SetDestination is used. My guess would be that they use the method that snaps to nearest point on navmesh with a small radius and if that doesn't work, they raycast down and try again.

[charred token]

Hey, I'm having some trouble with some NavMeshAgents. Basically, they sort of go inside each other, even though they all have colliders. I even tried freezing their x position, but it's still happening. Have you guys experienced this?
https://cdn.discordapp.com/attachments/849461722175897602/867978785776087040/Screen_Recording_2021-07-22_at_11.56.25_PM.mov

[noble fog]

Agents have their own (size) parameters you need to set. They ignore actual colliders because dynamic physics is not used.

[turbid panther]

I played around with the NavMesh system some more and turns out that there's a NavMesh.CalculatePath and it does exactly what I need (I was using NavMeshAgent.CalculatePath at first).
It still has some weirdness, like status is never a PathPartial, and always PathComplete, but it doesn't matter since I need to perform a distance check for the last point on the path in any case.

[charred token]

Where do I set these?

The only thing I see relating to size is in "Obstacle Avoidance"

[noble fog]

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

[charred token]

Oh i had the quality set to "none" for some reason

thanks

@noble fog can I make it so that it avoids some things but not others?

[noble fog]

Don't remember, should be in documentation.

[charred token]

I don't see a way. Is there any other way to make them not go inside each other than doing obstacle avoidance? Because that's messing everything else in my game up

[crisp quartz]

hey all, what is the best way to move a navmeshagent onto a new nav mesh? I tried warp but it seems to get stuck on the last one it was on (basically I load a new prefab room in with its own nav mesh and want to just reposition the character on to that.

[crisp quartz]

ah I think I solved it, i seem to have no navmesh in my prefab - can I not bake a navmesh into a prefab?

[crisp comet]

Hi guys, im brand new to unity, ive got this isometric view of player,managed to make him move and now i want to create an enemy to push him off the platform. I know how to create an enemy that will detect the player and follow him so I just need some help on how I would go about coding a script to tell the enemy to push player off platform?

[dire basalt]

Sorry WildRover, I do not know the best way to approach that.

Our Level Designer recently came into a problem with NavMesh in that it did not connect multiple areas on flat terrain (as shown in the image). From what I see from the scene, NavMesh is having problems with generating connections over overlapping meshes. It creates one area on the first mesh, one on the overlap and one on the other mesh, but doesn't connect these areas even though together they form a flat terrain. Do any of you know how we can combine these areas into one big surface as it should be?

[real light]

How would you guyd handle traffic light systems for cars and peds i usually see people use colliders/triggers im not sure if there are other methods

[alpine glacier]

how do I make enemy ai shoot player?

[fossil orbit]

have the enemy move to a position between player and platform center. When the enemy is near the designed position, have it pushing the player.

[fossil orbit]

You can keep tracks of all available traffic lights, then check if a car is about to enter said traffic light by using get distance or something else
But still, I think using triggers is more effective.

[tawdry fog]

is there a navmesh function for rotating only the 'head' part of an animal to face the player?

but only clamped within certain feasible regions to prevent it from bugging out and having a broken neck

[real sonnet]

That's not really a navmesh-suited problem

You can write a little script to have the head track and lookAt the player

I'm sure you can find plenty tutorials on that from YouTube for example

[alpine glacier]

.

[zealous mantle]

Enemy sees player. Enemy shoots bullet in player's direction.

[crisp quartz]

hey all, is there a way I can force a navmeshagent not to cut corners? and take the full turn

lazy navmeshagent

[atomic spruce]

Hi! I have this patrol script[https://hastebin.com/adiremorux.csharp] which moves the character in a fixed path, I have it attached in 1 object that spawns every 15sec.

how can I set a condition between these elements/array?
that if an object is particularly occupying a waypoint the others will stay put until the next waypoint is unoccupied

base on the script

[crisp quartz]

have a flag on the waypoint "occupied", check this before moving anything towards it - if its occupied then just dont move them

[clever flame]

Hi guys, can anybody help me with an issue Im having? Navmesh agents arent moving, full detail is here https://forum.unity.com/threads/generic-navemesh-agents-not-moving-post-now-with-added-detail.1147616/

[lament rapids]

is there a free ai app that I can use for more then a week? everything I see in the app store is $50 or free for 1 week then you must pay so how can I actually use unity for free is I must have an ai app to use it and there is not one for free? 😦

[kind sedge]

https://arongranberg.com/astar/download @lament rapids select the free one

[lament rapids]

thanks 😄 thats what i need

[crisp quartz]

he yall im struggling to pull out these edges more on my nav mesh, what setting should I alter? (basically want more space between navable mesh and wall - thanks

ah it was my agent radius I think, it just wasnt updating properly

[crisp quartz]

can you have tigher edges around the outside but more gap inside?

[kind sedge]

atm my enemys can go through eachother if they have rotation disabled

they like to stack up

nvm rotation doesnt affect it

[weak moon]

How come when I load a scene all other scene's navmeshes are shown?

Is there any way I can just see the navmesh for that particular scene?

[crimson oxide]

That should be the default behavior. Are you loading scenes additively or just opening fresh? Can you try clearing the navmeshes?

[weak moon]

I'm double-clicking them in the Project explorer 🙂

[crimson oxide]

That is odd behavior. As far as I'm aware, when you load up a new scene, things from other scenes shouldn't be loaded up unless they have Don't Destroy On Load, but that's from loading through script

[weak moon]

Yeah I don't know why it's doing it

The navmesh out there is from my TestScene I was building the gameplay in

[crimson oxide]

Have you tried clearing the navmesh and then rebuilding it?

[rancid brook]

hey im encountering some issues with navmeshagent i know knothing about ai and i followed a tutorial from brackys. The tutorial was about an rpg game and ep10 was called ai so didnt watch ep 1-9 i made a nav mesh agent but there wasw no nav mesh so i watched ep1 and brackys made a nav mesh now when i write in the script NavMeshAgent agent; the word agent becomes darker, which mean the it cant find the navmesh agent?

here is my code

https://gdl.space/lipafeheyi.cpp

and the instanse

[weak moon]

Yep, I've tried that. Come to think of it, could it be due to having duplicated the first scene instead of creating a new one?

[rancid brook]

https://gdl.space/usoqirixok.cs

pls help me

im confused

[weak moon]

You need a NavMeshAgent component on your GameObject

[rancid brook]

i have added it

and tried removing it and readding it

3 times

[weak moon]

@rancid brook You're not using the agent variable in the code

You're just getting the component

[rancid brook]

agent is a variable?

[weak moon]

You use it in your code as any other variable

[rancid brook]

ok so i change what exactly

[weak moon]

Depends what you want your NavMeshAgent to do

[rancid brook]

ok

[weak moon]

This for example sets its destination

[rancid brook]

thx a lot

ill try it

lol i hadnt save the code

🤣

its working!!!!!

😀

[weak moon]

Even with a brand new scene I still see the NavMesh from the other scene

@crimson oxide I found the culprit. I checked this little rascal and cleared the navmeshes and then baked again.

[rose pebble]

Yo, does anyone know how to make ranged navmesh agents fan out and attack ala starcraft?

Like surround the enemy instead of trying to fit through

[alpine glacier]

hey guys I wanted make an npc that is killable, but whenever i kill him this console shows up:

he has a nav mesh agent

also this is the technique im using to kill him

basically he turns into a ragdoll and navmesh agent is disabled

but i keep getting this error

btw im doing this for a gamejam so any help is appreciated

[fiery geode]

Sounds like it's either listening to a manager/acting as a hive-mind, or recognizing that a certain spot is range of the target is already taken and then pathing to a new position. NavMesh should be able to detect other units and pick a new position close to it.

[fiery geode]

How often is this being called? Just once when it dies? If so, try disabling the NavMeshAgent first.

[pure oriole]

Hey, I'm creating a procedural world made up of chunks, and I'm creating nav meshes on each chunk during runtime.
When the two chunks are loaded next two each other, the seems look to be connected, but my AI still can't pass to the other nav mesh.
For each nav mesh, I'm using NavMesh.AddNavMeshData(m_NavMesh); and passing in my chunk's nav mesh data, but maybe that's not enough?

The red line shows the border between the two chunks, and the red circle on the right is the agent's destination, but there's "no path".

[atomic spruce]

I have this patrol script[https://gdl.space/amujepunek.cs] which moves an object in a fixed path of waypoints, I wanted to figure out how to disable the movement of the objects that are recently spawned if the current waypoint is occupied by an object.

Ive added some codes (starts at line 114) but I find it really confusing since im new to C#. any tips?

[finite trellis]

Does anyone know of any good resources (open source libraries, tutorials/courses, even cheap paid assets) or suggestions for AI movement patterns? -- Beyond just basic avoidance of obstacles and chase. Things like hiding, circling, formations, and just having a generally interesting / dynamic movement pattern when - in my case - shooting at the player? I'm currently using A* Pathfinding but beyond just getting from point a to point b in an efficient manner I'm a little hung up.

[supple rock]

Sorry to interrupt but I have a short question. I have this map (green terrain), this water (blue, no collider), and the gray plane which is invisible at runtime.
The goal is to make it so terrain above the gray plane is walkable, and below isn't I've tried using modifier, modifier volume and obstacle and none of them work. The walkable terrain ends up below the gray plane not above. Any way to fix this?

Not the intended result

[crisp quartz]

hey all how can I stop the navmesh generating like this - looks broken/wonky

[noble fog]

Could probably add an invisible collider as a railing along the steps.

[fast sail]

if those are steps meant to be walked up, that looks correct. if it's not, adjust your step height

in the agent settings

[true lava]

Hey guys, is there a way to have a NavMeshAgent not attempt to move towards the path? I'm trying to make a physics vehicle take on the Agent's info and navigate through it instead of having the agent move itself

[true lava]

I have already tried just forming a path and storing that (no Agent involved), but the thing is too strict. if the source or destinations are high off the navmesh, they don't find a path below

[timid wigeon]

I'm trying to "find the point with the shortest path including obstacle avoidance" and Google is saying I can calculate the path and then use NavMeshAgent.remainingDistance but doesn't that mean I have to assign the path to the Agent and can't really just compare 3 different points and their paths?

[swift roost]

is there any way of setting a NavMeshAgent in code by finding a GameObject with the script?

[timid wigeon]

Yes to all the interpretations of what you just said.

You might not want to do that because that sounds super costly, but yes you can

@swift roost https://docs.unity3d.com/ScriptReference/Object.FindObjectsOfType.html This is one way

[gaunt geode]

It can be a dumb question but, did you marked the grey plane as a NavMesh Static object? 🤔

[supple rock]

No I haven't.
Should I have?

[sharp pollen]

does A* pathfinding have to involve diagonal movement?

[zealous mantle]

Nope

[gaunt geode]

When you bake the NavMesh walkable area, it only takes into consideration Navigation Static objects

You either mark them Navigation Static through the Navigation Window, or doing it manually on the Inspector. https://docs.unity3d.com/Manual/nav-BuildingNavMesh.html

[supple rock]

Ahh

Thank you, gonna try this

[fringe raft]

Hi Folks, is it possible to calculate a NavMeshPath using an editor script? I keep getting: - "CalculatePolygonPath" can only be called on an active agent that has been placed on a NavMesh. when trying

I suspect it is because I'm in Edit mode, but would be nice to run it in Editor mode to tweak what I'm doing

[keen wigeon]

Hello! I'm trying to make a patrolling ai, I "stole" the code from a video. I'm having problems at rotating it...

 if (currentWaypoint < waypoint.Length)
        {
            Vector3 target = waypoint[currentWaypoint].position;
            target.y = self.transform.position.y;
            Vector3 moveDirection = target - self.transform.position;
            if (moveDirection.magnitude < 3)
            {
                currentWaypoint++;
            }
            else
            {
                transform.rotation = Quaternion.LookRotation(moveDirection, Vector3.up);
                controller.Move(moveDirection * speed * Time.deltaTime);
            }
        }
        else
        {
            if (loop)
            {
                currentWaypoint = 0;
            }
        }

All variables are defined, no errors. Just the rotation looking off.

(I need help asap it is 00:12 rn)

[alpine glacier]

I don't think its AI problem

it's rotation problem for #archived-code-general

[keen wigeon]

oh sorry

[tribal marten]

I'm having a really odd problem

The NPC in editor view is placed at a certain position, but when I go into Play Mode, they spawn in a completely different position, but it's the same every time

and I can't even move them during play mode in scene view

ignore this, I just didn't do my due diligence with animations and it's fixed now

[clever flame]

Does anybody wanna help me figure out why my agents refuse to move?

Like does an agent always need to do a Ray/Raycasthit check just to move or nah?

[fiery geode]

Can you share what your scene setup is? Or maybe the code you're using to move them?

[clever flame]

Created a forum post with just that in fact https://forum.unity.com/threads/can-someone-explain-why-agents-dont-move-without-a-raycasthit-condition.1151552/

[fiery geode]

Even with what you've added I can't find a reason why it shouldn't work. Can you provide more screenshots of the setup of your game object for example? What components it has?

[clever flame]

Would this suffice? I cache the agent in the start of UnitObject script

[fiery geode]

I see that the Patrol Points list is empty. Was this already the case?

[clever flame]

Patrol points only get populated when the function is called, in essence it goes as follows

called when patrol button is clicked

inputmanager checks on update and if patrolfunction is true it starts the Patrol method setup

Patrol method grabs all selectedUnits to pass the 2 Vector3 params to their single entity execution

until finally the array is populated by setting empty child objects with the required positions (1 where the unit is currently at and the other where the mouse was clicked in InputManager)

Then in the unitObject update it checks if patrol is true and if so its supposed to execute the UnitObject Patrol method

[fiery geode]

Why does the MonoBehaviour class UnitObject also have a parameterless Patrol() method? And what does that do? Does UnitObject derive from BehaviorFunctions?

[fiery geode]

I am referring to this code snippet where objecctClass is of type UnitObject, and Patrol() is called

[clever flame]

This is to execute a recurring loop of moving agents from point A to B and back and forth, I just cant get it to move at all yet, this method is called from the update function

Objectclass in just a reference from inside the InputManager to call and interact with individual entities holding the UnitObject script

And nothing really derives from BehaviourFunctions since its a seperate class thats supposed to exeute specific button action logic

[fiery geode]

Ok, I see

[clever flame]

I understand it might be confusing with 2 methods called Patrol, 1 is in the inputmanager script where the user configs the selected entities and the other is the single entity making attempts to move

[fiery geode]

That is indeed a bit confusing

[clever flame]

so any idea why the agents arent moving?

[fringe crag]

Hey all, Does NavMesh.CalculatePath() not work with the component based NavMeshSurfaces? I've tried both navmesh.calculatepath and navmeshagent.calculatepath and they return false. If i swap over to the standard navmesh system it returns true. Anyone have any insight?

[fringe crag]

@clever flame have you tried manually entering a known valid position in your set destination as a test instead of grabbing from your patrol points

[tame crag]

Hey does anyone here know of any Advanced Enemy AI tutorials out there? The ones im finding on youtube dont seem to explain something like an enemy Cover System and such just Idle and Shoot

[molten sequoia]

You are more likely to find better resources if you split things up. Trying to search for a complete AI solution that also has a solid cover implementation is a lot less likely to yield results than searching for a general AI design and then looking up approaches for (AI) cover systems.

[timid wigeon]

The easiest implementation of "cover" I've seen only works for "low cover". They usually put the objects either on a separate layer or cache a list of all the objects. Each piece of cover will provide a "safe spot" or "safe zone" that is on whatever edge is facing away from the player, so basically it's whatever the center position is of the cover piece + (a Vector with the same direction as from the player to the cover center).normalized * magnitude related to radius /thickness of the cover.

Side cover, however, is a whole other beast I've been tackling for a couple weeks

@tame crag I'm about here

[smoky jasper]

anyone here ?

[alpine glacier]

Does anyone know what could be the problem here?

I'm very very new to AI, so it may be obvious.

I wrote a script that simply tells the AI to go from one point to another, I baked the area and it all looks fine but for some reason the AI won't go up or down stairs no matter what.

I've messed around with the Agent/Bake settings to try fix this, but nothing has worked

The Ai Code: https://paste.ofcode.org/37iTEDDjSb9czjqTP8dSEZa

[quaint gazelle]

@alpine glacier hmm, have u tried voxel size

@alpine glacier

https://answers.unity.com/questions/1069078/how-to-allow-navmesh-agent-climb-up-stairs.html

[sinful mist]

Im creating a npc AI thats moves when im not looking and stops moving when looking, is there any method or other thing that i can use to know if im looking or not to the npc AI? it works simular to scp 173
or do i need to put a manual trigger on the camera?

[fringe crag]

Vector3.Dot product will take 2 forward vectors and give you a number between 1 and -1 whether or not they are facing each other

and if they are you can do a ray cast to see if there's a wall in the way

@sinful mist

https://docs.unity3d.com/ScriptReference/Vector3.Dot.html

[blazing pilot]

I'm new to game making, any tips on pros and cons of using NavMesh vs. a* for pathfinding? Are they suited for different things? Right now I'm making a 2d game

[tardy junco]

A* is just an algorithm. There are many implementations, including unity's Navmesh.

[blazing pilot]

aahh thank you! i'd completely misunderstood!

[safe basalt]

the direction.y doesnt work

the chasing doesnt always work

and i only wants him to attack the player not the walls and object when it enters the trigger

[modern orbit]

hello

i just want to ask where i can find a reference for making a code where the code detects if all of the prefabs are gone

like some sort of a victory mode

[alpine glacier]

does anyone know if its fine to leave the markup list blank when using NavMeshBuilder.CollectSources() ?

well empty

[hot siren]

Is there a way to make the AI ignore a large block (ground below the river) but not the underwater zones? Images to follow

I got a giant obstacle that clears all navmesh underwater, but I'd like to not have to make a trillion of those to cover all underwater without covering the underwater baths

[wary ledge]

Anyone have any recommendations of topdown / isometric AI? I'm leaning towards the style of like, Enter the Gungeon I suppose.

Specifically, tutorials of how to handle different attacks I suppose.

[near tusk]

hm

i am trying to implement auto-run on my player; the player is using click-to-move. ray cast to a point on the terrain and set that point as the destination. simple.

I am really struggling to get the character to keep moving forward at a constant rate. The closest I have come is the following, but it fails when terrain is not flat or if there are obstacles; the player bumps around, changes direction, or stops in those cases. Once I hit an invalid patch of the navmesh it stops working. This makes sense, but what I am really trying to do is just keep the player plowing forward no matter what and never changing direction; just like basically every MMORPG that has an auto-run function.

Is there another approach to do this using the navMeshAgent and click-to-move-like movement?

{
     playerNavMeshAgent.SetDestination( this.transform.position + this.transform.forward );
}```

[fickle ravine]

you can use Astar pathfinding projecting

but usually you have an invisible object using the pathfinding

and your player moving towards its current direction

then do instant changes on the agent

its very hard to get a nice result with an agent on your visible character

[wary ledge]

Yeah, so back to my question from earlier, I'm trying to figure out the best way to go about AI.

I think I'm over complicating things. But I was tyring to think of a way for each enemy to be Dynamic in the sense of their attacks. I already got movement down, for the most part at least.

But I sort of wanted Grunts (Small guys) to run towards the player, hit, then move out.

I was thinking about doing a weighted system, but honestly at this point I think thats just over complicating it, as at the end of the day I think all I'd need is: Run Towards player, Attack player, back up, repeat.

[tardy junco]

A simple state machine would do, no?

[wary ledge]

Sort of what I figured, but I was also leaning towards a system where I could scale it to boss fights in the future.

Which, in that case, I'd like to have a weighted grab bag.

But again, probably over complicating things.

[tardy junco]

weighted grab bag?🤔

What prevents you from scaling state machine to boss fights?

[wary ledge]

Yeah. Are you questioning what a weighted grab bag is, or quesitoning why I'd do it?

[tardy junco]

Both

[wary ledge]

Basically lets say a boss can do like.. 4 Attakcs.

Basic hit the player is weighted at say, 50% of the time.
Super Awesome One Shot Ability of Doom is 10%
FireBall 20%
Ice Storm 20%.

Concept being, you reach into the bag, pull out one of the above, and action it.

[tardy junco]

So just weighted random action. I don't see why you can't implement it with a state machine...🤔

[wary ledge]

You could, technically.

https://www.youtube.com/watch?v=gZUDzlKdtf4

This was the tutorial I was originally referencing. Well, not tutorial, just the concept.

[tardy junco]

Simplest way would just be a state that decides the next state out of a list for example based on their weights.

OR you could modify the whole system to use weights when transiting between states.

[wary ledge]

Yeah, that's sort of what I was thinking Something like:

Chase Player -> In Range of ONE of many Attacks AND Attack is Off Cooldown -> Use Attack -> Otherwise keep chasing.

[near tusk]

Interesting! I will see if I can get that to work! Thanks.

[quick zealot]

I am using a procedural dungeon generator to create random dungeons and I am kinda confused how to create the NavMesh of the level dynamically.

[tardy junco]

You'll need to use nav mesh components from github

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

Although it seems like they moved it to a package recently.🤔

[wary ledge]

So, first time doing a state machine: https://gist.github.com/Toonic/9e84f6d8fc08ed044df2f06861a6d390

Does this look about right?

TBH, I'm 90% winging it as I've made similar before, but never based around AI.

[wary ledge]

Also, better yet... I'm not sure if I should do local avoidance/flocking or not... Currently, I am.

But I'm thinking maybe I shouldn't.

[tardy junco]

I'd split it into several classes and avoid putting anything specific in the fsm itself. States would be separate classes with one base(abstract?) And contain all the logic.

[tribal bolt]

Yup, so to dig deeper on that, have one base class that manages the running state. It'll basically constantly call the state that's currently being run, then make a base class for the States, and have all the states inherit or implement an interface from the class.

In the State classes you'll have it do the if conditions, so while you're in that state class and it's doing that state by the end of the function, return the same state to the state manager. However, if your state meets the conditions to change to a different state, then return that state to the state manager instead. Then the state manager will start running that states class conditionals and the loop forever continues

This allows you to easily add more states to your code without breaking anything in the process

[wary ledge]

Cool, thanks for the reply's guys. I got a rough idea from what you were saying.

First time coding AI, it's deff something thats for sure.

[half heart]

Any ai tutorials for pedestrians/cars in 2d games

[covert jetty]

Guys, I kindly need your help... I'm trying to use two navmeshes areas for two different agents, but I can't figure it out how to make them work

I set the agents physical properties here

I'm not sure this was the correct way to use areas

unfortunately only the humanoid meshagents work, as I set "FinalBoss" in the Nav Mesh Agent component in a specific prefab, its instances will behave as they are not on a navmesh

Do I need the NavMeshSurface from the NavMeshComponents repo?

[tall viper]

Has anyone experience with A* and may help me with enemies not wanting to pathfind outside the screen?

[worn sorrel]

can anyone explain the best way to do a* pathfinding

[tall viper]

The "best" way really depends on your game

[tall viper]

I could help you with the basics

[wary ledge]

On the topic of State Machines etc.

I got something down pat now. I'd consider it a hell of a lot better then what I had.

However, I'm just wondering what'd be the best way to now going about creating attacks? Obviously I'd want something like: public abstract class Attack : MonoBehaviour {};

But would my best option to be basically, code an "Attack" then attach that Script to the Monster/Whatever. Then from there, have my Attack State/Action call the attack that way?

[worn sorrel]

Im making a top down shooter 3d

[tall viper]

Im actually working on a topdown shooter as well

I used the A* Pathfinding project

Brackeys once explained it in on of his videos

[wary ledge]

If you're really lazy, theres a good A* Plugin available for free.

[worn sorrel]

i need to code it as im doing a school project

ive got the basics done

[wary ledge]

You need to code A* from scratch for a school project?

[worn sorrel]

Yeah

like within unity

[hot patrol]

Honestly there’s so many resources about a* creating one on your own is not very hard…

[fickle ravine]

I recommend, modified it myself. Solved all of unity nav mesh problems

even got special points like jump points going

[covert jetty]

In case anybody else had this problem, the NavMeshComponent repo assets solved my problem

[quick zealot]

I saw this

Q: Can I bake a NavMesh at runtime?
A: Yes.

On the link that you share... Is there any tutorial that you recommend me to follow?

[finite trellis]

Anyone know: Do NavMesh agents move via transform.translate?

[still minnow]

Can anyone tell em what's going on here? I've tried everything, I just want these to bake fully without gaps

https://imgur.com/td9iEIh

[tardy junco]

I'm sure there are at least a few related tutorials on YouTube, can't recommend anything specific. I think they also have a few demo scenes/examples of use case on the GitHub page.

[tardy junco]

Yeah, they most likely move via the transform.

[tardy junco]

I think there was a "step height"(or something) parameter either in the baking or agent settings.🤔

[still minnow]

yea, ive maxed or reduced that. Done all the obvious stuff

[tardy junco]

What's the current value?

[still minnow]

it's in the screenshot, which is 2, but ive adjusted this and all of the other values to all kinds of ranges

this is actually the BEST case scenario in the screenshot

I've considered using a collider instead, but I dont want to use colliders across the entire navmesh

[tardy junco]

Hmm

Could try making a simplified mesh in blender and build the Navmesh from that.🤔

[still minnow]

like, for the entire map?

[tardy junco]

If that problem is recurring, then yeah.

@still minnow actually

[still minnow]

the map is large, with buildings, doors, stairs, floors of those buildings

[tardy junco]

Ah nvm

[tardy junco]

Is it modular? Made from repeating modules?

[still minnow]

yes

a few of them

i mean, i know i can do workarounds to fix it. Im just trying to keep a decent workflow and understand unitys navmesh

[tardy junco]

Yeah, I donno. Step height should be able to fix that.

[still minnow]

I think its because the render mesh has a curve, that exceeds the max slope of 60

https://i.imgur.com/FwVylN0.png

and i'd love to script something on a modifier to ignore the max slope or something

but i dont see anything about the slope in the code

[tardy junco]

Hmm... From what I understand, step height should have higher priority than slope angle. Otherwise stairs wouldn't have a Navmesh.🤔

[still minnow]

i've put every value into this thing i could thing of lol

[tardy junco]

Yeah, I believe you. Did you try building a regular Navmesh(not a surface)?

[still minnow]

what do you mean?

[tardy junco]

I mean in the navigation tab.

[still minnow]

i did before, but ill try it again

what exactly is height mesh?

[tardy junco]

From what I understand it's something to place your character precisely on steps and stuff, but it's something that comes in addition, not to replace the Navmesh itself.