#⚛️┃physics | Official Unity Discord | Page 37

lapis plaza Apr 11, 2019, 6:00 PM

#

This all depends if they expose the api enough or share the integration sources

#

It is missing raycasts atm :D

tropic hamlet Apr 11, 2019, 6:03 PM

#

I just want to use unity's physics but lack of primitive stacking (note: don't need to stack across the network, just for local) would be just right for me

fallow gorge Apr 11, 2019, 6:14 PM

#

seems backwards not to have something which seems so basic for a modern physics engine, really hoping they reconsider or that the havok part is affordable

lapis plaza Apr 11, 2019, 6:20 PM

#

I wish they'd put the Unity Physics on github

#

users could then add caching easier if Unity wont

fallow gorge Apr 11, 2019, 6:45 PM

#

as great as user stuff is, id feel better about the long term availability if unity themselves implemented that

lapis plaza Apr 11, 2019, 6:47 PM

#

of course, but when it comes to physics, I dunno if it's worth waiting for years to maybe get the functionality you need

fallow gorge Apr 11, 2019, 7:10 PM

#

well yeah id use it and be thankful. But I'd forever be left wondering about what exciting frontiers an official solution could lift me towards, not to mention the vast multitude of possibilities for user solutions built on top of that!

lapis plaza Apr 12, 2019, 2:16 AM

#

Unity Physics docs:
https://docs.unity3d.com/Packages/com.unity.physics@0.0/manual/index.html
https://docs.unity3d.com/Packages/com.unity.physics@0.0/api/Unity.Physics.html

#

I don't actually know how you are supposed to find the latter link by site navigation, I just know it exists :p

tired pawn Apr 12, 2019, 7:28 AM

#

Thanks Olento

peak panther Apr 12, 2019, 6:26 PM

#

Hey so I'm looking to create a 2D character controller and I'm really at a loss as to which method to go with.
idk if I should use transform.Translate or Rigidbody.movePosition or Rigidbody.velocity or which way I should be detecting collision.
any suggestions?

sly violet Apr 12, 2019, 7:28 PM

#

@peak panther Probably need to start with this one:
https://docs.unity3d.com/ScriptReference/Rigidbody2D.Cast.html

peak panther Apr 12, 2019, 7:29 PM

#

@sly violet Is this the standard thing to use with a kinematic rigidbody?

#

also thanks for responding

sly violet Apr 12, 2019, 7:34 PM

#

I think the standard thing to use is the existing character controller, but if you're dead set on making your own, you'll need to know what you're going to hit before you hit it so you can adjust the movement appropriately, and that requires a "cast" before you move.

tropic hamlet Apr 12, 2019, 7:45 PM

#

I made my own rigidbody based one but I think people will need actual years of experience in doing that to do it well. Raycast based ones are far safer

peak panther Apr 12, 2019, 7:56 PM

#

Hmm is that so?

#

Is there any good character controller then (rigidbody preferably, but I'll look at some raycast based ones) that I can look at?

sleek prairie Apr 13, 2019, 12:51 AM

#

I'm experimenting with using the Physics system to avoid doing between 1 to 10 million distance checks in code to check if objects are within range of each other

#

for now I've implemented a system that uses box colliders as triggers, but I've been reading and people say using sphere colliders is better for performance than box colliders

#

is that correct? and if so, why?

#

do box colliders check per vertex and sphere colliders check based on a radius?

lapis plaza Apr 13, 2019, 1:13 AM

#

@stuck bay it's not just the vertical stacking, I mean, it's issue in all directions

#

📎 bounce.gif

#

and yeah, I have situations where that can happen

#

here's a thread about it https://forum.unity.com/threads/dots-physics-dynamic-objects-will-not-settle-on-other-dynamic-objects.646939/

#

sleep would require caching

#

you can't just sleep with stateless thing

#

because you can't just put things to sleep if the velocity is 0

#

it needs to be close to 0 for multiple physics steps

#

you don't need tons of past states tho

#

even having previous frame can help a lot on the stability in these kinds of scenarios

#

this is why verlet integration is more stable vs naive euler solver

#

as it uses previous steps data as well

#

http://codeflow.org/entries/2010/nov/29/verlet-collision-with-impulse-preservation/

#

the final example on that isn't super stable looking either but it does eventually stabilize 😄

#

I do plan to investigate the options to fix that kind of jittering on Unity Physics if Unity doesn't beat me to it, it's not a priority but it's very solvable issue

#

Unity Physics in general?

#

I really wish they'd put it on github

#

but I'm afraid they don't as they don't put any ECS packages there

#

mathematics is exception

#

but it's not only for ECS either

#

nah

#

it's proprietary

#

I mean you get the c# scripts on the package but it's mostly native code there

#

it's also one of the few packages Unity actually ships with native binaries in it

#

usually they tend to put native code in engine side

#

but this lets them upgrade Burst without it being tied to engine releases

#

and by the native part, I mean the actual Burst compiler

#

📎 unknown.png

sly violet Apr 13, 2019, 1:43 AM

#

@sleek prairie : Yeah, sphere-to-sphere collisions are cheap to calculate because all you need to know is the distance squared.

#

No geometry, no rotation...

lapis plaza Apr 13, 2019, 1:46 AM

#

@stuck bay tbh, that's something I have no idea about, better ask in #archived-dots 😃

sleek prairie Apr 13, 2019, 2:02 AM

#

@sly violet I see, thanks! I'll probably try OverlapSphere on the 1st set of objects and a sphere collider on the 2nd set. Right now I'm doing box collider on both the 1st set and 2nd set and seeing some performance spikes sometimes with more than 10k objects

sly violet Apr 13, 2019, 2:14 AM

#

Hmm, you'll probably see a small across-the-board improvement but the spikes will probably still happen. Spheres are quick to check but doesn't help with the, what's it called, sorting out which checks can be automatically culled and not even checked.

#

Just guessing/theorizing, though, I'd be interested in hearing how it works for you.

sleek prairie Apr 13, 2019, 2:42 AM

#

I will run some tests tomorrow to see how it turns out. I can post the results here along with my previous tests. The good thing about OverlapSphere is that I think I can "space out" the calls since I only need to do that once per object and I have a time window of about 1 second from the time the objects are created to the time that I actually need the results of the distance checks

#

I have a set of 10k objects that are not moving and another set of around 1000 objects that are activated at runtime and then moved

#

I think some of the Physics performance spikes come from moving those 1000 objects that have colliders, but I might be wrong on that one

#

@sly violet I'll tag you here once I post the results 🙃

tropic hamlet Apr 13, 2019, 4:24 AM

#

They know way better than I, by far.

#

:D

lapis plaza Apr 13, 2019, 4:27 AM

#

I'd love to see different broadphases as option, like physics engines tend to have

lapis plaza Apr 13, 2019, 5:00 AM

#

yeah but Unity Physics is stateless, so they kinda want everything to be generated from scratch each time

#

I don't know what's involved on on SAP algo tho

#

like, could you make it deterministic by storing the previous state as cache

#

yeah, it does limit things

#

yeah but I mean now if you need to roll back to previous state

#

you have to have the cache rolled back in similar state for determinism to work

#

should read how it's implemented to really understand this 😃

#

well, wikipedia can do that too ;D

#

https://en.wikipedia.org/wiki/Sweep_and_prune

Sweep and prune

In physical simulations, sweep and prune is a broad phase algorithm used during collision detection to limit the number of pairs of solids that need to be checked for collision, i.e. intersection. This is achieved by sorting the starts (lower bound) and ends (upper bound) of ...

#

but I'm terrible at understanding text in this form, I understand equations, formulas and code way better 😃

#

describe algo in text = takes 100x more effort to understand that to examine it in snippet with description

#

don't they do that already for Hybrid Renderers lod system?

#

I think they have some basic AABB setup there

#

or "basic" they did exploit a lot of things for it, Mike Acton did a talk about it on Unite LA

#

they pretty much relied on every building being static there tho

#

https://www.youtube.com/watch?v=k_ORJXmPu9M

YouTube

Unity

ECS Track: LOD and Culling Systems That Scale - Unite LA

How we used ECS to display a large environment in the Unite LA MegaCity streaming demo. Slides: https://www.slideshare.net/unity3d/lod-and-culling-systems-th...

▶ Play video

#

📎 unknown.png

#

well, they used that for culling

#

well, I don't really know what the whole thing would be called

#

it's not SAP

#

it's custom setup

#

the talk does describe it 😄

#

I totally expected them to come up with something like that in the Unity Physics, was pretty surprising to see BVH instead

#

but it could have been some thing from Havok devs, if they wanted to make it that way

#

now that I watch that LOD talk again, I totally understand why they couldn't use this 😄

#

they could have used this for static colliders

#

but that's about it, the chucks are hardcoded for the bounds

#

so you can't do that for anything dynamic

#

wonder if the static BVH is still expensive

#

as this talks setup would be super cheap

#

I'll worry about the broadphase if it turns out to be too high cost in the final setup

#

in my case, it's mostly about static colliders

#

https://forum.unity.com/threads/ecs-packages-on-github.656512/

Unity Forum

ECS packages on github?

@Joachim_Ante Is Unity planning to put Entities, Hybrid, Unity Physics, AI Planner, upcoming DSPGraph etc on github? Especially Entities and Hybrid...

#

I tried to get some answer but less surprisingly that got ignored

#

have asked for that physics repo on every possible thread, it's not even possible they haven't seen it

#

they just ignore it 😄

#

the havok guy even follows me on the forum but it's not his call either 😄

#

I'm guessing they want to keep the DOTS code out of github on purpose

#

like, they don't hide it from Unity users as it's on the packages but they don't want to share it openly to the whole world

#

now basically they can limit the users to ones who use it through Unity editor

#

well, technically you can download the packge from the bintray directly too but it's not something most people would do

#

well, UE4 source code requires github linking too

#

IMO they could do same for Unity repo

#

if that game more access to the actual devs, that would be fine

tropic hamlet Apr 13, 2019, 2:42 PM

#

May be too soon

sleek prairie Apr 13, 2019, 6:02 PM

#

So I've been doing experiments to determine what is the most efficient way to check game objects within a certain range in my game

#

Tests Conditions: 2 sets of objects (A and B) with 1000 objects each

#

Tests Objectives: Check what objects of set A are in range with objects of set B

#

Test1

📎 test1.png

#

Test2

📎 test2.png

#

Test3

📎 test3.png

#

@sly violet you were right, box colliders and sphere colliders don't make a significant difference

#

sphere colliders are slightly more performant. Edit* on the Test3 I forgot to change the last line, it's actually SphereCollider --- OverlapSphere: 7ms

lapis plaza Apr 13, 2019, 6:05 PM

#

missing Test4 where you use the new math lib, burst compiler and job system for the Test1 😃

#

but brute force isn't good

sleek prairie Apr 13, 2019, 6:06 PM

#

@lapis plaza hehe I still haven't used yet any of the new features with ECS, Burst, Job System or any of that

#

would it make a significant difference for my intended use?

lapis plaza Apr 13, 2019, 6:07 PM

#

well, test1's case would get way better results but it's still kinda wrong way to do it

sleek prairie Apr 13, 2019, 6:07 PM

#

I should point that my my tests don't actually test against ALL 1000 objects on each set, they are not all positioned at the same point, instead they are spread out to simulate the actual implementation that I need for my game

lapis plaza Apr 13, 2019, 6:07 PM

#

the overlap sphere is best pick IMO

#

and then run the burst job for it 😄

#

you don't need to do ECS for this

#

just do regular job and use new mathematical and burst

#

it'll give you speedup for that part for sure

#

it'll multithread and SIMD optimize the math

#

you can also read the transforms in job systems worker threads

#

so it's not all tied to main thread

sleek prairie Apr 13, 2019, 6:09 PM

#

@lapis plaza what about test1 is wrong? I thought of implementing some kind of spatial segmenting so I wouldn't need to test all objects, but I didn't think it was worth it because it would overcomplicate the system and I still need to add more things on top of it

lapis plaza Apr 13, 2019, 6:09 PM

#

your test3 already does that

#

physics engines do this already for you, so why not take advantage of it

#

how costly is the overlapsphere itself?

sleek prairie Apr 13, 2019, 6:10 PM

#

exactly, that's what I got from reading around the forum and the recommendations of some peeps 😃

#

one sec I'll test the overlapSphere without populating the arrays

lapis plaza Apr 13, 2019, 6:11 PM

#

it would tell if there's room for optimization on the array processing

sleek prairie Apr 13, 2019, 6:13 PM

#

this is what I get on the profiler

📎 fbd5968bc48703dffad2b3aad41fac5c.png

#

so the OverlapSphere itself is 2.39 and the method that calls it 4.32ms

lapis plaza Apr 13, 2019, 6:13 PM

#

so 2.4ms?

sleek prairie Apr 13, 2019, 6:14 PM

#

yeah I guess so

lapis plaza Apr 13, 2019, 6:14 PM

#

what the method does?

sleek prairie Apr 13, 2019, 6:14 PM

#

on that profiler screenshot is empty

#

just the Physics.OverlapSphere

#

Physics.OverlapSphere(transform.position, 2.5f);

lapis plaza Apr 13, 2019, 6:14 PM

#

weird that it's almost double cost than the actual check

sleek prairie Apr 13, 2019, 6:15 PM

#

I call it on the Update method with if (Input.GetKeyDown(KeyCode.S))

#

maybe it's that?

lapis plaza Apr 13, 2019, 6:15 PM

#

ah

#

you should measure the cost in some isolated function

#

you will not get right results if you have extras there

sleek prairie Apr 13, 2019, 6:17 PM

#

so should I isolate the OverlapSphere in it's own method?

#

and then call that from the update with that input check?

lapis plaza Apr 13, 2019, 6:17 PM

#

for benchmarking, it does make it easier, yes

sleek prairie Apr 13, 2019, 6:17 PM

#

ok, now that I'm at it I'll sample the method for the profiler too

lapis plaza Apr 13, 2019, 6:17 PM

#

I'd recommend keeping Update slim anyway

#

there's also jobified spherecast but it might be overkill if you only need one overlap check (there's no jobifed overlap checks)

#

what's nice about the jobified thing is that it returns the results in native array that can then be easily processed with another job

#

but I guess the overlapsphere's array works too for that purpose

#

anyway, this should get you started on jobs if you want to try it out: https://docs.unity3d.com/Manual/JobSystem.html

#

(for processing the overlap results)

sleek prairie Apr 13, 2019, 6:26 PM

#

📎 1d912b33064b7c3c66843c9d2be9a189.png

#

with deep profiling shows different results but more consistent

#

seems the other cost comes from getting the transform and position

#

I don't think I'm populating the array too inefficiently

#


triggeredObjects = new GameObject[triggeredColliders.Length];
for (int i = 0; i < triggeredColliders.Length; i++)
{
        triggeredObjects[i] = triggeredColliders[i].gameObject;
}```

#

the difference between calling only OverlapSphere and calling OverlapSphere + populating the array doesn't seem to significant

#

it's pretty much 1 or 2ms in difference with 1000 objects calling the method

#

I must mention that the arrays themselves aren't too big, 100 items at most

#

I'll be sure to check out that link, it's been on my list for a while to learn about ECS, Job System and the Burst Compiler

lapis plaza Apr 13, 2019, 7:01 PM

#

oh, so you don't even batch process the objects in range there? yeah, burst wouldn't help you there 😃

#

also the spherecastcommand (jobified spherecast) will not help here either as spherecasts don't get the initial overlap apparently

#

@sleek prairie

sleek prairie Apr 13, 2019, 7:05 PM

#

@lapis plaza sorry but what do you mean by "batch process" ?

#

I'm an artist reconverted to programmer, some concepts are still foreign to me 😅

lapis plaza Apr 13, 2019, 7:06 PM

#

I just mean processing all the elements through one by one and doing some math on them

#

what do you do with those gameobjects you get in the array?

#

like, why do you even need this?

sleek prairie Apr 13, 2019, 7:08 PM

#

I need to know what objects are in range to apply other systems on top of them. For example a procedural generation system that needs to act on arrays of points that are contained inside those objects

#

also do some shader stuff with the objects in range

#

I give a unique identifier (an int) to each of the objects and then I need the other systems to do stuff with them

lapis plaza Apr 13, 2019, 7:09 PM

#

sounds like you'd love the ECS 😄

sleek prairie Apr 13, 2019, 7:13 PM

#

probably, I've heard good things about it! 🙃

#

but for now it's not a priority since I haven't hit any roadblocks with the "regular" Unity way of doing code

lapis plaza Apr 13, 2019, 7:18 PM

#

it's just, if you process a ton of data, it makes that kinda of tasks a lot of faster

sleek prairie Apr 13, 2019, 7:22 PM

#

If it offers a considerable improvement in performance and I can combine it with the Monobehaviour way of coding..

#

I will probably use ECS for my procedural points system

sharp sentinel Apr 14, 2019, 12:52 PM

#

@lapis plaza I am just reading through the conversation to learn, but pardon me, isnt batch processing essentially processing elements into batches of similar attributes to be stored to then be processed in one go? What's this about processing one by one?

Just a curious question because I am still learning, I dont meant to challenge. Does batch processing mean something else in unity or am I just wrong?

sleek prairie Apr 14, 2019, 1:01 PM

#

@sharp sentinel same here! heh 😅 all I know about "batches" is related to Unity's rendering pipeline.

lapis plaza Apr 14, 2019, 3:59 PM

#

@sharp sentinel I was referring to https://docs.unity3d.com/2018.3/Documentation/ScriptReference/SpherecastCommand.html and Unity's Job System for processing transforms if needed

#

I dunno if the batch processing is somehow right or wrong term, I just use it for processing a lot of data in similar fashion

sharp sentinel Apr 14, 2019, 4:01 PM

#

right. I see. All good. Batch processing meant something different from what I learnt, so I was curious. Thanks 👌

lapis plaza Apr 14, 2019, 4:01 PM

#

afaik, it's pretty common term tho, like if you process similar inputs, with same operations, many softwares call it batch operation

#

imagine like image editor that runs series of operations in script for each file you enter for the batch processing

sharp sentinel Apr 14, 2019, 4:06 PM

#

I kind of get the gist of what you are getting at. Right. Makes sense for the name. It's almost the same as what I know except for the organizing the data into separate scheduled payloads a step prior, what you are saying is the entire input is already in a "batch."

sleek prairie Apr 14, 2019, 8:28 PM

#

I'm trying to improve performance on a system for my game that is based on OverlapSphere()

#

so I have 500 objects, each one with a child that has a trigger collider and a rigidbody

#

both parent and child are on different physics layers set to not interact with each other

#

I call the OverlapSphere on all 500 parent objects in 1 frame

#

with the child's colliders active, the OverlapSphere calls take around 40ms

#

but with the child's collliders NOT active, it takes around 1ms to 2 ms

#

what's the explanation for that?

#

I even call the method to only act on a specific layer and with QueryTriggerInteraction set to ignore

#

Physics.OverlapSphere(transform.position, overlapRadius, LayerMask.GetMask("OriginalObject"), QueryTriggerInteraction.Ignore)

#

the OverlapSphere is working as intended and it's only getting colliders on the layer I provide

#

Profiling with child collider active

📎 OverlapChildrenColliderActive.png

#

Profiling with child collider NOT active

📎 overlapChildConlliderNOTactive.png

sly violet Apr 14, 2019, 8:49 PM

#

...I've always wondered if the collision matrix is resolved before or after the principle calculations for casts and such. Kind of looks like after, doesn't it?

#

Still seems too big.

#

Like, even if they share the same underlying matrix, the increase should be proportional, and that seems geometric.

sleek prairie Apr 14, 2019, 8:50 PM

#

My knowledge of how the Physics engine works is quite limited so most of this stuff just flies over my head

#

I would assume the collision matrix is doing what I want it to do

#

because the OverlapSpheres don't return the colliders of the child objects

sly violet Apr 14, 2019, 8:51 PM

#

Hmm, sorry, mine too, in this case. I don't really know how it's optimized under the hood except in the most general sense.

#

Yeah, thing is, there's two ways to not return it, right? They could not find it in the first place, or they could find it but discard it.

sleek prairie Apr 14, 2019, 8:52 PM

#

even with both those ways I still get a massive cost in performance

#

the workaround I have now is just keep the child collider inactive until the OverlapSpheres have finished doing their thing

sly violet Apr 14, 2019, 8:53 PM

#

Heh. Is it cheap to activate them all?

#

I'm wondering if you can put them in a separate physics instance instead of a separate Layer. Might help.

#

...But right now I can't even find that option.

#

I'd've sworn that was something in beta at least.

#

Oh, they call it "Multi-scene physics"

sleek prairie Apr 14, 2019, 8:56 PM

#

activating the child's colliders is not costing any time at all, it's so fast that I can't even find it in the profiler based on the graph because I don't activate them all at once

#

I progressively activate them based on other conditions

#

which are procedural and can be the case that most of them activate at the same time, but highly unlikely

sly violet Apr 14, 2019, 8:56 PM

#

Hmm. Sounds good, then.

sleek prairie Apr 14, 2019, 8:57 PM

#

That stuff about "physics instances" sounds interesting

#

is that a 2019-only feature?

sly violet Apr 14, 2019, 8:57 PM

#

https://blogs.unity3d.com/2018/11/12/physics-changes-in-unity-2018-3-beta/

Unity Technologies Blog

Physics Changes in Unity 2018.3 Beta – Unity Blog

The main change Unity 2018.3 brings to physics is the 3D physics engine upgrade, from PhysX 3.3.3 to PhysX 3.4.2. It’s the first time we ship the latest av...

#

Unfortunately it's tied to scenes, which means you can't use it for child objects.

sleek prairie Apr 14, 2019, 8:58 PM

#

thanks for the link, I'm gonna take a quick read

sly violet Apr 14, 2019, 8:59 PM

#

Batched queries might help you, too, although they don't include Overlap so maybe not. I think we discussed that earlier in this thread.

sleek prairie Apr 14, 2019, 8:59 PM

#

ah! Seems this is somewhat related to what I'm experiencing

#

By default, all the changes to Unity transforms were synced to the physics engine right before they might be needed. That is, every time a raycast is run, every time a simulation is run and so on. says that blog page

#

on the screenshot I posted for the profiler that shows 40ms

#

the bigger performance costs come from: Physics.SyncRigidbodyTransform and Physics.SyncColliderTransform

sly violet Apr 14, 2019, 9:00 PM

#

Huhhh. Yeah, that's a huge improvement, although it doesn't explain the discrepancy.

#

I wonder if this ever made it out of beta?

sleek prairie Apr 14, 2019, 9:02 PM

#

I have 2018.3

#

maybe it didn't

#

My (extremely uneducated) guess as to what is happening with my OverlapSphere is that the Physics engine is somehow feeding and syncing the child's colliders and rigidbodies data into the physics calculations even tho I have the collision matrix set to ignore it

#

but it is weird since I am not getting any performance cost increase from moving the objects when the child colliders are not active

#

I clearly see a performance cost on the profiler graph as soon as the child colliders are getting activated

#

If someone more experienced with Unity Physics can shed some light on the situation so I can understand better what is happening I would greatly appreciate it 😊

#

on the meantime I'm gonna do some reading on transform syncing with the Physics engine

ocean horizon Apr 15, 2019, 7:41 AM

#

@sleek prairie Simply turn off "AutoSyncTransforms" in physics settings. This should be off by default on new projects.

#

The collision matrix is used early, not as a filter after the query has run i.e. it's used during the broadphase detection, not narrowphase.

sleek prairie Apr 15, 2019, 8:16 AM

#

@ocean horizon oh thanks! Now the OverlapSphere takes the same amount of time

#

I still end up needing to keep the objects inactive while their movement starts

#

otherwise I get a huge spike in Physics.Processing of about 100ms with 500 moving objects

#

If I activate the child objects in the middle of the movement Physics.Processing doesn't even reach 5ms at worst

ocean horizon Apr 15, 2019, 8:19 AM

#

You should not be activating/deactivating objects. Make sure you're not moving physics objects by modifying the Transform. Use Rigidbody.MovePosition/MoveRotation.

sleek prairie Apr 15, 2019, 8:21 AM

#

@ocean horizon I would, the thing is I am moving the parent object and it doesn't have a collider or rigidbody

#

I need the child object to detect whenever it collides with a surface

#

I move the parent with this ```while (elapse_time < duration)
{
discoverer.transform.Translate(0, (Vy - (gravity * elapse_time)) * Time.deltaTime, Vx * Time.deltaTime);
}

ocean horizon Apr 15, 2019, 8:21 AM

#

So why not place the body on the parent?

#

Then you can drive the body which will modify the transform thus avoiding any transform-sync prior to the simulation stepping.

sleek prairie Apr 15, 2019, 8:23 AM

#

that's a good question 🤔

#

I organized it this way for ease of use

#

since the parent does something completely different to what the child does

#

but I might aswell follow your advice

#

and put all together

#

also my previous setup used a collider on the parent for distance checks so that was one of the reasons

ocean horizon Apr 15, 2019, 8:29 AM

#

Prior to the simulation stepping, we have to perform a check to see if any transforms with RB or colliders changed. If so, we update the appropriate RB/colliders. The best case is where there are no transform changes and it's actually far quicker for this case than a single transform change. So for best performance, let the RB drive the transform, it's what it's there for.

sleek prairie Apr 15, 2019, 8:30 AM

#

oooohhh

ocean horizon Apr 15, 2019, 8:30 AM

#

For ad-hoc cases where you don't do this then it won't matter. But if you're doing hundreds/thousands then it becomes important.

sleek prairie Apr 15, 2019, 8:30 AM

#

so by moving the object through the RB it doens't need to sync transforms and colliders at all?

ocean horizon Apr 15, 2019, 8:30 AM

#

No. Transform sync is Transform -> RB/Collider.

#

i.e. an indirect way to modify the RB or collider. Doing it directly avoids that completely.

sleek prairie Apr 15, 2019, 8:31 AM

#

oh! I'm really thankful for the advice 😊

ocean horizon Apr 15, 2019, 8:32 AM

#

You're most welcome. Rule is, don't modify transform with a RB/collider on it if you can help it.

sleek prairie Apr 15, 2019, 8:32 AM

#

will do! 😃

ocean horizon Apr 15, 2019, 8:32 AM

#

The only reason it's supported is because we cannot ignore it.

#

As much as we'd like to sound loud sirens when it happens. 😃

sleek prairie Apr 15, 2019, 8:33 AM

#

question about Syncing if I disable "Auto Sync Transform" is there any undesired behavior I might encounter when leaving it disabled?

#

or any stuff that I should be mindful of?

ocean horizon Apr 15, 2019, 8:36 AM

#

Prior to that option, when you modified the transform we immediately updated any RB/collider. The transform system changed a while back and we no longer got notifications of transform changes. This meant that to keep backwards compatibility with "changing RB/collider when a Transform changes" we had to implement the "AutoSyncTransform" stuff. This means that anything that can read a RB/collider state that depended on knowing about the transform has to perform a transform-sync first to ensure it's up to date. So with it on, it changed very little. With it off however, the RB/collider are only updated once prior to the simulation step (or if you call "Physics.SyncTransforms" yourself).

#

So with it off, the behaviour change is that the RB/collider are still where they were prior to the Transform change. Spatial queries "see" them at the old position.

#

The advantage is that the transform change is decoupled from physics and you don't pay a physics cost, even if you change it thousands of times per frame.

#

we only sync-up when it matters to the simulation.

sleek prairie Apr 15, 2019, 8:38 AM

#

hmmm, I think I get the gist of it. So if I want to do it "the wrong way" and move by transform even when I have a RB I need that on

#

does it apply to moving by transform when only having a collider and not an RB?

ocean horizon Apr 15, 2019, 8:39 AM

#

We automatically sync transforms prior to the simulation so no, you don't need it on.

sleek prairie Apr 15, 2019, 8:40 AM

#

so I don't really need Auto Sync on in any case?

#

doesn't matter what I do and how I fudge the physics and transforms

ocean horizon Apr 15, 2019, 8:41 AM

#

You should avoid having it on if performance is important. Knowing that when you change a transform, it does not immediately update the rb/collider is important.

#

If you change transform then perform a query, it won't see it at the new position.

#

but that is potentially a bad way to do things anyway

sleek prairie Apr 15, 2019, 8:42 AM

#

I got it now, thanks for bearing with me 🙃

ocean horizon Apr 15, 2019, 8:42 AM

#

often it's query then adjust things for next simulation step.

sleek prairie Apr 15, 2019, 8:42 AM

#

sets a reminder to himself: DO NOT MOVE PHYSICS-DEPENDENT OBJECTS VIA TRANSFORM

ocean horizon Apr 15, 2019, 8:43 AM

#

Not a problem. It's all hidden stuff so confusing. Mainly this "feature" was a backwards compatibility thing. Moving forward it should be off.

sleek prairie Apr 15, 2019, 8:43 AM

#

thanks a lot for taking the time

#

may the gamedev gods bless you 🙏🏽

lapis plaza Apr 15, 2019, 9:00 AM

#

what makes this twice the fun is that you need to relearn all the fancy tricks when you move to ECS physics as old rules don't apply 😄

#

I mean, for the transform update magic etc

#

(it's fine though)

sleek prairie Apr 15, 2019, 9:01 AM

#

OH GOD

lapis plaza Apr 15, 2019, 9:01 AM

#

I kinda never got the transform autoupdate benefit unless you made a game that didn't utilize physics sims more

#

for example, if you make a custom vehicle, you need the updated transforms each fixedupdate anyway

#

so that ruins the optimization afaik?

#

or can you somehow fetch only one transform?

ocean horizon Apr 15, 2019, 9:02 AM

#

transform autoupdate benefit

#

It was a backwards compatibility option.

#

The transform system changed and it was "imposed" on all teams in Unity.

lapis plaza Apr 15, 2019, 9:02 AM

#

yeah, but I mean the benefit of that optimization now

ocean horizon Apr 15, 2019, 9:02 AM

#

Which optimisation?

lapis plaza Apr 15, 2019, 9:03 AM

#

for not updating all gameobject transforms on fixedupdates

ocean horizon Apr 15, 2019, 9:03 AM

#

We do update them then and only then. We always do that once.

#

Actually when you perform a simulation step. Whether in fixed-update (auto) or manual

lapis plaza Apr 15, 2019, 9:03 AM

#

I thought the idea was to update the transforms only for next update unless the user needs the data on fixed?

ocean horizon Apr 15, 2019, 9:04 AM

#

The idea is to simply not update physics when the transform changes. Only update it when it's required which is prior to the simulation stepping.

lapis plaza Apr 15, 2019, 9:04 AM

#

oh, that way around

#

heh, it's been a while since I watched the talk

ocean horizon Apr 15, 2019, 9:05 AM

#

The transform system itself changed to not produce callbacks. This was done due to its integration into the Job system

lapis plaza Apr 15, 2019, 9:05 AM

#

there was some handy snippet that made this all obvious how you should take it into account

ocean horizon Apr 15, 2019, 9:05 AM

#

So the trusted callback we had just went away. 😃

#

In-fact, the amount of projects I saw where the transform was changed many, many times between a simulation step was scary. Even doing position, rotation and scale changes separately and multiple times.

#

Each one caused a physics update for that GO.

#

No longer with the option off obvs.

#

Only thing to remember is that changing a transform does not affect physics.

#

fast but potentially inconvenient if you expect all state to be in-sync all the time.

#

For new projects, this is off so fast but can catch you out if you expect to always be in sync.

#

There's a "Physics.SyncTransform" call which is what we call prior to a simulation step. This is also called (if you have "Physics.AutoSyncTransforms" on) prior to any query.

#

2D/3D are identical in this regard.

#

<message ends> 😉

lapis plaza Apr 15, 2019, 9:10 AM

#

heh, thanks for the insight 😃

#

I hope I don't have to use the built-in physics anymore though, I really enjoy the freedom with source code access on ECS Unity Physics 😃

#

finally get those contact callbacks exposed I've asked access for years on stock physics integration

ocean horizon Apr 15, 2019, 9:13 AM

#

Yep and we'd expose them if we could but there's no way to stop things happening during such a callback that could cause serious problems.

#

We've asked for a "sandboxed" callback on our side too.

lapis plaza Apr 15, 2019, 9:13 AM

#

this isn't saying the built-in integration wouldn't be good, it's mainly about the access and enabling more things

ocean horizon Apr 15, 2019, 9:13 AM

#

Yep. It's good for you and us too.

lapis plaza Apr 15, 2019, 9:13 AM

#

not having access to physics code has always been driving me to custom physics solutions on unity

#

used to use bullet at some point

#

yeah, I imagine it'll be easier to maintain in the long run

ocean horizon Apr 15, 2019, 9:14 AM

#

Being able to point to source code in a chat like this is far easier.

#

I can do that to some degree with Box2D but there's still a large "hidden" layer to contend with.

#

So yeah, more stuff on Github the better.

lapis plaza Apr 15, 2019, 9:17 AM

#

btw, talking about github

#

you know is Unity Physics is going to be in Github?

#

right now it's only distributed via Package Manager / bintray

#

would be fancy to get it on github so would be easier to fork and share modifications to it

#

right now, there's no DOTS packages on the github at all, so wondering if this is some policy?

#

well, there's mathematics package there, but that's about it

ocean horizon Apr 15, 2019, 9:24 AM

#

Unfortunately it's a separate team that deals with that so I have no idea what the plans are there. I don't think there's a specific policy though. The physics stuff is so new that I think it's just "how it is" for now.

lapis plaza Apr 15, 2019, 9:25 AM

#

yeah, I hope there's not 😃

#

but in case it's Unity not wanting to share the packages fully open, I'd be fine even with private repos with requirement to link your github account like many companies do

#

main thing would be to get access to the repos, not the visibility

#

so yeah, will wait and see what happens

hollow echo Apr 15, 2019, 11:42 AM

#

https://twitter.com/AnthonyYakovlev/status/1117736994883416064

Anthony Yakovlev (@AnthonyYakovlev)

Exposing a thread-safe Physics.BakePhysicsMesh(meshInstanceId, isConvex) function to bake the data for MeshCollider on-demand from any thread. Usage example with C# job system (generate N meshes and bake off the main thread): https://t.co/vWIQ64yVqG

tropic hamlet Apr 15, 2019, 1:52 PM

#

What's the typical use cases? Dynamic destruction?

hollow echo Apr 15, 2019, 1:56 PM

#

Perhaps - I was hoping someone would tell me 😛

tropic hamlet Apr 15, 2019, 1:57 PM

#

Terrain mesh collider ?

#

shrugs

hollow echo Apr 15, 2019, 1:57 PM

#

I know someone working on a skinned mesh slicer and it'd be handy for that I imagine

tropic hamlet Apr 15, 2019, 1:57 PM

#

yeah just seems odd Unity would want to focus time on this thing especially

#

order probably came from zeus

hollow echo Apr 15, 2019, 1:59 PM

#

It may not have taken much work, as it could have been written for internal systems and is now just being exposed

tropic hamlet Apr 15, 2019, 1:59 PM

#

I don't have any blocking issues with physx, but the new unity physics lack of even 1 stacked object means I pretty much need to keeo both physx and new physics running in the same project - which is bad since I need to double colliders up on everything...

pearl warren Apr 15, 2019, 1:59 PM

#

why you need the new physics?

tropic hamlet Apr 15, 2019, 2:00 PM

#

for networking it will be a no-brainer

#

but I can cope just using it for character-to-character collision and char to env plus hitscan stuff

#

stacking and more is for player side dynamics mostly, that is it won't break the game being slightly inaccurate

#

just doubling up on colliders seems a bit off.

ocean horizon Apr 15, 2019, 2:02 PM

#

@tropic hamlet So you know, Anthony isn't working on the DOTS physics system at all.

tropic hamlet Apr 15, 2019, 2:02 PM

#

I suspected as much

ocean horizon Apr 15, 2019, 2:03 PM

#

A feature such as mesh baking isn't prioritised again DOTS physics. It's completely separate.

tropic hamlet Apr 15, 2019, 2:03 PM

#

What's it for?

ocean horizon Apr 15, 2019, 2:04 PM

#

Better wait for Anthony to turn up. I'll ping him.

tropic hamlet Apr 15, 2019, 2:05 PM

#

Thanks! I don't anticipate giving up Physx any time soon given the limitations of dots-physics

ocean horizon Apr 15, 2019, 2:05 PM

#

I wouldn't expect anyone to. It's so hot-off-the-press and still in an early release state.

tropic hamlet Apr 15, 2019, 2:06 PM

#

I think really the adjustment to make is that "Unity is sharing things that are only a month or two old" vs before

ocean horizon Apr 15, 2019, 2:07 PM

#

Yep. They are drinking in all the feedback though and are eager to improve. High velocity on it from what I can see.

tropic hamlet Apr 15, 2019, 2:07 PM

#

Nice energy but can also lead to pressure for things that are basically bad to ask for because most of your customers (me included) will not be specialised like you are

#

I just hope "no" is a valid choice on Unity's side :D

ocean horizon Apr 15, 2019, 2:08 PM

#

I don't see that, seems a little pessimistic if you don't mind me saying sir. There's a lot of smart and open minds on it from what I can see.

#

Not saying I'm correct. Just that it's early days, they are logging all the feedback from what I can see.

#

Taking forums posts and making notes of them etc. Certainly not ignoring stuff. The drive seems to be coming directly from that feedback.

tropic hamlet Apr 15, 2019, 2:09 PM

#

I think you misunderstood me: I mean it's easy for someone like me to demand and ask for things (or many like me) but Unity should feel OK about saying no to that and explaining why if it's a common request because it's easy to chase customer demands down dead ends.

ocean horizon Apr 15, 2019, 2:10 PM

#

Oh sorry, I misinterpreted. No offense.

tropic hamlet Apr 15, 2019, 2:10 PM

#

course not, I can't be offended by text :P

ocean horizon Apr 15, 2019, 2:10 PM

#

Glad to hear it, some are though. 😃

tropic hamlet Apr 15, 2019, 2:10 PM

#

I know, sadly. despite it being characters on a hard drive and not really there in the room :P

#

Was it your work that enabled GC free collision events like OnCollisionStay?

ocean horizon Apr 15, 2019, 2:11 PM

#

Yes.

tropic hamlet Apr 15, 2019, 2:12 PM

#

Thanks a lot. Made my codebase so much cleaner and allowed me to make a better game.

ocean horizon Apr 15, 2019, 2:12 PM

#

Awesome. It's sad that it was like that in the first place.

tropic hamlet Apr 15, 2019, 2:13 PM

#

Technical debt :) btw I did a ball-popping game in Torque2D once but sold it so I had to delete all my torque posts :( but thanks for back then too :D

ocean horizon Apr 15, 2019, 2:13 PM

#

Hey hey. T2D was fun to work on although I never liked TorqueScript.

#

The engine was a real mess too.

#

But the GG guys were top-notch.

tropic hamlet Apr 15, 2019, 2:14 PM

#

Yeah... and if you dared ask for documentation then you'd be told to read a mess, basically.

#

I did have fun with it back then, for me cool to use a fully-fledged commercial engine

#

T2D was your pet project before them wasn't it?

ocean horizon Apr 15, 2019, 2:16 PM

#

T2D was my pet project. I started on it just as my first daughter was born. Spent late nights coding with a baby on my lap. I have no idea how the hell I managed to do that.

tropic hamlet Apr 15, 2019, 2:16 PM

#

:D

ocean horizon Apr 15, 2019, 2:17 PM

#

T2D wasn't the best thing in the world but it was fun to do. I learned a lot of things, mostly of what not to do.

#

but isn't the way of programming? 😉

tropic hamlet Apr 15, 2019, 2:17 PM

#

I enjoyed it immensely but also the legacy of that project is kind of living on right here isnt it?

ocean horizon Apr 15, 2019, 2:18 PM

#

In some ways I guess it is.

#

Glad I've not got a baby on my lap these days

lapis plaza Apr 15, 2019, 2:18 PM

#

@tropic hamlet baking/cooking physx colliders is way to turn your collision data into more optimized physx structure, you need to bake the data if you do procedurally generated meshes runtime for example

#

this is traditionally slow and lagging part of the proc gen stuff as it usually happens in the main thread in game engines

#

being able to offload it to threads is a big help if you do this kind of stuff

#

you have to bake the data too, it's not optional afaik

tropic hamlet Apr 15, 2019, 2:20 PM

#

Haha I'll never be able to read your posts again Melv, without imagining you with a nappy in one hand, a laptop in another and a bit of chaos at 4am!

ocean horizon Apr 15, 2019, 2:33 PM

#

It was pretty much that right there sir.

tropic hamlet Apr 15, 2019, 2:40 PM

#

I think that sounds familiar to be honest.

molten raft Apr 15, 2019, 4:50 PM

#

I have an object that can snap into place on another object. I use a fixed joint to connect them, but because one has gravity enabled it hangs a bit lower than the other one. Is there a better way to connect two rigidbodies?

sly violet Apr 15, 2019, 7:28 PM

#

I'm surprised it's noticeable. Does it need to be a dynamically breakable joint (i.e. does it have a break force/torque that isn't infinite)? If not, you could just make them one rigidbody while snapped together...

molten raft Apr 16, 2019, 7:04 AM

#

i decided to manually set the position and rotation to the same as that made it look right

molten raft Apr 16, 2019, 8:29 PM

#

I have a ship where i have objects i can move around. When i accelerate the ship, the objects fly around. Is there a way to prevent this without manually setting the position every frame?

lapis plaza Apr 16, 2019, 8:31 PM

#

common trick would be to simulate the ship inside parts on separate static location and just mirror the physics into visual target

molten raft Apr 16, 2019, 8:36 PM

#

im not quite sure what you mean, but using multiple cameras?

lapis plaza Apr 16, 2019, 8:38 PM

#

by putting physics part of the ship interior into different location, you simulate it normally there, then just send the transform changes to the moving ship for visual representation which can tilt however it wants

#

basically there's two places then where you sim physics: at the static location that has the inside parts and the ship itself that is moved by physics

molten raft Apr 16, 2019, 8:39 PM

#

i see, but wouldnt it be a lot of work?

lapis plaza Apr 16, 2019, 8:39 PM

#

depends how you view it

molten raft Apr 16, 2019, 8:39 PM

#

and having double of everything wouldnt be that efficient i suppose

lapis plaza Apr 16, 2019, 8:39 PM

#

you don't really double everything but gameobjects, but even still the gameobjects on the other part has meshes and other part has the colliders and rigidbodies

#

so it's not really duplicate, just split visual/simulation

#

there's one simpler trick that many fps game have done for a moving train but it's bit harder for a boat that rocks

#

on train, they just move the world around the train, train stays static so you'll not have issues moving around it

molten raft Apr 16, 2019, 8:42 PM

#

i thought a bit about that, but it seems like a not so good workaround for unity physics

#

what if i add another ship cause of coop or something like that

lapis plaza Apr 16, 2019, 8:43 PM

#

for the first example, that would work, you'd just have to swap the chars between physics spaces on fly

#

like if players jumps from ship to another, it would temporarily be in world space then again in ship space

molten raft Apr 16, 2019, 8:44 PM

#

i have a vr game and the user has to move stuff around when the ship gets hit, but when i accelerate everything goes nuts

lapis plaza Apr 16, 2019, 8:44 PM

#

it's not super hard to do but it's hacky alright

molten raft Apr 16, 2019, 8:44 PM

#

so i was thinking that using multiple cameras could work

lapis plaza Apr 16, 2019, 8:45 PM

#

how the cameras help there?

#

also, game physics are hacked together a lot

#

you just need to know how the physics engine work to bend their rules to your will

molten raft Apr 16, 2019, 8:46 PM

#

you have one camera at where the player would be in the real physics ship

#

and another one in the "fake" ship with all the movable objects

#

draw the first one and then on top draw the second one

#

if there are holes in the 2. one it will be filled by the 1. one

#

wow it seems very easy in unity

#

its working now :-)

lapis plaza Apr 16, 2019, 9:00 PM

#

that works with built-in renderer now, LWRP and HDRP doesn't support that anymore

#

but there's some camera stacking coming to LWRP at some point

molten raft Apr 16, 2019, 9:18 PM

#

~~Whats that?~~

tardy spear Apr 17, 2019, 1:40 PM

#

Just in case anyone wants to try some new stuff, here is an update to the experimental physics branch: https://forum.unity.com/threads/physx-4-0-in-unity-experimental-builds.634960/page-2#post-4439872

Unity Forum

PhysX 4.0 in Unity - experimental builds

That said, why would some be unhappy.

Yeah, i don't want to sound too negative here. The new physics engine looks amazing! There is a bunch of stuff...

#

(physx 4.1, articulations, Physics.BakePhysicsMesh)

lapis plaza Apr 17, 2019, 1:41 PM

#

so, physx 4 is currently coming to 2019.3?

tardy spear Apr 17, 2019, 1:41 PM

#

well, if I manage to get it stable enough for production - yes

lapis plaza Apr 17, 2019, 1:42 PM

#

I guess 2019.3 alphas will then be first ones that can build the player?

tardy spear Apr 17, 2019, 1:42 PM

#

so any feedback helps really

lapis plaza Apr 17, 2019, 1:42 PM

#

if it makes it that is

tardy spear Apr 17, 2019, 1:42 PM

#

players - yes

lapis plaza Apr 17, 2019, 1:42 PM

#

is it really worth making 2019.1 version since 2019.3 alphas are this close?

#

or do you need it for testing?

tardy spear Apr 17, 2019, 1:43 PM

#

I hope to attract more people. It's always a balance. I could have published builds off trunk on day 1, but then many complain nothing works there, so no reason to care.

#

That's been my experience with alphas over the years. "Oh, a new alpha? Nice. I'll check when it's stable in a year" At which point the feedback is too late

lapis plaza Apr 17, 2019, 1:44 PM

#

heh

#

yeah, I've noticed

tardy spear Apr 17, 2019, 1:45 PM

#

So this time around I'll spend more time testing in older versions rather than in the newer ones. I tried that when updating to PhysX 3.4 in Unity 2018.3 and was happy with the results.

lapis plaza Apr 17, 2019, 1:45 PM

#

yeah, I guess 2019.1 makes sense then 😃

tardy spear Apr 17, 2019, 1:46 PM

#

what about you? do you personally need 19.3 alpha to try out the new stuff?

lapis plaza Apr 17, 2019, 1:46 PM

#

I'm pretty much on the bleeding edge due to HDRP

#

2019.1 is first I could use in production but there are constantly updates that make it better on upcoming versions, and considering Unity's current TECH / LTS plan, IMO it makes sense to work toward the LTS if you can't stick to previous years LTS

#

on HDRP, it's targeted to get out of preview at the 2019.3 so I don't really see issues on using bleeding edge until Unity reaches that version (and 2019 LTS later on)

tropic hamlet Apr 17, 2019, 1:49 PM

#

@tardy spear awesome thanks for your work!

lapis plaza Apr 17, 2019, 1:49 PM

#

btw, is there any sample / programmer doc for the articulations?

#

or does it work similarly to existing joints ?

#

I've never setup these on physx despite them being around

tardy spear Apr 17, 2019, 1:50 PM

#

yeah, if you check the post I mentioned above there is some programmer-quality doc there

lapis plaza Apr 17, 2019, 1:50 PM

#

ah, thanks 😃

#

will try it out

tardy spear Apr 17, 2019, 1:51 PM

#

thanks. let me know what you think.

lapis plaza Apr 17, 2019, 1:51 PM

#

oh nice, sample project too

#

doc looked informative

tardy spear Apr 17, 2019, 1:52 PM

#

there will be proper docs just a bit later. for now I wrote a few pages of ideas dump + physx links to get ppl started

#

Interesting to see if anyone's opinionated about the mesh cooking stuff.

lapis plaza Apr 17, 2019, 1:55 PM

#

yeah, that's fine at this point IMO 😃

#

people who use this kind of preview probably know a bit better what to expect from physics in general

lapis plaza Apr 17, 2019, 2:16 PM

#

I'd want to try something like this

📎 unknown.png

#

it's using reduced coordinate articulations, but since these can't make loops, how did they implement the dual wishbone suspension?

#

here's another sample which I believe is using same setup: https://www.youtube.com/watch?v=6dATi4-wb3o&feature=youtu.be&t=278

YouTube

Pierre Terdiman

PhysX 4.0 in PEEL 2.0 showreel

A live-demo (not edited!) of PhysX 4.0 in PEEL 2.0. This is CPU PhysX. The GPU version would be faster but it currently has a big lag at startup time which...

▶ Play video

#

in my experience the regular joints just don't cut it if you need a stable vehicle in such setup, but articulations could work

#

wonder if that first one is on some physx sample

tardy spear Apr 17, 2019, 4:28 PM

#

I asked nvidia and the reply is basically they combined articulations and usual joints

#

(this is not yet supported in my custom build I shared earlier today)

lapis plaza Apr 17, 2019, 4:39 PM

#

I kinda suspected that as you need to have link

#

thanks for asking 😃

#

I just tried making something with this but the articulation joint setup is pretty unintuitive now without any visual cue where the anchors are and how they align

#

I haven't tried connecting pvd but you really should be able to conf the anchors without connecting a debugger 😃

tardy spear Apr 17, 2019, 4:42 PM

#

in my test project there is a Readback script that shows the anchors

lapis plaza Apr 17, 2019, 4:43 PM

#

ah, I'll check it out, I haven't really used regular joints in ages with Unity physics even, maybe it's always been like this 😃

#

I did check the sample, just didn't realize there were a script for this

#

ah, found it

tardy spear Apr 17, 2019, 4:49 PM

#

you may have to enable the gizmo drawing stuff in the gizmo settings box

#

but you're right, I'll add some anchor points editor niceness in the upcoming builds

#

Just noticed some misunderstanding happening around Physics.BakePhysicsMesh above in the thread. I wanted to clarify that it's been on the todo list for ages, and let me assure there are plenty of people wondering about the options to have mesh baking off the main thread.

#

You can pre-bake meshes and then instantiate objects faster

lapis plaza Apr 17, 2019, 5:01 PM

#

hmmmm, scaling transforms is also not supported atm, not sure if you mentioned it on the doc

#

after scaling my cubes, my visuals and physics don't match anymore 😃

tardy spear Apr 17, 2019, 5:02 PM

#

It was a day of work, and I hope for some great impact actually on everyone who deals with MeshCollider really.

lapis plaza Apr 17, 2019, 5:02 PM

#

yeah, the off main thread baking sounds like a nice thing, especially for proc gen

tardy spear Apr 17, 2019, 5:03 PM

#

For those who don't know, I've been on a mission to improve mesh cooking time: https://twitter.com/AnthonyYakovlev/status/1031940663447900161, because I care, and plenty of devs care too.

#

So for that reason, it's weird for me to read something like "why would you prioritise this vs object stacking in Unity.Physics". You have to know it's happening in different galaxies really.

lapis plaza Apr 17, 2019, 5:05 PM

#

people don't necessary know it's different team as it would be logical assumption it's the same people (if you don't know better)

tardy spear Apr 17, 2019, 5:07 PM

#

I agree about "people", but I'm not sure why would that be coming from hippocoder, that leaves me puzzled.

lapis plaza Apr 17, 2019, 5:13 PM

#

tbh, the stacking isn't the most critical issue on Unity Physics atm even

#

to get back to the wishbone suspension thing, you said nvidia used combination of regular joints.. you can do that with articulations?

#

I guess it's just part of the outside physics from the articulation system then

tardy spear Apr 17, 2019, 5:17 PM

#

yes you can. An articulation is a type of rigidbody so one can attach the joints to it. However that's not yet possible with the build on forums

#

"You can add normal pxjoints connecting links in an articulation. Just be aware that those joints are solved by the pgs/tgs solver so, unlike the rest of the articulation's mounts, they can exhibit error. Also be cautious about over-constraining the system because articulations don't like it when joints try to constrain already constrained axes. If the joint axes drift due to normal fp rounding errors, the solver can calculate enormous impulses to correct tiny drifts due to articulations being infinitely stiff. We are pretty good at catching/removing these degenerate constraints in the solver, but it's still something to bear in mind. (edited) "

lapis plaza Apr 17, 2019, 5:20 PM

#

ah, so much to learn

#

In addition to applying impulses and setting joint positions and velocities, it is possible to interact with the articulation links through the application of joint forces/torques, which behave like an actuator. forces are not implemented at this point?

#

I did see the comment about drive parameters but it's different I think

tardy spear Apr 17, 2019, 5:24 PM

#

joints will come in the next builds

lapis plaza Apr 17, 2019, 5:24 PM

#

I'm mainly thinking how to apply custom forces to the articulated bodies now

tardy spear Apr 17, 2019, 5:26 PM

#

that will also come.

#

for now you have drives

#

maybe I can add forces if you need them and rebuild tomorrow 😃

lapis plaza Apr 17, 2019, 5:29 PM

#

forces would let me bypass the physx friction mainly, but it's not really urgent for testing this

#

also, even with forces and regular joints allowed with articulations, I don't think it would be feasible to build anything but slow moving vehicles with this setup anyway as we can't filter the collisions on the wheels (they would still explode on some collision seams for sure)

#

but I'm genuinely interested on making a proto like on those physx videos on articulations as I think it's still cool demo

#

btw, did physx let you update those drive values after the thing was created?

#

like, could you update the drive params at runtime?

#

I didn't see a way to set those from code on current build

tardy spear Apr 17, 2019, 5:38 PM

#

I think this should actually work right now

#

📎 Screenshot_2019-04-17_at_20.38.50.png

lapis plaza Apr 17, 2019, 5:40 PM

#

there's nothing on the current builds metadata for that drive

#

📎 unknown.png

#

I mean, I can access that drive object but can't do anything with it

#

or I'm just missing something 😃

tardy spear Apr 17, 2019, 6:10 PM

#

damn! you're right.

#

I copied code from C where fields are public by default in structs. Well, shit happens, sorry

lapis plaza Apr 17, 2019, 6:19 PM

#

np, that's what the test builds are for, right? 😄

fallow gorge Apr 17, 2019, 6:44 PM

#

what does the ? signify with

                RigidBody? hitBody = null;

fallow gorge Apr 17, 2019, 7:05 PM

#

sorry in conjunction with the new physics package

#

in the ecs physics samples MousePickBehaviour.cs, theres a query for a rigidbody, doing RigidBody hitBody = null isnt the same as adding the ?(nor does it work)

lapis plaza Apr 17, 2019, 7:17 PM

#

seem to be https://docs.microsoft.com/en-us/dotnet/csharp/language-reference/operators/null-conditional-operators

Null-conditional operators - C# Reference

#

but I don't get it

#

returns null if the left-hand operand evaluates to null

#

that returns null either way, so what's the point?

#

oh, it's probably used later on

fallow gorge Apr 17, 2019, 7:20 PM

#

ah I see, thanks @lapis plaza

sly violet Apr 17, 2019, 8:11 PM

#

I think that makes it nullable? Except Rigidbody should already be nullable so I dunno?

opaque ledge Apr 17, 2019, 8:47 PM

#

It's technically a ternary operator. The normal usage is -> (if a) ? (do b) : (if not a, do c)
In this case the ternary (?) is acting as a null check. so -> (if a == null) ? (then b)
The other end of that would have been the else case of the ternary the x ? a : b.. but if it's null then you just don't do anything soo x ? a

#

commonly its just referred to as a null check

#

so in this instance, RigidBody? hitBody = null; is the same as if(Rigidbody != null) { hitbody = null}

#

The thing that doesn't make sense in that code sample is that they are checking against the class, not an instance of that class

lapis plaza Apr 17, 2019, 8:59 PM

#

ternary operator would require that : to be there

#

the ms doc page I linked seems to fit into this case way better

sly violet Apr 17, 2019, 9:00 PM

#

A '?' after a Type makes it a nullable type.

lapis plaza Apr 17, 2019, 9:00 PM

#

seems so

sly violet Apr 17, 2019, 9:01 PM

#

Not unusual for say "int?", but doesn't seem meaningful for a Class?

lapis plaza Apr 17, 2019, 9:01 PM

#

could be just some logic error

sly violet Apr 17, 2019, 9:02 PM

#

https://docs.microsoft.com/en-us/dotnet/csharp/programming-guide/nullable-types/using-nullable-types

Using nullable types - C# Programming Guide

Learn how to work with C# nullable types

#

It says you can't create a nullable Reference type, so it should error, I think?

#

https://docs.microsoft.com/en-us/dotnet/csharp/programming-guide/nullable-types/

Nullable types - C# Programming Guide

Learn about C# nullable types and how to use them

lapis plaza Apr 17, 2019, 9:08 PM

#

ah, it's actually not the Unity's old rigidbody class, it's Unity.Physics.Rigidbody struct

#

the old class based RB would have been Rigidbody, not RigidBody 😃

maiden grove Apr 17, 2019, 9:08 PM

#

wait - is that a convention?

lapis plaza Apr 17, 2019, 9:09 PM

#

@sly violet so this explains why they do that (as structs are value type)

#

almost everything in Unity's ECS side are done in structs afaik

hollow echo Apr 17, 2019, 10:56 PM

#

It's perhaps worth noting that in C#8 update enables you to use nullable reference types, and that nothing is nullable unless you specify it to be. Which is going to be strange to see how Unity deals with that change when we get that update. (Probably should discuss further in #💻┃code-beginner if interested)

bleak folio Apr 17, 2019, 11:24 PM

#

You have to turn that on explicitly, so I guess Unity is going to handle it just as it does with unsafe code

tropic hamlet Apr 18, 2019, 12:02 AM

#

it's not coming from me. I know already you have nothing to do with unity physics so I assume you mis-read me.

#

@tardy spear

#

likely I was replying to something else offtopic

#

AND! I really love your work man. I'm using Physx (your work) in my game that will ship. I can't do it without your work so far, thank you.

fallow gorge Apr 18, 2019, 4:36 PM

#

stacking!!! https://forum.unity.com/threads/unity-physics-discussion.646486/page-6#post-4444927

Unity Forum

Unity Physics Discussion

So an update:

I have managed to get the errors to go away by adding the ConvertToEntity.cs-scripture to the root gameObject, but the skinned mesh...

lapis plaza Apr 18, 2019, 5:18 PM

#

heh, I was just going to post that 😄

#

Specific issues to be addressed in the near term include:

A solution for stable stacking that does not require adding extra state
 - We are currently seeing reasonable stacking on pyramids up to 5 blocks high and across a range of scales.
 - The seesaw sample is now stable and can be more complex.
 - Work is still ongoing, balancing stack stability with behavioural anomalies.

A faster and more robust convex hull generator
 - The convex hull generator has been improved, though work is still progressing on removing unnecessary coplanar triangles.
 - A faster hull generator will also allow visualisation in the editor and potential runtime usecases.
 - More use case examples are being worked on across the board (e.g. trigger volumes)

In the medium term:
 - We continue to work on getting the Havok Physics integration ready for its first release in summer.
 - Performance improvements in Unity Physics are being investigated, in many of the areas raised here in the forum (e.g. removing Complete calls to scheduled Jobs were possible)
 - APIs are planned to allow a full simulation step within a single job (e.g. for predictions / mini simulations)

#

I hope the perf part will still happen sooner than later as current perf isn't all that great 😃

#

but the plan is looking good

#

Breakable or deformable joints are not directly on the roadmap, as such. That said, we do need to add joint impulse reporting so your own systems can check for violations and remove the associated Joint entity (breaking the Joint), or change the Joint parameters (deforming the Joint). Expect a new Joint demo when reporting is added (though I don't have an ETA just yet). Articulations are not planned.

tropic hamlet Apr 18, 2019, 5:30 PM

#

I see.

#

good.

fallow gorge Apr 18, 2019, 5:30 PM

#

what did articulations refer to in your question?

lapis plaza Apr 18, 2019, 5:30 PM

#

this definitely means there will be community extensions: Sleeping is currently out of scope for Unity physics. The design pillars for Unity Physics are for it to be stateless and also simple enough for users to confidently modify to their unique needs. Sleeping is something that feels simple on the surface but actually adds lots of complexity to a physics engine implementation, so does not fit the design goals. A primary goal of the Havok Physics integration will exist to address the use cases where a stateful engine really does becomes necessary - such as scenes that have enough settled rigid bodies, such that sleeping the inactive ones becomes important.

tropic hamlet Apr 18, 2019, 5:30 PM

#

I will allow Unity to continue without punishment for a bit longer.

fallow gorge Apr 18, 2019, 5:31 PM

#

I expect you to crack the whip if the havok license cost is outrageous 😄

lapis plaza Apr 18, 2019, 5:31 PM

#

I have no plans to use Havok atm 😄

#

it's yet another closed source thing

#

even PhysX is more open now 😄

#

and for articulations, I mean algos that can solve a chain of joints at once

#

like https://en.wikipedia.org/wiki/Featherstone's_algorithm for example

Featherstone%27s_algorithm

#

PhysX and Bullet both implement this

#

but yeah, I didn't expect them to support it as Unity Physics will be bare bones setup

fallow gorge Apr 18, 2019, 5:34 PM

#

so far perf seems good enough for me but the lure of cached performance sounds too good for me to resist if havok is affordable(for me)

#

my maths is too terrible to make compelling use of the source 😦

lapis plaza Apr 18, 2019, 5:35 PM

#

you can do naive sleeping and caching on your own

tropic hamlet Apr 18, 2019, 5:35 PM

#

Well I would be aggressively sleeping most things anyway so being able to rewind and resim, it's too powerful to ignore really.

lapis plaza Apr 18, 2019, 5:35 PM

#

like, if you do large world, you can just opt out simulation for parts that are not nearby for example

tropic hamlet Apr 18, 2019, 5:36 PM

#

Yeah it's ECS so your entire planet will probably cost no more than a couple of memcpy since most of it would be sleeping, right?

lapis plaza Apr 18, 2019, 5:38 PM

#

also, trivial sleeping would just keep track of physics steps the rigidbody is below the set velocity threshold and then have total amount after it would sleep, this would add only like one byte of data for each stored state so you can just rewind and set this counter yourself along with the other values when you rewind and resim

#

it would still be fully deterministic

#

of course everything could be made more optimized and complicated but I would expect one to get pretty far with simple and naive optimizations like that

tropic hamlet Apr 18, 2019, 5:40 PM

#

ok how about disabling?

#

after time

#

i dont agree that sleeping is out of scope.

lapis plaza Apr 18, 2019, 5:41 PM

#

you can't disable the collisions

#

as it would never wake up then

#

but when it's sleeping, you just ignore it

fallow gorge Apr 18, 2019, 5:41 PM

#

i guess i dont know how you would wake it up via physics if you were to sleep a rigidbody, like if its removed from the sim how do you use the sim to wake it up?

lapis plaza Apr 18, 2019, 5:41 PM

#

you could use a component to tag the live ones

#

and use that for the simulation

#

it would never run the sim on the sleeping ones then

#

well, you do need to make collision checks for the sleeping bodies

#

but you don't have to run simulation on them

#

like, you know sleeping ones are like static collisions while not active

#

so put it on that pile while sleeping

#

I think Unity Physics treats these separately already

#

dynamic vs static

#

@tropic hamlet I feel that's some politics involved here, it's not what is feasible or in scope, they just want to keep Unity Physics simple to maintain and have a reason for people to pay for Havok

#

if they made Unity Physics better than Havok, nobody would get Havok

#

it's probably also question of resources Unity is willing to put in maintaining the physics module

fallow gorge Apr 18, 2019, 5:45 PM

#

definitely sounds like politics for the github/source answer

lapis plaza Apr 18, 2019, 5:45 PM

#

yeah, I really wonder if we'll ever see this on github

#

I sure hope it happens

#

also wonder if they do things like OverlapSphere

#

now they only support overlap check for AABB box

#

collider casts work with different physics shapes so that's not an issue

tropic hamlet Apr 18, 2019, 5:53 PM

#

posted a reply

#

regarding sleeping

#

I think we can ask them to provide the neccessary groundwork to cover most of our optimisation requirements without sleeping

#

just provide the management for it so we can decide on simulation state for groups or individuals

#

its good enough

#

"if they made Unity Physics better than Havok, nobody would get Havok" I don't want f-ing havok :)

lapis plaza Apr 18, 2019, 5:55 PM

#

yeah, I agree

tropic hamlet Apr 18, 2019, 5:55 PM

#

some stacking was necessary and its enough. but now the sleeping issue... so i need to do it myself, but require public api, then i do not need to hack source.

lapis plaza Apr 18, 2019, 5:56 PM

#

just need the repo on github

#

there's no real reason for each implement this separately :p

tropic hamlet Apr 18, 2019, 5:56 PM

#

yes but ... why if a simple api of management is public, it's enoiugh by default for most people

lapis plaza Apr 18, 2019, 5:56 PM

#

ah, if they make the api public

#

people can share example projects using it

tropic hamlet Apr 18, 2019, 5:56 PM

#

most people (all) will want some form of sim on/off public yes

#

going to source for that is ridiculous

#

and we dont need sleep api then

#

it's a workaround so the havok guy can satisfy the business end, u see?

lapis plaza Apr 18, 2019, 5:57 PM

#

I'll do some testing on this at some point, but it's not a priority atm

#

yeah, I saw what you were going for there on the post

tropic hamlet Apr 18, 2019, 5:57 PM

#

sleeping is not allowed, for business purposes

#

really even physx has overkill for sleeping. most of us would be more than happy just turning on things locally :D

#

otherwise it beomes not very useful for local non mp games without source

#

and people may want to simulate some locally and some in mp

lapis plaza Apr 18, 2019, 11:37 PM

#

New demo to be added to the list then ;-) ```

fallow gorge Apr 19, 2019, 12:05 AM

#

i was actually just trying this and removing the velocity component constantly crashes the editor

lapis plaza Apr 19, 2019, 12:06 AM

#

you probably have Burst enabled

#

try without

#

I mean, I think I saw mention that Burst doesn't support adding or removing components on fly yet

fallow gorge Apr 19, 2019, 12:06 AM

#

no it was a componentsystem using postupdatecommands

lapis plaza Apr 19, 2019, 12:07 AM

#

the thing that didn't work?

fallow gorge Apr 19, 2019, 12:07 AM

#

yeah, super simple mainthread

lapis plaza Apr 19, 2019, 12:08 AM

#

it could require other changes too

fallow gorge Apr 19, 2019, 12:11 AM

#

public class ToggleSystem : ComponentSystem
{
    private EntityQuery hasvel;
    private EntityQuery novel;

    protected override void OnCreateManager()
    {
        base.OnCreateManager();
        

        hasvel = GetEntityQuery(new EntityQueryDesc
        {
            All = new ComponentType[] { typeof(PhysicsVelocity)  }
        });
        
        novel = GetEntityQuery(new EntityQueryDesc
        {
            All  = new ComponentType[] { },
            None = new ComponentType[] { typeof(PhysicsVelocity) }
        });
    }


    protected override void OnUpdate()
    {
        Entities.With(hasvel).ForEach((Entity entity) =>
        {
            
            if (Input.GetKey(KeyCode.Space))
            {
                PostUpdateCommands.RemoveComponent<PhysicsVelocity>(entity);
            }
        });
        
        Entities.With(novel).ForEach((Entity entity) =>
        {
            if (Input.GetKey(KeyCode.Delete))
            {
                PostUpdateCommands.AddComponent(entity, new PhysicsVelocity());
            }
        });
    }
}

#

shouldnt be anything wrong with that right?

lapis plaza Apr 19, 2019, 12:13 AM

#

did you check that without Burst?

#

it could still crash if the Unity Physics uses it

fallow gorge Apr 19, 2019, 12:15 AM

#

oh in another system?

#

unity really needs a recently opened scenes menu list

lapis plaza Apr 19, 2019, 12:18 AM

#

this sounds like it's crashing if you do the add/remove in a job tho

📎 unknown.png

#

but worth checking still

#

I mean, it's just toggling the setting on/off from the main menu

fallow gorge Apr 19, 2019, 12:22 AM

#

well i suppose its not crashing when I use this in the physics samples github project

#

but it spits out a ton of errors and the world goes wonky

#

did try without burst and also adding
[UpdateAfter(typeof(BuildPhysicsWorld))] to the top

#

postupdatecommands should be the same as using a command buffer with the EndSimulationEntityCommandBufferSystem right?

lapis plaza Apr 19, 2019, 12:32 AM

#

I wouldn't know 😃

fallow gorge Apr 19, 2019, 12:59 AM

#

could you link xoofx' post thread?

lapis plaza Apr 19, 2019, 1:06 AM

#

@fallow gorge https://forum.unity.com/threads/release-of-burst-1-0.662536/#post-4434778

Unity Forum

Release of burst 1.0

Hi All,

We have just published the final version of burst 1.0, more than 90 releases since the first package published last year, that's quite a...

fallow gorge Apr 19, 2019, 1:10 AM

#

thanks

fallow gorge Apr 19, 2019, 2:53 AM

#

ah finally got it working with a jobcomponent system with a proper job handle dependency to the physics world. now I cant remember what I planned on doing with it

peak panther Apr 19, 2019, 2:59 AM

#

I'm getting some oddities with Rigidbody.MovePosition

#

this is the code

#

📎 unknown.png

#

and even with no y, the rigidbody is moving down slightly when rubbing against a perfectly perpendicular wall

#

https://gyazo.com/ea036731247bdd42389c5f50a7936865

Gyazo

▶ Play video

#

this is the scene

#

📎 unknown.png

#

any reason as to why this is happening?

fallow gorge Apr 19, 2019, 3:10 AM

#

is that gravity acting on it? im not familiar with 2d physics(is gravity at 0?)

peak panther Apr 19, 2019, 3:11 AM

#

ya gravity scale is at 0

#

oh nvm then

#

I freezed rotation and all of the sudden no variation

#

coolio

sullen saddle Apr 19, 2019, 4:25 AM

#

With the new ECS/DOTS and the new Unity Physics, how can I lock the rotation? Do I need to write a system for that? No simple checkboxes anymore? Same thing to lock position in Z for example.

lapis plaza Apr 19, 2019, 7:18 AM

#

@sullen saddle see this post https://forum.unity.com/threads/unity-physics-discussion.646486/page-6#post-4445458

stuck bay Apr 19, 2019, 8:36 AM

#

comparing new ue4 physics to new unity physics is quite interesting, they seem to be on the opposite ends of the spectrum, new unity physics is specifically stateless, while new ue4 physics seem to extremely depend on states to keep things performant

fallow gorge Apr 19, 2019, 3:34 PM

#

full featured vs barebones, hoping the full havok integration competes with chaos

frail warren Apr 20, 2019, 7:24 PM

#

Hi guys. Does anyone know of problems with Physics.OverlapSphere at the moment? I've seen a few reports of unexpected behaviour, and I've just wasted an afternoon trying to implement something only to conclude it just doesn't seem to be working, at all?

#

For example, I place a primitive sphere of radius 0.5 with 0.5 radius sphere colliders at x, y, z, and perform OverlapSphere at x, y, z with a reasonable radius, and get nothing. With a very large radius, I detect the sphere.
Performing an OverlapSphere at 0,0,0 with seemingly any radius at all (including 0!) is detecting ALL colliders in the scene.

frail warren Apr 20, 2019, 7:47 PM

#

After a little more looking around it appears that Overlap is only working if the object has been created in a previous Physics tick/frame...?

lapis plaza Apr 20, 2019, 7:49 PM

#

probably

#

because it polls the result from existing physics scene

frail warren Apr 20, 2019, 7:52 PM

#

I mean it makes sense

#

I just can't find any documentation of this.

#

I've just set up a simple script to create a gameobject with collider on start, and then do a sphere and boxoverlap test on it then, at the first FixedUpdate, and the second FixedUpdate.

#

And I only start getting the results from the second.

#

The first 2 calls return nothing. 🤔

#

Closest I can see is If you move colliders from scripting or by animation, there needs to be at least one FixedUpdate executed so that the physics library can update it's data structures, before a SphereCast will hit the collider at it's new position.

#

Unfortunately this includes creating them.

sly violet Apr 21, 2019, 12:15 AM

#

@frail warren Have you tried Physics.SyncTransforms ? Not sure it works with collider creation... But I think it should.

frail warren Apr 21, 2019, 9:04 AM

#

I'll have to take a look

#

Thanks

frail warren Apr 21, 2019, 9:44 AM

#

@sly violet Yeah that seems to be working. Significantly faster I'd expect. Thanks.

rare cedar Apr 21, 2019, 8:19 PM

#

Hello every body good morning. Can I ask a question abaut gravity in 2D games here?

stuck bay Apr 21, 2019, 8:20 PM

#

sure, what do you want to ask? @rare cedar

rare cedar Apr 21, 2019, 8:22 PM

#

Okay thanks @stuck bay . Can i send you a private message in order to not disturb here?

stuck bay Apr 21, 2019, 8:22 PM

#

okay, i'm not good with gravity(especally with 2d) but i can try and help

fallow gorge Apr 21, 2019, 8:24 PM

#

its too quiet, you wont be disturbing anything @rare cedar

lapis plaza Apr 21, 2019, 8:39 PM

#

in general it's recommended to ask help here, you get more eyes on your problem + you don't take individual peoples time who may or may not be able to help you

#

most efficient way to get help is just to post the full question here

stuck bay Apr 21, 2019, 9:46 PM

#

mind if someone give me some help with steering a car with unity?, i ask in a different server but the response i got was someone start going on about drifting and linked to advance math stuff for realistic drifting

stuck bay Apr 22, 2019, 4:26 AM

#

anyone?

latent spoke Apr 22, 2019, 5:52 AM

#

anyone know what's wrong with terrian tree leaf that dont move at all after save and close

stuck bay Apr 22, 2019, 7:47 AM

#

ahhh, is that even about physics? @latent spoke

lapis plaza Apr 22, 2019, 7:53 AM

#

@stuck bay your question is super vague

#

you could get anything as answer with that base info

stuck bay Apr 22, 2019, 7:54 AM

#

well i haven't even ask the question yet

lapis plaza Apr 22, 2019, 7:54 AM

#

ah, it's usually best just to ask the actual question

stuck bay Apr 22, 2019, 7:54 AM

#

okay

#

here's a couple of my questions

#

or some problems i need help with, the first problem with my car is that when i steer and let go of the W key the forward momentum moves in the original direction of when the W key was pressed, so basically when i turn with some momentum the car moves sideways, another problem is that there's too much sideway momentum, like the car is on ice, those are the main two i am having trouble about

lapis plaza Apr 22, 2019, 7:59 AM

#

you use wheelcolliders?

stuck bay Apr 22, 2019, 7:59 AM

#

nope

#

i hate them, it just gives me alot of problems

lapis plaza Apr 22, 2019, 8:00 AM

#

well, then you have to describe how your vehicle physics is assembled for anyone to have even slightest idea on what's going on there 😃

stuck bay Apr 22, 2019, 8:00 AM

#

here's my code

#

    {
       Gear1,
       Gear2,
       Reverse

    };

    public GameObject FrontLeftWheel;
    public GameObject FrontRightWheel;

    public float Acceleration;
    public float Steering;
    public float TurnSpeed;
    public float RearWheelKick;
    public float MaxSpeed;
    public float CurSpeed;

    public Gears Gear;
    

    public Rigidbody m_rigidbody;

    float V;
    float H;
    

    void GetInput()
    {
        H = Input.GetAxis("Horizontal");
        V = Input.GetAxis("Vertical");
    }
    void GearChange()
    {
        if (CurSpeed <= 0)
        {
            Gear = Gears.Reverse;
        }
        if (CurSpeed <= 5 && CurSpeed > 0)
        {
            Gear = Gears.Gear1;
        }else if (CurSpeed <= 8 && CurSpeed >= 5)
        {
            Gear = Gears.Gear2;
        }
    }
    void Steer()
    {
        float Yangle = -Input.GetAxis("Horizontal");

        transform.Rotate(0, -Yangle * Steering * Time.fixedDeltaTime, 0);

        m_rigidbody.AddRelativeForce(-Vector3.right * CurSpeed * Yangle / 1);

    }


    private void FixedUpdate()
    {
        GetInput();
        GearChange();
        Steer();
        CurSpeed = transform.InverseTransformDirection(m_rigidbody.velocity).z;

        if (CurSpeed < MaxSpeed)
        {
            m_rigidbody.AddRelativeForce(Vector3.forward * V * Acceleration * Time.fixedDeltaTime);
        }
        
    } ```

#

btw i know some variables aren't use for anything, they where just for testing

#

also i want to know a good way of making steering effected by speed, i can't for the life of me find a tutorial on how to do this without making turning feel weird

lapis plaza Apr 22, 2019, 8:02 AM

#

if you make physics based vehicle, you don't do things like transform.Rotate 😃

#

you do everything with physics forces

stuck bay Apr 22, 2019, 8:03 AM

#

i did, but it was too slippery

lapis plaza Apr 22, 2019, 8:03 AM

#

what kind of handling is your target?

stuck bay Apr 22, 2019, 8:03 AM

#

handling?

lapis plaza Apr 22, 2019, 8:03 AM

#

what you need your vehicles for?

stuck bay Apr 22, 2019, 8:04 AM

#

for a game?

lapis plaza Apr 22, 2019, 8:05 AM

#

ok, so something very basic then

#

you know this tutorial? https://www.youtube.com/watch?v=LG1CtlFRmpU

YouTube

SpaceDustStudios

Space Dust Racing UE4 Arcade Vehicle Physics Tour

A rundown of the general principles used for arcade-style vehicle physics using Unreal Engine 4 for the upcoming party combat racer Space Dust Racing. Dev bl...

▶ Play video

#

(it has nothing ue4 specific in it)

stuck bay Apr 22, 2019, 8:06 AM

#

nope

lapis plaza Apr 22, 2019, 8:06 AM

#

that explains super bare bones arcade handling raycast vehicle

stuck bay Apr 22, 2019, 8:06 AM

#

i am trying to get a feel more with something like the Dust storm from burnout paradise

lapis plaza Apr 22, 2019, 8:07 AM

#

that's a lot harder then 😃

stuck bay Apr 22, 2019, 8:09 AM

#

okay, but can you help me with my code?, like how to make the turn effected by speed?, or the physics problems i listed? because so far none of this answers my problems that i been trying to solve for a while?

stuck bay Apr 22, 2019, 8:26 AM

#

well i fixed the sliding problem by lowering the X velocity when H is not held down

lapis plaza Apr 22, 2019, 8:26 AM

#

well, you can look at the video I linked and see how they did the turning there

#

it's basically similar solution to yours but they do turning via torque

wary geode Apr 22, 2019, 8:27 AM

#

How i can change position using RigidBody2D? I don't remember. AddForce?

lapis plaza Apr 22, 2019, 8:27 AM

#

but they also add extra force there based on velocity too

stuck bay Apr 22, 2019, 8:28 AM

#

@wary geode you can use alot of solutions like MovePosition(i dunno if it's in Unity 2d tho) but yeah AddForce can help make your character move

#

can someone at least help me with making turning effected by Speed?, because that's one thing the tutorial didn't answer and i can't find a solution too

lapis plaza Apr 22, 2019, 8:45 AM

#

@stuck bay
for rotating via torque: https://youtu.be/LG1CtlFRmpU?t=248
for keeping traction while turning (even without pressing throttle: https://youtu.be/LG1CtlFRmpU?t=285

stuck bay Apr 22, 2019, 8:45 AM

#

.......

lapis plaza Apr 22, 2019, 8:46 AM

#

you said your vehicle keeps moving in old direction if you stop hitting W

#

this is what fixes that

stuck bay Apr 22, 2019, 8:46 AM

#

NO

#

IT DOESN'T EVEN ANWSER MY QUESTION, I AM TURNING BY TORQUE

#

m_rigidbody.AddTorque(Vector3.up * H * Steering * Time.fixedDeltaTime, ForceMode.Acceleration);

lapis plaza Apr 22, 2019, 8:47 AM

#

typing on caps will not make people help you more

stuck bay Apr 22, 2019, 8:47 AM

#

well i am barely getting any help anyways so what's the point?

lapis plaza Apr 22, 2019, 8:47 AM

#

good luck 😃

stuck bay Apr 22, 2019, 8:48 AM

#

ugh, fking hell

latent spoke Apr 23, 2019, 1:29 AM

#

@JellyBelly#5965 I think i am in wrong place? I dont even know which will help with terrian tree issue?

median frost Apr 23, 2019, 8:24 PM

#

can I somehow find out in the collision even what kind of type the other collider has?
like if(collision.type == Circlecollider)?
and then just cast it as CircleCollider

hollow echo Apr 23, 2019, 10:31 PM

#

You can cast it with

CircleCollider circleC = collision as CircleCollider;
if(circleC != null){
//is CircleCollider
}```
@median frost if that's what you mean

sly violet Apr 23, 2019, 11:42 PM

#

Hmm. Usually you check, then cast:

if (other.collider is CircleCollider) {
    CircleCollider otherCircle = other.collider as CircleCollider;
    ...
}

hollow echo Apr 24, 2019, 12:25 AM

#

That's more expensive

#

Also you're casting safely after a check that tells you that you can cast unsafely

#

With the newer C# you can do the is check with a pattern match

median frost Apr 24, 2019, 6:51 AM

#

@hollow echo @sly violet thanks

#

I am doing a top down 2d rpg

#

do you guys know a simple way to make me "not able to walk through walls"?

#

this is my code: https://pastebin.com/Ww9ESXM0

Pastebin

[C#] private void OnTriggerEnter2D(Collider2D collision) ...

#

I have a small circle collider on the feet of my player and on the "foot" of a tree

#

I want that on collision just to set the player a bit back s.t. he doesn't move through

hollow echo Apr 24, 2019, 6:53 AM

#

You cannot use transform.Translate, you need to use a Rigidbody2D and apply forces

median frost Apr 24, 2019, 6:53 AM

#

this kind of works but it is a bit bumpy, sometimes jumps into a bit of the wrong direction

#

but I wanted to do the movement without forces since it seemed easier to me

#

but probably the reimplementation of exactly these issues makes it hard again, right?

hollow echo Apr 24, 2019, 6:54 AM

#

without forces you're not interacting with physics and hence walls are meaningless

median frost Apr 24, 2019, 6:54 AM

#

I mean I have no problem with forces or the physics behind it, just that if I then stop, I have to set forces to 0 etc instead of just not moving

#

yeah that is why I tried to reimplement the functionality myself but probably a bad idea

hollow echo Apr 24, 2019, 6:55 AM

#

You can do it manually with casts, and it might work better for you. There's an example here https://unity3d.com/learn/tutorials/topics/2d-game-creation/player-controller-script

median frost Apr 24, 2019, 6:56 AM

#

@hollow echo thanks! I'll take a look 😃

#

yup that was exactly what I needed

#

calculating the velocity yourself is a much more clever way than what I did

vapid crane Apr 24, 2019, 7:15 AM

#

anyone had a chance to mess with the new unity physics?

lapis plaza Apr 24, 2019, 8:13 AM

#

Pretty sure many have

#

If you have actual question, better ask that instead :)

zinc furnace Apr 24, 2019, 11:25 AM

#

anyone got five minutes to talk through an idea?

#

I have a project which is set in space, and involves very fast movement at times, and i am looking for a cast iron way to calculate collisions.

#

At times, it is clearly possible for an object to move far enough during a frames processing that the physics system is not going to calculate a collision because the objects dont technically move simultaneously. If the bullet moves say 5 units in one frame, and the ship also moves five units, but is three units long... theyll pass through the same space but wont collide.

#

I am thinking of a method using ecs.

#

Due to the nature of my game, a large portion of other aspects of the physics are unnecesary, i can then code my own as seems necesary really with ecs anyways.

#

So, i am thinking of a method where as the ship moves, during the frame, it leaves like a trail of bounding boxes of broad simplistic composition, which the bullets then calculate their own contacts with.

#

Whilst this does not provide an ideally accurate solution with regards to accuracy... if they are moving at the kind of speed it possibly would miss... but frankly its probably better to encourage more incorrect hits than allow too many incorrect misses. That is, assuming the gun is at least pointed at the right angle hehe

#

How does it sound?

lapis plaza Apr 24, 2019, 3:46 PM

#

I'd just use sweeps/raycasts

sly violet Apr 24, 2019, 3:50 PM

#

Two sweepcasts won't catch each other.

#

Will two rigidbodies with continuous dynamic collision detection catch each other in that scenario?

lapis plaza Apr 24, 2019, 4:14 PM

#

was more of thinking to sweep in front, like most character controllers do to detect if they are about to hit a wall

#

CCD helps but it's not always catching everything

sly violet Apr 24, 2019, 4:30 PM

#

CharacterControllers/sweeps are great for fast colliders hitting slow or stationary colliders. It seems inapplicable to the problem described, which is detecting multiple fast bodies on an intercepting course.

#

@zinc furnace : Can you live with them all being spheres? It's not too hard to calculate the closest approach of two velocity vectors.

lapis plaza Apr 24, 2019, 5:16 PM

#

ah, if you mean like two objects moving in completely different directions then sure

#

that kind of setup would catch head-on collisions tho

#

if you don't have a ton of objects, you could sort nearby ships that could potentially collide and the the point of collision from their trajectories, then calculate the moment both ships would be at that point and check if the time the overlap that point is within some threshold

zinc furnace Apr 24, 2019, 8:10 PM

#

Im using ecs ... no access to the usual physics, if i use movement systems, they'll just overwrite the physics every frame. The movement system would move the object then the physics would try to apply drag and such... All to no avail as the movement overrides the physics next frame.

#

Im also fairly sure that if i do sweeps then its a) not gonna catch each other, and b) gonna basically force me to use a hybrid system

#

@Olento, its a wide empty gulf of space, most of it.

#

I can, for instance, only calculate the physics to such detail when bullets are nearby...

#

Or reduce calculations to nil when nothing is close.

#

Excluding momentum and drag based stuff of course.

#

I have interest to use ecs as it will allow a lot of other important behaviours on the backend server, so makes sens on the client too.

#

Im thinking, after more consideration, that the ship will take its bounding box, and place it in interpolated positions. The result being the simplest collider shape would be used to cover its trail, and it would use a number of copies equal to how many needed to make the length continuos.

Also... The speeds required to meet the criteria of the issue im trying to avoid are such that the player's chance of really seeing if the hit was a hit are very low.

#

In my experience tho, if you fired bullets at more than 300 units per second, then theyd easily skip the target on a slow frame. If you try 600 they'd skip most frames, and if you tried 1000, you'll never hit. Altho, of course, i expect this to scale with your cpu/gpu power, which makes the whole scenario even less acceptable.

sly violet Apr 24, 2019, 8:43 PM

#

If you process on a fixed time step like most physics, then the skippage shouldn't be affected by the frame time, so that's one thing. 😉

#

I would again recommend finding the closest approach. It's not too difficult and will give you a very good answer.
First, you set one object as the reference point (probably the ship) and treat the other (bullet) as its position and velocity relative to the ship.
Then you solve for the closest approach, which tells you where and when that is:
https://en.wikipedia.org/wiki/Distance_from_a_point_to_a_line
From that you can determine whether two spheres collide in the given frame without using external physics at all.

Distance from a point to a line

The distance (or perpendicular distance) from a point to a line is the shortest distance from a fixed point to any point on a fixed infinite line in Euclidean geometry. It is the length of the line segment which joins the point to the line and is perpendicular to the line. Th...

#

Finally, if there's a lot of bullets, you'll probably want some way to only check the ones that are reasonably nearby.

rugged beacon Apr 25, 2019, 5:51 AM

#

Would this be a good spot for physics related code issues of a very basic level, or would that be better for #💻┃code-beginner

rugged beacon Apr 25, 2019, 6:10 AM

#

Right now I'm working on building out a basic movement player controller, back to basics here. The end case I am wanting to have is directional input system that generates relative force as a constant on the player.
this.GetComponent<Rigidbody>().AddRelativeForce(new Vector3(0, 0, verticalSpeed));

The issue at this current moment is that by accelerating the player in a direction I seem to be negating my downward/gravity based velocity, essentially causing a gliding behavior.

sly violet Apr 25, 2019, 6:11 AM

#

Odd. AddForce shouldn't have that sort of effect.

rugged beacon Apr 25, 2019, 6:13 AM

#

It's very odd, this is on FixedUpdate and I am only inputting velocity in the chosen forward or horizontal axes relative to the player

#

Other than a simple Jump AddRelativeForce I'm not touching my vertical velocity

sly violet Apr 25, 2019, 6:14 AM

#

Is your gravity the built-in gravity?

rugged beacon Apr 25, 2019, 6:15 AM

#

At this point in time I believe so, the gravity is currently set to Y -9.81

#

It's the default gravity

#

Would the solution here be to just keep radically increasing my mass and the force I'm applying to get the downward velocity I want?

sly violet Apr 25, 2019, 6:33 AM

#

No. Gravity is mass-independent, anyway.

#

Well, the velocity imparted by it is mass independent, I mean.

#

Really, what you've told me doesn't make sense. Something else is up.

rugged beacon Apr 25, 2019, 6:34 AM

#

Right, my tired brain should have told me that

#

That's really the entirety of my movement code itself. My jump is essentially the same line but with the velocity applied in the up direction

#

Changing the drag also isn't changing my behavior here

tropic hamlet Apr 25, 2019, 1:00 PM

#

Add relative will be relative the the current orientation. Practically any kind of relative force will alter orientation.

marsh bobcat Apr 30, 2019, 11:10 AM

#

Weird problem...
I have this ship deck.
Has a Convex, Non-Trigger Mesh Collider.
Has a Non-Kinematic Rigid Body.
I have a Character Controller attached to my object.
My Character Controller can stand above the deck (falls off when the deck is made a Trigger)
But, I can't fire the OnCollision/OnTrigger events on my script (Ship Floor)
Am I missing something here?

📎 unknown.png

gusty sleet Apr 30, 2019, 11:12 AM

#

Make your Convex, Non-Trigger Mesh Collider a kinematic rigidbody

#

Or a static rigidbody

marsh bobcat Apr 30, 2019, 11:18 AM

#

that doesn't work either

#

📎 unknown.png

gusty sleet Apr 30, 2019, 11:19 AM

#

Are the layers correct?

marsh bobcat Apr 30, 2019, 11:21 AM

#

should be, my character controller can sit on top of it. and if i disable the deck it falls through

#

(normal Player/Floor layer interaction for my game)

gusty sleet Apr 30, 2019, 11:23 AM

#

CharacterController isn't a normal physics object

#

That's why it works

marsh bobcat Apr 30, 2019, 12:43 PM

#

hmm.. ok. so i just reversed the mechanic by checking for the floor manually by raycasting from the character controller

ripe jacinth Apr 30, 2019, 12:49 PM

#

OnTrigger takes Collider as a parameter, not Collision

#

And yeah, CharacterController is not a rigidbody. Usually it'

#

it's better to use a controller with rigidbody than character controller

sly violet May 1, 2019, 2:41 AM

#

...I'm genuinely curious what would happen to your CharacterController if that ship starts rocking. My gut instinct is that it won't work well, but I'm not sure.

sleek prairie May 1, 2019, 5:08 PM

#

Is there a way to make an object recognize a collision on the same frame it's activated?

#

on frame 500 the collider of my object is enabled like this someObject.GetComponent<Collider>().enabled = true;

#

but until frame 501 the OnTriggerEnter is not called

sly violet May 1, 2019, 6:15 PM

#

No. It won't get triggered until the next physics step. You might be able to use an Overlap function to detect the situation.

#

https://docs.unity3d.com/ScriptReference/Physics.OverlapSphere.html

#

Unfortunately, Unity 3d physics only has Overlap functions for primitive shapes, and not a generic Collider version.

sleek prairie May 1, 2019, 6:20 PM

#

@sly violet I see, thanks anyways!

#

Unless theres a way to force a physics step manually I think I'm out of luck on this one 😅

sly violet May 1, 2019, 6:21 PM

#

Well... There is:
https://docs.unity3d.com/ScriptReference/Physics.Simulate.html

#

Note that this will move everything with a velocity and all that an extra frame.

sleek prairie May 1, 2019, 6:25 PM

#

yeah I tried that one today, didn't seem to work

#

unless I did something wrong, which is entirely possible

#

oh wait

#

I had to turn off Physics.autosimulation

#

It seems to work now

#

Physics.autoSimulation = false;

//Enable colliders

if (//OnTriggerEnters have been called)
{
        //My code
}

Physics.Simulate(Time.fixedDeltaTime);```

#

so how does Physics.Simulate(float step); work?

#

the bigger the step the further the calculations are made?

sly violet May 1, 2019, 7:11 PM

#

It's the equivalent of the physics Fixed Timestep project setting in Time.

sleek prairie May 1, 2019, 7:18 PM

#

oh I see, I think I get it now

sleek prairie May 2, 2019, 5:40 PM

#

I'm trying to detect if an object with a Sphere Collider is inside of a Mesh Collider but they don't detect each other

#

OnTriggerStay only fires when the sphere collider is intersecting with the "surface" of the mesh collider

#

but if it's completely inside, nothing happens

#

Any idea about how to solve this? I need my mesh collider to be non-convex

sly violet May 2, 2019, 7:45 PM

#

A non-convex mesh collider cannot be volumetric (filled in) so you're basically out of luck. Can you put some sphere colliders in it such that your triggerer won't be able to avoid them all?

timid pebble May 4, 2019, 5:09 AM

#

I have a job that produces a NativeArrray<RaycastCommands> for batching later

        NativeArray<RaycastHit> rayHits = new NativeArray<RaycastHit>(commands.Length, Allocator.TempJob);
        JobHandle batchJob = RaycastCommand.ScheduleBatch(commands, rayHits, 1, setupJob);

        batchJob.Complete();
      
        for(int i = 0; i < rayHits.Length; i++)
        {
            if (rayHits[i].collider)
            {
                Debug.LogWarning("Hit " + rayHits[i].collider.name);
                continue;
            }

            break;
        }

but the batched ray never hits anything despite that I have checked the values with a normal ray

        RaycastHit hit;
        if(Physics.Raycast(commands[0].from, commands[0].direction, out hit, commands[0].distance))
        {
            if(hit.collider)
            {
                Debug.Log(hit.collider.name);
            }
        }

which does hit.

#

So I have no idea if I am doing something wrong

lapis plaza May 5, 2019, 4:45 AM

#

@timid pebble hard to tell from that snippet

#

you didn't show your setupjob

#

I mean, you clearly have commands[0] setup if the latter code runs

#

but there could be something that would explain this on the setup stage

#

batched raycasts do work for me and I don't immediately spot anything wrong on the part you shared here

timid pebble May 5, 2019, 2:58 PM

#

    [BurstCompile]
    public struct SetupCommandsJob : IJobParallelFor
    {
        [DeallocateOnJobCompletion]
        [ReadOnly] public NativeArray<PreviousTranslation> prevPositions;
        [DeallocateOnJobCompletion]
        [ReadOnly] public NativeArray<Translation> curPositions;
        [DeallocateOnJobCompletion]
        [ReadOnly] public NativeArray<Rotation> rotations;

        [WriteOnly] public NativeArray<RaycastCommand> cmds;

        public void Execute(int index)
        {
            cmds[index] = new RaycastCommand
            {
                from = prevPositions[index].Value,
                direction = math.forward(rotations[index].Value),
                distance = math.distance(prevPositions[index].Value, curPositions[index].Value)
            };
        }
    }

I did it with IJobForEach<T> originally but in a scramble to resolve the issues its now using IJobParallelFor. All I know is that I have some refactoring to do

#

But for now I just left it to a forloop with Physic.Raycast until I figure out what the issue was

blazing scarab May 5, 2019, 8:23 PM

#

why does ecs physics not have a ColliderOverlap (other than aabb) but it does have a ColliderCast?

#

and ColliderCast with 0 length gives an assertion failure

fallow gorge May 6, 2019, 3:05 AM

#

those questions might be better asked in the forum thread so the havok guy could answer

strong plaza May 6, 2019, 10:21 AM

#

is there any information on when unity will put out physics2d for ecs?
I wanted to rewrite the system with slightly different data types, but not sure yet if it is worth it,
if anyone has looked through the new unity physics code, would it be difficult to replace the solvers and raycasting solution with 2d shapes? I think it would be?

timid pebble May 6, 2019, 1:41 PM

#

I haven't seen a whole lot of related to 2D anything for ECS

#

Maybe this will yield some results:
https://forum.unity.com/threads/ecs-physics-2d-collisions.661411/

Unity Forum

Ecs Physics 2d Collisions.

hey Everyone,
im making a 2.5D game and i need the third dimension only for rendering, all my physics part need a 2D logic : Environement Collisions,...

strong plaza May 6, 2019, 1:56 PM

#

thanks!

strong plaza May 6, 2019, 4:54 PM

#

okay, so for anyone following this topic and having similar questions,
in another thread there was an official statement by an unity employee, was from april,

Unity Tiny is being ported to c# as you might know, it already has some 2d systems in place,
those systems are expected to arrive somewhere with 2019.2, prediction was Q2 2019, 2d physics should be included

#

in case I misinterpreted something, here is a link
https://forum.unity.com/threads/c-update-for-project-tiny.643000/

lapis plaza May 7, 2019, 8:07 PM

#

new experimental 2019.1 / PhysX 4.1 build: https://forum.unity.com/threads/physx-4-1-in-unity-experimental-builds.634960/page-2#post-4512052

Unity Forum

PhysX 4.1 in Unity - experimental builds

That said, why would some be unhappy.

Yeah, i don't want to sound too negative here. The new physics engine looks amazing! There is a bunch of stuff...

lapis plaza May 7, 2019, 8:36 PM

#

heh, Unity Hub 1.6.2 can't deal with custom editor if there's same number engine version installed already 😄

#

luckily there's been an update to 2019.1 so I can differentiate these again

#

it's still an oversight as there will always be experimental editor versions and 2019.1+ doesn't even come with it's own launcher anymore

fallow gorge May 7, 2019, 11:47 PM

#

thought that was ecs for a second 😦

lapis plaza May 8, 2019, 11:05 AM

#

oh snap, I got a use case where I would have needed those trigger vs trigger collisions on that that physx 4 build

#

I know I can workaround it by duplicating each trigger with kinematic collider of the same shape but it's pretty ugly setup

tropic hamlet May 8, 2019, 7:56 PM

#

Why not just do overlap check?

lapis plaza May 8, 2019, 8:56 PM

#

because I'd need kinematic bodies then still

#

plus I'd expect you lose some of the optimization if you do manual overlap checks vs use built-in triggers, especially since you have to go through interop layer with physx for each poll

strong plaza May 8, 2019, 11:03 PM

#

manual overlap checks should be pretty expensive in comparison,
the regular physics update should do broadphase in one iteration and has probably some other optimizations, instead of checking each body individually against the whole world

timid pebble May 9, 2019, 3:09 AM

#

Whats the proper way to move a physics body in its forward direction? I have tried physicsVelocity.Linear = math.forward(rotation.Value) * (speed *deltaTime). It ends up just pushing itself through other physics bodies or falling through the floor. Using the preview Unity.Physics and this is phone coding so it may be slightly off

lapis plaza May 9, 2019, 6:56 AM

#

@timid pebble well, you are hardcoding it's velocity so that's kinda expected, it can't slow down as you reset it each update back to full speed

#

use forces to move it if you want it to interact

strong plaza May 9, 2019, 9:59 AM

#

if you hard code everything or do everything manually, you also have to react to the collision resolution manually - so for example when a penetration happens you probably want to change the speed + change the forward rotation for example to the hit normal or sometihng

#

not the best idea to do it like I described, but generally that could be to keep in mind I guess

timid pebble May 9, 2019, 2:00 PM

#

Well the reason I did that is because I could only find ApplyImpulse nothing for consistently applying a force

timid pebble May 9, 2019, 3:24 PM

#

Ah, I see now. Unity.Physics.Extensions is where I should be looking

strong plaza May 10, 2019, 3:31 PM

#

📎 chika_unity.png

tropic hamlet May 11, 2019, 12:42 PM

#

the manual overlap is indeed more expensive but how many of those are you doing? 1 query vs world is nothing.

#

for example you might do it for weapons, and so how many weapoms are requiring collision at a given moment? it usually boils down to a handful in most games

stuck bay May 12, 2019, 3:17 AM

#

so what does it mean when an impulse is applied to a contact pair? Looking at https://docs.unity3d.com/ScriptReference/Collision-impulse.html whats the best way to get the impulse applied directly to my object?

sly violet May 12, 2019, 6:54 AM

#

Theoretically it should be just half the impulse, AFAIK. Every action has an equal and opposite, conservation of momentum, and all that. I'm not sure if that's correct if one of the colliders is kinematic or static, though.

stuck bay May 12, 2019, 7:03 AM

#

thats not my problem. i understand what impulse is and the math behind this stuff. I cannot understand what body this impulse is supposed to be applied to. while debugging the direction of the impulse using Debug.DrawRay it doesnt seem to be the impulse of either bodies. The magnitude seems to be fine but is the direction just completely useless? I launched two balls coming in from opposing sides and each give me an impulse in a nearly same direction... I cannot imagine how this could be useful to me.

sly violet May 12, 2019, 7:06 AM

#

...What are you trying to do?

stuck bay May 12, 2019, 7:06 AM

#

real time mesh deformation on impact

sly violet May 12, 2019, 7:07 AM

#

Ah. You might try getting the magnitude from the impulse (the direction of the impulse won't help for that I don't think) and using the contact normals for direction.

#

https://docs.unity3d.com/ScriptReference/ContactPoint.html

stuck bay May 12, 2019, 7:09 AM

#

the normal is just the normal of the vertex collided, not the direction in which force will be applied. a good example of when this will give me inaccurate results would be a ball sort of skimming on top of a sphere

#

i do use contact point and have successfully gotten mesh deformation on impact but only coming from a specific direction, this is because of the results that i dont understand from the impulse and lack of other things to use. i dont really understand why unity hides so many details about these collisions

sly violet May 12, 2019, 7:22 AM

#

Well, one last thing you can do is just track the change in velocity. Keep in mind IIRC you'll need to compare with not the immediately previous FixedUpdate, but the one before.

stuck bay May 12, 2019, 7:23 AM

#

adding relative velocity and the normal might actually work. debugging the rays it seems perfect but i wanna do more research and check to see if this is what would actually be used because at this point i just need to know

#

maybe im just completely thinking about this wrong

#

impulse is definitely a wacky value im pretty confident in that at this point

#

ima do some research and update here once i figure it all out

fallow gorge May 13, 2019, 6:47 PM

#

in ecs physics, if I use a colliders CollisionFilter for a raycast, it should ignore that particular collider right?

#

hmm maybe I was wrong with this assumption? is there a way to ignore a specific collider with raycasting(like so a character doesnt raycast itself)?

hybrid garden May 13, 2019, 6:58 PM

#

You have to use DOTS physics shape

fallow gorge May 13, 2019, 6:58 PM

#

found the answer to my question, its currently unsupported https://forum.unity.com/threads/unity-physics-discussion.646486/page-7#post-4473808

Unity Forum

Unity Physics Discussion

Sorry for the dumb question, but how to apply constrain a phyisic body? I'm using a prismatic joint to simulate a suspension but i dont know how to fix...

hybrid garden May 13, 2019, 7:01 PM

#

are you sure? I've got raycasts ignoring specific bodies in my scene

#

            // Now cast the ray and see if the it hits any colliders.
            var rayCastInput = new RaycastInput
            {
                Ray = ray, Filter = new CollisionFilter
                {
                    CategoryBits = ~0u << 2, // Belongs to group Raycasts
                    MaskBits = ~0u << 2 // Only collides with Raycasts group
                }
            };

fallow gorge May 13, 2019, 7:04 PM

#

are the ray casters sharing the same filters as what is being raycast against? that forum post says this currently doesnt exist?

hybrid garden May 13, 2019, 7:11 PM

#

Hmm. Mine is casting the ray from the camera, so its probably not the same case, sorry.

fallow gorge May 13, 2019, 7:13 PM

#

np my situation is collider to collider that share the same filter(but I want to ignore the originating collider)

lapis plaza May 13, 2019, 7:51 PM

#

@fallow gorge you probably saw my question on this topic here https://forum.unity.com/threads/unity-physics-discussion.646486/page-6#post-4457455 (and the answer two posts below)

fallow gorge May 13, 2019, 7:51 PM

#

yeah I did 😃 just gonna offset the raycast position for now

hybrid garden May 13, 2019, 7:54 PM

#

@lapis plaza did you ever explore GroupIndex on collision filters over the category/mask bits?

lapis plaza May 13, 2019, 7:54 PM

#

I didn't have time to try it out yet

#

it's not a huge issue right now, can workaround it by abusing the categories

hybrid garden May 13, 2019, 7:55 PM

#

                CollisionFilter filter = new CollisionFilter
                {
                    CategoryBits = 1u << (int)PhysicsLayer.RayCast, // (int)PhysicsLayer.RayCast == 0
                    MaskBits = 1u << (int)PhysicsLayer.Ships, // (int)PhysicsLayer.RayCast == 1
                    GroupIndex = col.ColliderPtr->Filter.GroupIndex // entities (ships and bullets) on the same team share the same negative int
                };

lapis plaza May 13, 2019, 7:56 PM

#

this is what Unity does on the vehicle sample scene atm

hybrid garden May 13, 2019, 7:56 PM

#

per an example from that thread

lapis plaza May 13, 2019, 7:56 PM

#

(they just put each vehicle to their own category)

#

I mean, it does work when you have only few vehicles 😄

#

not that great if you have like open world with hundreds of them

#

could still pool and reuse the categories for visible objects only that need it but it gets really hacky quickly

zinc furnace May 14, 2019, 3:23 AM

#

Not sure if this is the best place to ask - but a question about quaternions...
Been trying to understand whats going on, and, so far, it would appear that the xyz component is a unit vector, defining an axis for the magnitude to work around. The rotation quaternion used to describe the current rotation of an object could be said to be the result of taking quaternion.identity and rotating it around the axis defined by its xyz, an amount equivalent to say 360 degrees divided by the magnitude? A magnitude of 0.5 around an axis would provide then 180 degree rotation around said axis?

#

please @me if anyone can confirm (or deny) my understanding.

hollow echo May 14, 2019, 3:28 AM

#

https://eater.net/quaternions

#

They're complex numbers so your intuitions may be a little incorrect - it's hard to tell based on your description
Basically they're unit-length axis-angle rotations.

#

For rotations the quaternion is a unit quaternion, and that's what you're dealing with in Unity
To use them you don't need to understand their makeup practically. Just use the API, AngleAxis is an extremely helpful one.
Using quaternion multiplication is also a good thing to understand. A * B => A "rotates" B

zinc furnace May 14, 2019, 4:19 AM

#

Im currently needing to hardcode the transfers for ecs...

#

Trying to code in rotational inertia, and rotational thrust.

hollow echo May 14, 2019, 4:22 AM

#

Not sure I understand, there should be good maths functions in the Mathematics library, they have AxisAngle, LookRotation...

#

You construct the rotation using a function like that and then apply it with multiplication 🤷

#

If you for some reason want to construct them via Euler angles you can also use the Euler functions

hollow echo May 14, 2019, 5:36 AM

#

@zinc furnace to attempt to clarify what your original explanation could have meant, a quaternion is not: axis, angle, it's magnitude is always 1. That way any resultant multiplication also has the magnitude of 1.
A quaternion is instead structured like: (axis * sin(angle * 0.5), cos(angle * 0.5))
seeing as the axis is always a unit length, this equation means that the axis and angle are scaled inversely and contribute as such to the magnitude.
Because of this though it's difficult to intuitively look at a unit quaternion's values and understand what the resultant rotation may be, only that if there is a magnitude in the xyz component, or xyz and w, then there is a rotation; and if w is 1 then there is zero rotation (ie. it is the identity).

zinc furnace May 14, 2019, 8:29 AM

#

Thanks for the answer @hollow echo

#

(axis * sin(angle * 0.5), cos(angle * 0.5)) would only give two values tho?

hollow echo May 14, 2019, 8:29 AM

#

axis is a 3 component vector

zinc furnace May 14, 2019, 8:29 AM

#

oh i see

#

so its kinda backwards when the w comes at the end in the quat objects then?

#

no ... at the beginning...

#

im very tired hehe

hollow echo May 14, 2019, 8:31 AM

#

Well, from what I wrote it aligns fine with unity's makeup, but it's actually not a standard and you often find them as:
w, x, y, z

zinc furnace May 14, 2019, 8:31 AM

#

would i be right in thinking that when you try to apply a quaternion with a magnitude != 1, is when your mesh will disappear?

hollow echo May 14, 2019, 8:32 AM

#

Uh, I'm honestly not sure what happens, with the standard transform Unity complains loudly

zinc furnace May 14, 2019, 8:33 AM

#

ok, in ecs which is what im using atm, it makes the meshes disappear i think... not 100 % sure, but it definitely happens if a NaN creeps in

#

Would you be up for a quick discussion, a bit more of an indepth topic, very much physics tho.

#

?

#

(I tend to ask a lot of questions 😉 )

hollow echo May 14, 2019, 8:34 AM

#

Nah, I'm not really much of a physics person 😛 I just know a little about quaternions 🤷

zinc furnace May 14, 2019, 8:34 AM

#

ah, ok fair enough 😃

#

Anyone free to give any comments on modelling the torque and rotational drag in a physics system, pure newtonian physics, without gravity. Quite arcady, i think, but full 3d, its spaceships!

#

Its for ecs, and i can have my rotations accessable as Eulers, so i think i will try to use those, and then it should be easy to keep a set of speeds for each axis, and modify them with acceleration rather than directly controlling the rotation...?

lapis plaza May 16, 2019, 1:43 AM

#

latest (second) Unity Physics package doesn't seem very happy when building with IL2CPP 😃 ```Temp\StagingArea\Data\il2cppOutput\Il2CppAttributes.cpp(58784): error C2039: 'numeric_limits': is not a member of 'std'

#

it's coming from [SoftRange(0f, 1f, TextFieldMax = float.MaxValue)] tag on PhysicsMaterialCoefficient struct

#

if I remove it, it does build without errors

#

but it also crashes on standalone immediately now 😄

#

crash is access violation on cpp // System.Void Unity.Collections.LowLevel.Unsafe.UnsafeUtility::InternalCopyPtrToStructure<Unity.Transforms.Translation>(System.Void*,T&)

#

oh well... I'll try on 2019.1 next, this was on 2019.3 😃

#

2019.3 also creates wrongly named data directory for IL2CPP when you use SLN creation option (it's proper if you build it fully on editor)

#

will see what all is 2019.3 issues and what not and try to file issues for these

cosmic mica May 16, 2019, 11:37 AM

#

How do raycasts and other queries work with rigidbody interpolation enabled? Our projectile system works with small linecasts each frame, but our fixedupdate is significantly slower (20fps). Do queries check against the interpolated collider or against the uninterpolated?

#

The npcs are vehicles that can go quite fast, more than 100 m/s

#

So there is a big jump from fixedupdate to fixedupdate

lapis plaza May 16, 2019, 11:42 AM

#

I haven't tested this but I've implemented this on other game engines. Normally you wouldn't actually alter the collider locations to the interpolated locations as it changes the physics engine state

#

in some games you do need to trace in regular Update tho and then you kinda have to do this

#

I haven't tested how this works in Unity, but I'd be surprised if they didn't just keep the latest physics state on Update() instead of feeding the interpolated state back to PhysX during update

#

@tardy spear could probably confirm this if he's around

#

but you can also easily test it yourself

#

just put big deltatime for fixed update, put some debug draw to your physics simulated objects and ray hit points

#

I think physics debugger on unity would be fine for visualizing where the colliders go

ocean horizon May 16, 2019, 1:50 PM

#

Yes, interpolation for 2D/3D physics is done by keeping the last physics body pose then interpolating to the current physics body pose. Extrapolation is done by interpolating from the current physics body pose to a pose based upon the last lin/and vel at that pose. It drives the Transform not physics state per frame update in between simulation steps.

#

So interpolation is always behind, moving to the current pose. Extrapolation is always ahead moving from the current pose to some potential future pose.

#

Queries always check the current physics state as it's not related to transforms. Interpolation/Extrapolation is a rendering thing, not physics.