PhysX 4.1 in Unity - experimental builds

@yant Any news on Physics.IgnoreCollision getting fixed for 2019.3 using the Temporal Solver?

EDIT: Disregard, I'm wrong.

 

http://fogbugz.unity3d.com/default.asp?1238440_jo39evpehsg3fnb7
Here is a video of what exactly is happening and how I reproduce it 100% of the time:

 

Unfortunately I don't have a smaller project for repro, but I do have the project up on collab if you want to try that.

EDIT: Oh darn, i'm at my seat limit for the collab project.

 

I'll see if I can make a small repro project.

 

Hmm, i'm stumped. It's actually not Physics.IgnoreCollision, sorry about wrongly accusing it! The joints seem to jitter violently in certain cases and I can't figure out why. If I disable my arm colliders it mitigates it a bit but not totally. If I enable prepocessing it also mitigates it a bit.

The mass differences for the joints isn't very high (not more than 2x) so I don't think it can be that. Do you have any idea why the issue only crops up with TGS?

Sorry again for jumping to conclusions.

 

I have same issues with Temporal Solver; when I start using it, whole Physics go crazy after a few seconds. I am using lots of Configurable Joints, which are often chained.

 

Maybe TGS is not an actual improvement for already working setups? I recall not finding any significant difference when I first tested it in an early experimental build.

 

I reported case 1239507 on this, as the same issue with "[Physics.PhysX] BV4 midphase only supported on Intel platforms" happens on HoloLens 2 on Universal Windows Platform ARM64 target in 2019.3.10f1.

 

Thank you for the explanation

I would really appreciate this fix being implemented in 2019.3 (or 4 LTS).

Is there any workarounds to disable it globally in a Unity project that we could use? (All prefabs and scenes with mesh colliders break and must be modified :|)

 

I am wondering how to increase buffer size in Physx 4.1 unity package?

 

I have a simple scene with a ball with a random speed and a script with physicsSim.Simulate(Time.fixedDeltaTime) to know where the ball will bounce on the ground, then a new random speed is applied and new simulated bounce are calculated and so on. Randomly the simulated ball is stucked onto the ground at the first bounce, but the real ball bounces more the 2 times. This doesn't happen if the initial speed is not random but it's always the same. I noticed that if i change the default contact offset from 0.01 to 0.1 or another value the bug always happens. The simulated ball touches the ground the first time and it doesn't bounce. The real ball bounces normally.

 

As the bugreport is closed I cannot track the progress of this fix into Unity 2019.3, do you have any updated ETA @yant ?

-
See the closed bug for Android where a lot of HL developers are giving feedback on the same issue, so the issue should be addressed: https://issuetracker.unity3d.com/is...-platforms-appears-when-exporting-for-android

 

Hi @yant, Unity 2019.4 (LTS) was released today, and PhysX BV4 bug is still there. Can you provide an approximate date when it will be ready for 2019.4.x? On Unity 2020.1.0b11 and 2020.1.0b10 everything works.

 

Thank you for the quick reply.

 

Just wanted to say that it would be great for the next physX version to have more primitive shapes, such as Cylinder (same as capsule, just with flat ends) , Pyramid (4 vertices total) , TriangleWithHeight (6 vertices total) and a ConeCast / ConeCastAll / ConeCastNonAlloc method. Everyone needs this, absolutely everyone. Would be awesome if they were included some time in the future.

Afaik primitive shapes are performant because the overlap/raycast check is hardcoded (e.g. sphere collider as distance check) - for the shapes mentioned above, there are mathematical "representations"/hit-tests ... I'd even try to write them myself if it would be possible to extend the given implementation.

Reason why they are useful:
Cylinders:
pretty obvious - (explosive) barrels, water bottle, other dynamic cylindrical objects, which should "stand straight" without flipping over

Pyramid:
Projectiles, spears, tip of a sword, they are a great addition to compound colliders

TriangleWithHeight:
Connection between two perpendicular box colliders, ramps, perfect for compound colliders which represent a curved plane such as a pipe (hanging from a crane where you can walk through --> dynamic body with curved surface)

All these are cases where convex mesh colliders are not suitable or just bad for performance.

And ConeCast is useful in many cases, especially AI, light Cone, PlayerVisionCone or just to check if something is below a CircleSprite on the UI with a perspecivic camera.

 

@Hannibal_Leo unfortunately there isn’t much that can be done from Unitys side. PhysX is developed by nvidia, Unity is providing access to physx through their API. Your request should be pointed at Nvidia (or Havok or Unity Physics).

 

ECS Physics as well as Havoc actually offer a cylinder Shaped Collider but using ECS as solo programmer on an indie studio right now? ECS is in constant change and not finished - animation, sound, particles are missing.
And by far too much work for one person.

And I just searched on the NVidia website and someone asked them about Cylinders this year, end of jannuary .. they say "use convex colliders, they are fast enough" ... well .. I've had pretty bad results on my building destruction game with MeshColliders :S - once I switched to compund colliders, everything was running smoothly again so ..
But I don't think they would change anything just for one user - if unity would ask them, I see a greater probability that the feature(s) could be implemented.