Penetration problems

[Bird]

Hi Julio,

I'm seeing some penetration problems in the latest 4.0 from github. There's 800 convex hull shape instances in the stack and the bowl is a static mesh. As you can see in the video, it doesn't happen with all objects. I think I used to solve this by using CCD in 3.14 but I don't think you're using that anymore. Correct?

https://youtu.be/A-XR_uwOvpE

[Julio Jerez]

with the jeep, I was going to say, is that a bug, is it a darn good bug, but the I saw the tortoise.
it is me or are the wheels spinning? or maybe some video low rest effect.

is not using ccd now, what that there before?
few questions:

are the shape compound?
what happen if you more sub steps?

maybe this could be a good case to enable the CCD.

[Bird]

are the shape compound?

No. Just single convex hull shapes

what happen if you more sub steps?

Definite improvement for the turtle but I can still get penetrations with other objects with sub-steps set to 8. There are no penetrations with any objects that I tried with sub-steps set to 10

it is me or are the wheels spinning? or maybe some video low rest effect

Yeah, I see what you mean but no the wheels aren't spinning. I wish they were.

[Julio Jerez]

Ok so at 8 sub step the penetration goes away.

That's our clue to re implement the code.

Basically is has to do an Island partitions that separate the fas moving o ject an solve the in a pre pass. Before the general solver.

All the logic is in but thre are few collision funtion that has to be written.

Let us put that in the list of the thing to add.
Meantime try with more sub step until I get around to implement it.

It has to be there because can not afford such high cost for something that is very rare. Plus the substep has to be adaptive.