Page MenuHome

Increasing cloth simulator collision quality parameter leads to chaotic behaviour in trivial cases
Closed, ArchivedPublic

Description

System Information
Debian 8, nVidia GTX970, intel 4770K

Blender Version
Blender 2.7.4

Short description of error
Increasing the cloth simulator collision quality parameter leads to chaotic behaviour in trivial cases. It seems to work best with 1.

Exact steps for others to reproduce the error
I made some unit test cases for the cloth simulator because it showed strange behaviour. The one thing which really doesn't work is the collision quality parameter. The cloth simulation works best when this is 1 and the higher it is set, the worse things get, even in trivial cases.

Test case:
Ground plane at 0, size 1x1 BU, collision activated with outer distance 0.01 BU

Cloth: vertical rectangular mesh, 0.005 BU wide, 0.04 BU high, subdivided into 8x1 faces of 0.005 BU square. The lower end is at Z=0.
The two lowest vertices at Z=0 are put into a vertex group.
Cloth physics is activated with collision distance = 0.01 BU, and pinning set to the one vertex group.

Observations:
With collision quality = 1, it works as expected.
With increasing collision quality it gets worse. Also the distance of the lowest unpinned vertices from the ground plane increase almost proportionaly to the quality.
From quality=7 on, the behaviour is completely chaotic.
From quality=15 on, the cloth is confined BELOW the ground plane.

I will append a .blend file with animated collision quality which clearly shows the behaviour.

On IRC Keito indicated that currently there is nobody taking care of the cloth simulator. I could probably fix this bug and a few others, so if indeed there is nobody who feels that the cloth simulator is his baby, please feel free to assign the bug to me.

Best regards,

Michael

Event Timeline

Uploaded a sample file which increases the collision quality by 1 every 20 frames here:

http://www.pasteall.org/blend/36789

Sergey Sharybin (sergey) changed the task status from Unknown Status to Unknown Status.Jul 7 2015, 12:19 PM
Sergey Sharybin (sergey) claimed this task.

Surely collision/cloth simulation could be improved, but even then there'll be cases when physics might fail.

What happens here is that object is originally inside the collision cage, which makes physics to apply force to move that object outside of the collision, when ends up in quite unstable solution because of the pinning.

In more sane setups increasing quality actually helps, so would not consider this a bug. Thanks for the report anyway!

Dear Sergey,

what in my view is not ok is that several quality steps lead to a much worse results then several normal steps. Why should quality steps have a stronger effect than normal steps? If anything they should have a softer effect, because the effect of forces should be less with a smaller time step.

I don't think that this can be accounted to usual integrator defficiencies.

I think thinks would be perefectly fine if the quality steps would be identical to normal steps, just with the effect of forces adjsuted to the reduced time step.

Best regards,

Michael

It depends on how solver works and handles constraints. The way how physics works in blender is on the list for improving, but this happens outside of the bug tracker.