[Bf-committers] It's time to get rid of cloth

[bjornmose]

Well,

my last contribution was ' old man grumbling'

to be more constructive I'd like point to the issues to be focused on:

We do have stiff PDEs ODEs to solve. We know very well, that common
forward solvers perform very poor under 'stiff' conditions. David Baraff
proposed using 'implicit backward solvers' some days (~ 10 * 356) ago.
Even 'numerical recipes' knows stiff differential equations.

How deal with it?

1. Implement solvers that can do .. expensive but possible in open source.

2. Think about volume energy constraints ( see 3. )

3. Think over the model .. a mesh is a mesh .. blender as grown from
past knows 2D objects in 3D .. a skin of faces. Works very fine if is
forward driven. But there is no reliable concept if we want to have
vertex to vertex interaction deep inside. (Think of the cube having
springs  from (1,1,1) to (-1,-1-1))   The fist thing on the task list
IMHO should be to add 'edges' that is vertex to vertex connections that
do not modify the surface but represent physical interaction.



On 02.11.20 19:07, Brecht Van Lommel via Bf-committers wrote:
> The paper shows how to make SDF collision more accurate by using the SDF
> directly rather than sampling it. We have triangle mesh based collision in
> Blender, which in a state-of-the-art implementation would avoid those
> problems already.
>
> The point of switching to SDFs would be performance, at the cost of detail
> loss compared to triangle mesh based methods.
>
> On Mon, Nov 2, 2020 at 5:38 PM Aaron Carlisle via Bf-committers <
> bf-committers at blender.org> wrote:
>
>> Hi looking at this thread I wonder if anyone has looked into implementing
>> SDF
>> based collision detection in Blender. It seems to be quite a bit more
>> stable
>> than sample based detection. Here is a recent paper on the topic [1]
>> <https://mmacklin.com/sdfcontact.pdf>
>>
>> 1. https://mmacklin.com/sdfcontact.pdf <
>> https://mmacklin.com/sdfcontact.pdf>
>>
>> On Wed, Oct 21, 2020, 21:56 bjornmose via Bf-committers <
>> bf-committers at blender.org> wrote:
>>
>>> Might it be, that old experienced coders knowing physics and all the
>>> other things about stiff differential equations turned their back to
>>> blender development?
>>>
>>> May be because lack of time .. may be an ignored patch proposal .. for
>>> those do not have the time at hand to provide elaborated solutions?
>>>
>>> Well ...
>>>
>>> bjornmose
>>>
>>>
>>> On 20.10.20 20:35, Joe Eagar via Bf-committers wrote:
>>>> Brecht, you can avoid the worst numerical instability issues of CCD by
>>>> adding a small repulsion force when cloth is in close proximity with
>>>> geometry.  IIRC most commercial cloth sims do this.
>>>>
>>>> Ton, I'm sorry I blew up.  Cloth collision is something of a
>>>> special nightmare for me.
>>>>
>>>> Sergej, that's fair.
>>>> Sebastian, sounds like you're doing good work, just remember that
>> robust
>>>> collisions are important :)
>>>> _______________________________________________
>>>> Bf-committers mailing list
>>>> Bf-committers at blender.org
>>>> https://lists.blender.org/mailman/listinfo/bf-committers
>>> _______________________________________________
>>> Bf-committers mailing list
>>> Bf-committers at blender.org
>>> https://lists.blender.org/mailman/listinfo/bf-committers
>>>
>> _______________________________________________
>> Bf-committers mailing list
>> Bf-committers at blender.org
>> https://lists.blender.org/mailman/listinfo/bf-committers
>>
> _______________________________________________
> Bf-committers mailing list
> Bf-committers at blender.org
> https://lists.blender.org/mailman/listinfo/bf-committers