[ODE] pressure soft body

[Mateusz Kula]

On Tue, 17 Apr 2007, Patrick Enoch wrote:
Hi,

> Hi,
>
> the link is:   http://www.smorfland.uni.wroc.pl/~toczka/2ndtype/index.html
> None of your applets works here (Windows-firefox || macosx-safari).
>

I think it could be becuse I have used the most recent version of JDK 
(6.0), you should have JRE 6.0 installed to run them.

>
> what do you mean with "3d-softbody" - including collisions? how complete is 
> your engine already? (would be probably answered on your site)
>

on my site i have put a 2d soft body, means not a ball but a circle.
How complete is it ... it's hard to say, some things have been done, and 
there are still some to do.
I hope it will be working in a couple of weeks, depending on amounts of my 
free time.

>
> about integration into ODE:
> - i dont know if the collider can handle "expanding/contracting" objects - 
> the AABB might change while the body is at rest! - TODO
> - the collider must handle concave bodies, in case a rigid object hits the 
> soft object. - AFAIK DONE

I don't know it too. Thath's why I asked question about how to intergrate 
it with ode.

> you probably use implicit integration? (or explicit with a *very small* 
> timestep). there is a SOR and a CG solver already in ODE, that could be raped 
> for that purpose. it is able to solve (constant-) linearized box-constrained 
> problems. though i dont see a problem for a non-constant matrix. the CG would 
> have to restart a couple of times, SOR should handle that w/o probs.
>

I'm using Verlet integration.

>
> cloth would be a nice addition. i have played with simcloth3 which gives good 
> results but not perfect. problem is, there is no "front" of the cloth. using 
> double-sided polygons didnt really resolve my problems. i had mainly problems 
> with cloth-cloth intersections that did not get resolved - even with very 
> small timesteps or increased penalty forces. (how) did you solve that?
>

I haven't resolved it yet.

>
> I thought a lot about coupling simcloth3 (implicit) with ODE (explicit). but 
> with no real solution.
> a primitive solution would ping-pong the two simulations into a steady state, 
> which is then the next timestep. that is, integrate from t_0 to t_1 until 
> both engines are satisfied (in terms of collisions, body-volume, 
> constraints). in simcloth, collisions at t_1 are used in order to implicitly 
> integrate from t_0 to t_1.
>
> ----
>
> Please find the attached "quickstep.cpp" with a working CG solver. You need 
> to replace some stuff in your files.
>
> Looking forward to your applets!

Mathew Kula