[ODE] Trimesh problems

[Anders Olofsson]

CFM at 1e-3 makes the sphere sink through the triangles (the plane too, but 
they pop up again ofcourse). I had it at 1e-8 to prevent that. 1e-5 works 
sort of but the sphere still starts to jump on large triangles. Its 
annoying that it works so great on planes and small triangles but not large 
triangles.

I tried to turn on autodisable but no change, guess I've too fiddle with 
thresholds.

I dont know why large triangles causes problems, I tried an even larger 
triangle and the spheres just went through it!.. Feels like some accuracy 
problem.

The only solutions I see now is to split up my mesh in smaller triangles (a 
large triangle will be splitted up to many) or somehow detect 
geoms<->triangles collisions myself and use the plane that the triangle 
lies on as a collision geom somehow.

Not so nice either one.

On the upside, just tried boxes instead of spheres, they seems to work fine 
on large triangles. (No jumping)

--

I'll search the archive for sphere penetrations. (I assume its about 
sphere<->trimeshes, right?!). Someone ought to have notice this problem before.

I'm using the prebuilt ODE 0.5, single with trimesh. I think I'd mess 
things up even more if I built it myself from the CVS. Isn't it time to 
release a new version soon anyhow? :)

I'm hoping that my other problem is related to this too (spheres making 
sharp turns with no reason or something like that).

Thanks for your help!

--Anders Olofsson

At 21:12 2006-01-09, you wrote:

>A CFM of 1e-8 is pretty darn small for a floating-point value. You can 
>only represent about 1e-5 for something that's scaled at "1" with floats. 
>I'd try setting CFM to 1e-3 and see how that goes.
>
>Also, ODE isn't perfect. For resting bodies, you really want to turn on 
>auto-disable, else they may jitter forever, and the jitter could 
>potentially cause energy additions to the system.
>
>The size of the triangles is interesting, though -- might there be some 
>force that's scaled by a side length, or un-normalized cross product, or 
>something?
>
>Cheers,
>
>                         / h+
------
>I recall some recent changes in CVS UNSTABLE for sphere penetrations.
>
>Are you using UNSTABLE, MAIN or a tar ball?
>Could you try the two CVS branches, and see if there is a difference
>between the 2?