[ODE] box stacking instability

[Pierre Terdiman]

> I've noticed that the "boxstack" test is unstable as well, although not to
> the degree that the test program I'm using does.  If one box comes to rest
> on another, while it doesn't pop up (as in this other demo) it does appear
> to vibrate and will move about -- I presume this is due to impulses
> correcting for penetration of the top box into the lower box.
>
> I'm talked to a professor who knows something about dynamics systems, and
> his understanding is that the impulse-based method in inherently unstable
> in this configuration -- using this algorithm it is almost impossible for
> the system to come to equilibrium.  This seems to be an enormous problem,
> and one which makes ODE rather useless for my purposes.  It looks like I
> shall have to just use collision detection with some basic physics hacked
> on...  I need for it to be possible for objects to be supported by others,
> such as platforms on which other objects rest.  It seems that ODE handles
> the case of stability on a plane quite well, but that is rather limited
> use compared to the general case :-(

The boxstack test is unstable because the collision detection routine
reports a single contact point, period.

Pierre