[ODE] Re: LCP

[Adam Moravanszky]

> if you are using acceleration based equations of motion, then you can
> indeed find friction problems that have many LCP solutions, or none at
> all. some of baraff's papers give a good explanation of this problem.


Russ,

my dynamics sim uses a separate "custom" friction model for this reason.
Thus, the friction forces are external forces as far as the LCP solver is
concerned.  I believe this worked quite well.  (I have to admit that the
reason I resorted to this in the first place was that I was unable to get a
stable LCP solver when also doing dynamic+static friction with it.
Apparently, this was no problem for you.)  While I never saw your sim
visibly fail, and believe this to be more of a theoretical problem, perhaps
you would be interested to try my alternative.  I have not touched this
stuff for at least a year now, so I can't remember if without the friction,
the LCP becomes faster or not, which would be desireable considering its
high time complexity.  My friction hack, which is a simple implementation of
NormalForce * frictionCoeff in the contact sliding direction (for dynamic
friction) (plus some more hacks to minimize oscillations when you have
assimetric contract force configurations) is of course linear time.


--
-- Adam Moravanszky
http://n.ethz.ch/student/adammo/