[ODE] Silly theoretical question
vedrank at croteam.com
Thu Feb 24 09:44:08 MST 2005
The idea behind Baraff's work was to have easy and clean method for modeling
constraints (joints) and also to be robust at the same time. The problem is
how to solve situations when you have closed loops formed by constraints
equations. In that kind of situations any direct or iterative LCP solver
becomes unstable. To make it stable you then use time-stepping scheme to
solve equations of motion in several frames. Everything is fine at high
framerates. When your simulation is running at interactive framerates, we're
speaking of 20 to 30 FPS, then also order of integration becomes an issue.
So, to balance simulator you need to tune siffness parameter and get
plausable solution of your system.
Using penalty methods would help you in some situations, that is for sure.
The question is, is that approach good enough to use it for all kind of
constraint models ? If it is, people would use it long time ago !
It is your compromise which method will you use. It depends on requests you
are up to.
Let's not forget that physics simulation methods are not from yesterday or 2
years ago. Barraf wrote his papers back there in 1994, 95'... and now we are
10 years later not having anything "much" better. All LCP solving methods,
of whatever kind, are dating since middle 70' and 80' of the last century.
(not mentioning simplex method and its roots from WW II :) )
vedrank at croteam.com
> ODE was done on the basis of David Baraff works. As far as I can remember,
> (and confirmed right now by reading his thesis, part 1, 2.4 The penalty
> he considered not to use penalty-based (where constraints are substituted
> by penalty forces) equation of motion because of stiffness of resulting
> equations of motion.
> So, supposing we get a solver that could relatively inexpensively solve
> set of such equations what kind of benefit we could obtain? Should we
> search for such a solver? And, to triple confusion of the reader and
> possibler answerer, why ODE mentions in documentation that stiff systems
> are bad?
> I read overview of Havoc physics library and they mention that using stiff
> springs for ragdolls might be beneficial comparing to constrained ragdolls
> (cannot find that overview right now, though).
> It seems that not only ODE is based on Baraff's work. As far as I can see,
> everyone and his dog done their physics library that way.
More information about the ODE