[ODE] Questions about QuickStep

[Nguyen Binh]

Hi, 
 <snip>
As I understand it, the new iterative solver deals with each
constraint as if it were the only one in the system.
 </snip>
Nope! There are two types of iterative solver in ODE:
 a) StepFast (Sergey and David) : like what you think, it iterates
through each constraints and try to satisfy it with the assumption
that under very small timestep, that constraint only has local effects
b) QuickStep (Alen and Russ) : it use iterative method to solve the
LCP system, exactly it uses SOR method (and I noticed there is also an
experiment CG version of Russ in the code but commented out)
 <snip>
 I was wondering how does the solver cope with the common problems of
solving just one constrained at a time, like the classic case of a
block creeping down a plane - the block never actually stops. Or a
stack of boxes that should be frozen due to static friction, but
actually they jitter and slowly slide. Does a sleeping / deactivation
mechanism deal with this?
 </snip>
The sleeping  mechanism surely lessons the jitter but does not solve
it completely :(. But in my experience, a smaller timestep, a better
scale of units, the increase in numer of step, a proper value of
linear/angular sleep threshold value and the use of damping (refer to
John) will almost solve the problem.

<snip>
Also, how does it deal with static friction? Does it use a hack for
situations vhere relative velocity is below some threshold or is it
true static friction, as in the classical LCP formulation?
</snip>
IMHO,ODE uses classical LCP formulation to model static friction. You
could take a look at the code or the document for more details.

<snip>
I've implemented this iterative method some time ago but I never got
passed these problems. My implementation was based on velocity
constraints though. I know ODE uses acceleration constraints. Does
this minimize the problem?
</snip>  
No idea! But I do think acceleration constraints approach is much more
popular now.
_______________________________________________
ODE mailing list
ODE at q12.org
http://q12.org/mailman/listinfo/ode

Nguyen Binh,
www.glassegg.com