[ODE] A 'real' demo of ODE

[Gregor Veble]

Hi Russ et al. :)

That's interesting to hear, that a single more complex link is faster than the sum of 
its parts. I was under the impression that the numerical demand is predominantly 
dependent upon the number of degrees of freedom removed. To use an example: if you 
model the same sort of a link by either a single hinge joint, or by using two 
ball-socket type of links, could you please explain where the difference in numerical 
complexity comes about?

I see your point about the correction parameter doing their stuff if links are moved, 
since it is a slow motion relative to others I bet for a steering type of a link such 
an approach would turn out verz well.

About the 'pseudo algorithm' of mine; I studied the constrained systems a bit more 
(Baraff) and realized that using the principle of virtual work makes the calculations 
very much simpler (the matrices to be solved are much smaller than what I had in 
mind). Therefore the first part of the message I sent you can easily be discarded :). 
The approach towards correction of errors, though, does seem to differ from the usual 
one, however it is slightly more numerically demanding (but probably more robust). 
When you have time, let me know what you think of it.

Greetings,

-Gregor

---- Original Message ----
From:		Russ Smith
Date:		Wed 11/14/01 5:21
To:		Gregor Veble
Cc:		ODE list
Subject:	Re: [ODE] A 'real' demo of ODE

a single joint that captured the kinematics of the situation would be
faster, but obviously it is not a simple joint to create. when i was
at mathengine we looked at the problem a bit, the goal was to have
a unified way of creating joint constraints from various pieces, so
for example we could just specify that "the joint has these links" and
the resulting constraint would be properly computed. some headway
was made, but the problem is that real joints have too many exceptions
to a unified scheme. so we just did it the hard way. this is an area
where some very useful work could be done.

> One thing I wanted to ask, though; does ODE support time varying
> constraints (where constraint functions depend explicitly on time)?

sort of, yes - you can easily vary the constraint parameters per
timestep, and bias the error reduction terms to compensate for the
various extra d/dt terms you will get in the constraint - so in the end
you can achieve this.

p.s. i have not yet had a chance to look at your "pseudo-algorithm"
yet gregor, too busy...

russ.

--
Russell Smith
http://www.q12.org
_______________________________________________
ODE mailing list
ODE at q12.org
http://q12.org/mailman/listinfo/ode