[ODE] notes on the quickstep method

Russ Smith russ at q12.org
Thu May 20 19:46:36 MST 2004


the quickstep method works well so far for stacks of boxes and for some
mechanisms and structures, but not very well for articulated creatures
with motors. i've figured out the reason why: the SOR method that is
used performs very poorly for ill-conditioned (near singular) systems -
and with ODE it's very easy to make near singular systems.

for example, you can get a singular system if two motors are fighting
each other, e.g. if there are two creature legs on the ground, friction
is high and the legs are being forced against each other. also, too many
contact points with nonzero friction will cause problems too.

i also implemented a conjugate gradient LCP method, but it performs
about the same.

anyway, i'm still pondering how this situation can be improved.
for some light reading about the performance of various LCP methods,
check out this paper:

	http://www.ep.liu.se/ecp/010/004/ecp01004.pdf

(the author is a former colleague of mine, very smart guy).

russ.

-- 
Russell Smith
http://www.q12.org


More information about the ODE mailing list