[ODE] Stacking objects
Adam D. Moss
adam at gimp.org
Mon Feb 23 20:13:59 MST 2004
Jeffrey Smith wrote:
> I've noticed that during normal, "happy" steps this error is between
> 10^-3 and 10^-1. When stacking objects, however, this error shoots
> up to 10^2 to 10^4. Clearly, a bad thing.
> How can I reduce this error?
> Is there any way to do the step "more
> carefully" (presumably with a higher cost) if I detect that the error
> is climbing?
ODE's main idea of the amount of 'care' to take is mostly derived
from how short your timestep is.
> Would fixing ODE to use doubles instead of floats for
> dReals affect this?
Depends on where your error is 'coming from' (numerical
error vs contradictory constraints, etc), but double doesn't hurt.
> Is there a more robust LCP solution technique
> that I could implement to fix this?
Not really... StepFast() is generally more error-prone, though
it does deal with certain uncommon types of stability problems
more gracefully than the standard solver (i.e. fighting
Adam D. Moss . ,,^^ adam at gimp.org http://www.foxbox.org/ co:3
More information about the ODE