[ODE] quickStep doesn't work for me.

[jon klein]

On May 18, 2004, at 7:15 PM, Russ Smith wrote:

> you are right, quickstep will have a little more trouble making a 
> system
> look right when there are motors present. there are actually many 
> things
> that can be done to the algorithm to improve convergence in these 
> cases,
> but upping the number of iterations works reasonably well for me. for
> example, in my own ant simulator:
>
>    http://ode.org/tmp/ants.jpg
>
> ...where there are motors at every joint, the default 20 iterations
> makes the ant look a bit springy and spongy. at 100 iterations the
> problems go away.
>
> there should only be motor related problems though, not contact related
> problems ...?
>
> an interesting test is to take the boxstack demo and reduce the number
> of iterations to ONE. surprisingly it still kind of works, but the 
> boxes
> behave like little cubes of jello.

I'm also having problems with quickstep.  Here's a sample:

http://artificial.com/~jk/bouncy.mpeg

This is with 100 iterations and a step size of .005, which works nicely
with dWorldStep.  If I use a smaller number of iterations, the 
simulation
blows up completely.

Any tips?

- jon klein