[ODE] Re: Possible solution to unstable rotating bodies problem

Tyler Streeter tylerstreeter at gmail.com
Fri Apr 22 15:15:12 MST 2005


The method I implemented is more or less a temporary solution to a
nagging problem.  I needed some way to avoid the instability without
incurring a huge performance penalty, so I'm pretty happy with the
results.  I don't suggest that this method be added to ODE (in the
long term, a more permanent solution that improves the integrator's
stability would of course be much better), so I just added it to my
own library.

Concerning the fact that it only works with freely-spinning bodies...
that's the only situation where I've noticed problems.  I haven't seen
anything spin out of control while in contact with the ground.

Tyler

On 4/22/05, Matt Hancher <mdh at email.arc.nasa.gov> wrote:
> > Again, most people would rarely need this many freely-spinning objects
> > at once.
> 
> I think this is exactly this issue.  The question for me is *not*
> what the performance hit is for spinning bodies relative to the
> current situation in ODE.  The current situation in ODE is unusable
> in this situations, so this is not the interesting question.  The
> important questions revolve around the relative performance impact
> of various techniques for addressing the problem, and also the
> impact those techniques have on the performance of all the *other*,
> *non-spinning* bodies.
> 
> My concern with your solution is that it *only* works for freely-
> spinning objects.  It is possible to get bodies spinning themselves
> into oblivion while in contact with the ground, for instance.  Any
> solution you use to solve *that* problem will probably also prevent
> your freely-spinning bodies from exploding, at which point adding
> additional code to handle that case becomes silly.
> 
> The only reason I can think of to handle freely-spinning bodies
> separately is if you wanted their physics to be more accurate than
> a forward-Euler velocity integrator can provide.  Your approach
> does improve this behavior, but if that's really what you want
> then you're probably better separating them from the ODE integrator
> entirely and using a more stable simulation technique to handle
> those objects.
> 
> Matt
> 
> Matthew D. Hancher
> Intelligent Systems Division
> NASA Ames Research Center
> mdh at email.arc.nasa.gov
> 
> _______________________________________________
> ODE mailing list
> ODE at q12.org
> http://q12.org/mailman/listinfo/ode
>



More information about the ODE mailing list