[ODE] Performance comparisons between different physics engines..

[Nate W]

On Fri, 9 Jan 2004, Henrik Karlsson wrote:

> The following engines would have been interesting to compare with:
> Havok (http://www.havok.com/)
> MathEngine (http://www.mathengine.com/)
> Meqon  (http://www.meqon.com)

Novodex: www.novodex.com
dynamechs: www.google.com :-)

There's one other that I don't recall at the moment, not open source but 
with a pretty liberal license.  Hopefully someone else will think of it.

> If anyone knows any physics engines on the net that are using implicit
> integration (backwards euler etc) please send me an mail.. would have
> been interesting to look at and see if you could "feel" any difference
> compared to engines using explicit integration.

I think you will find that all libraries "feel" different because there
are no standard units for parameters like ERP and CFM (heck, those
parameters aren't even standard).  You can make ODE behave very
differently by tweaking those; I suspect you could tune parameters to make
any two libraries behave somewhat similarly in most respects, but it may
be difficult to tell whether the differences are due to parameters or the
algorithms themselves.

I did some side-by-side comparisons with ODE and Novodex (Juice supports
both, the Novodex support is not complete) and found some interesting
differences, for example Novodex is not prone to explosions after deep
penetrations - penetrations are resolved gradually rather than
explosively.  But generally speaking I was never sure whether the
differences had to do with tunable parameters or, as I said, the
algorithms themselves.

-- 

Nate Waddoups
Redmond WA USA
http://www.natew.com/