[ODE] Re: ODE

[Russ Smith]

Adam Moravanszky wrote:
> 
> Hi,
> 
> I decided to use ODE for the foundations of my diploma thesis.  That
> in turn will be the basis of a Christoph Niederberger's Ph.D. on
> virtual creatures. This implies that I will be working on improving
> ODE so that is not only as stable as it is now, but also more
> accurate.  The thesis work will integrate the improved dynamics
> engine into Garage Games' Torque engine (www.garagegames.com), which
> will also receive a graphical rehaul, and add controllers for the
> animation of legged virtual creatures.
>
> So I'd like to offer the coordination of my work with yours:  instead
> of branching off with the current version and developing in parallel
> with you, with eventual overlaps, I could let my work be part of your
> 'official' future versions of ODE.

that's great, help is definitely appreciated.

> The first significant contribution I offer is the integration of my,
> IMO very robust OBB collision and contact determination library
> 'Crater'.  You can see it being used by my physics engine 'Corpus' in
> the demo I have on my website at
> http://www.n.ethz.ch/student/adammo/ProResting.htm

where can i find source code for crater? it would need to provide
contact normal and penetration depth as well as contact point.
the current box-box function can only find up to one contact point,
but if only one is required is finds that point *very* quickly (at
least compared to most box-box routines i have seen) - so perhaps
merging the two techniques would be interesting.

> This will find all the relevant contacts between pairs of boxes, and
> solve the problem you mention below.  Any other part of the so far
> closed source Corpus is also fair game for integration into ODE if
> that makes sense.
> 
> Higher order integration schemes,

this would be useful - note that i am using a 1st order semi-implicit
integration scheme right now, so higher order schemes should also be
as stable. several people have written to me recently about energy
conserving integrators, we might also want to consider those.

> ...partition dynamics (as described in
> the  Baraff paper of the same name), detecting stationary clusters
> and 'freezing' them, and looking into Baumgarte stabilization to
> combat drifting joints (if it ends up being a problem and you don't
> have something like that yet -- I didn't look) would be worthwhile
> directions for my development.

detecting stationary clusters: on the TODO list, and definitely a
requirement. Baumgarte stabilization: not necessary, ODE has its own
stabilization scheme built into the timestepper.

> Of course I will be happy to help
> with other topics too if you can convince me that they are important.
> :-)  I have allocated just under two months of full time development
> for dynamics, in any case.

see my prevoius email in the ODE list on this subject - the most
important things to do right now are the boring non-physics things
like configuration management. good configuration management will
make ODE immediately useful to more people than extra features at
this point.

russ.

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