[ODE] hierarchial spaces

Erwin de Vries erwin at vo.com
Tue Oct 1 13:19:42 2002


Catching up.

IIRC its even worse than a lot of AABB tests for a large group. Its calling
a LOT of specialized collision functions for a large group.

As i reported earlier, i've written a library which fixes these issues. Most
code has been written. I'm using it for our own apps now, but i want to fix
several issues before i publically release it. And i need to fix the
tricollider issues first.

Erwin

----- Original Message -----
From: "David McClurg" <dmcclurg@pandemicstudios.com.au>
To: <ode@q12.org>
Sent: Friday, September 27, 2002 03:24
Subject: [ODE] hierarchial spaces


> currently, spaces don't treat geom groups any differently from other
geoms.  this has the following dis-advantages...
>
> - in a hash space, the geoms in a group don't get the same treatment as
the geoms in the space.  that means many more AABB tests for a large group.
> - when a geom group is processed by the space, the near callback must
decide whether to collide with the entire group or not at all.
>
> i'm using a group for all my static tri-lists so they do not
self-intersect.  this is different from the application suggested by the
user guide (eg, a collection of geometry objects that are located together
in space) but i still think it's a good use of groups.  here is my
recommended solution...
>
> - implement hash space testing in groups.  perhaps flag it on a per-group
level.  maybe a space is really just a group!
>
> - as you traverse the tree of groups and geoms in dSpaceCollide(), issue
the near callback at each node.  that gives the caller the ability to filter
out which geoms are passed to dCollide().
> _______________________________________________
> ODE mailing list
> ODE@q12.org
> http://q12.org/mailman/listinfo/ode
>
>