Graphics Reference
In-Depth Information
Chapter
12
Geometrical
Robustness
Just as it is important with numerical robustness, it is important for collision geometry
on which tests are performed to be well formed. If not, the presence of small cracks
between neighboring faces could cause objects to fall through what is intended to be
solid. In fact, several geometrical features may cause problems for collision detection
code, including the following.
Redundant vertices, edges, and faces
●
Degenerate faces, including nonconvex or self-intersecting faces and faces of
zero or near-zero area
●
Incorrect orientation of single-sided faces
●
Unintended holes, cracks, gaps, and t-junctions between neighboring faces
●
Nonplanar faces
●
In addition, interpenetrating faces may or may not present a problem, depending
on the type of collision approach used. Overlapping faces may sometimes also be
a problem, specifically if they have different surface properties assigned. A large list
of the errors afflicting models in computer-aided design (CAD) is given in [Gu01].
Many of the errors listed also affect geometry for collision detection.
A lot of research is available on the cleanup and repair of geometry. Most of it
is focused on boundary representations of solid models, as used in CAD (see, for
example, [Barequet97] and [Murali97]). Unfortunately, although collision geometry
is sometimes constructed as a solid model it is more frequently modeled as a mesh
or a polygon soup. Unlike CAD geometry, such geometry cannot be expected to be
either connected or manifold. That is, faces can be both disjoint and interpenetrating,
and more than two faces may share an edge. The closed meshes of CAD geometry
are not supposed to have holes, and the cleanup processing usually assumes that all
465
