This paper describes a method to generate data structures for
large motion data sets so it is easier to query logically
similar motions parts. This paper explains how match webs are
created and used. Also describes various techniques to reduce
the time & space complexity of creating and storing the match web.
The paper also suggests finding numerically similar motions
and using them to find more motions to be able to reach
logically similar but numerically dissimilar motions.
This paper also describes how to parameterize the blend space
to create more user intuitive parameters.
Like in the regcurve paper, I'm thinking if using del local
orient and using the distance between the del local orient
might possibly allow more efficient motion web creation
by allowing to directly find logically similar motions
instead of needing to find some intermediate motions
to be able to reach a numerically different motion.