This chapter goes over the technical issues of editing motion capture data. Many key ideas are in here: 

The data's properties:
* motion capture data is often in a format that doesn't lend itself to easy editing
* it doesn't contain any information about "why", or even important properties
* if you have an "angry walk", the property that makes it angry can be hard to describe, but easy to destroy

Poses:
* Skeletons are the way to go, though other formats can make certain constraints (like footplants) easier to achieve, though it usually requires that the system resolve connectivity constraints.

Motion:
* Curve representation has a lot to do with how changes manifest themselves -- dense values that come from motion capture may not be the most convenient form to make some changes. Curve-fitting reduces the number of keys, which allows easier editing. Various techniques (both linear and non-linear) are discussed.

* Cleaning up motion can be easily done by removing high frequency noise, but this will remove crisp motions (as well as some of the subtleties that give a motion its unique character). So low-pass filtering should be carefully applied.

* Aliasing in time scaling can produce odd resulting motions. So we should filter out higher frequency content before scaling.


Filtering:
* Adding motions together works in practice but not in theory.
* Time warps should be used to align motion before blending.


Motion Warping:
* Figure out displacement map for desired changes, interpolate map, then add back into original motion. Works surprisingly well.