1. This paper introduces a data structure called a registration curve that expands 
the class of motions that can be successfully blended without manual input. 
Registration curves can improve upon existing automatic blending methods.

2. Motion blending produces new motions  by combining multiple clips according to 
time-varying weights. it will not produce realistic results unless the input motions 
are chosen with some care.

3. A registration curve consists of a timewarp curve, a coordinate alignment curve, 
and a set of constraint matches. A registration curve can be created in 
the following steps:
(1) Construct a time warping curve S(u)using time-warping algorithm.
(2) Build an alignment curve A(u) around S(u).
(3) Identify constraint matches by using S(u) to map constraint intervals into
    a standard time frame.

4. The fundamental technique to the approach is a function that simultaneously determines
the distance between two frames of motion and an aligning rigid 2D transformation. 
The difference is calculated through point cloud comparison.

5. Blending with registration curves involve the following 4 steps:
(1) Determine a position S(u_i)on the timewarp curve.
(2) Position and orient the frames at S(u_i).
(3) Combine frames based on the blending weights.
(4) Determine the constraints on the resulting frame.

6. Advantages
(1) Explicitly address differences in timing, path, and constraint state.
(2) Expand the range of motions that can be successfully blended. 
(3) offer a simple interface for common blending operations like transitions, 
    interpolations, and continuous control.
(4) Apply to other motions besides motion capture data.

7. Limitations
(1) Assume that structurally related parts of motions look more similar than unrelated parts. 
(2) Do not enforce any physical constraints like balance.