1. This paper presented a novel method for creating realistic, controllable motion from motion capture database.

2. Realistic animation of human motion is a challenging task, and a popular solution is motion capture. But motion capture data is hard to edit and modify.

3. This paper presented a framework for generating realistic, controllable motion through a database of motion capture. This approach involves automatically constructing a graph that encapsulates connections among different pieces of motion in the database and then searching this graph for motions that satisfy user constraints.

4. Motion graph
Motion graph encodes how the motion capture data may be re-assembled in different ways. It is a directed graph wherein edges contain either pieces of original motion data or automatically generated transitions. The nodes then serve as choice points where these small bits of motion join seamlessly.  Motion graphs transform the motion synthesis problem into one of selecting sequences of nodes, or graph walks. 

5. Key technique points for Motion Graph Construction
(1) Detecting candidate transitions.
First, compute the distance for every pair of frames in the database, forming a sampled 2D error function. 
Second, select the local minima as candidate transitions.

(2) Selecting transition points
Transition points are selected as the local minima with low error.

(3) Creating transitions through blending

(4) Pruning the graph 
The graph is pruned such that, starting from any edge, it is possible to generate arbitrarily long streams of motion of the same type such that as much of the database as possible is used. 

6. Key technique points for creating motion from Graph
(1) Converting graph walks to motion
(2) Graph walk can be found as a path in the graph which meets user requirments and minimizes the error. 

7. A promising application of motion graph is path synthesis: given a path P specified by the user, generate motion such that the character travels along P.

8.  Limitations and Potential Improvments
(1) Automatically determine the transition thresholds.
(2) Incorporate parameterizable motions  into motion graph.
(3) Apply motion graphs to problems other than path synthesis.