Interactive Motion Generation from Examples
------------------------
The paper describes an animation system in which the users feed the system  
with examples of motion captures. And they specify some constraints for  
the system to follow (like initial position, final position, some poses along  
the path). The system proceeds to build the best blend of the given motions to  
obey the constraints.  
Some of the constraints should be followed verbatim (hard constraints), whilst  
other should be followed as much as possible (soft constraints).  
  
The authors also use motion graphs, even though their graph is not really the  
same as the other "motion graph" papter. Anyway, to search the graph, they use  
a summarizing method aimed at finding the broad level of the path, and the  
higher details can be found by digging deeper into the graphs.
  
Since finding the best path required an exponential time, they use an incremental  
random searching method, which starts by n random paths, and proceeds to mutate  
them and selecting the best ones as a basis to further mutations. The user can  
stop solving at any time. It would take about 10 seconds to reach a solution that  
is acceptable enough.