### Summary
This paper introduces match webs as a method for searching a large database of motions for motions semantically similar to a reference motion
### Problem
Small motion databases are relatively simple to search, but are too sparse to blend existing motions into new ones.  Large databases are useful for creating new blended motions, but are difficult to search.  This paper tries to address this problem by proposing a method for efficiently searching large motion databases.
### Methods Used
1. Frames are compared using a numerical distance metric
2. Every pair of frames is compared and given a similarity score
3. These scores are plotted on a grid, and chains of local minima are found
4. Bridges are made between nearby chains, and stored as a match web

### Key Ideas
In addition to the searching mechanism mentioned above, this paper goes on to suggest a method for sampling blend weights to map from input blend weights to output positions