The reason I wanted to read this paper was  
because I wasn't fully satisfied by the   
weird way Baraff and Witkins decided to  
handle collisions. 
   
This paper introduces some nice ideas.  
The use of repulsions to avoid collisions   
seems to be a really good idea. This is the  
idea I like the most in this paper.  
  
There are several other nice ideas, including   
using a hierarchical structure to detect   
collisions, which reduces the complexity of  
finding collisions from O(n^2) to O(n lg(n))  
(I think).  
  
Another nice idea is turning problematic areas   
into rigid objects! When there are just so  
many collisions in one area that there is no  
way to handle them, we just turn them to a  
rigid object and call it a day. This is one  
of those weird ideas that just work. I would  
have loved to see an example of what things  
look like at the moment they do this.  
  
Aside from collisions, there are some other nice  
ideas in the paper, like using subdivision to  
add more realism to the scene, integrating  
friction into their model and limiting the   
the amount of change on each triangle per  
frame to 10%.  
  
It was a nice read, even though I get the feeling  
that there is no systematic approach to cloth   
simulation. I just had the impression that it is  
chaos, and every author just does whatever they  
want!