The tl;dr of this article is that the author summarizes some really interesting leads on automated letter spacing, but so far hasn't come up with a winner. There are a number of loose ends in there that might be interesting for others to pick up and run with since he's made much of his code available.

Most promising to me is the concept of bubble kerning, which not only appears to work in the limited testing the author does here, but also meshes with my intuition about how to space letters.