I read the usual CS papers, plus lots of PhDs thesis. I follow the references in the bibliographies like crazy, and I also look up the writer's on the 'Net and see what other stuff they've done/written.
It feels like detective work, actually: the most-cited papers are frequently not the best. There's tons of researchers working in obscurity but who nevertheless have useful results.
I see my role as taking advances from research and bringing them into practical use.
What are some examples of the best research that is obscure?
In general, stack languages are great for graphics in both 2 dimensions (PostScript) and 3 dimensions (GML)[1].
[0] http://digisrv-1.biblio.etc.tu-bs.de:8080/docportal/receive/...
[1] http://en.wikipedia.org/wiki/Generative_Modelling_Language
I have a couple of times found myself inventing solutions to problems I had in the business, only to see them appear as cutting edge open source at the same or a little later (ie Python deployment tools)
I have always assumed one needs to be working in an area, pushing hard and then finding that "what no-one has an answer to this!?" moment.
I would have trouble believing I could find a solution at the cutting edge, and find the people who needed it, without being knee-deep myself.