When I was a sophomore at Cornell University I did some research on the (now defunct) Cougar Project. The idea was to create an embedded database over a sensor network. This work was very similar to the work in TinyDB being done at Berkley around the same time.
Our group began to lose interest, especially after an Intel official told us, "there's no killer app for this stuff." I thought he was wrong then, and time seems to have supported my thinking.