Viktor Leis's working group consistently produces interesting fundamental database research. AdaptiveRadixTries, HeightOptimisedTrees, and the Umbra database system are all done by them.
ART is an elegant way of thinking about the problem of index representation. I've been using ART variants for almost 15 years now. The algorithm was somewhat common in supercomputing/HPC (used to index some types of sparse data models), which is where I originally came across it. In addition to being performant, ART is considerably more expressive and flexible than people might assume.
