I'm interested in the study of truly autonomous systems - agents that make all their decisions for themselves - and how we can use such entities to do useful things in the world. Once our robots no longer need to hold our hands, we can send them to places we are too afraid to go into ourselves. To this end, I've been a part of the CMU Mine Mapping Project, which sent an autonomous groundhog into an abandoned mine at the end of May.
There are a number of sub-issues involved with creating an autonomous agent. Currently, I'm investigating pretty quick ways of getting our robots to make sensible plans in uncertain environments. When almost all the information you had about a place turns out to be wrong, what do you do? And how can you do it quickly? It turns out that robots gain little by throwing tantrums.