I currently work with (at least) two advisors. My original group was Mike Stilman's Humanoids lab here at Georgia tech. Research in that lab primarily centers on the Golem project, a platform for investigating challenges in learning and planning on robots that match humans in size, strength, and dexterity. Broadly speaking, I got into this project to explore machine learning problems from an agent-based perspective. Within this area, I focus on learning and control methods to allow our robot to discover how constrained objects behave, through autonomous exploration. I think robots should (and will!) be capable of learning sophisticated and transferable models of the world for themselves, and that this problem lies at the nexus of research in machine learning, cognitive science, and robotics.
My de facto lab, however, is Charles Isbell's Pfunk group. We're an eclectic bunch, but overall we focus on agent-based machine learning. This includes reinforcement learning, of course, but also some game theory, supervised learning, and programming languages work. With this, my core interest is in exploring bayesian methods for learning dynamics with robots and virtual agents.
My thesis will eventually be called something to the effect of "Physics-Based Reinforcement Learning for Mobile Manipulation", and the basic idea is to incorporate physical-based models and planning representations from robotics into the Reinforcement Learning framework. I think this is a good idea because physics-based methods actually work on real robots (which is more than I can say for most of the models and planners in RL), but currently require a lot of engineering know-how to implement properly. By contrast, the RL framework provides a firm theoretical foundation for the problem of designing agents which can learn how to behave despite under-specified world knowledge. Presto!
Well, not really. For more information about my struggle to bring this idea to fruition, and for other projects I work on for school or fun, click here.
Update: I just successfully defended my thesis proposal, and the slides (with movies) can be found here.