I am interested in all aspects of sequential decision making under uncertainty, from philosophical to practical questions.

In particular, I am trying to enable robots to learn models of the world and use them for control (i.e. model-based reinforcement learning). This gives rise to many interesting research questions:

What a priori knowledge about the world is required and how do we have to explore to guarantee identifiability?
How can we tractably learn such models online, from a continuous, long stream of sensory data?
How can such models be used for computationally efficient optimal control?
Previously, I have mainly worked on Bayesian inference in dynamical systems (i.e. filtering). I have worked on general questions in approximate filtering, as well as methods exploiting the structure of specific problems, such as inferring the poses of all relevant objects and the robot manipulator from a depth sensor.