My primary research areas are perception and learning, in particular for real-world robotic systems, typically using multi-view computer vision for 3D perception, but often including other modalities such as force-torque sensing. Usually, the robotic system has been seen as part of a larger environment that also include human co-workers, where the focus has been on trying to make human-robot interaction work as seamlessly as possible by interpreting the intentions of others and adapt accoridingly.
In recent years, my focus has been more on modelling sequential data in general. These studies have involved not just reinforcement-based control of robotic systems, but also prediction and generation of human movement, modelling and control of power systems, as well as online learning of dynamical systems for vehicle control. I am also interested in understanding factors of implicit regularization of neural networks and how these affect the networks' ability to generalize.