My research sits across machine learning, computer vision, and robotics - three areas that are increasingly difficult, and increasingly unproductive, to treat as separate. I am interested in how models represent and reason about the world, visually, linguistically, and physically, and in where those representations break down.

I have a deep fascination with reinforcement learning - not just as a training method, but as a framework for thinking about how models learn to reason and act. That thread runs through most of what I work on.