Research

I develop learning and planning methods for sample-efficient, generalizable robot manipulation, with a focus on dexterous and long-horizon tasks. I am interested in how robots can build useful models of the physical world and use them to reason, plan, and improve through interaction. My goal is to develop world models that co-evolve with the agent’s experience and capabilities, enabling robots to reuse prior knowledge and continually learn over time. I am particularly interested in world models, inference-time adaptation, and self-directed learning.