I work on embodied intelligence at the point where models meet the physical world. My current focus is building agents that can understand goals, interact with robots, and improve through feedback from real environments.

My work spans the full loop from human demonstrations and robot data to vision-language models, vision-language-action policies, and embodied agents. I am interested in how models can connect perception, reasoning, planning, and action to make robots more capable and adaptable.

I also explore systems that combine agents, tools, and real robots in a continuous process of learning and refinement. This includes robot manipulation, navigation, task decomposition, memory, tool use, and the interfaces that allow models to work reliably with physical systems.

Ultimately, I want to build embodied systems that can learn from experience, reason about their own behavior, and gradually become more useful in the real world.