My research focuses on robot learning, with a particular emphasis on leveraging internet-scale data and foundation models for robotics.
My ultimate goal is to free humans from dangerous or mundane tasks by entrusting them to machines.
Previously, I received my B.S. in Computer Science from Columbia University,
where I worked with Peter Allen on brain-signal guided robot learning
and Tony Dear on reinforcement learning for snake robot locomotion.
We introduce ZeroMimic, a system that distills robotic manipulation skills from egocentric human web videos for zero-shot deployment in diverse environments with a variety of objects.
POCR chains pre-trained "what" and "where" foundation models to create object-centric representations for robotics.
The "where" model identifies object candidates with segmentation masks, which are then bound to slots and encoded by the "what" model,
enabling robots to learn policies over these representations.
