Research Statement
My research focuses on robot learning and planning for embodied agents that must act over long horizons in open-world environments. I study how structured abstractions, world models, and decision-time reasoning can make robot behavior more generalizable and efficient.
A recurring theme is turning structure in the world, such as symmetry, compositionality, hierarchy, and uncertainty, into practical learning and planning algorithms.
Learning and Planning with Structured Abstraction for Embodied Decision-Making
PDFResearch Areas
Abstraction and Representation
Learning structured representations (symmetry, compositionality) for efficient generalization.
Planning and World Modeling
Developing world models and planning algorithms for long-horizon reasoning.
Decision-time Scaling
Scaling decision quality with more compute and interaction.
Mobile Manipulation
Enabling diverse manipulation skills on mobile robot platforms.
Research Impact
My work aims to bridge the gap between classical planning and modern learning approaches, enabling robots to operate robustly in the real world.