I study how agents decide what experience to acquire, use feedback to construct and improve robot policies, and retain what they learn as reusable physical skills. My interests span demonstrations, simulation, and real-world interaction, with an emphasis on learning efficiently under limited physical experimentation.
I am a core contributor to ACE-CaP-0: Real-World Code-as-Policies in the Era of Experience. The project explores how a coding agent can request robot demonstrations and use simulation on demand to develop reusable code policies through real-world feedback. The resulting programs execute with runtime perception and control, without a human or coding agent in the loop.
This direction centers on three connected questions: what experience to acquire, how to use it to improve a policy, and how to retain the resulting capability for reuse. Learning can take place through program revision, skill accumulation, or parameter optimization.
