Speaker: Linfeng Zhao Northeastern University’s Khoury College
Time: 2025-06-13 09:00-2025-06-13 10:00
Venue: Seminar Room 2, 19th Floor, Tower C, TusPark (//hku.zoom.us/j/94295508050?pwd=hpBrebmLVPZSleX8Xjh3mwjDR1Agyt.1)

Abstract:

Embodied decision‑making tasks pose novel, long‑horizon, and partially observable challenges that need rich internal world models and robust planning pipelines. I leverage learning‑based methods to introduce a structured pipeline for embodied agents. First, I develop modular learning components that construct internal models via symmetry, compositionality, and hierarchical abstractions, then use these representations for planning. Second, I investigate planning strategies that actively gather information at inference time to refine the agent’s state estimate. Together, these modular design and components enable flexible deployment, scalable compute use during inference, and adaptation to diverse embodied decision‑making tasks.

Short Bio:

Linfeng Zhao is a final-year Ph.D. student in Computer Science at Northeastern University’s Khoury College, advised by Prof. Lawson L.S. Wong. He’s been closely working with LIS group at MIT (with Profs. Leslie Kaelbling and Tomas Lozano-Perez) and GLL group at Northeastern (with Prof. Robin Walters). Previously, he interned at Meta, Boston Dynamics AI Institute, Amazon, and Microsoft, and worked with Prof. Hao Su at UC San Diego. His research focuses on learning-based approaches for embodied decision-making for open environments and long-horizon tasks and experiment with mobile manipulation robots. He specifically studies learning for planning, world modeling, perception, and policy learning.