【AI Seminar】2026.10.06 World Models and World-Action Models: Learning to Predict, Plan, and ActProf. Chun-Yi Lee
Release date：2026/09/30
Read ： 113
2. Date & Time: Tuesday, October 6, 2026, 13:00-15:00
3. Venue: 11th Floor, Management Building AI Lecture Hall,
4. Meeting Link: https://reurl.cc/DYKy9e or scan the QR code on the poster
About the Speaker:
Chun-Yi Lee is a Professor in the Department of Computer Science and Information Engineering at National Taiwan University and the supervisor of Elsa Lab. He holds a Ph.D. in Electrical Engineering from Princeton University. His research spans deep reinforcement learning, intelligent robotics, computer vision, and generative AI. He is a recipient of the 2026 Academia Sinica Scholar Award and the K. T. Li Breakthrough Award.
Lecture Abstract：
World models learn to predict how an environment evolves and are fast becoming a foundation for physical AI. This talk reviews recent progress, from model-based reinforcement learning (DreamerV3) and generative interactive world models (Genie, NVIDIA Cosmos) to self-supervised, prediction-based models such as Meta's V-JEPA and V-JEPA 2. It then turns to world-action models, which couple a learned world model with action for planning and control. Through examples such as V-JEPA 2-AC and Cosmos Policy, it shows how these models enable zero-shot robot planning, before closing with the main open challenges and opportunities.
附件一 AI Seminar 1006 李濬屹教授 海報_page-0001.jpg