Monday, April 14
4:00 pm to 5:30 pm
Gates Hillman Center 6115
4:00 pm to 5:30 pm
Gates Hillman Center 6115
Learning Humanoid Robots from Simulation to Real to Simulation
Abstract: How do we teach humanoid robots to move like humans—and do so reliably in the real world? In this talk, I’ll share my journey in building a learning-based pipeline that closes the loop between simulation and reality for humanoid whole-body control.
Starting from real-time teleoperation (H2O), to scalable data humanoid collection (OmniH2O), to learning versatile controllers across tasks (HOVER), and finally to sim-to-real fine-tuning with real-world feedback (ASAP), I’ll walk through how each piece builds toward agile, expressive, and reliable humanoid skills. Central to this journey is a shift from sim-to-real transfer to a real-to-sim feedback loop—where real-world rollouts inform better simulators and better policies.
Through this lens, I’ll discuss key ideas like motion retargeting, sim2real transfer, multi-simulator training infra, delta dynamics learning, and real-world deployment, aiming to show how data, learning, and simulation co-evolve to push the frontier of humanoid skills.
Committee:
Prof. Guanya Shi (advisor)
Prof. Changliu Liu (advisor)
Prof. Kris Kitani
Zhengyi Luo