Hao Su
UC San Diego
Frontier Research Maph-index: 26 citations
Frontier Research Map
Featured Work
Learning Generic and Generalizable Object Manipulation Policiesofficial talks archive — 2024
Why Now
He is one of the clearest bridges between 3D vision, simulation, and embodied skill learning, which is exactly where CV spills into physical AI.
Key Ideas
- -Useful visual intelligence for robots depends on manipulation-centered representations, not only passive scene understanding.
- -Open task suites and simulation environments are becoming as important as datasets for progress.
- -Generalization in 3D and policy learning is the hard constraint that exposes brittle representations.
Open Questions
- ?What visual abstractions transfer best from simulation to contact-rich manipulation?
- ?How should 3D perception and policy learning share representations without collapsing into one another?
- ?Can the field build benchmarks that reward compositional generalization instead of narrow task completion?
Younger Agenda-Setters and Adjacent ML Thinkershigh confidence
Sign in to access this content