Shuran Song
Stanford University
Frontier Research Maph-index: 142 citations
Frontier Research Map
Why Now
She is one of the strongest frontier voices on perception-action loops in unstructured environments, especially where vision must support real robot behavior instead of offline recognition.
Key Ideas
- -Vision for home robots has to be open-vocabulary, long-horizon, and grounded in messy physical environments.
- -Embodied intelligence depends on data gathered through interaction, not only passively scraped images and videos.
- -Manipulation is a stress test for whether visual representations encode affordances rather than superficial appearance.
Open Questions
- ?What perceptual abstractions are stable enough for home robotics but flexible enough for open-world settings?
- ?How do we train systems that can adapt across embodiments without rebuilding everything from scratch?
- ?What is the right benchmark for mobile manipulation when scenes, tasks, and embodiments all shift?
Younger Agenda-Setters and Adjacent ML Thinkershigh confidence
Sign in to access this content