The authors developed EgoX, a framework that organizes demonstrations around corresponding kinds of observations and actions, using coordinates relative to the acting body.
Robots handle unfamiliar conditions better after learning from different bodies
Combining human, physical-robot and simulated demonstrations improved manipulation progress across six robot types, with larger gains under conditions outside the training distribution.
Big Tech
Yaru Niu · Binghong Chen · Changwei Yao · Shangtao Li · Zhenlong Fang · Revanth Krishna Senthilkumaran · +16 more
Carnegie Mellon University · University of California, Berkeley · Bosch Center for Artificial Intelligence · The University of Texas at Arlington · NVIDIA
Research Digest··2 min read
Niu and colleagues report that robots can learn manipulation more effectively by sharing experience across humans, physical robots and simulated agents, while retaining body-specific sensing and control interfaces.
The Daily Edition
The front page in your inbox each weekday morning: the lead, the picks, and why each one earned its place.