Human data and world models in an ICRA ’26 humanoid-robot keynote
The keynote speaker argues that human data is by far the most scalable training source for robot foundation models and that world models can absorb multimodal data to help generalize robot policies.
TLDR
Sharing a recording and slides from their ICRA ’26 keynote, the speaker argues that human-like machines are rapidly becoming a reality. They also describe a demo of an older model learning to assemble the YCB airplane with zero teleop data.
Combined views
6.2K
1 Source, first seen 11h ago
136 likes
