The Closed Loop of Robot Intelligence: Model to Data

YuXiang_IRVL · x · 2026-07-16

In Karen Liu's closing keynote on Robot Intelligence, the core takeaway is summarized in one phrase: Model → Data → Policy → Data.

The main focus isn't merely chasing better models, but rather establishing and continuously running a data flywheel. This allows models, data, and policies to iterate against each other, forming a much more robust closed loop for robot learning.

Original post →

More from Embodied

Embodied channel →