Skip to content

Comment on S1: In-Context Learning for Robotics

Comments

Robotics data remains scarce and expensive, leaving many important tasks outside the pre-training distribution. Out-of-distribution ICL is therefore essential for general-purpose robot policies. S1 ICL can drive ultimate success in novel tasks 2/3rds of the time.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.