OpenAI's Astra shows in-context learning for mobile robot manipulation from video alone

npew · x · 2026-09-11

OpenAI researchers demoed Astra doing in-context learning on mobile manipulation: no text prompt — it infers the task from video alone, generalizes across different environments, camera angles and layouts, and decides autonomously between end-effector and joint-space control, with full trace recording.

Original post →

More from Embodied

Embodied channel →