Robot Astra shows in-context learning for mobile manipulation, adapting via video alone
Scobleizer · x · 2026-09-12
A demo of the GPT-6-powered robot Astra shows in-context learning on mobile manipulation: across a different environment, camera angle, and layout, it infers the task from video with no text prompt, and even decides when to use end-effector vs joint space control, with trace recording afterwards. Scobleizer notes that after major robotics companies achieved ICL last month, Astra catching up is great for deployment speed but frustrating for foundation model makers.
Related event: OpenAI's Astra Robot Demonstrates Video In-Context Learning(2 posts)→
More from Embodied
- Tesla Optimus 'robot university' leak: 40-camera mocap suits, 24/7 data ops for months — Scobleizer · 2026-10-28
- Blind Robot Hand Spins Baoding Balls 225 Times With Only Touch — Scobleizer · 2026-10-28
- Predicting Uber Has 3 Years Left in the US as Waymo Takes Over — JHochderffer · 2026-10-28
- Rollback: World Models Rewind Robot Decisions to Explore Counterfactual Actions — agibot-world · 2026-10-28
- GPT-6.1 Terra beats Opus 5.5 at real robot control, 3.7x better than GPT-5.6 from 3 months ago — ycombinator · 2026-10-28
- Robots gearing up for a roboparty: researcher teases policy still in training — andrew_n_carr · 2026-10-28