GPT-6 Astra drives a robot arm on first try via physical ICL; Ken Goldberg touts Agentic Robotics
zhaoran_wang · x · 2026-09-11
- A viral demo claims GPT-6 Astra supports physical in-context learning out of the box: the team dropped a human recording of a novel task into the Codex app, prompted it to drive a robot arm the same way, and it worked on the first pass.
- Ken Goldberg (UC Berkeley / AUTOLab) amplified it, arguing Agentic Robotics is "on fire": by bridging model-based and model-free methods it combines the best of both. To tackle latency, AR now generates executable ROS2 code, and he says the frontier is now harness design.
More from Embodied
- Tesla Cybercab carefully passes cyclist on narrow hillside road in viral clip — EricETesla · 2026-09-11
- FCC Supply-Chain Rules Could Shape How US Robots Are Built and Sold — Rewkang · 2026-09-11
- First Rabbit R1 units arrive; early user promises a rethink of human-machine interaction — SimonBalmain · 2026-09-11
- Rabbit pushes surprise OTA update fixing R1 battery drain and SIM connectivity — SimonBalmain · 2026-09-11
- Masked Mimic now playable on Miniverse with sparse point control mode — carlosdponx · 2026-09-11
- Shanghai team open-sources running, jumping DIY humanoid robot with full hardware and RL training stack — tom_doerr · 2026-09-11