RAX at CoRL2026: robots learn 0.1mm insertion with zero demos by scaling online exploration
du_yilun · x · 2026-11-08
At CoRL2026 the team will present RAX, an agentic robotics system that scales online exploration instead of demonstration data. A coding agent connects directly to the robot's perception and control APIs, learns from feedback, and builds perceive-plan-act programs while the model stays frozen. Results include zero teleoperation or action demonstrations on real robots, from 0.1mm-clearance insertion to deformable objects and long-horizon mobile whole-body control.
More from Embodied
- Dimensional-Funded ESP32 Treasure Hunt Workshop Pushes American Open Source in Robotics — StewartalsopIII · 2026-11-08
- Samsung, SK Hynix and Micron in talks to standardize fab robots for fully automated fabs by 2030 — zephyr_z9 · 2026-11-08
- The brain as the next UI: upcoming article charts BCI-meets-agentic-AI roadmap via earbuds — melnykowycz · 2026-11-08
- Tesla to Launch More Efficient Cybercab in 2025; Analyst Sees Sub-$15k COGS at Scale — JOBhakdi · 2026-11-08
- Blogger ditches folding iPhone Duo for iPhone Air: attention finally back on content — op7418 · 2026-11-08
- While US debates datacenter bans, PUDU robots are already deployed at scale — Scobleizer · 2026-11-08