RAX open-sourced: frozen coding agents learn robot tasks via online exploration, zero demos
du_yilun · x · 2026-10-17
A Harvard/NVIDIA/Berkeley/MIT/Stanford team releases RAX (Robot Agent eXploration): a coding agent connects directly to a robot's perception and control APIs, learning tasks through online exploration — the model stays frozen while exploration scaling improves programs and builds reusable skills.
Results: ZERO teleoperation or action demonstrations on real robots — from insertion with 0.1mm clearance and air-fryer manipulation to deformable objects and long-horizon mobile whole-body control.
Fully open-sourced: harness code, a community skill library, and a Hugging Face dataset with complete real robot logs including Claude sessions.
More from Embodied
- Perplexity CEO builds interactive physics reconstruction of Tesla Roadster jump demo — AravSrinivas · 2026-10-17
- Professor recruiting one PhD student for Fall 2027 on agentic wearables — jasonwuishere · 2026-10-17
- Sandbar's Stream Ring nails whisper-quiet dictation across iOS and Mac in demo — thesephist · 2026-10-17
- U. Tokyo's TwinX robot: one wire links two wheeled-legged bots for morphing and teamwork — LexiLove · 2026-10-17
- Snapchat Specs pop-up in LA: tech is cool but the $2k glasses 'look ridiculous' — SimonBalmain · 2026-10-17
- IEEE profiles roboticist Peter Stone: robot soccer, service robots and being a generalist — PeterStone_TX · 2026-10-17