EmbodiedSWE: Coding Agents Already Solve Complex Robot Tasks, Distilled into Fast VLA Policies
PeterHndrsn · x · 2026-09-24
EmbodiedSWE, shared by Haoxiang You, systematically benchmarks frontier coding agents on precise, dexterous, long-horizon robot tasks:
- The agents could already autonomously solve remarkably complex physical tasks — surprising the team
- But running a coding agent per robot action is expensive, slow, and hard to deploy safely
- The proposed path: use coding-agent solutions as teachers for VLAs — a single successful solution can be expanded into a large, diverse training dataset at surprisingly low cost
This suggests a viable route to general-purpose robots: a GPT-like "brain" distilled into fast System 1 robot policies.
Related event: EmbodiedSWE: Coding Agents Tackle Long-Horizon Robotic Tasks(4 posts)→
More from Embodied
- Homebrew robot arm build log Day 7: geartrain CAD, wiring and gripper sizing — _Stocko_ · 2026-11-11
- Perceptron AI launches EgoHome, an open benchmark for sub-second egocentric hand understanding — rohanpaul_ai · 2026-11-11
- Runway open-sources Praxis-1, turning video pretraining into real robot control — runwayml · 2026-11-11
- Unitree G1 humanoid wearing VR goggles pilots an A1 drone in land-air robot demo — CyberRobooo · 2026-11-11
- CoRL Spotlight Meta-Ctrl predicts robot plan failures, Llama 3 8B beats GPT-5.5 on VirtualHome — guyvdb · 2026-11-11
- Luna, a jewelry-style AI wearable that speaks before you ask, launches on Kickstarter — nikola_mr64990 · 2026-11-11