EmbodiedSWE distills coding agents into robot policies for long-horizon control
PeterHndrsn · x · 2026-09-24
- EmbodiedSWE, a collaboration across Princeton, Yale, CMU and Stanford, asks whether coding agents can solve complex long-horizon robotics control tasks and whether those solutions can train VLAs.
- Key findings: GPT-6 can handle robotics tasks far more long-horizon and dexterous than any VLA; the team shows how far coding agents go and scalably converts their solutions into large amounts of diverse demonstrations, yielding better generalization and real-robot transfer.
- Co-author PeterHndrsn adds that Astra performs very well on this benchmark.
Related event: EmbodiedSWE: Coding Agents Tackle Long-Horizon Robotic Tasks(4 posts)→
More from Embodied
- UBTECH sells 400 U1 Pro humanoids to Yongda Auto in ~$10M deal — CyberRobooo · 2026-11-12
- Baidu World 2026 to showcase Apollo Go robotaxi rides and Apollo World physical AI model — Baidu_Inc · 2026-11-12
- NVIDIA's open-source Alpamayo 2 Edge VLA driving model lands in Autoware — drmapavone · 2026-11-12
- A closer look at Astra's robot gripper design (and a joke at its expense) — _Stocko_ · 2026-11-12
- Robot startup passes first presale milestone with 10+ units claimed — R4nsack · 2026-11-12
- Runway launches Praxis-1, a 27B MoE world action model, with a 73ms Fast variant — ChongZzZhang · 2026-11-12