Coding agents generate robot training data: VLA finetuned on agent demos runs on real hardware
Haoxiang You · hf · 2026-09-24
Researchers present EmbodiedSWE, a framework studying whether coding agents can solve long-horizon dexterous robotics tasks and turn verified solutions into scalable supervision for general robot policies.
Key contributions:
- EMBODIEDSWE-BENCH: a simulation benchmark spanning contact-rich manipulation, deformable objects, and long-horizon tasks requiring up to 30 minutes of continuous interaction. Frontier coding agents solve complex tasks and transfer solutions across tasks and embodiments.
- EMBODIEDSWE-GEN expands a single agent solution into large, diverse trajectory datasets for training VLAs. Performance scales with more demonstrations, and agent-aided diversification improves generalization to held-out task variations.
- A VLA finetuned solely on coding-agent-generated simulation demonstrations completes a long-horizon task on a real robot.
The framework uses coding agents to solve robotics tasks and convert verified solutions into scalable policy supervision.
Related event: EmbodiedSWE: Coding Agents Tackle Long-Horizon Robotic Tasks(4 posts)→
More from coding & agent
- User automates entire video clipping workflow with Nous Research's Hermes agent — Teknium · 2026-11-12
- As models improve, writing good evals gets harder and more important — amankhan · 2026-11-12
- Prime agent ships with day-one multi-agent communication, runs in Minecraft unmodified — sarahcat21 · 2026-11-12
- Autonomous agent edits just 1,024 weights to lift Qwen3.5 0.8B from 7.4% to 37.4% on MATH-500 — VictorSanh · 2026-11-12
- AI game jam with Anthropic, Google, OpenAI teams: cash prizes plus $50K continuation grants — chrisfirst · 2026-11-12
- Free AI game jam in SF Nov 14–15 with teams from Anthropic, Google, Meshy, OpenAI — chrisfirst · 2026-11-12