EmbodiedSWE: Coding Agents for Long Horizon Dexterous Robotics
EmbodiedSWE uses coding agents to generate scalable simulation supervision for long-horizon robot policies.
EmbodiedSWE studies whether coding agents can supply scalable supervision for general robot policies. EmbodiedSWE-Bench tests agents on contact-rich manipulation, deformable objects, and long-horizon tasks requiring up to half an hour of interaction. Frontier agents solve complex tasks and transfer prior solutions, but the results are instance-specific and need heavy iteration. EmbodiedSWE-Gen turns one verified solution into diverse trajectories; VLA performance rises with more demonstrations, and a model finetuned only on those simulation demos completes a long-horizon task on a real robot.
- Benchmark spans contact-rich, deformable, and half-hour tasks
- Coding agents transfer solutions across tasks and embodiments
- Generated demonstrations improve VLA generalization to held-out variations
- A simulation-only finetuned VLA completes a real-robot task
Full article176 words · extracted from huggingface.co · click to collapse
We study coding agents for long-horizon, dexterous robotics and ask whether their solutions can provide scalable supervision for learning general robot policies. To test this, we develop EMBODIEDSWE-BENCH, a simulation benchmark for coding agents spanning contact-rich manipulation, deformable objects, and long-horizon tasks requiring up to half an hour of continuous interaction. We find that frontier coding agents can solve complex long-horizon tasks and transfer prior solutions across both tasks and embodiments. We also design supporting tools that help agents more effectively solve these tasks. However, the resulting solutions require substantial iterative interaction and are typically specialized to individual task instances. We therefore introduce EMBODIEDSWE-GEN, which expands a single solution from coding agent into large diverse trajectories for training a VLA. VLA performance improves with more generated demonstrations, and agent-aided diversification improves generalization to held-out task variations. We also show that a VLA finetuned solely on coding-agent-generated simulation demonstrations completes a long-horizon task on real robot. Together, our framework uses coding agents to solve complex robotics tasks and turn verified solutions into scalable supervision for robot policies.
Text extracted automatically; images, tables and formatting may be missing. Original: https://huggingface.co/papers/2609.27308