Metraipolis › archive › dc_a98ef761f340

Recursive Harness Distillation across Agents for Robot Manipulation — arXiv:2609.33378

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.33378
License (as archived)
CC0-1.0
Archived
2026-10-07T21:01:20.721Z
SHA-256 of the archived text
5e0ca9c1f8871484fa77c81d3e81e55ee4e64e8e4fcefa37348ed3b3cb593c5d
Stable URL of this page
https://metraipolis.com/documento/dc_a98ef761f340

How to cite

Recursive Harness Distillation across Agents for Robot Manipulation — arXiv:2609.33378 — arXiv (en), https://arxiv.org/abs/2609.33378 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_a98ef761f340 (Metraipolis archive dc_a98ef761f340, sha256 5e0ca9c1f8871484fa77c81d3e81e55ee4e64e8e4fcefa37348ed3b3cb593c5d)
BibTeX
@misc{dca98ef761f340,
  title = {Recursive Harness Distillation across Agents for Robot Manipulation — arXiv:2609.33378},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.33378},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_a98ef761f340; sha256 5e0ca9c1f8871484fa77c81d3e81e55ee4e64e8e4fcefa37348ed3b3cb593c5d},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.33378. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

A central goal in robotics is to enable manipulation across changing tasks and environments. Vision-language-action (VLA) models provide broad manipulation capabilities but can struggle when execution requires diagnosing failures and adapting behavior. Strong agents can discover effective interventions through interaction with these policies. We propose Recursive Harness Distillation to accumulate this experience as reusable guidance across agents. A strong agent distills its experience into a playbook for a light agent, then recursively refines the playbook using the light agent's execution feedback. The resulting playbook enables agents to reuse accumulated intervention knowledge in new task instances without updating model parameters. In real-world manipulation, the harness improves success from 37.3% to 64.0%. On SimplerEnv Bridge, the light agent with the playbook achieves 66.7% success, compared with 41.7% for the GR00T-only baseline, and outperforms the strong agent without a playbook. The same playbook also benefits the strong agent, which reaches 79.2% success. These results demonstrate the feasibility of harness distillation for robotics: intervention experience can be accumulated, refined through execution, and reused across agents to improve manipulation.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.AI).
Categorie arXiv: Robotics (cs.RO) ; Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.33378 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-07T20:29:02.627Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).