Metraipolis › archive › dc_1673369c2db8
Beyond Policy Patches: Distilling RL Specialists into Vision-Language-Action Policies — arXiv:2611.02777
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2611.02777
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-03T18:57:19.652Z
- SHA-256 of the archived text
37eb8928e1c029d5b46cb811d9b6f41d1028d834934a1e45fcd8aa9407e78a4e- Stable URL of this page
https://metraipolis.com/documento/dc_1673369c2db8
How to cite
Beyond Policy Patches: Distilling RL Specialists into Vision-Language-Action Policies — arXiv:2611.02777 — arXiv (en), https://arxiv.org/abs/2611.02777 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_1673369c2db8 (Metraipolis archive dc_1673369c2db8, sha256 37eb8928e1c029d5b46cb811d9b6f41d1028d834934a1e45fcd8aa9407e78a4e)
BibTeX
@misc{dc1673369c2db8,
title = {Beyond Policy Patches: Distilling RL Specialists into Vision-Language-Action Policies — arXiv:2611.02777},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2611.02777},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_1673369c2db8; sha256 37eb8928e1c029d5b46cb811d9b6f41d1028d834934a1e45fcd8aa9407e78a4e},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.02777. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Vision-language-action (VLA) policies can handle noncritical stages of long-horizon manipulation yet fail during short precision-critical phases. Lightweight reinforcement learning (RL) specialists can address these failures without directly applying RL to the full VLA, but retain auxiliary controllers and switching mechanisms at deployment. We study how to transfer their capabilities into a single flow-matching VLA. Direct imitation of specialist rollouts can reproduce action jitter and leave the VLA unprepared for states encountered during its own execution. We introduce REPAIR, a value-guided distillation framework that combines offline initialization with on-policy refinement. At each critical-phase observation, REPAIR queries the corresponding specialist for multiple action candidates and weights them using its critic to construct a supervision distribution. A shared weighted flow-matching objective trains the VLA first on observations from specialist-assisted rollouts, then on observations collected through its own execution. Specialists provide supervision without taking control during on-policy collection and are removed at deployment. Across six real-robot tasks with 30 evaluation trials per task, REPAIR achieves 100% success on all six tasks, with lower mean end-effector jerk than specialist-assisted execution on every task. Tema degli esploratori: statistica e scienza dei dati (elenco arXiv cs.RO, nuovi lavori del giorno). Categorie arXiv: Robotics (cs.RO). Pagina del lavoro: https://arxiv.org/abs/2611.02777 — 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 2026-11-03T18:17:07.756Z dalla pagina https://arxiv.org/list/cs.RO/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).