Metraipolis › archive › dc_aaa8a5e66ef1

ANCHOR: KL-Anchored Adaptation for Stable Offline-to-Online RL — arXiv:2611.02177

Source: arXiv (en)
Original: https://arxiv.org/abs/2611.02177
License (as archived): CC0-1.0
Archived: 2026-11-03T10:48:11.281Z
SHA-256 of the archived text: 336537622382be1f621150ed78df95dc9926d0a87baa9fb769a69b66b44ca87b
Stable URL of this page: https://metraipolis.com/documento/dc_aaa8a5e66ef1

How to cite

ANCHOR: KL-Anchored Adaptation for Stable Offline-to-Online RL — arXiv:2611.02177 — arXiv (en), https://arxiv.org/abs/2611.02177 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_aaa8a5e66ef1 (Metraipolis archive dc_aaa8a5e66ef1, sha256 336537622382be1f621150ed78df95dc9926d0a87baa9fb769a69b66b44ca87b)

BibTeX

@misc{dcaaa8a5e66ef1,
  title = {ANCHOR: KL-Anchored Adaptation for Stable Offline-to-Online RL — arXiv:2611.02177},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.02177},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_aaa8a5e66ef1; sha256 336537622382be1f621150ed78df95dc9926d0a87baa9fb769a69b66b44ca87b},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.02177. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Offline-to-online reinforcement learning (RL) follows the pretrain-finetune paradigm: an agent is first pretrained on historical data and then adapted through online interaction. We study offline-to-online adaptation without offline replay, where the agent must improve using only the pretrained model and newly collected experience. A key challenge is to avoid catastrophic failure, a sharp early performance collapse during online fine-tuning, while still achieving strong asymptotic performance. Through controlled experiments, we show that removing offline replay can induce catastrophic failure, while retaining strong conservative regularization from offline RL can suppress final online performance. We also find that common stabilization strategies do not reliably prevent early collapse, and that the critic-side mechanisms we test are insufficient on their own. Motivated by these findings, we propose ANCHOR, a simple adaptation method that anchors the actor to the offline-pretrained policy early in online fine-tuning, anneals this constraint over time, and retains only a small residual conservative critic regularizer. We provide a theoretical interpretation showing that KL anchoring helps preserve critic reliability across nearby occupancy distributions, with a two-phase regret bound explaining when anchoring should be maintained or relaxed. Across FrankaKitchen, AntMaze, and Adroit tasks, ANCHOR substantially reduces catastrophic failure while maintaining strong asymptotic performance. Ablations show that KL annealing and the residual conservative weight are important for balancing early stability and online adaptation, while additional analysis shows that ANCHOR is most beneficial when the pretrained policy contains useful behavior to preserve. Code is available at this https URL .

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2611.02177 — 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 2026-11-03T10:39:50.548Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).