Metraipolis › archive › dc_c908bf4191df
BLPO: Beyond-Local Policy Optimization for Long-Horizon LLM Agents — arXiv:2611.03005
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2611.03005
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-03T10:33:37.526Z
- SHA-256 of the archived text
3715f487cbe27ffc553ae4745aba72fb8104c32de13bc8b824d6c48a3ae34978- Stable URL of this page
https://metraipolis.com/documento/dc_c908bf4191df
How to cite
BLPO: Beyond-Local Policy Optimization for Long-Horizon LLM Agents — arXiv:2611.03005 — arXiv (en), https://arxiv.org/abs/2611.03005 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_c908bf4191df (Metraipolis archive dc_c908bf4191df, sha256 3715f487cbe27ffc553ae4745aba72fb8104c32de13bc8b824d6c48a3ae34978)
BibTeX
@misc{dcc908bf4191df,
title = {BLPO: Beyond-Local Policy Optimization for Long-Horizon LLM Agents — arXiv:2611.03005},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2611.03005},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_c908bf4191df; sha256 3715f487cbe27ffc553ae4745aba72fb8104c32de13bc8b824d6c48a3ae34978},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.03005. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Long-horizon tasks require large language model (LLM) agents to coordinate interdependent actions, a capability central to automating complex workflows. Stepwise group-based reinforcement learning provides finer credit than trajectory-level supervision, but reliable comparisons must account for differences in interaction history. Enforcing this consistency within the limited rollouts of each task can leave insufficient support for estimating action preferences. We propose Beyond-Local Policy Optimization (BLPO), which estimates task-progress advantages from experience shared across tasks and training updates. We summarize trajectory prefixes into task-progress states and abstract actions by their function, allowing corresponding decisions from different tasks to contribute to shared estimates. To control bias and noise from heterogeneous tasks, we accumulate context-relative returns in a global memory with balanced aggregation and reliability weighting. These global advantages complement local comparisons over normalized actions and trajectory-level credit without a learned critic or additional rollouts. Experiments on ALFWorld and WebShop with 1.5B and 7B policies show improved task success over the evaluated baselines under matched training budgets. Further analyses show faster learning and shorter trajectories with little computational overhead. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2611.03005 — 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 2026-11-03T10:29:48.614Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).