EvoCUA-1.5: Online Reinforcement Learning for Multi-turn Computer-Use Agents

EvoCUA-1.5 trains multi-turn desktop agents through online reinforcement learning in executable sandboxes. It combines step-level policy optimization, task filtering and calibration, an adaptive curriculum, and asynchronous rollout handling with staleness control.

AI-assisted editorial review against a primary source. This is a catalog annotation, not an independent reproduction or endorsement.

What is this resource?

Resource type

Research papers

Environment

desktop

Topics

reinforcement-learning, long-horizon, sandbox, data-generation

Record publication date

2026-07-07

Evidence retrieved

2026-10-16T06:36:32Z

Resource ID

arxiv:2607.09773

Recorded paper version

arXiv v2

Version updated: 2026-09-04T15:47:10Z.

This identifies the retained arXiv metadata snapshot; newly observed changes require separate review.

Authors

Mianqiu Huang, Taofeng Xue, Chong Peng, Jinrui Ding, Jie Yang, Sicheng Fan, Jiale Hong, Yufei Gao, Xiaocheng Zhang, Linsen Guo, Xin Yang, Dengchang Zhao, Yuchen Xie, Peng Pei, Xunliang Xie, Xipeng Qiu

Cite this paper

Download BibTeX · Download CSL JSON

Citations preserve the selected source edition, its authors and recorded date precision. Unrecorded venues and identifiers are omitted.

Explore related topics

Scope and limitations

The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.

Primary evidence and verification

Source retrieval is separate from experiment reproduction. Use the original publication or publisher page when citing claims about the resource.

1. arxiv api

   Retrieved 2026-10-16T06:36:32Z. Supports: name, authors, published_at, updated_at, identifiers.arxiv, details.arxiv_snapshot.

   Content fingerprint

   13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285
2. editorial primary review

   Retrieved 2026-10-16T06:36:32Z. Supports: name, kind, summary, environments, topics, scope, notes.

   Content fingerprint

   13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285

Suggest a correction

Open a source-backed correction. Include this record ID and the original source. Read our editorial policy and coverage boundaries.