Metraipolis › archive › dc_1133634705cb

ComputerSD: Online Self-Distillation from Real-Time Feedback for Computer-Use Agents — arXiv:2609.40253

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.40253
License (as archived)
CC0-1.0
Archived
2026-10-07T12:46:17.623Z
SHA-256 of the archived text
245995e237a2e75c6b0fe97997b83e76705b324d45c6fdc76d09dc406df9cceb
Stable URL of this page
https://metraipolis.com/documento/dc_1133634705cb

How to cite

ComputerSD: Online Self-Distillation from Real-Time Feedback for Computer-Use Agents — arXiv:2609.40253 — arXiv (en), https://arxiv.org/abs/2609.40253 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_1133634705cb (Metraipolis archive dc_1133634705cb, sha256 245995e237a2e75c6b0fe97997b83e76705b324d45c6fdc76d09dc406df9cceb)
BibTeX
@misc{dc1133634705cb,
  title = {ComputerSD: Online Self-Distillation from Real-Time Feedback for Computer-Use Agents — arXiv:2609.40253},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.40253},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_1133634705cb; sha256 245995e237a2e75c6b0fe97997b83e76705b324d45c6fdc76d09dc406df9cceb},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.40253. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Online training enables computer-use agents (CUAs) to improve through interaction with executable environments. However, existing methods primarily rely on sparse outcome rewards, which provide no supervision for intermediate actions. On-policy self-distillation (OPSD) offers token-level learning signals through privileged rescoring, but directly applying it to CUA online training presents two challenges: fixed guidance may become misaligned with the student's current state, and guidance-induced probability shifts may conflict with step-level correctness. We introduce ComputerSD, an online self-distillation method for CUAs that converts real-time feedback from executed GUI transitions into guidance for policy learning. A fine-tuned GUI analyzer produces guidance and a step-level value score after each action; the guidance provides privileged context, while the score regulates the resulting OPSD signals. ComputerSD jointly optimizes token-level OPSD and trajectory-level GRPO in a fully asynchronous training framework. On OSWorld-Verified, ComputerSD outperforms outcome-only GRPO by 1.9 and 4.1 percentage points on the general-purpose Qwen3-VL-8B-Thinking and specialized EvoCUA-8B backbones, respectively. Evaluation in out-of-distribution settings further supports the generalizability of ComputerSD. These results demonstrate the effectiveness of learning from real-time feedback through online self-distillation for CUAs.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-10-01 in cs.AI).
Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.40253 — 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 2026-10-07T12:14:14.240Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-10-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).