Metraipolis › archive › dc_19e8501aa09a
SCA: Spatial Credit Assignment for Reinforcement Learning of GUI Agents — arXiv:2609.36939
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.36939
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T16:23:19.115Z
- SHA-256 of the archived text
67ecf1958bccc1ee21dc5d93e73f7ac5f1508f113fac43ac356cbc8ecfa774a9- Stable URL of this page
https://metraipolis.com/documento/dc_19e8501aa09a
How to cite
SCA: Spatial Credit Assignment for Reinforcement Learning of GUI Agents — arXiv:2609.36939 — arXiv (en), https://arxiv.org/abs/2609.36939 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_19e8501aa09a (Metraipolis archive dc_19e8501aa09a, sha256 67ecf1958bccc1ee21dc5d93e73f7ac5f1508f113fac43ac356cbc8ecfa774a9)
BibTeX
@misc{dc19e8501aa09a,
title = {SCA: Spatial Credit Assignment for Reinforcement Learning of GUI Agents — arXiv:2609.36939},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.36939},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_19e8501aa09a; sha256 67ecf1958bccc1ee21dc5d93e73f7ac5f1508f113fac43ac356cbc8ecfa774a9},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.36939. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
GUI agents automate tasks on digital devices by grounding language instructions in visual interfaces. Existing group-relative reinforcement learning improves GUI action prediction by comparing the rewards of multiple responses sampled from the same GUI state. However, binary evaluation treats spatially different failed clicks as identical and provides no relative signal when all sampled clicks fail. To address these limitations, we propose Spatial Credit Assignment (SCA), which uses the screen coordinates of sampled clicks to refine group-relative credit. Specifically, SCA predicts each held-out response's reward from the other responses in groups containing both successes and failures, then uses the prediction residual to adjust credit. When all sampled clicks fail, SCA instead orders them by distance to the annotated target. These spatial references are used only to construct the training update; the deployed policy remains unchanged. We evaluate whether this correction improves the policy update itself by comparing its error and directional alignment with the exact return gradient in a controlled synthetic study. Across GUI grounding and offline action-prediction benchmarks, SCA improves grounding across professional domains and achieves the strongest results among reinforcement-fine-tuned models on most action-prediction metrics, with consistent gains across the reported GUI suites. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-30 in cs.AI). Categorie arXiv: Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2609.36939 — 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 2026-10-07T15:54:53.570Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-30?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).