OS-Themis: A Scalable Critic Framework for Generalist GUI Rewards
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866917353665593344 |
|---|---|
| author | Li, Zehao Wu, Zhenyu Zhao, Yibo Yang, Bowen Xie, Jingjing Liu, Zhaoyang Liu, Zhoumianze Jin, Kaiming Liang, Jianze Li, Zonglin Wu, Feng Zhou, Bowen Wang, Zun Ding, Zichen |
| author_facet | Li, Zehao Wu, Zhenyu Zhao, Yibo Yang, Bowen Xie, Jingjing Liu, Zhaoyang Liu, Zhoumianze Jin, Kaiming Liang, Jianze Li, Zonglin Wu, Feng Zhou, Bowen Wang, Zun Ding, Zichen |
| contents | Reinforcement Learning (RL) has the potential to improve the robustness of GUI agents in stochastic environments, yet training is highly sensitive to the quality of the reward function. Existing reward approaches struggle to achieve both scalability and performance. To address this, we propose OS-Themis, a scalable and accurate multi-agent critic framework. Unlike a single judge, OS-Themis decomposes trajectories into verifiable milestones to isolate critical evidence for decision making and employs a review mechanism to strictly audit the evidence chain before making the final verdict. To facilitate evaluation, we further introduce OmniGUIRewardBench (OGRBench), a holistic cross-platform benchmark for GUI outcome rewards, where all evaluated models achieve their best performance under OS-Themis. Extensive experiments on AndroidWorld show that OS-Themis yields a 10.3% improvement when used to support online RL training, and a 6.9% gain when used for trajectory validation and filtering in the self-training loop, highlighting its potential to drive agent evolution. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_19191 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | OS-Themis: A Scalable Critic Framework for Generalist GUI Rewards Li, Zehao Wu, Zhenyu Zhao, Yibo Yang, Bowen Xie, Jingjing Liu, Zhaoyang Liu, Zhoumianze Jin, Kaiming Liang, Jianze Li, Zonglin Wu, Feng Zhou, Bowen Wang, Zun Ding, Zichen Artificial Intelligence Reinforcement Learning (RL) has the potential to improve the robustness of GUI agents in stochastic environments, yet training is highly sensitive to the quality of the reward function. Existing reward approaches struggle to achieve both scalability and performance. To address this, we propose OS-Themis, a scalable and accurate multi-agent critic framework. Unlike a single judge, OS-Themis decomposes trajectories into verifiable milestones to isolate critical evidence for decision making and employs a review mechanism to strictly audit the evidence chain before making the final verdict. To facilitate evaluation, we further introduce OmniGUIRewardBench (OGRBench), a holistic cross-platform benchmark for GUI outcome rewards, where all evaluated models achieve their best performance under OS-Themis. Extensive experiments on AndroidWorld show that OS-Themis yields a 10.3% improvement when used to support online RL training, and a 6.9% gain when used for trajectory validation and filtering in the self-training loop, highlighting its potential to drive agent evolution. |
| title | OS-Themis: A Scalable Critic Framework for Generalist GUI Rewards |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2603.19191 |