Adaptive Milestone Reward for GUI Agents
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866915794235949056 |
|---|---|
| author | Zheng, Congmin Mo, Xiaoyun Ma, Xinbei Lin, Qiqiang Zhao, Yin Zhu, Jiachen Lou, Xingyu Wang, Jun Wang, Zhaoxiang Liu, Weiwen Zhang, Zhuosheng Yu, Yong Zhang, Weinan |
| author_facet | Zheng, Congmin Mo, Xiaoyun Ma, Xinbei Lin, Qiqiang Zhao, Yin Zhu, Jiachen Lou, Xingyu Wang, Jun Wang, Zhaoxiang Liu, Weiwen Zhang, Zhuosheng Yu, Yong Zhang, Weinan |
| contents | Reinforcement Learning (RL) has emerged as a mainstream paradigm for training Mobile GUI Agents, yet it struggles with the temporal credit assignment problem inherent in long-horizon tasks. A primary challenge lies in the trade-off between reward fidelity and density: outcome reward offers high fidelity but suffers from signal sparsity, while process reward provides dense supervision but remains prone to bias and reward hacking. To resolve this conflict, we propose the Adaptive Milestone Reward (ADMIRE) mechanism. ADMIRE constructs a verifiable, adaptive reward system by anchoring trajectory to milestones, which are dynamically distilled from successful explorations. Crucially, ADMIRE integrates an asymmetric credit assignment strategy that denoises successful trajectories and scaffolds failed trajectories. Extensive experiments demonstrate that ADMIRE consistently yields over 10% absolute improvement in success rate across different base models on AndroidWorld. Moreover, the method exhibits robust generalizability, achieving strong performance across diverse RL algorithms and heterogeneous environments such as web navigation and embodied tasks. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_11524 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Adaptive Milestone Reward for GUI Agents Zheng, Congmin Mo, Xiaoyun Ma, Xinbei Lin, Qiqiang Zhao, Yin Zhu, Jiachen Lou, Xingyu Wang, Jun Wang, Zhaoxiang Liu, Weiwen Zhang, Zhuosheng Yu, Yong Zhang, Weinan Machine Learning Artificial Intelligence Computation and Language Reinforcement Learning (RL) has emerged as a mainstream paradigm for training Mobile GUI Agents, yet it struggles with the temporal credit assignment problem inherent in long-horizon tasks. A primary challenge lies in the trade-off between reward fidelity and density: outcome reward offers high fidelity but suffers from signal sparsity, while process reward provides dense supervision but remains prone to bias and reward hacking. To resolve this conflict, we propose the Adaptive Milestone Reward (ADMIRE) mechanism. ADMIRE constructs a verifiable, adaptive reward system by anchoring trajectory to milestones, which are dynamically distilled from successful explorations. Crucially, ADMIRE integrates an asymmetric credit assignment strategy that denoises successful trajectories and scaffolds failed trajectories. Extensive experiments demonstrate that ADMIRE consistently yields over 10% absolute improvement in success rate across different base models on AndroidWorld. Moreover, the method exhibits robust generalizability, achieving strong performance across diverse RL algorithms and heterogeneous environments such as web navigation and embodied tasks. |
| title | Adaptive Milestone Reward for GUI Agents |
| topic | Machine Learning Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2602.11524 |