Code, data, and trained models will be released at <a href=\"https://mobile-forge.github.io\" rel=\"nofollow\">https://mobile-forge.github.io</a>.</p>\n","updatedAt":"2026-06-24T02:07:05.401Z","author":{"_id":"64d761b98ebc40443831f82a","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64d761b98ebc40443831f82a/DHBOtOstiFp2-lDY6b9gb.png","fullname":"Guangyi Liu","name":"lgy0404","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8735991716384888},"editors":["lgy0404"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/64d761b98ebc40443831f82a/DHBOtOstiFp2-lDY6b9gb.png"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2606.19930","authors":[{"_id":"6a349baf4c5c5e0d69bf1b9e","name":"Guangyi Liu","hidden":false},{"_id":"6a349baf4c5c5e0d69bf1b9f","name":"Pengxiang Zhao","hidden":false},{"_id":"6a349baf4c5c5e0d69bf1ba0","name":"Gao Wu","hidden":false},{"_id":"6a349baf4c5c5e0d69bf1ba1","name":"Yiwen Yin","hidden":false},{"_id":"6a349baf4c5c5e0d69bf1ba2","name":"Mading Li","hidden":false},{"_id":"6a349baf4c5c5e0d69bf1ba3","name":"Liang Liu","hidden":false},{"_id":"6a349baf4c5c5e0d69bf1ba4","name":"Congxiao Liu","hidden":false},{"_id":"6a349baf4c5c5e0d69bf1ba5","name":"Zhang Qi","hidden":false},{"_id":"6a349baf4c5c5e0d69bf1ba6","name":"Mengyan Wang","hidden":false},{"_id":"6a349baf4c5c5e0d69bf1ba7","name":"Liang Guo","hidden":false},{"_id":"6a349baf4c5c5e0d69bf1ba8","name":"Yong Liu","hidden":false}],"publishedAt":"2026-06-18T00:00:00.000Z","submittedOnDailyAt":"2026-06-24T00:00:00.000Z","title":"MobileForge: Annotation-Free Adaptation for Mobile GUI Agents with Hierarchical Feedback-Guided Policy Optimization","submittedOnDailyBy":{"_id":"64d761b98ebc40443831f82a","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64d761b98ebc40443831f82a/DHBOtOstiFp2-lDY6b9gb.png","isPro":false,"fullname":"Guangyi Liu","user":"lgy0404","type":"user","name":"lgy0404"},"summary":"MLLM-based mobile GUI agents have made substantial progress in UI understanding and action execution, but adapting them to real target apps remains costly because mobile apps are numerous, frequently updated, and hard to cover with human-written tasks, demonstrations, or reward labels. Existing annotation-free GUI learning reduces manual supervision, yet lacks a unified substrate connecting target-app exploration, curriculum mining, rollout execution, and feedback, while policy optimization often relies on isolated rollouts and coarse rewards that are hard to convert into reliable improvement signals. We present MobileForge, an annotation-free adaptation system for mobile GUI agents. MobileForge consists of MobileGym, which grounds task generation and rollout evaluation in real mobile app interaction, and Hierarchical Feedback-Guided Policy Optimization (HiFPO), which turns trajectory outcomes, step-level process feedback, and corrective hints into hint-contextualized step-level GRPO updates. Using only automatically generated annotation-free adaptation data, MobileForge adapts Qwen3-VL-8B to 67.2% Pass@3 on AndroidWorld, close to the closed-data GUI-specialized GUI-Owl-1.5-8B base model at 69.0%. The MobileForge-adapted ForgeOwl-8B further reaches 77.6% Pass@3 on AndroidWorld and 41.0% success on the out-of-domain MobileWorld GUI-only split, establishing the strongest open-data mobile GUI agent in our evaluation. Code, data, and trained models will be released at https://mobile-forge.github.io/.","upvotes":22,"discussionId":"6a349baf4c5c5e0d69bf1ba9","projectPage":"https://mobile-forge.github.io","githubRepo":"https://github.com/kwai/MobileForge","githubRepoAddedBy":"user","ai_summary":"MobileForge enables efficient adaptation of mobile GUI agents through annotation-free learning by combining real app interaction grounding with hierarchical feedback-guided policy optimization.","ai_keywords":["MLLM-based mobile GUI agents","annotation-free GUI learning","MobileGym","Hierarchical Feedback-Guided Policy Optimization","GRPO updates","Pass@3","success rate"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","githubStars":1,"organization":{"_id":"69bcbf46685c38830c5f8892","name":"kwaiAI","fullname":"kwai","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/6882dccd3dbdaf621b683333/jmnA7jSbcQby728JAArIj.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"64d761b98ebc40443831f82a","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64d761b98ebc40443831f82a/DHBOtOstiFp2-lDY6b9gb.png","isPro":false,"fullname":"Guangyi Liu","user":"lgy0404","type":"user"},{"_id":"620783f24e28382272337ba4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/620783f24e28382272337ba4/zkUveQPNiDfYjgGhuFErj.jpeg","isPro":false,"fullname":"GuoLiangTang","user":"Tommy930","type":"user"},{"_id":"677cd488b35098e1340c940e","avatarUrl":"/avatars/f3b41ecc994ecc1f08ea1ba7e6467ab4.svg","isPro":false,"fullname":"Wu Gao","user":"Wugao02","type":"user"},{"_id":"694b8c49d7e02d8a1c1d8ebb","avatarUrl":"/avatars/4b6008ccf0562a8f1cf85f0d62ec2650.svg","isPro":false,"fullname":"jack","user":"113tom","type":"user"},{"_id":"643429be546e16f17a133929","avatarUrl":"/avatars/fe82c49367ac05d5decab5ffcda62441.svg","isPro":false,"fullname":"Wooo Taylor","user":"Wooo0","type":"user"},{"_id":"6458ce236fa580137af5aa95","avatarUrl":"/avatars/db65a7332e375eb5daad5c1b076b1e3b.svg","isPro":false,"fullname":"Yuxiang Chai","user":"Yuxiang007","type":"user"},{"_id":"666aa99cd1652853e4f9a8b9","avatarUrl":"/avatars/7cd5a0c34b5ccb8eff5a353d88d15a93.svg","isPro":false,"fullname":"HanXiao","user":"HanXiao1999","type":"user"},{"_id":"6779c21c76d1c8d9cf03fbab","avatarUrl":"/avatars/6efab949d19515926015f191f31392c1.svg","isPro":false,"fullname":"XiangChen","user":"Soever","type":"user"},{"_id":"676127cf11b19ea602bb202a","avatarUrl":"/avatars/dfd802a24bd63e509728159ebb1769f6.svg","isPro":false,"fullname":"Zhengxi Lu","user":"LZXzju","type":"user"},{"_id":"663e1cc209862e819b9e694c","avatarUrl":"/avatars/005a2ed070f0c65223a17c88b18f8e93.svg","isPro":false,"fullname":"Yaozhen Liang","user":"asot2887","type":"user"},{"_id":"66e01f65f147db9777c74aa7","avatarUrl":"/avatars/c2cc265a27f88bccdcfd43ce9909529d.svg","isPro":false,"fullname":"Zhixin Lin","user":"Zhixin-L","type":"user"},{"_id":"646def60df618b303b419323","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/646def60df618b303b419323/JLJGYen4-5M8ivsLsSk0w.jpeg","isPro":false,"fullname":"Lei Wang","user":"demolei","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":3,"organization":{"_id":"69bcbf46685c38830c5f8892","name":"kwaiAI","fullname":"kwai","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/6882dccd3dbdaf621b683333/jmnA7jSbcQby728JAArIj.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2606/2606.19930.md","query":{}}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2606.19930 in a model README.md to link it from this page.

Cite arxiv.org/abs/2606.19930 in a Space README.md to link it from this page.