Hugging Face Daily Papers · June 10, 2026 · 3 min read

Role-Agent: Bootstrapping LLM Agents via Dual-Role Evolution

Mirrored from Hugging Face Daily Papers for archival readability. Support the source by reading on the original site.

Like Read original ↗

Role-Agent: Bootstrapping LLM Agents via Dual-Role Evolution<br>Github: <a href=\"https://github.com/AMAP-ML/roleagent\" rel=\"nofollow\">https://github.com/AMAP-ML/roleagent</a></p>\n","updatedAt":"2026-06-10T06:32:43.065Z","author":{"_id":"69a52b6bfc49cf45f641d563","avatarUrl":"/avatars/71be1cf443f24755f2a69801aeb5c451.svg","fullname":"wangxucong","name":"xuc865","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.6034516096115112},"editors":["xuc865"],"editorAvatarUrls":["/avatars/71be1cf443f24755f2a69801aeb5c451.svg"],"reactions":[{"reaction":"🔥","users":["xiaochonglinghu"],"count":1}],"isReport":false}},{"id":"6a290f3be1279a2eec75695b","author":{"_id":"66d255e3947594430c723ff6","avatarUrl":"/avatars/c56e4792332a01bf34085a75ee64916e.svg","fullname":"xiaochonglinghu","name":"xiaochonglinghu","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":9,"isUserFollowing":false},"createdAt":"2026-06-10T07:16:11.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Interesting!","html":"<p>Interesting!</p>\n","updatedAt":"2026-06-10T07:16:11.421Z","author":{"_id":"66d255e3947594430c723ff6","avatarUrl":"/avatars/c56e4792332a01bf34085a75ee64916e.svg","fullname":"xiaochonglinghu","name":"xiaochonglinghu","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":9,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7847728729248047},"editors":["xiaochonglinghu"],"editorAvatarUrls":["/avatars/c56e4792332a01bf34085a75ee64916e.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2606.10917","authors":[{"_id":"6a28c5ffe7d78ea7587e5319","name":"Xucong Wang","hidden":false},{"_id":"6a28c5ffe7d78ea7587e531a","name":"Ziyu Ma","hidden":false},{"_id":"6a28c5ffe7d78ea7587e531b","name":"Shidong Yang","hidden":false},{"_id":"6a28c5ffe7d78ea7587e531c","name":"Tongwen Huang","hidden":false},{"_id":"6a28c5ffe7d78ea7587e531d","name":"Pengkun Wang","hidden":false},{"_id":"6a28c5ffe7d78ea7587e531e","name":"Yong Wang","hidden":false},{"_id":"6a28c5ffe7d78ea7587e531f","name":"Xiangxiang Chu","hidden":false}],"publishedAt":"2026-06-09T14:28:07.000Z","submittedOnDailyAt":"2026-06-10T00:00:00.000Z","title":"Role-Agent: Bootstrapping LLM Agents via Dual-Role Evolution","submittedOnDailyBy":{"_id":"69a52b6bfc49cf45f641d563","avatarUrl":"/avatars/71be1cf443f24755f2a69801aeb5c451.svg","isPro":false,"fullname":"wangxucong","user":"xuc865","type":"user","name":"xuc865"},"summary":"Although Large Language Model (LLM) agents have demonstrated strong performance on complex tasks, their learning is often limited by inefficient interaction feedback and static training environments, which hinder broader generalization. To address these limitations, this paper introduces Role-Agent, black{a framework} that harnesses a single LLM to function concurrently as both the agent and the environment, enabling a bootstrapped co-evolution. Role-Agent comprises two synergistic components: World-In-Agent (WIA) and Agent-In-World (AIW). In WIA, the LLM acts as the agent and predicts future states after each action; the alignment between predicted and actual states is then used as a process reward, encouraging environment-aware reasoning. In AIW, the LLM analyzes failure modes from failed trajectories and retrieves tasks with similar failure patterns, thereby reshaping the training data distribution for targeted practice. Experiments on multiple benchmarks show that Role-Agent consistently improves performance, yielding an average gain of over 4\\% over strong baselines.","upvotes":73,"discussionId":"6a28c5ffe7d78ea7587e5320","githubRepo":"https://github.com/AMAP-ML/roleagent","githubRepoAddedBy":"user","ai_summary":"Role-Agent framework enables LLM agents to function as both agent and environment through bootstrapped co-evolution, improving performance via environment-aware reasoning and targeted practice.","ai_keywords":["Large Language Model","LLM agents","bootstrapped co-evolution","World-In-Agent","Agent-In-World","environment-aware reasoning","targeted practice"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","githubStars":48},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"66d255e3947594430c723ff6","avatarUrl":"/avatars/c56e4792332a01bf34085a75ee64916e.svg","isPro":false,"fullname":"xiaochonglinghu","user":"xiaochonglinghu","type":"user"},{"_id":"64d1dc5273174cecdffc97d3","avatarUrl":"/avatars/6564e6b68fee9673f75b6366adf39a3b.svg","isPro":false,"fullname":"Wang Yong","user":"seashell11","type":"user"},{"_id":"6929389f2d4c5b1204c4eacd","avatarUrl":"/avatars/98c6037f92ea193492db0c6dd6f73386.svg","isPro":false,"fullname":"Chens","user":"JupiterWis","type":"user"},{"_id":"650758da9622235d7dcba97e","avatarUrl":"/avatars/258802da8dfe3182e7f57288d6249f09.svg","isPro":false,"fullname":"Jianhao Zeng","user":"JianhaoZeng","type":"user"},{"_id":"68c261077e9a08a552ebcca4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/6G9CY69HxZZXAbAyueJ54.png","isPro":false,"fullname":"Bryan","user":"Bryan-A","type":"user"},{"_id":"6936d25459fd31452bd03ba4","avatarUrl":"/avatars/b15dca21fc5828a0593b68b393719c33.svg","isPro":false,"fullname":"rujingdang","user":"rujingdang","type":"user"},{"_id":"6459f2ae896f285ceb2384f0","avatarUrl":"/avatars/6bf4ed9ddd8f4dd45a97bec29274ae38.svg","isPro":false,"fullname":"wf","user":"Olivia0","type":"user"},{"_id":"661de9defdbc9c247f159d15","avatarUrl":"/avatars/38e21e78327cc908201122405c48f41b.svg","isPro":false,"fullname":"Rui Dai","user":"DerryD","type":"user"},{"_id":"64904c353be5db53615bd38a","avatarUrl":"/avatars/44296f0155fef0833aaf79201b5e344b.svg","isPro":false,"fullname":"chen zhihao","user":"mrbug","type":"user"},{"_id":"660ae36c58941eba04907543","avatarUrl":"/avatars/0f59544d578d6dafa19492386551e9f2.svg","isPro":false,"fullname":"yuyongjia","user":"yyjok","type":"user"},{"_id":"689d4a717fb1c6267bb59acc","avatarUrl":"/avatars/13b5117a0108e40cbffa8114c112cfad.svg","isPro":false,"fullname":"peter","user":"peterlrm","type":"user"},{"_id":"663bbb61ec1aafe3d6c05558","avatarUrl":"/avatars/b2f593d0ae0adbaad9a7a99b490e3a2b.svg","isPro":false,"fullname":"Ziyu Ma","user":"poiuytrewq123","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2606/2606.10917.md"}">

Papers

arxiv:2606.10917

Role-Agent: Bootstrapping LLM Agents via Dual-Role Evolution

Published on Jun 9

· Submitted by

wangxucong on Jun 10

Upvote

Authors:

Abstract

Role-Agent framework enables LLM agents to function as both agent and environment through bootstrapped co-evolution, improving performance via environment-aware reasoning and targeted practice.

Generated by Qwen/Qwen2.5-Coder-32B-Instruct

Although Large Language Model (LLM) agents have demonstrated strong performance on complex tasks, their learning is often limited by inefficient interaction feedback and static training environments, which hinder broader generalization. To address these limitations, this paper introduces Role-Agent, black{a framework} that harnesses a single LLM to function concurrently as both the agent and the environment, enabling a bootstrapped co-evolution. Role-Agent comprises two synergistic components: World-In-Agent (WIA) and Agent-In-World (AIW). In WIA, the LLM acts as the agent and predicts future states after each action; the alignment between predicted and actual states is then used as a process reward, encouraging environment-aware reasoning. In AIW, the LLM analyzes failure modes from failed trajectories and retrieves tasks with similar failure patterns, thereby reshaping the training data distribution for targeted practice. Experiments on multiple benchmarks show that Role-Agent consistently improves performance, yielding an average gain of over 4\% over strong baselines.

View arXiv page View PDF GitHub 48 Add to collection