\n\t\t\n\t\n\t\n\t\t🌍 UI-Simulator: LLMs as Scalable, General-Purpose Simulators For Evolving Digital Agent Training\n\t\n\n

\n","updatedAt":"2025-10-17T01:50:18.203Z","author":{"_id":"634e4670a51d5df8c2d92fce","avatarUrl":"/avatars/c52d7150b4de6a2eb2d83b345d35cbc2.svg","fullname":"Da Yin","name":"DaYin","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.38861608505249023},"editors":["DaYin"],"editorAvatarUrls":["/avatars/c52d7150b4de6a2eb2d83b345d35cbc2.svg"],"reactions":[],"isReport":false}},{"id":"68f2ef5af790cae567c5e9ea","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":318,"isUserFollowing":false},"createdAt":"2025-10-18T01:37:30.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"This is an automated message from the [Librarian Bot](https://huggingface.co/librarian-bots). I found the following papers similar to this paper. \n\nThe following papers were recommended by the Semantic Scholar API \n\n* [Scaling Synthetic Task Generation for Agents via Exploration](https://huggingface.co/papers/2509.25047) (2025)\n* [GUI-ReWalk: Massive Data Generation for GUI Agent via Stochastic Exploration and Intent-Aware Reasoning](https://huggingface.co/papers/2509.15738) (2025)\n* [Watch and Learn: Learning to Use Computers from Online Videos](https://huggingface.co/papers/2510.04673) (2025)\n* [Auto-scaling Continuous Memory for GUI Agent](https://huggingface.co/papers/2510.09038) (2025)\n* [Generalizable End-to-End Tool-Use RL with Synthetic CodeGym](https://huggingface.co/papers/2509.17325) (2025)\n* [R-WoM: Retrieval-augmented World Model For Computer-use Agents](https://huggingface.co/papers/2510.11892) (2025)\n* [MATRIX: Multimodal Agent Tuning for Robust Tool-Use Reasoning](https://huggingface.co/papers/2510.08567) (2025)\n\n\n Please give a thumbs up to this comment if you found it helpful!\n\n If you want recommendations for any Paper on Hugging Face checkout [this](https://huggingface.co/spaces/librarian-bots/recommend_similar_papers) Space\n\n You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: `@librarian-bot recommend`","html":"

This is an automated message from the Librarian Bot. I found the following papers similar to this paper.

The following papers were recommended by the Semantic Scholar API

Please give a thumbs up to this comment if you found it helpful!

If you want recommendations for any Paper on Hugging Face checkout this Space

You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: \n\n@librarian-bot\n\t recommend

\n","updatedAt":"2025-10-18T01:37:30.367Z","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":318,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.6885868906974792},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2510.14969","authors":[{"_id":"68f19da26e0bef323a68fc68","name":"Yiming Wang","hidden":false},{"_id":"68f19da26e0bef323a68fc69","name":"Da Yin","hidden":false},{"_id":"68f19da26e0bef323a68fc6a","name":"Yuedong Cui","hidden":false},{"_id":"68f19da26e0bef323a68fc6b","name":"Ruichen Zheng","hidden":false},{"_id":"68f19da26e0bef323a68fc6c","name":"Zhiqian Li","hidden":false},{"_id":"68f19da26e0bef323a68fc6d","name":"Zongyu Lin","hidden":false},{"_id":"68f19da26e0bef323a68fc6e","name":"Di Wu","hidden":false},{"_id":"68f19da26e0bef323a68fc6f","name":"Xueqing Wu","hidden":false},{"_id":"68f19da26e0bef323a68fc70","name":"Chenchen Ye","hidden":false},{"_id":"68f19da26e0bef323a68fc71","name":"Yu Zhou","hidden":false},{"_id":"68f19da26e0bef323a68fc72","name":"Kai-Wei Chang","hidden":false}],"mediaUrls":["https://cdn-uploads.huggingface.co/production/uploads/634e4670a51d5df8c2d92fce/hP00XFC26Wxs0TvDfDWas.mp4"],"publishedAt":"2025-10-16T17:59:38.000Z","submittedOnDailyAt":"2025-10-17T00:20:18.195Z","title":"LLMs as Scalable, General-Purpose Simulators For Evolving Digital Agent\n Training","submittedOnDailyBy":{"_id":"634e4670a51d5df8c2d92fce","avatarUrl":"/avatars/c52d7150b4de6a2eb2d83b345d35cbc2.svg","isPro":false,"fullname":"Da Yin","user":"DaYin","type":"user"},"summary":"Digital agents require diverse, large-scale UI trajectories to generalize\nacross real-world tasks, yet collecting such data is prohibitively expensive in\nboth human annotation, infra and engineering perspectives. To this end, we\nintroduce UI-Simulator, a scalable paradigm that generates\nstructured UI states and transitions to synthesize training trajectories at\nscale. Our paradigm integrates a digital world simulator for diverse UI states,\na guided rollout process for coherent exploration, and a trajectory wrapper\nthat produces high-quality and diverse trajectories for agent training. We\nfurther propose UI-Simulator-Grow, a targeted scaling strategy that\nenables more rapid and data-efficient scaling by prioritizing high-impact tasks\nand synthesizes informative trajectory variants. Experiments on WebArena and\nAndroidWorld show that UI-Simulator rivals or surpasses open-source agents\ntrained on real UIs with significantly better robustness, despite using weaker\nteacher models. Moreover, UI-Simulator-Grow matches the performance of\nLlama-3-70B-Instruct using only Llama-3-8B-Instruct as the base model,\nhighlighting the potential of targeted synthesis scaling paradigm to\ncontinuously and efficiently enhance the digital agents.","upvotes":8,"discussionId":"68f19da26e0bef323a68fc73","projectPage":"https://ui-simulator.notion.site/llms-as-scalable-digital-world-simulator","githubRepo":"https://github.com/WadeYin9712/UI-Simulator","githubRepoAddedBy":"auto","ai_summary":"UI-Simulator generates diverse UI trajectories for digital agents using a scalable paradigm, improving robustness and performance with targeted scaling strategies.","ai_keywords":["UI-Simulator","digital world simulator","guided rollout process","trajectory wrapper","UI-Simulator-Grow","WebArena","AndroidWorld","Llama-3-70B-Instruct","Llama-3-8B-Instruct"],"githubStars":19,"organization":{"_id":"60b492a78e9589ce25930346","name":"uclanlp","fullname":"UCLA NLP","avatar":"https://cdn-uploads.huggingface.co/production/uploads/1622446751930-5ec0135ded25d76864d553f1.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"634e4670a51d5df8c2d92fce","avatarUrl":"/avatars/c52d7150b4de6a2eb2d83b345d35cbc2.svg","isPro":false,"fullname":"Da Yin","user":"DaYin","type":"user"},{"_id":"6270324ebecab9e2dcf245de","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6270324ebecab9e2dcf245de/cMbtWSasyNlYc9hvsEEzt.jpeg","isPro":false,"fullname":"Kye Gomez","user":"kye","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":"684d57f26e04c265777ead3f","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/cuOj-bQqukSZreXgUJlfm.png","isPro":false,"fullname":"Joakim Lee","user":"Reinforcement4All","type":"user"},{"_id":"61d53df2062444ea769d3b79","avatarUrl":"/avatars/fa771202368b6b2626a8fdf1c4369239.svg","isPro":false,"fullname":"Ming Zhong","user":"MingZhong","type":"user"},{"_id":"64ba6e227789c0a80d740969","avatarUrl":"/avatars/4bf1f59121741713461a53514b4e85fe.svg","isPro":false,"fullname":"Xueqing Wu","user":"xqwu","type":"user"},{"_id":"65025370b6595dc45c397340","avatarUrl":"/avatars/9469599b176034548042922c0afa7051.svg","isPro":false,"fullname":"J C","user":"dark-pen","type":"user"},{"_id":"686db5d4af2b856fabbf13aa","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/6BjMv2LVNoqvbX8fQSTPI.png","isPro":false,"fullname":"V bbbb","user":"Bbbbbnnn","type":"user"}],"acceptLanguages":["*"],"dailyPaperRank":0,"organization":{"_id":"60b492a78e9589ce25930346","name":"uclanlp","fullname":"UCLA NLP","avatar":"https://cdn-uploads.huggingface.co/production/uploads/1622446751930-5ec0135ded25d76864d553f1.png"}}">

Papers

arxiv:2510.14969

LLMs as Scalable, General-Purpose Simulators For Evolving Digital Agent Training

Published on Oct 16, 2025

· Submitted by

Da Yin on Oct 17, 2025

UCLA NLP

Upvote

Authors:

Abstract

UI-Simulator generates diverse UI trajectories for digital agents using a scalable paradigm, improving robustness and performance with targeted scaling strategies.

AI-generated summary

Digital agents require diverse, large-scale UI trajectories to generalize across real-world tasks, yet collecting such data is prohibitively expensive in both human annotation, infra and engineering perspectives. To this end, we introduce UI-Simulator, a scalable paradigm that generates structured UI states and transitions to synthesize training trajectories at scale. Our paradigm integrates a digital world simulator for diverse UI states, a guided rollout process for coherent exploration, and a trajectory wrapper that produces high-quality and diverse trajectories for agent training. We further propose UI-Simulator-Grow, a targeted scaling strategy that enables more rapid and data-efficient scaling by prioritizing high-impact tasks and synthesizes informative trajectory variants. Experiments on WebArena and AndroidWorld show that UI-Simulator rivals or surpasses open-source agents trained on real UIs with significantly better robustness, despite using weaker teacher models. Moreover, UI-Simulator-Grow matches the performance of Llama-3-70B-Instruct using only Llama-3-8B-Instruct as the base model, highlighting the potential of targeted synthesis scaling paradigm to continuously and efficiently enhance the digital agents.