Hugging Face Daily Papers · · 4 min read

Recovering Policy-Induced Errors: Benchmarking and Trajectory Synthesis for Robust GUI Agents

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

We observe that existing GUI agent benchmarks focus on overall success rates or robustness to external perturbations, while real-world failures are predominantly caused by the agent's own policy — and current training data significantly underrepresents planning and progress-perception errors that dominate real-world failures. Motivated by this gap, we propose GUI-RobustEval, a fine-grained benchmark for measuring error awareness and recovery across 11 error types and 4 error depths, and RoTS, a tree-based online synthesis framework that generates 800K error-recovery trajectories by actively exploring failure modes and synthesizing long-horizon recovery data. RoTS-trained models achieve strong performance on OSWorld with substantially smaller degradation under compounding errors compared to existing baselines, demonstrating improved robustness to long-horizon policy-induced failures.</p>\n","updatedAt":"2026-06-01T06:10:40.330Z","author":{"_id":"651390af641b14c330ef85dd","avatarUrl":"/avatars/1e6e1a5dbb16a1d4e02a08e82456f7fa.svg","fullname":"Tianpeng Bu","name":"smallnono","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8694667220115662},"editors":["smallnono"],"editorAvatarUrls":["/avatars/1e6e1a5dbb16a1d4e02a08e82456f7fa.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2605.29447","authors":[{"_id":"6a18ecf756b4bb14ec65cddb","user":{"_id":"651390af641b14c330ef85dd","avatarUrl":"/avatars/1e6e1a5dbb16a1d4e02a08e82456f7fa.svg","isPro":false,"fullname":"Tianpeng Bu","user":"smallnono","type":"user","name":"smallnono"},"name":"Tianpeng Bu","status":"claimed_verified","statusLastChangedAt":"2026-05-29T08:51:28.520Z","hidden":false},{"_id":"6a18ecf756b4bb14ec65cddc","user":{"_id":"687768739ce4d7aff04ea968","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/9ys-kzPAWIb5GghP0tkuA.png","isPro":false,"fullname":"LiuXin","user":"444515liuxin","type":"user","name":"444515liuxin"},"name":"Xin Liu","status":"claimed_verified","statusLastChangedAt":"2026-05-29T08:51:24.363Z","hidden":false},{"_id":"6a18ecf756b4bb14ec65cddd","name":"Qihua Chen","hidden":false},{"_id":"6a18ecf756b4bb14ec65cdde","name":"Hao Jiang","hidden":false},{"_id":"6a18ecf756b4bb14ec65cddf","name":"Shurui Li","hidden":false},{"_id":"6a18ecf756b4bb14ec65cde0","name":"Hongtao Duan","hidden":false},{"_id":"6a18ecf756b4bb14ec65cde1","name":"Lu Jiang","hidden":false},{"_id":"6a18ecf756b4bb14ec65cde2","name":"Lulu Hu","hidden":false},{"_id":"6a18ecf756b4bb14ec65cde3","name":"Bin Yang","hidden":false},{"_id":"6a18ecf756b4bb14ec65cde4","name":"Minying Zhang","hidden":false}],"publishedAt":"2026-05-28T00:00:00.000Z","submittedOnDailyAt":"2026-06-01T00:00:00.000Z","title":"Recovering Policy-Induced Errors: Benchmarking and Trajectory Synthesis for Robust GUI Agents","submittedOnDailyBy":{"_id":"651390af641b14c330ef85dd","avatarUrl":"/avatars/1e6e1a5dbb16a1d4e02a08e82456f7fa.svg","isPro":false,"fullname":"Tianpeng Bu","user":"smallnono","type":"user","name":"smallnono"},"summary":"While GUI agents have advanced rapidly, they often lack the robustness to recover from their own errors, hindering real-world deployment. To bridge this gap at both the evaluation and data levels, we introduce GUI-RobustEval and propose Robustness-driven Trajectory Synthesis. GUI-RobustEval contains 1,216 executable test cases that systematically measure error recovery capabilities across a broad and realistic spectrum of error modes. At the data level, RoTS is a scalable synthesis framework that creates 800k high-quality data via a tree-based pipeline that proactively discovers diverse error modes and synthesizes corresponding recovery steps. Our two models, RoTS-7B and RoTS-32B, fine-tuned on our dataset, both demonstrate significant gains on GUI-RobustEval and traditional GUI benchmarks. Notably, RoTS-32B achieves state-of-the-art performance on OSWorld, with a 47.4% success rate and a 33.8% All-Pass@4 score, suggesting that improved long-horizon error recovery ability contributes to both robustness and overall performance. Our code is available at https://github.com/AlibabaResearch/RoTS.","upvotes":14,"discussionId":"6a18ecf756b4bb14ec65cde5","githubRepo":"https://github.com/AlibabaResearch/RoTS","githubRepoAddedBy":"user","ai_summary":"GUI agents lack robust error recovery capabilities, which this work addresses through GUI-RobustEval and Robustness-driven Trajectory Synthesis, demonstrating improved performance on real-world benchmarks.","ai_keywords":["GUI agents","error recovery","GUI-RobustEval","Robustness-driven Trajectory Synthesis","OSWorld","All-Pass@4"],"githubStars":4},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"687768739ce4d7aff04ea968","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/9ys-kzPAWIb5GghP0tkuA.png","isPro":false,"fullname":"LiuXin","user":"444515liuxin","type":"user"},{"_id":"65324f49453e0c5fb6ca6837","avatarUrl":"/avatars/af82e193ccd5699303f23203865cb92e.svg","isPro":false,"fullname":"WuShaoyu","user":"Saoyu","type":"user"},{"_id":"6821b167c90d45ee4130c299","avatarUrl":"/avatars/547ffe1050d00ea253f83f157318c13d.svg","isPro":false,"fullname":"jiashuo zhang","user":"jiajiaShuo","type":"user"},{"_id":"670fcb3ffe84ac0ce43a8507","avatarUrl":"/avatars/6c6e7cf19b0b71f32e7dae8f1521976b.svg","isPro":false,"fullname":"Jifang Wang","user":"PigCatchingExpert","type":"user"},{"_id":"666feeb511d9dadd24cf1166","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/666feeb511d9dadd24cf1166/DgC4LoG_SX4Ckn60QAFr6.jpeg","isPro":false,"fullname":"Hao Jiang","user":"Lutalica","type":"user"},{"_id":"65ae052786f88a686b01820e","avatarUrl":"/avatars/2d49e980057d6aa364bbdd5d0058910c.svg","isPro":false,"fullname":"haozhou","user":"hao36918","type":"user"},{"_id":"69b6bedb14f287b912c1c7ee","avatarUrl":"/avatars/07cf89d9bdeb402876d1a8c8ee078de3.svg","isPro":false,"fullname":"Yang Yong","user":"yangyong1234","type":"user"},{"_id":"66297e6ddb8773b8d2896204","avatarUrl":"/avatars/3a68656f691f439970efe68621f7e2e0.svg","isPro":false,"fullname":"SRL","user":"APterosaur","type":"user"},{"_id":"67767c85829ba1c6cb2eaaa7","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/O-UIRXJeJYhNMEZQl7soO.png","isPro":false,"fullname":"lee","user":"coco0119","type":"user"},{"_id":"651390af641b14c330ef85dd","avatarUrl":"/avatars/1e6e1a5dbb16a1d4e02a08e82456f7fa.svg","isPro":false,"fullname":"Tianpeng Bu","user":"smallnono","type":"user"},{"_id":"6391e0a484afa726d66106cc","avatarUrl":"/avatars/ab39ba83bc9ce0da387b8d5026d5742c.svg","isPro":false,"fullname":"plmsmile","user":"plmsmile","type":"user"},{"_id":"6a17d75d90e20f5029f879b2","avatarUrl":"/avatars/e7c8aab0a5eecfa791635e76bea3aac7.svg","isPro":false,"fullname":"pupudi","user":"pupudi","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2605/2605.29447.md"}">
Papers
arxiv:2605.29447

Recovering Policy-Induced Errors: Benchmarking and Trajectory Synthesis for Robust GUI Agents

Published on May 28
· Submitted by
Tianpeng Bu
on Jun 1
Authors:
,
,
,
,
,
,
,

Abstract

GUI agents lack robust error recovery capabilities, which this work addresses through GUI-RobustEval and Robustness-driven Trajectory Synthesis, demonstrating improved performance on real-world benchmarks.

AI-generated summary

While GUI agents have advanced rapidly, they often lack the robustness to recover from their own errors, hindering real-world deployment. To bridge this gap at both the evaluation and data levels, we introduce GUI-RobustEval and propose Robustness-driven Trajectory Synthesis. GUI-RobustEval contains 1,216 executable test cases that systematically measure error recovery capabilities across a broad and realistic spectrum of error modes. At the data level, RoTS is a scalable synthesis framework that creates 800k high-quality data via a tree-based pipeline that proactively discovers diverse error modes and synthesizes corresponding recovery steps. Our two models, RoTS-7B and RoTS-32B, fine-tuned on our dataset, both demonstrate significant gains on GUI-RobustEval and traditional GUI benchmarks. Notably, RoTS-32B achieves state-of-the-art performance on OSWorld, with a 47.4% success rate and a 33.8% All-Pass@4 score, suggesting that improved long-horizon error recovery ability contributes to both robustness and overall performance. Our code is available at https://github.com/AlibabaResearch/RoTS.

Community

Paper author Paper submitter about 5 hours ago

We observe that existing GUI agent benchmarks focus on overall success rates or robustness to external perturbations, while real-world failures are predominantly caused by the agent's own policy — and current training data significantly underrepresents planning and progress-perception errors that dominate real-world failures. Motivated by this gap, we propose GUI-RobustEval, a fine-grained benchmark for measuring error awareness and recovery across 11 error types and 4 error depths, and RoTS, a tree-based online synthesis framework that generates 800K error-recovery trajectories by actively exploring failure modes and synthesizing long-horizon recovery data. RoTS-trained models achieve strong performance on OSWorld with substantially smaller degradation under compounding errors compared to existing baselines, demonstrating improved robustness to long-horizon policy-induced failures.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images

· Sign up or log in to comment

Get this paper in your agent:

hf papers read 2605.29447
Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash

Models citing this paper 0

No model linking this paper

Cite arxiv.org/abs/2605.29447 in a model README.md to link it from this page.

Datasets citing this paper 0

No dataset linking this paper

Cite arxiv.org/abs/2605.29447 in a dataset README.md to link it from this page.

Spaces citing this paper 0

No Space linking this paper

Cite arxiv.org/abs/2605.29447 in a Space README.md to link it from this page.

Collections including this paper 1

Discussion (0)

Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.

Sign in →

No comments yet. Sign in and be the first to say something.

More from Hugging Face Daily Papers