Librarian Bot. I found the following papers similar to this paper.

The following papers were recommended by the Semantic Scholar API

CLAP: Contrastive Latent Action Pretraining for Learning Vision-Language-Action Models from Human Videos (2026)
Robotic VLA Benefits from Joint Learning with Motion Image Diffusion (2025)
ConLA: Contrastive Latent Action Learning from Human Videos for Robotic Manipulation (2026)
Latent Reasoning VLA: Latent Thinking and Prediction for Vision-Language-Action Models (2026)
Motus: A Unified Latent Action World Model (2025)
MVP-LAM: Learning Action-Centric Latent Action via Cross-Viewpoint Reconstruction (2026)
Spatial-Aware VLA Pretraining through Visual-Physical Alignment from Human Videos (2025)

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":"2026-02-12T01:44:09.220Z","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":0.692431628704071},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2602.10098","authors":[{"_id":"698bf4886052d3bed9630a8b","name":"Jingwen Sun","hidden":false},{"_id":"698bf4886052d3bed9630a8c","name":"Wenyao Zhang","hidden":false},{"_id":"698bf4886052d3bed9630a8d","name":"Zekun Qi","hidden":false},{"_id":"698bf4886052d3bed9630a8e","name":"Shaojie Ren","hidden":false},{"_id":"698bf4886052d3bed9630a8f","name":"Zezhi Liu","hidden":false},{"_id":"698bf4886052d3bed9630a90","name":"Hanxin Zhu","hidden":false},{"_id":"698bf4886052d3bed9630a91","name":"Guangzhong Sun","hidden":false},{"_id":"698bf4886052d3bed9630a92","name":"Xin Jin","hidden":false},{"_id":"698bf4886052d3bed9630a93","name":"Zhibo Chen","hidden":false}],"publishedAt":"2026-02-10T18:58:01.000Z","submittedOnDailyAt":"2026-02-11T01:04:16.621Z","title":"VLA-JEPA: Enhancing Vision-Language-Action Model with Latent World Model","submittedOnDailyBy":{"_id":"6039478ab3ecf716b1a5fd4d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6039478ab3ecf716b1a5fd4d/_Thy4E7taiSYBLKxEKJbT.jpeg","isPro":true,"fullname":"taesiri","user":"taesiri","type":"user"},"summary":"Pretraining Vision-Language-Action (VLA) policies on internet-scale video is appealing, yet current latent-action objectives often learn the wrong thing: they remain anchored to pixel variation rather than action-relevant state transitions, making them vulnerable to appearance bias, nuisance motion, and information leakage. We introduce VLA-JEPA, a JEPA-style pretraining framework that sidesteps these pitfalls by design. The key idea is leakage-free state prediction: a target encoder produces latent representations from future frames, while the student pathway sees only the current observation -- future information is used solely as supervision targets, never as input. By predicting in latent space rather than pixel space, VLA-JEPA learns dynamics abstractions that are robust to camera motion and irrelevant background changes. This yields a simple two-stage recipe -- JEPA pretraining followed by action-head fine-tuning -- without the multi-stage complexity of prior latent-action pipelines. Experiments on LIBERO, LIBERO-Plus, SimplerEnv and real-world manipulation tasks show that VLA-JEPA achieves consistent gains in generalization and robustness over existing methods.","upvotes":18,"discussionId":"698bf4896052d3bed9630a94","projectPage":"https://ginwind.github.io/VLA-JEPA/","githubRepo":"https://github.com/ginwind/VLA-JEPA","githubRepoAddedBy":"user","ai_summary":"VLA-JEPA is a JEPA-style pretraining framework that improves vision-language-action policy learning by using leakage-free state prediction in latent space, enhancing generalization and robustness in manipulation tasks.","ai_keywords":["Vision-Language-Action","JEPA","latent-action objectives","pixel variation","action-relevant state transitions","appearance bias","nuisance motion","information leakage","target encoder","student pathway","latent representations","future frames","current observation","latent space","dynamics abstractions","camera motion","background changes","JEPA pretraining","action-head fine-tuning","generalization","robustness"],"githubStars":50},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"620783f24e28382272337ba4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/620783f24e28382272337ba4/zkUveQPNiDfYjgGhuFErj.jpeg","isPro":false,"fullname":"GuoLiangTang","user":"Tommy930","type":"user"},{"_id":"6407e5294edf9f5c4fd32228","avatarUrl":"/avatars/8e2d55460e9fe9c426eb552baf4b2cb0.svg","isPro":false,"fullname":"Stoney Kang","user":"sikang99","type":"user"},{"_id":"663b8ee3d6acd1aa1223eb09","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/663b8ee3d6acd1aa1223eb09/IshSiAz2hZ5gY2iILRCHY.jpeg","isPro":false,"fullname":"Sasha","user":"Catlilface","type":"user"},{"_id":"63c3e8abc7d7f4c63a515a02","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63c3e8abc7d7f4c63a515a02/npMHnVP2hHLbvoUGe7C4O.jpeg","isPro":false,"fullname":"Zekun Qi","user":"qizekun","type":"user"},{"_id":"65f9533b136fb8ddbd14e1fa","avatarUrl":"/avatars/d88f75da0448093ccd1babba2a37d73f.svg","isPro":false,"fullname":"Zhang","user":"WenyaoZhang","type":"user"},{"_id":"683ff9810278714998705654","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/xhvcLacOLbx27WiHY0BZf.png","isPro":false,"fullname":"chenxin ding","user":"chenxin-ding","type":"user"},{"_id":"683ff8138c64413c0d7bfbba","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/ZmBD-cyEuag_Y6JHrd5Il.png","isPro":false,"fullname":"Xukun Cai","user":"ikun-kun","type":"user"},{"_id":"66839c7f7b0920f40d7ff3e0","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/66839c7f7b0920f40d7ff3e0/nuOh5acjmb9QZK3wNQxt1.jpeg","isPro":false,"fullname":"Dairu Liu","user":"dairuliu","type":"user"},{"_id":"683ffeb9967d02230641c8ad","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/z2WZJkllrCKTCliLEMrr-.png","isPro":false,"fullname":"linkai wang","user":"linkai-wang","type":"user"},{"_id":"683ff8b1f9f5cf052b4db216","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/ugYm9NE_FIGM2Wgp98UQ7.png","isPro":false,"fullname":"jiaqima","user":"jiaqi-ma","type":"user"},{"_id":"68cb1b1ff283dbb922dc8e9c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/zylXL3XGKy5E8tISylviX.png","isPro":false,"fullname":"Yangbowen","user":"Bowen-Yang","type":"user"},{"_id":"65c20ee58aedd6edd2b89000","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/65c20ee58aedd6edd2b89000/LtS4YTbmxiCFqHSGHfdC8.png","isPro":false,"fullname":"Chmielewski","user":"Eryk-Chmielewski","type":"user"}],"acceptLanguages":["*"],"dailyPaperRank":0}">

Papers

arxiv:2602.10098

VLA-JEPA: Enhancing Vision-Language-Action Model with Latent World Model

Published on Feb 10

· Submitted by

taesiri on Feb 11

Upvote

Authors:

Abstract

VLA-JEPA is a JEPA-style pretraining framework that improves vision-language-action policy learning by using leakage-free state prediction in latent space, enhancing generalization and robustness in manipulation tasks.

AI-generated summary

Pretraining Vision-Language-Action (VLA) policies on internet-scale video is appealing, yet current latent-action objectives often learn the wrong thing: they remain anchored to pixel variation rather than action-relevant state transitions, making them vulnerable to appearance bias, nuisance motion, and information leakage. We introduce VLA-JEPA, a JEPA-style pretraining framework that sidesteps these pitfalls by design. The key idea is leakage-free state prediction: a target encoder produces latent representations from future frames, while the student pathway sees only the current observation -- future information is used solely as supervision targets, never as input. By predicting in latent space rather than pixel space, VLA-JEPA learns dynamics abstractions that are robust to camera motion and irrelevant background changes. This yields a simple two-stage recipe -- JEPA pretraining followed by action-head fine-tuning -- without the multi-stage complexity of prior latent-action pipelines. Experiments on LIBERO, LIBERO-Plus, SimplerEnv and real-world manipulation tasks show that VLA-JEPA achieves consistent gains in generalization and robustness over existing methods.