Current world models face a fundamental tension: faithful long-horizon simulation demands deep computation, but deeper models are expensive to deploy and prone to compounding errors. We resolve this by introducing Looped World Models (LoopWM), which are the first looped architectures for world modelling. Our method iteratively refines latent environment states through a parameter-shared transformer block. This yield up to 100× parameter efficiency over conventional approaches with adaptive computation that automatically scales depth to match the complexity of each prediction step. Orthogonal to scaling model size and training data, LoopWM establishes iterative latent depth as a new scaling axis for world simulation, which might significantly push the community forward.</p>\n","updatedAt":"2026-06-17T01:57:00.124Z","author":{"_id":"645b949b4093d5c140e12849","avatarUrl":"/avatars/b296f072f2578b846517517645bf9267.svg","fullname":"lu","name":"hongyyyyy","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8635534048080444},"editors":["hongyyyyy"],"editorAvatarUrls":["/avatars/b296f072f2578b846517517645bf9267.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2606.18208","authors":[{"_id":"6a31fea4bc818ff14e453d0c","user":{"_id":"645b949b4093d5c140e12849","avatarUrl":"/avatars/b296f072f2578b846517517645bf9267.svg","isPro":false,"fullname":"lu","user":"hongyyyyy","type":"user","name":"hongyyyyy"},"name":"Hongyuan Adam Lu","status":"claimed_verified","statusLastChangedAt":"2026-06-17T11:21:17.185Z","hidden":false},{"_id":"6a31fea4bc818ff14e453d0d","name":"Z. L. Victor Wei","hidden":false},{"_id":"6a31fea4bc818ff14e453d0e","name":"Qun Zhang","hidden":false},{"_id":"6a31fea4bc818ff14e453d0f","name":"Jinrui Zeng","hidden":false},{"_id":"6a31fea4bc818ff14e453d10","name":"Bowen Cao","hidden":false},{"_id":"6a31fea4bc818ff14e453d11","name":"Lingwei Meng","hidden":false},{"_id":"6a31fea4bc818ff14e453d12","name":"Mocheng Li","hidden":false},{"_id":"6a31fea4bc818ff14e453d13","name":"Zezhong Wang","hidden":false},{"_id":"6a31fea4bc818ff14e453d14","name":"Haonan Yin","hidden":false},{"_id":"6a31fea4bc818ff14e453d15","name":"Naifu Xue","hidden":false},{"_id":"6a31fea4bc818ff14e453d16","name":"Minyu Chen","hidden":false},{"_id":"6a31fea4bc818ff14e453d17","name":"Cenyuan Zhang","hidden":false},{"_id":"6a31fea4bc818ff14e453d18","name":"Zefan Zhang","hidden":false},{"_id":"6a31fea4bc818ff14e453d19","name":"Hao Wei","hidden":false},{"_id":"6a31fea4bc818ff14e453d1a","name":"Jiawei Zhou","hidden":false},{"_id":"6a31fea4bc818ff14e453d1b","name":"Haoran Xu","hidden":false},{"_id":"6a31fea4bc818ff14e453d1c","name":"Hao Yang","hidden":false},{"_id":"6a31fea4bc818ff14e453d1d","name":"Ronglai Zuo","hidden":false},{"_id":"6a31fea4bc818ff14e453d1e","name":"Tongda Xu","hidden":false},{"_id":"6a31fea4bc818ff14e453d1f","name":"Yonghao Li","hidden":false},{"_id":"6a31fea4bc818ff14e453d20","name":"Jian Chen","hidden":false},{"_id":"6a31fea4bc818ff14e453d21","name":"Hebin Wang","hidden":false},{"_id":"6a31fea4bc818ff14e453d22","name":"Zeyu Gao","hidden":false},{"_id":"6a31fea4bc818ff14e453d23","name":"Yang Li","hidden":false},{"_id":"6a31fea4bc818ff14e453d24","name":"Wei Zhao","hidden":false},{"_id":"6a31fea4bc818ff14e453d25","name":"Qimin Zhong","hidden":false},{"_id":"6a31fea4bc818ff14e453d26","name":"Siqi Liu","hidden":false},{"_id":"6a31fea4bc818ff14e453d27","name":"Yumeng Zhang","hidden":false},{"_id":"6a31fea4bc818ff14e453d28","name":"Leyan Cui","hidden":false},{"_id":"6a31fea4bc818ff14e453d29","name":"Zhangyu Wang","hidden":false},{"_id":"6a31fea4bc818ff14e453d2a","name":"Wai Lam","hidden":false}],"publishedAt":"2026-06-16T00:00:00.000Z","submittedOnDailyAt":"2026-06-17T00:00:00.000Z","title":"Looped World Models","submittedOnDailyBy":{"_id":"645b949b4093d5c140e12849","avatarUrl":"/avatars/b296f072f2578b846517517645bf9267.svg","isPro":false,"fullname":"lu","user":"hongyyyyy","type":"user","name":"hongyyyyy"},"summary":"Current world models face a fundamental tension: faithful long-horizon simulation demands deep computation, but deeper models are expensive to deploy and prone to compounding errors. We resolve this by introducing Looped World Models (LoopWM), which are the first looped architectures for world modelling. Our method iteratively refines latent environment states through a parameter-shared transformer block. This yield up to 100x parameter efficiency over conventional approaches with adaptive computation that automatically scales depth to match the complexity of each prediction step. Orthogonal to scaling model size and training data, LoopWM establishes iterative latent depth as a new scaling axis for world simulation, which might significantly push the community forward.","upvotes":8,"discussionId":"6a31fea4bc818ff14e453d2b","ai_summary":"Looped World Models introduce iterative latent state refinement through shared transformer blocks, achieving 100x parameter efficiency while adapting computational depth to prediction complexity.","ai_keywords":["world models","looped architectures","latent environment states","parameter-shared transformer block","adaptive computation","iterative refinement","computational efficiency"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","organization":{"_id":"69d4cb0e4de7a919dd8dd1df","name":"FaceMind","fullname":"FaceMind","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/645b949b4093d5c140e12849/aRqbb8VStMQ6LZzfAJaqI.jpeg"}},"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":"60c94c629cacafb192d805fc","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60c94c629cacafb192d805fc/nA0jBaVl7Ti5LGoZ2myAs.png","isPro":false,"fullname":"TimeLordRaps","user":"TimeLordRaps","type":"user"},{"_id":"677272184d148b904333e874","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/5dUau7gxLk4Wm1TiiJJri.jpeg","isPro":false,"fullname":"Efstathios Karypidis","user":"Sta8is","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":"60c8d264224e250fb0178f77","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/60c8d264224e250fb0178f77/i8fbkBVcoFeJRmkQ9kYAE.png","isPro":false,"fullname":"Adam Lee","user":"Abecid","type":"user"},{"_id":"645b949b4093d5c140e12849","avatarUrl":"/avatars/b296f072f2578b846517517645bf9267.svg","isPro":false,"fullname":"lu","user":"hongyyyyy","type":"user"},{"_id":"65995de7f82b15d663011d3c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/65995de7f82b15d663011d3c/S2Vw3G2QTkurEOUDL6ArW.png","isPro":false,"fullname":"Baalateja Kataru","user":"bkataru","type":"user"},{"_id":"683b8a28e1526aee261ac442","avatarUrl":"/avatars/3e24a17422d1fd8aa3fae5fc209c32aa.svg","isPro":false,"fullname":"Jared Leonard","user":"jmaleonard","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"69d4cb0e4de7a919dd8dd1df","name":"FaceMind","fullname":"FaceMind","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/645b949b4093d5c140e12849/aRqbb8VStMQ6LZzfAJaqI.jpeg"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2606/2606.18208.md","query":{}}">
Looped World Models
Published on Jun 16
· Submitted by lu on Jun 17 Authors: ,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,
Abstract
Looped World Models introduce iterative latent state refinement through shared transformer blocks, achieving 100x parameter efficiency while adapting computational depth to prediction complexity.
Current world models face a fundamental tension: faithful long-horizon simulation demands deep computation, but deeper models are expensive to deploy and prone to compounding errors. We resolve this by introducing Looped World Models (LoopWM), which are the first looped architectures for world modelling. Our method iteratively refines latent environment states through a parameter-shared transformer block. This yield up to 100x parameter efficiency over conventional approaches with adaptive computation that automatically scales depth to match the complexity of each prediction step. Orthogonal to scaling model size and training data, LoopWM establishes iterative latent depth as a new scaling axis for world simulation, which might significantly push the community forward.
Community
Current world models face a fundamental tension: faithful long-horizon simulation demands deep computation, but deeper models are expensive to deploy and prone to compounding errors. We resolve this by introducing Looped World Models (LoopWM), which are the first looped architectures for world modelling. Our method iteratively refines latent environment states through a parameter-shared transformer block. This yield up to 100× parameter efficiency over conventional approaches with adaptive computation that automatically scales depth to match the complexity of each prediction step. Orthogonal to scaling model size and training data, LoopWM establishes iterative latent depth as a new scaling axis for world simulation, which might significantly push the community forward.
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Cite arxiv.org/abs/2606.18208 in a model README.md to link it from this page.
Cite arxiv.org/abs/2606.18208 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2606.18208 in a Space README.md to link it from this page.
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.