Hugging Face Daily Papers · July 7, 2026 · 4 min read

GigaWorld-1: A Roadmap to Build World Models for Robot Policy Evaluation

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

Like Read original ↗

Code：<a href=\"https://github.com/open-gigaai/giga-world-1\" rel=\"nofollow\">https://github.com/open-gigaai/giga-world-1</a><br>Model：<a href=\"https://huggingface.co/open-gigaai/Giga-World-1\">https://huggingface.co/open-gigaai/Giga-World-1</a><br>Data：<a href=\"https://huggingface.co/datasets/open-gigaai/CVPR-2026-WorldModel-Track-Dataset\">https://huggingface.co/datasets/open-gigaai/CVPR-2026-WorldModel-Track-Dataset</a><br>Benchmark：<a href=\"https://huggingface.co/spaces/open-gigaai/CVPR-2026-WorldModel-Track-LeaderBoard\">https://huggingface.co/spaces/open-gigaai/CVPR-2026-WorldModel-Track-LeaderBoard</a></p>\n","updatedAt":"2026-07-07T02:44:55.893Z","author":{"_id":"6426616ea5ec4a5cbc535634","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6426616ea5ec4a5cbc535634/5IfSFYd9QOxz8K9QmBCst.png","fullname":"JeffWang","name":"Jeff-Wang","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":6,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.2569692134857178},"editors":["Jeff-Wang"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/6426616ea5ec4a5cbc535634/5IfSFYd9QOxz8K9QmBCst.png"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2607.02642","authors":[{"_id":"6a4c64d625849b193a834046","name":"GigaWorld Team","hidden":false},{"_id":"6a4c64d625849b193a834047","user":{"_id":"64f020a42a7a94b68c0df99c","avatarUrl":"/avatars/39315929c3675039f9a7662e7b60ad04.svg","isPro":false,"fullname":"maangyuan","user":"maay21","type":"user","name":"maay21"},"name":"Angyuan Ma","status":"claimed_verified","statusLastChangedAt":"2026-07-07T12:12:09.489Z","hidden":false},{"_id":"6a4c64d625849b193a834048","name":"Boyuan Wang","hidden":false},{"_id":"6a4c64d625849b193a834049","name":"Bohan Li","hidden":false},{"_id":"6a4c64d625849b193a83404a","name":"Chaojun Ni","hidden":false},{"_id":"6a4c64d625849b193a83404b","name":"Guo Li","hidden":false},{"_id":"6a4c64d625849b193a83404c","name":"Guan Huang","hidden":false},{"_id":"6a4c64d625849b193a83404d","name":"Guosheng Zhao","hidden":false},{"_id":"6a4c64d625849b193a83404e","name":"Hao Li","hidden":false},{"_id":"6a4c64d625849b193a83404f","name":"Hengtao Li","hidden":false},{"_id":"6a4c64d625849b193a834050","name":"Jingyu Liu","hidden":false},{"_id":"6a4c64d625849b193a834051","name":"Jiwen Lu","hidden":false},{"_id":"6a4c64d625849b193a834052","name":"Qiuping Deng","hidden":false},{"_id":"6a4c64d625849b193a834053","name":"Tingdong Yu","hidden":false},{"_id":"6a4c64d625849b193a834054","name":"Xuancheng Xu","hidden":false},{"_id":"6a4c64d625849b193a834055","name":"Xinyu Zhou","hidden":false},{"_id":"6a4c64d625849b193a834056","name":"Xiuwei Xu","hidden":false},{"_id":"6a4c64d625849b193a834057","name":"Xinze Chen","hidden":false},{"_id":"6a4c64d625849b193a834058","user":{"_id":"6426616ea5ec4a5cbc535634","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6426616ea5ec4a5cbc535634/5IfSFYd9QOxz8K9QmBCst.png","isPro":false,"fullname":"JeffWang","user":"Jeff-Wang","type":"user","name":"Jeff-Wang"},"name":"Xiaofeng Wang","status":"claimed_verified","statusLastChangedAt":"2026-07-07T12:12:13.375Z","hidden":false},{"_id":"6a4c64d625849b193a834059","name":"Xiaoyu Tian","hidden":false},{"_id":"6a4c64d625849b193a83405a","user":{"_id":"644012cf3e0374802e174f7c","avatarUrl":"/avatars/0f4a4bd6f96ce193871843e1d01439e8.svg","isPro":false,"fullname":"Yang Wang","user":"supermodelteam","type":"user","name":"supermodelteam"},"name":"Yang Wang","status":"claimed_verified","statusLastChangedAt":"2026-07-07T12:12:11.475Z","hidden":false},{"_id":"6a4c64d625849b193a83405b","name":"Yifan Chang","hidden":false},{"_id":"6a4c64d625849b193a83405c","name":"Yukun Zhou","hidden":false},{"_id":"6a4c64d625849b193a83405d","name":"Yun Ye","hidden":false},{"_id":"6a4c64d625849b193a83405e","name":"Zhenyu Wu","hidden":false},{"_id":"6a4c64d625849b193a83405f","name":"Zhanqian Wu","hidden":false},{"_id":"6a4c64d625849b193a834060","name":"Zheng Zhu","hidden":false}],"publishedAt":"2026-07-02T00:00:00.000Z","submittedOnDailyAt":"2026-07-07T00:00:00.000Z","title":"GigaWorld-1: A Roadmap to Build World Models for Robot Policy Evaluation","submittedOnDailyBy":{"_id":"6426616ea5ec4a5cbc535634","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6426616ea5ec4a5cbc535634/5IfSFYd9QOxz8K9QmBCst.png","isPro":false,"fullname":"JeffWang","user":"Jeff-Wang","type":"user","name":"Jeff-Wang"},"summary":"Evaluating embodied robot foundation models remains a critical bottleneck; unlike large language models efficiently assessed via digital benchmarks, robotic policies require slow, costly real-world rollouts limited by hardware and human supervision, which has driven interest in world models as surrogate policy evaluators, yet the key properties that make a world model reliable for policy assessment remain poorly understood. This work presents a systematic study of world models for robotic policy evaluation and introduces WMBench, a benchmark constructed from real-robot teleoperation data and matched policy rollouts covering diverse manipulation tasks to enable controlled comparisons across model families, action encodings, rollout horizons, and evaluation metrics. Using WMBench, we analyze 7 video world models, 4 action representation schemes, and over 324,000 simulated policy rollouts paired with real robot executions, further enriching our analysis with large-scale community submissions from the CVPR 2026 GigaBrain Challenge, curated synthetic trajectories, and a training videos spanning more than 12,000 hours. Our experiments deliver three core insights: evaluator quality is dominated by long-horizon, action-faithful rollout consistency rather than short-term visual realism; pretraining gains stem not only from data scale but from balancing general world knowledge with robot-specific controllability; and architectural choices including action encoding, memory design, and evaluator-focused post-training strongly determine alignment with real-world robot behavior. Drawing on these results, we derive a practical design roadmap and realize it in GigaWorld-1, a world model specially optimized for policy evaluation, and we fully release our code, models, datasets, and toolkits to advance scalable evaluation research for embodied foundation models.","upvotes":31,"discussionId":"6a4c64d625849b193a834061","projectPage":"https://open-gigaai.github.io/giga-world-1/","githubRepo":"https://github.com/open-gigaai/giga-world-1","githubRepoAddedBy":"user","ai_summary":"World models for robotic policy evaluation are systematically studied through a new benchmark, revealing that long-horizon rollout consistency and robot-specific controllability are more important than short-term visual realism for reliable policy assessment.","ai_keywords":["world models","robotic policies","policy evaluation","real-robot teleoperation","video world models","action representation schemes","rollout horizons","simulation","real-world robot behavior","GigaWorld-1"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","githubStars":145,"organization":{"_id":"68d6587936e2de9610d9f5f0","name":"open-gigaai","fullname":"GigaAI","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/68d6394328e169473e90e4a6/zUK7FKr_8XqrN0aFUgsD-.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"6426616ea5ec4a5cbc535634","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6426616ea5ec4a5cbc535634/5IfSFYd9QOxz8K9QmBCst.png","isPro":false,"fullname":"JeffWang","user":"Jeff-Wang","type":"user"},{"_id":"6039478ab3ecf716b1a5fd4d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6039478ab3ecf716b1a5fd4d/_Thy4E7taiSYBLKxEKJbT.jpeg","isPro":true,"fullname":"taesiri","user":"taesiri","type":"user"},{"_id":"64739aa271f07ae738d2d088","avatarUrl":"/avatars/a91d5b18864f8c42abd542eaf1f38b17.svg","isPro":false,"fullname":"JiangnanShao","user":"Andy31415","type":"user"},{"_id":"654c7f32386fc5525c1ebc12","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/654c7f32386fc5525c1ebc12/RIUS_nkbt9n5iu46lLQkb.jpeg","isPro":false,"fullname":"zhao guosheng","user":"f1yfish","type":"user"},{"_id":"6466e66314e059dde8bbf35c","avatarUrl":"/avatars/f2a27bb4e04749ec51f3a9fc36d87fb0.svg","isPro":false,"fullname":"I0u0I","user":"I0u0I","type":"user"},{"_id":"66b03ee1e48856bb7197ed62","avatarUrl":"/avatars/fdfed15190951d4b88bf37e419623be7.svg","isPro":false,"fullname":"Yang","user":"Jian0227","type":"user"},{"_id":"68917b7bb4c8617704deaae6","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/Il651f9440ic7oKFeGm78.png","isPro":false,"fullname":"XuanchengXu","user":"XuanchengXu","type":"user"},{"_id":"6909ce9b391d11cef04e17ee","avatarUrl":"/avatars/d3d4274c7610e2e66b0319075b34c7f4.svg","isPro":false,"fullname":"Daniel Harris","user":"codecodecode123","type":"user"},{"_id":"673de6768fc2cb3dee787d9b","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/mqFdHZoaHAzkmiC-62rUk.png","isPro":false,"fullname":"JoeDeng","user":"cbtogu","type":"user"},{"_id":"644012cf3e0374802e174f7c","avatarUrl":"/avatars/0f4a4bd6f96ce193871843e1d01439e8.svg","isPro":false,"fullname":"Yang Wang","user":"supermodelteam","type":"user"},{"_id":"66a1efaffbabdd7b0d2f1889","avatarUrl":"/avatars/c4ee6bb97a63f06515e30a088f9378e7.svg","isPro":false,"fullname":"zz","user":"zzhugging","type":"user"},{"_id":"686fc6d16ea5d5fb0a4a5b2d","avatarUrl":"/avatars/d7a793b03951c2a7682167e17deaf8db.svg","isPro":false,"fullname":"chong","user":"Phil1220","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"68d6587936e2de9610d9f5f0","name":"open-gigaai","fullname":"GigaAI","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/68d6394328e169473e90e4a6/zUK7FKr_8XqrN0aFUgsD-.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2607/2607.02642.md","query":{}}">

Papers

arxiv:2607.02642

GigaWorld-1: A Roadmap to Build World Models for Robot Policy Evaluation

Published on Jul 2

· Submitted by

JeffWang on Jul 7

GigaAI

Upvote

Authors:

Angyuan Ma ,

Xiaofeng Wang ,

Yang Wang ,

Abstract

World models for robotic policy evaluation are systematically studied through a new benchmark, revealing that long-horizon rollout consistency and robot-specific controllability are more important than short-term visual realism for reliable policy assessment.

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

Evaluating embodied robot foundation models remains a critical bottleneck; unlike large language models efficiently assessed via digital benchmarks, robotic policies require slow, costly real-world rollouts limited by hardware and human supervision, which has driven interest in world models as surrogate policy evaluators, yet the key properties that make a world model reliable for policy assessment remain poorly understood. This work presents a systematic study of world models for robotic policy evaluation and introduces WMBench, a benchmark constructed from real-robot teleoperation data and matched policy rollouts covering diverse manipulation tasks to enable controlled comparisons across model families, action encodings, rollout horizons, and evaluation metrics. Using WMBench, we analyze 7 video world models, 4 action representation schemes, and over 324,000 simulated policy rollouts paired with real robot executions, further enriching our analysis with large-scale community submissions from the CVPR 2026 GigaBrain Challenge, curated synthetic trajectories, and a training videos spanning more than 12,000 hours. Our experiments deliver three core insights: evaluator quality is dominated by long-horizon, action-faithful rollout consistency rather than short-term visual realism; pretraining gains stem not only from data scale but from balancing general world knowledge with robot-specific controllability; and architectural choices including action encoding, memory design, and evaluator-focused post-training strongly determine alignment with real-world robot behavior. Drawing on these results, we derive a practical design roadmap and realize it in GigaWorld-1, a world model specially optimized for policy evaluation, and we fully release our code, models, datasets, and toolkits to advance scalable evaluation research for embodied foundation models.