Learning broad world knowledge directly from raw visual data is a fundamental capability of intelligence. We introduce UniVR, the first investigation into simultaneously learning complex reasoning, fine-grained physical dynamics, and long-term planning from pure visual demonstrations. At its core, UniVR features VR-GRPO, a reinforcement learning paradigm with complementary global and step-level rewards. This approach enforces logical coherence and physical consistency throughout the reasoning process without requiring task-specific heuristics or image-text pairs. To train and evaluate UniVR, we construct VR-X, a large-scale benchmark curated from 16 diverse sources spanning long-horizon manipulation, spatial puzzles, and physical reasoning. It is the first comprehensive suite to assess these heterogeneous capabilities under a purely visual protocol. Remarkably, UniVR achieves up to a 25% improvement on VR-X, and its superior visual reasoning also boosts performance on various multimodal understanding benchmarks. These findings underscore the vast potential of reasoning within visual spaces, with all code, data, and models to be open-sourced for further research.</p>\n<p><a href=\"https://cdn-uploads.huggingface.co/production/uploads/64e1cabf12a5504dda7e4948/aakHxABTSj0nJvR7B9ZO4.png\" rel=\"nofollow\"><img src=\"https://cdn-uploads.huggingface.co/production/uploads/64e1cabf12a5504dda7e4948/aakHxABTSj0nJvR7B9ZO4.png\" alt=\"Clipboard_Screenshot_1784263692\"></a></p>\n","updatedAt":"2026-07-17T04:49:21.554Z","author":{"_id":"64e1cabf12a5504dda7e4948","avatarUrl":"/avatars/53851eddb4e1cae773f3e3607181094b.svg","fullname":"rzw","name":"maverickrzw","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":6,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8325405716896057},"editors":["maverickrzw"],"editorAvatarUrls":["/avatars/53851eddb4e1cae773f3e3607181094b.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2607.12800","authors":[{"_id":"6a59a9796c2e371e6ca381b4","name":"Zhongwei Ren","hidden":false},{"_id":"6a59a9796c2e371e6ca381b5","name":"Yunchao Wei","hidden":false},{"_id":"6a59a9796c2e371e6ca381b6","name":"Yao Zhao","hidden":false},{"_id":"6a59a9796c2e371e6ca381b7","name":"Weibo Gong","hidden":false},{"_id":"6a59a9796c2e371e6ca381b8","name":"Xiao Liu","hidden":false},{"_id":"6a59a9796c2e371e6ca381b9","name":"Anran Wang","hidden":false},{"_id":"6a59a9796c2e371e6ca381ba","name":"Xiangtai Li","hidden":false},{"_id":"6a59a9796c2e371e6ca381bb","name":"Xiaojie Jin","hidden":false}],"publishedAt":"2026-07-14T00:00:00.000Z","submittedOnDailyAt":"2026-07-17T00:00:00.000Z","title":"UniVR: Thinking in Visual Space for Unified Visual Reasoning","submittedOnDailyBy":{"_id":"64e1cabf12a5504dda7e4948","avatarUrl":"/avatars/53851eddb4e1cae773f3e3607181094b.svg","isPro":false,"fullname":"rzw","user":"maverickrzw","type":"user","name":"maverickrzw"},"summary":"Learning broad world knowledge directly from raw visual data is a fundamental capability of intelligence. We introduce UniVR, the first investigation into simultaneously learning complex reasoning, fine-grained physical dynamics, and long-term planning from pure visual demonstrations. At its core, UniVR features VR-GRPO, a reinforcement learning paradigm with complementary global and step-level rewards. This approach enforces logical coherence and physical consistency throughout the reasoning process without requiring task-specific heuristics or image-text pairs. To train and evaluate UniVR, we construct VR-X, a large-scale benchmark curated from 16 diverse sources spanning long-horizon manipulation, spatial puzzles, and physical reasoning. It is the first comprehensive suite to assess these heterogeneous capabilities under a purely visual protocol. Remarkably, UniVR achieves up to a 25% improvement on VR-X, and its superior visual reasoning also boosts performance on various multimodal understanding benchmarks. These findings underscore the vast potential of reasoning within visual spaces, with all code, data, and models are open-sourced for further research.","upvotes":20,"discussionId":"6a59a9796c2e371e6ca381bc","projectPage":"https://maverickren.github.io/UniVR.github.io/","githubRepo":"https://github.com/bytedance/UniVR","githubRepoAddedBy":"user","githubStars":6,"organization":{"_id":"653b817d32c97d0655575872","name":"ByteDance","fullname":"ByteDance","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/6535c9e88bde2fae19b6fb25/0clr54wj5Ly-RkYU9OXPp.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"64e1cabf12a5504dda7e4948","avatarUrl":"/avatars/53851eddb4e1cae773f3e3607181094b.svg","isPro":false,"fullname":"rzw","user":"maverickrzw","type":"user"},{"_id":"6902c0711cbf187eb72ff30b","avatarUrl":"/avatars/bc35ca66abf84b08fb3ab43fe201605c.svg","isPro":false,"fullname":"MLLM","user":"Anran-MLLM","type":"user"},{"_id":"63958b4414513eaf9029ebf1","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/U1g5H071pWRswGAG9UTpo.png","isPro":false,"fullname":"Xiangtai Li","user":"LXT","type":"user"},{"_id":"63ecf637684767daeca7c51d","avatarUrl":"/avatars/b5561f881c175868186ae079e3270b29.svg","isPro":false,"fullname":"Yan Fang","user":"YanFang","type":"user"},{"_id":"650115a0d522560505d8a47b","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/650115a0d522560505d8a47b/q-ELq1-lL6Isqw4WC8nwo.jpeg","isPro":false,"fullname":"Xiao","user":"Yang1213112131","type":"user"},{"_id":"650d632cf1eef416d583a793","avatarUrl":"/avatars/c11388ad6bed54cf067fd7def79ddd27.svg","isPro":false,"fullname":"yang","user":"codeyang123","type":"user"},{"_id":"6901e42c8611d14fa3a0bd19","avatarUrl":"/avatars/00e17f840ea40b75a4b443e6e7060813.svg","isPro":false,"fullname":"ltf","user":"ltf101","type":"user"},{"_id":"64226f268adc8881b9791ad2","avatarUrl":"/avatars/078c59d88f3d12186c05b258a6e71460.svg","isPro":false,"fullname":"Ethan Lin","user":"dafbgd","type":"user"},{"_id":"62e18304cb1f164f2cb38e41","avatarUrl":"/avatars/019fc954080120d94824645bb8a180d8.svg","isPro":false,"fullname":"Curt Smith","user":"curtsmith","type":"user"},{"_id":"673bcb3551d8d86ed0d0c484","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/673bcb3551d8d86ed0d0c484/RC7ytBAEN_C0q6AgF97xP.jpeg","isPro":false,"fullname":"虾仁芹菜","user":"xiarenqincai","type":"user"},{"_id":"65a28e129acab19980226731","avatarUrl":"/avatars/abc3828f807efc4e03837b0eae063f98.svg","isPro":false,"fullname":"Jiahao Meng","user":"marinero4972","type":"user"},{"_id":"693155f0597b51b1c022a3ed","avatarUrl":"/avatars/cb559da7a375f36199610092becbd9b0.svg","isPro":false,"fullname":"zhangyichi","user":"AllTheTimezyc2003","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"653b817d32c97d0655575872","name":"ByteDance","fullname":"ByteDance","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/6535c9e88bde2fae19b6fb25/0clr54wj5Ly-RkYU9OXPp.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2607/2607.12800.md","query":{}}">
UniVR: Thinking in Visual Space for Unified Visual Reasoning
Published on Jul 14
· Submitted by rzw on Jul 17 Abstract
Learning broad world knowledge directly from raw visual data is a fundamental capability of intelligence. We introduce UniVR, the first investigation into simultaneously learning complex reasoning, fine-grained physical dynamics, and long-term planning from pure visual demonstrations. At its core, UniVR features VR-GRPO, a reinforcement learning paradigm with complementary global and step-level rewards. This approach enforces logical coherence and physical consistency throughout the reasoning process without requiring task-specific heuristics or image-text pairs. To train and evaluate UniVR, we construct VR-X, a large-scale benchmark curated from 16 diverse sources spanning long-horizon manipulation, spatial puzzles, and physical reasoning. It is the first comprehensive suite to assess these heterogeneous capabilities under a purely visual protocol. Remarkably, UniVR achieves up to a 25% improvement on VR-X, and its superior visual reasoning also boosts performance on various multimodal understanding benchmarks. These findings underscore the vast potential of reasoning within visual spaces, with all code, data, and models are open-sourced for further research.
Community
Learning broad world knowledge directly from raw visual data is a fundamental capability of intelligence. We introduce UniVR, the first investigation into simultaneously learning complex reasoning, fine-grained physical dynamics, and long-term planning from pure visual demonstrations. At its core, UniVR features VR-GRPO, a reinforcement learning paradigm with complementary global and step-level rewards. This approach enforces logical coherence and physical consistency throughout the reasoning process without requiring task-specific heuristics or image-text pairs. To train and evaluate UniVR, we construct VR-X, a large-scale benchmark curated from 16 diverse sources spanning long-horizon manipulation, spatial puzzles, and physical reasoning. It is the first comprehensive suite to assess these heterogeneous capabilities under a purely visual protocol. Remarkably, UniVR achieves up to a 25% improvement on VR-X, and its superior visual reasoning also boosts performance on various multimodal understanding benchmarks. These findings underscore the vast potential of reasoning within visual spaces, with all code, data, and models to be open-sourced for further research.
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/2607.12800 in a model README.md to link it from this page.
Cite arxiv.org/abs/2607.12800 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2607.12800 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.