Multimodal agents in robotics, AR, and autonomous driving must reason about places and layouts from continuous egocentric streams, often using evidence outside the current view. Existing benchmarks either evaluate offline over full videos or target events rather than spatial structure. We introduce OVO-S-Bench, a fully human-annotated benchmark for streaming spatial intelligence, comprising 1,680 questions over 348 source videos. Annotation involves 12 trained annotators, each also serving as a blind cross-reviewer, across roughly 804 person-hours of multi-round quality assurance. Each question carries a query timestamp and an evidence interval, and at evaluation, the model sees only the prefix preceding the query. Questions span four levels of increasing abstraction: instantaneous egocentric perception, spatiotemporal context tracking, spatial simulation and reasoning, and allocentric mapping. Across 38 proprietary and open-source MLLMs, Gemini-3.1-Pro trails human experts by 27 points, 59.2 vs. 86.6, with allocentric mapping as the dominant bottleneck. Notably, streaming and spatially fine-tuned MLLMs underperform their own backbones. We further find that chain-of-thought reasoning amplifies spatial errors when ungrounded in the stream. By exposing these limitations, OVO-S-Bench establishes a demanding testbed for next-generation streaming spatial MLLMs.</p>\n","updatedAt":"2026-06-04T06:04:31.084Z","author":{"_id":"64adfeac4beffa272dfaef21","avatarUrl":"/avatars/883f6ba38b993476115dfafcef9ce3c1.svg","fullname":"Yifei Li","name":"JoeLeelyf","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":5,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8818176984786987},"editors":["JoeLeelyf"],"editorAvatarUrls":["/avatars/883f6ba38b993476115dfafcef9ce3c1.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2606.03890","authors":[{"_id":"6a202b7015100c5272a84261","name":"Yifei Li","hidden":false},{"_id":"6a202b7015100c5272a84262","name":"Pengyiang Liu","hidden":false},{"_id":"6a202b7015100c5272a84263","name":"Yuhang Zang","hidden":false},{"_id":"6a202b7015100c5272a84264","name":"Zhongyue Shi","hidden":false},{"_id":"6a202b7015100c5272a84265","name":"Qi Fu","hidden":false},{"_id":"6a202b7015100c5272a84266","name":"Hongye Hao","hidden":false},{"_id":"6a202b7015100c5272a84267","name":"Jiwen Lu","hidden":false}],"publishedAt":"2026-06-02T00:00:00.000Z","submittedOnDailyAt":"2026-06-04T00:00:00.000Z","title":"OVO-S-Bench: A Hierarchical Benchmark for Streaming Spatial Intelligence in Multimodal LLMs","submittedOnDailyBy":{"_id":"64adfeac4beffa272dfaef21","avatarUrl":"/avatars/883f6ba38b993476115dfafcef9ce3c1.svg","isPro":false,"fullname":"Yifei Li","user":"JoeLeelyf","type":"user","name":"JoeLeelyf"},"summary":"Multimodal agents in robotics, AR, and autonomous driving must reason about places and layouts from continuous egocentric streams, often using evidence outside the current view. Existing benchmarks either evaluate offline over full videos or target events rather than spatial structure. We introduce OVO-S-Bench, a fully human-annotated benchmark for streaming spatial intelligence, comprising 1,680 questions over 348 source videos. Annotation involves 12 trained annotators, each also serving as a blind cross-reviewer, across roughly 804 person-hours of multi-round quality assurance. Each question carries a query timestamp and an evidence interval, and at evaluation, the model sees only the prefix preceding the query. Questions span four levels of increasing abstraction: instantaneous egocentric perception, spatiotemporal context tracking, spatial simulation and reasoning, and allocentric mapping. Across 38 proprietary and open-source MLLMs, Gemini-3.1-Pro trails human experts by 27 points, 59.2 vs. 86.6, with allocentric mapping as the dominant bottleneck. Notably, streaming and spatially fine-tuned MLLMs underperform their own backbones. We further find that chain-of-thought reasoning amplifies spatial errors when ungrounded in the stream. By exposing these limitations, OVO-S-Bench establishes a demanding testbed for next-generation streaming spatial MLLMs.","upvotes":21,"discussionId":"6a202b7015100c5272a84270","projectPage":"https://internlm.github.io/OVO-S-Bench/","githubRepo":"https://github.com/InternLM/OVO-S-Bench","githubRepoAddedBy":"user","ai_summary":"OVO-S-Bench presents a comprehensive benchmark for evaluating streaming spatial intelligence in multimodal language models through human-annotated questions spanning multiple abstraction levels.","ai_keywords":["multimodal language models","streaming spatial intelligence","egocentric streams","spatial reasoning","allocentric mapping","chain-of-thought reasoning","multimodal agents","autonomous driving","augmented reality"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","githubStars":8,"organization":{"_id":"64a2d5fa81252883206f24c9","name":"internlm","fullname":"Intern Large Models","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/6432683407bad11484a68457/Q3Y0dL79GcsnaBCGRMooZ.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"64adfeac4beffa272dfaef21","avatarUrl":"/avatars/883f6ba38b993476115dfafcef9ce3c1.svg","isPro":false,"fullname":"Yifei Li","user":"JoeLeelyf","type":"user"},{"_id":"644616965691ca69b0e02e79","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/F2uQXO9SkQwHU6benSwQB.jpeg","isPro":false,"fullname":"Juncheng Yan","user":"JonsonYan","type":"user"},{"_id":"661cfae9a853782abad2a495","avatarUrl":"/avatars/39723a07bf9efed8278e009fe966d044.svg","isPro":false,"fullname":"Yanran Zhang","user":"Yanran21","type":"user"},{"_id":"69bd0f6f76fe7a5ea0accdfa","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/gCxQSdKWbvY2bnqSSlUdl.png","isPro":false,"fullname":"Charlotte YOUNG","user":"AverySanchez21","type":"user"},{"_id":"63859cf3b2906edaf83af9f0","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63859cf3b2906edaf83af9f0/kajwuVzd4pDucSPlwghxo.png","isPro":true,"fullname":"Yuhang Zang","user":"yuhangzang","type":"user"},{"_id":"650abbb71aece923f21d87fc","avatarUrl":"/avatars/f09ff031c278bc42bfd7a563853e142c.svg","isPro":false,"fullname":"Junbo Niu","user":"Niujunbo2002","type":"user"},{"_id":"66fd3ed1104850d17b2c4e7c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/66fd3ed1104850d17b2c4e7c/Tw6--5JsovuUQ5khJ6t2J.jpeg","isPro":false,"fullname":"Hejun Dong","user":"fickle1101","type":"user"},{"_id":"6770bedd1ecc151d7576bef5","avatarUrl":"/avatars/14b717997e5f18e862207a43ce1bebb6.svg","isPro":false,"fullname":"buaaplay","user":"buaaplay","type":"user"},{"_id":"68a46dd0e6af085ab75860de","avatarUrl":"/avatars/9d185c22d15871e09bfee6ce3ab4f1b1.svg","isPro":false,"fullname":"Steve Fauci","user":"SteveFauci","type":"user"},{"_id":"662259980195004eb7938e82","avatarUrl":"/avatars/685e7988c90dcf506e7cb3f022fa16f4.svg","isPro":false,"fullname":"YigeMao","user":"YigesMx","type":"user"},{"_id":"69b69e47714094f530db6d2d","avatarUrl":"/avatars/00aad326f19279012b70106cda964d24.svg","isPro":false,"fullname":"hhy","user":"hhy2306454","type":"user"},{"_id":"646cd947da8e99940b6e55cf","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/646cd947da8e99940b6e55cf/9c0P0WppFqNW9pdo8LgOS.jpeg","isPro":false,"fullname":"Shengyuan Ding","user":"ChrisDing1105","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":3,"organization":{"_id":"64a2d5fa81252883206f24c9","name":"internlm","fullname":"Intern Large Models","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/6432683407bad11484a68457/Q3Y0dL79GcsnaBCGRMooZ.png"}}">
OVO-S-Bench: A Hierarchical Benchmark for Streaming Spatial Intelligence in Multimodal LLMs
Abstract
OVO-S-Bench presents a comprehensive benchmark for evaluating streaming spatial intelligence in multimodal language models through human-annotated questions spanning multiple abstraction levels.
Multimodal agents in robotics, AR, and autonomous driving must reason about places and layouts from continuous egocentric streams, often using evidence outside the current view. Existing benchmarks either evaluate offline over full videos or target events rather than spatial structure. We introduce OVO-S-Bench, a fully human-annotated benchmark for streaming spatial intelligence, comprising 1,680 questions over 348 source videos. Annotation involves 12 trained annotators, each also serving as a blind cross-reviewer, across roughly 804 person-hours of multi-round quality assurance. Each question carries a query timestamp and an evidence interval, and at evaluation, the model sees only the prefix preceding the query. Questions span four levels of increasing abstraction: instantaneous egocentric perception, spatiotemporal context tracking, spatial simulation and reasoning, and allocentric mapping. Across 38 proprietary and open-source MLLMs, Gemini-3.1-Pro trails human experts by 27 points, 59.2 vs. 86.6, with allocentric mapping as the dominant bottleneck. Notably, streaming and spatially fine-tuned MLLMs underperform their own backbones. We further find that chain-of-thought reasoning amplifies spatial errors when ungrounded in the stream. By exposing these limitations, OVO-S-Bench establishes a demanding testbed for next-generation streaming spatial MLLMs.
Community
Multimodal agents in robotics, AR, and autonomous driving must reason about places and layouts from continuous egocentric streams, often using evidence outside the current view. Existing benchmarks either evaluate offline over full videos or target events rather than spatial structure. We introduce OVO-S-Bench, a fully human-annotated benchmark for streaming spatial intelligence, comprising 1,680 questions over 348 source videos. Annotation involves 12 trained annotators, each also serving as a blind cross-reviewer, across roughly 804 person-hours of multi-round quality assurance. Each question carries a query timestamp and an evidence interval, and at evaluation, the model sees only the prefix preceding the query. Questions span four levels of increasing abstraction: instantaneous egocentric perception, spatiotemporal context tracking, spatial simulation and reasoning, and allocentric mapping. Across 38 proprietary and open-source MLLMs, Gemini-3.1-Pro trails human experts by 27 points, 59.2 vs. 86.6, with allocentric mapping as the dominant bottleneck. Notably, streaming and spatially fine-tuned MLLMs underperform their own backbones. We further find that chain-of-thought reasoning amplifies spatial errors when ungrounded in the stream. By exposing these limitations, OVO-S-Bench establishes a demanding testbed for next-generation streaming spatial MLLMs.
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/2606.03890 in a model README.md to link it from this page.
Cite arxiv.org/abs/2606.03890 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2606.03890 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.