We present Zetta, a closed-loop embodied harness that evolves code-based runtime critics and recovery skills online while keeping the base policy frozen. Through three timescale-separated loops, Zetta provides action-frequency governance, rollout-level critic-recovery proposal, and validation-gated skill updates. Together with Z-Infra, a rollout infrastructure decoupling agent logic from heterogeneous execution resources, Zetta achieves state-of-the-art success on LIBERO-Pro and RoboCasa under our current rollout budget, reaching 90.8% and 93.6%, with an 11.1x inference speedup; success continues to scale with self-exploration experience; learned skills transfer zero-shot, and clear robotic \"Aha Moments\" emerge. These results show that closed-loop harness self-evolution opens a scaling path for reliable physical intelligence.</p>\n","updatedAt":"2026-08-20T03:55:50.326Z","author":{"_id":"6632ef2cb07d501395dc5210","avatarUrl":"/avatars/7cab9c60f6e18ca02e2d13ce4d09afaf.svg","fullname":"Xin Ding","name":"Xin64","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8736597299575806},"editors":["Xin64"],"editorAvatarUrls":["/avatars/7cab9c60f6e18ca02e2d13ce4d09afaf.svg"],"reactions":[{"reaction":"👍","users":["huangmz21","midagudo","PanAn1234","addaddgo"],"count":4}],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.16590","authors":[{"_id":"6a85134c536bdd3bdd48f767","user":{"_id":"6632ef2cb07d501395dc5210","avatarUrl":"/avatars/7cab9c60f6e18ca02e2d13ce4d09afaf.svg","isPro":false,"fullname":"Xin Ding","user":"Xin64","type":"user","name":"Xin64"},"name":"Xin Ding","status":"claimed_verified","statusLastChangedAt":"2026-08-19T08:45:04.346Z","hidden":false},{"_id":"6a85134c536bdd3bdd48f768","name":"Liang Mi","hidden":false},{"_id":"6a85134c536bdd3bdd48f769","name":"Mingzhe Huang","hidden":false},{"_id":"6a85134c536bdd3bdd48f76a","name":"Zixuan Wang","hidden":false},{"_id":"6a85134c536bdd3bdd48f76b","name":"Chao Zhang","hidden":false},{"_id":"6a85134c536bdd3bdd48f76c","name":"Zixu Hao","hidden":false},{"_id":"6a85134c536bdd3bdd48f76d","name":"Fu Chen","hidden":false},{"_id":"6a85134c536bdd3bdd48f76e","name":"Xiangyu Li","hidden":false},{"_id":"6a85134c536bdd3bdd48f76f","name":"Yikai Zheng","hidden":false},{"_id":"6a85134c536bdd3bdd48f770","name":"Yaoyu Guo","hidden":false},{"_id":"6a85134c536bdd3bdd48f771","name":"Weijun Wang","hidden":false},{"_id":"6a85134c536bdd3bdd48f772","name":"Kun Li","hidden":false},{"_id":"6a85134c536bdd3bdd48f773","name":"Hao Wu","hidden":false},{"_id":"6a85134c536bdd3bdd48f774","name":"Yunxin Liu","hidden":false},{"_id":"6a85134c536bdd3bdd48f775","name":"Ting Cao","hidden":false}],"publishedAt":"2026-08-17T00:00:00.000Z","submittedOnDailyAt":"2026-08-20T00:00:00.000Z","title":"Zetta ζ: An Efficient Closed-Loop Embodied Harness for Self-Evolving Physical Intelligence","submittedOnDailyBy":{"_id":"6632ef2cb07d501395dc5210","avatarUrl":"/avatars/7cab9c60f6e18ca02e2d13ce4d09afaf.svg","isPro":false,"fullname":"Xin Ding","user":"Xin64","type":"user","name":"Xin64"},"summary":"Embodied agents are increasingly used to close the gap left by end-to-end policy models. Yet the agentic path has not realized closed-loop learning in physical execution: existing harnesses remain largely open-loop, following fixed skills during rollout and reflecting only after an episode completes. Such post-hoc reflection cannot govern execution as it unfolds, because physical interaction requires decisions to track rapidly changing robot-environment states at a frequency beyond today's large agentic models. We present Zetta, a closed-loop embodied harness that evolves code-based runtime critics and recovery skills online while keeping the base policy frozen. Through three timescale-separated loops, Zetta provides action-frequency governance, rollout-level critic-recovery proposal, and validation-gated skill updates. Together with Z-Infra, a rollout infrastructure decoupling agent logic from heterogeneous execution resources, Zetta achieves state-of-the-art success on LIBERO-Pro and RoboCasa under our current rollout budget, reaching 90.8% and 93.6%, with an 11.1x inference speedup; success continues to scale with self-exploration experience; learned skills transfer zero-shot, and clear robotic \"Aha Moments\" emerge. These results show that closed-loop harness self-evolution opens a scaling path for reliable physical intelligence.","upvotes":89,"discussionId":"6a85134d536bdd3bdd48f776","projectPage":"https://air-embodied-brain.github.io/zetta/","githubRepo":"https://github.com/air-embodied-brain/Zetta-Embodiment","githubRepoAddedBy":"user","ai_summary":"Zetta is a closed-loop embodied harness that evolves runtime critics and recovery skills online to govern physical execution at action frequency, achieving high success on robot benchmarks with faster inference and scaling self-exploration.","ai_keywords":["closed-loop embodied harness","code-based runtime critics","recovery skills","action-frequency governance","rollout-level critic-recovery","validation-gated skill updates","Z-Infra","LIBERO-Pro","RoboCasa","self-exploration","zero-shot skill transfer"],"ai_summary_model":"thinkingmachines/Inkling-Small","githubStars":152},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"6632ef2cb07d501395dc5210","avatarUrl":"/avatars/7cab9c60f6e18ca02e2d13ce4d09afaf.svg","isPro":false,"fullname":"Xin Ding","user":"Xin64","type":"user"},{"_id":"68b6c78cb3c52cf8ca942e09","avatarUrl":"/avatars/d4edb397c53493afaec5c8537258ddc8.svg","isPro":false,"fullname":"Zishwen","user":"Cyan9061","type":"user"},{"_id":"662852bf8bfc90408a97acdf","avatarUrl":"/avatars/008e6faf51e1be4f0c0d17008dfe2f49.svg","isPro":false,"fullname":"cao","user":"tingcao","type":"user"},{"_id":"6607c0cbc201aee890d3c27c","avatarUrl":"/avatars/40ca5f38f4fa40e85ca8b42e9a85d216.svg","isPro":false,"fullname":"liang mi","user":"midagudo","type":"user"},{"_id":"64f1b4ebb45e86c2be5b3818","avatarUrl":"/avatars/24ab1b864d93a3fb0b3ff54886beb53f.svg","isPro":false,"fullname":"Zixu Hao","user":"haozixu0","type":"user"},{"_id":"664422fe34fff844a5204d4f","avatarUrl":"/avatars/bce39fbb484834d8f7c9046c939502a3.svg","isPro":false,"fullname":"Chen Gong","user":"gongchenooo","type":"user"},{"_id":"686f8417326edb5fba4598ef","avatarUrl":"/avatars/900b6e006788a739d318febf9879c5a0.svg","isPro":false,"fullname":"Kejun Gao","user":"Arran1025","type":"user"},{"_id":"6690e9ad5f663d4d4c2116c5","avatarUrl":"/avatars/d7293ee7dff03dd55d9fef49d27b39e9.svg","isPro":false,"fullname":"黄明哲","user":"huangmz21","type":"user"},{"_id":"65d5ce8fcff388f15589a7c9","avatarUrl":"/avatars/a3747afc67618d87fe97978d68d85661.svg","isPro":false,"fullname":"SII-ChenChen","user":"orange0318","type":"user"},{"_id":"64fed18735e95605e7b43e19","avatarUrl":"/avatars/30d609b0b978f6cfac731f426900e9fc.svg","isPro":false,"fullname":"yulei96","user":"Yulei96","type":"user"},{"_id":"6a867d8052f53f9d1caaa9eb","avatarUrl":"/avatars/7415eac9cf4ea93e5c9ab5798ee634d6.svg","isPro":false,"fullname":"fbichao","user":"fbichao","type":"user"},{"_id":"69ec49d9242c42522855df5d","avatarUrl":"/avatars/0ee3ef6c94c55c72fd2f54306a1d75d4.svg","isPro":false,"fullname":"ANPAN","user":"PanAn1234","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":1,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.16590.md","query":{}}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2608.16590 in a model README.md to link it from this page.

Cite arxiv.org/abs/2608.16590 in a dataset README.md to link it from this page.

Cite arxiv.org/abs/2608.16590 in a Space README.md to link it from this page.