Hugging Face Daily Papers · June 16, 2026 · 4 min read

Geometric Action Model for Robot Policy Learning

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

Like Read original ↗

<a href=\"https://cdn-uploads.huggingface.co/production/uploads/686fb8a66ea5d5fb0a4953a9/Fw26Fdk9VDgVH4lhBw6Og.webp\" rel=\"nofollow\"><img src=\"https://cdn-uploads.huggingface.co/production/uploads/686fb8a66ea5d5fb0a4953a9/Fw26Fdk9VDgVH4lhBw6Og.webp\" alt=\"teaser3\"></a></p>\n","updatedAt":"2026-06-16T05:19:17.171Z","author":{"_id":"686fb8a66ea5d5fb0a4953a9","avatarUrl":"/avatars/b4aa187b82dd04a5a7ece3b922d86657.svg","fullname":"Sunghwan Hong","name":"hongsunghwan","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.5287291407585144},"editors":["hongsunghwan"],"editorAvatarUrls":["/avatars/b4aa187b82dd04a5a7ece3b922d86657.svg"],"reactions":[{"reaction":"👍","users":["Ryoo72","nandometzger","SeonghuJeon","adelitzas","jiho31"],"count":5}],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2606.17046","authors":[{"_id":"6a30dc5ba0d4daae4286021d","user":{"_id":"64886f020a30bd0e7bd26dc3","avatarUrl":"/avatars/f02e7996a95968ba5cf9d5ba1de6d8b0.svg","isPro":false,"fullname":"Jisang Han","user":"onground","type":"user","name":"onground"},"name":"Jisang Han","status":"claimed_verified","statusLastChangedAt":"2026-06-16T12:05:40.104Z","hidden":false},{"_id":"6a30dc5ba0d4daae4286021e","user":{"_id":"67861f4658328c475597e540","avatarUrl":"/avatars/ff3d7b7912544cd0799d289e6c51db7a.svg","isPro":false,"fullname":"Seonghu Jeon","user":"SeonghuJeon","type":"user","name":"SeonghuJeon"},"name":"Seonghu Jeon","status":"claimed_verified","statusLastChangedAt":"2026-06-16T12:05:46.838Z","hidden":false},{"_id":"6a30dc5ba0d4daae4286021f","name":"Jaewoo Jung","hidden":false},{"_id":"6a30dc5ba0d4daae42860220","name":"René Zurbrügg","hidden":false},{"_id":"6a30dc5ba0d4daae42860221","user":{"_id":"67565151e704deb871d99c95","avatarUrl":"/avatars/b7f7b0f29f6879de3c9586fe7ba9d6b9.svg","isPro":false,"fullname":"honggyu An","user":"honggyuAn","type":"user","name":"honggyuAn"},"name":"Honggyu An","status":"claimed_verified","statusLastChangedAt":"2026-06-16T12:05:44.617Z","hidden":false},{"_id":"6a30dc5ba0d4daae42860222","name":"Tifanny Portela","hidden":false},{"_id":"6a30dc5ba0d4daae42860223","name":"Marco Hutter","hidden":false},{"_id":"6a30dc5ba0d4daae42860224","name":"Marc Pollefeys","hidden":false},{"_id":"6a30dc5ba0d4daae42860225","name":"Seungryong Kim","hidden":false},{"_id":"6a30dc5ba0d4daae42860226","user":{"_id":"686fb8a66ea5d5fb0a4953a9","avatarUrl":"/avatars/b4aa187b82dd04a5a7ece3b922d86657.svg","isPro":false,"fullname":"Sunghwan Hong","user":"hongsunghwan","type":"user","name":"hongsunghwan"},"name":"Sunghwan Hong","status":"claimed_verified","statusLastChangedAt":"2026-06-16T12:05:42.297Z","hidden":false}],"publishedAt":"2026-06-15T00:00:00.000Z","submittedOnDailyAt":"2026-06-16T00:00:00.000Z","title":"Geometric Action Model for Robot Policy Learning","submittedOnDailyBy":{"_id":"686fb8a66ea5d5fb0a4953a9","avatarUrl":"/avatars/b4aa187b82dd04a5a7ece3b922d86657.svg","isPro":false,"fullname":"Sunghwan Hong","user":"hongsunghwan","type":"user","name":"hongsunghwan"},"summary":"Generalist robot policies must follow user instructions while reasoning about how objects, cameras, and robot actions interact in the 3D physical world. Recent vision-language-action models (VLAs) and video world-action models (WAMs) inherit strong semantic or temporal priors from large-scale foundation models, but they still operate primarily on 2D image frames or 2D-derived latent spaces, leaving implicit the 3D geometry required for contact-rich manipulation. We propose the Geometric Action Model (GAM), a language-conditioned manipulation policy that directly repurposes a pretrained geometric foundation model (GFM) as a shared substrate for perception, temporal prediction, and action decoding. GAM splits the GFM at an intermediate layer: the shallow layers serve as an observation encoder, and a causal future predictor inserted at the split layer forecasts future latent tokens conditioned on language, proprioception, and action history. The predicted future tokens are then routed through the remaining GFM blocks for feature propagation and decoding, allowing a single backbone to produce both future geometry and actions. This design equips the GFM with language-conditioned temporal world modeling through minimal architectural modification while preserving its rich geometric priors. Across a broad suite of simulation and real-robot manipulation benchmarks, GAM is more accurate, more robust, faster, and lighter than current foundation-model-scale baselines.","upvotes":78,"discussionId":"6a30dc5ba0d4daae42860227","projectPage":"https://cvlab-kaist.github.io/Geometric-Action-Model/","githubRepo":"https://github.com/cvlab-kaist/Geometric-Action-Model","githubRepoAddedBy":"user","ai_summary":"A geometric action model leverages pretrained geometric foundation models to enable language-conditioned manipulation policies with improved accuracy, robustness, and efficiency in 3D physical environments.","ai_keywords":["vision-language-action models","video world-action models","geometric foundation models","language-conditioned manipulation policy","causal future predictor","latent tokens","action decoding","temporal world modeling","3D physical world","contact-rich manipulation"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","githubStars":27,"organization":{"_id":"63dd02d5f37111482523565a","name":"ETHZurich","fullname":"ETH Zürich","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1675428475378-63dcff68a8877129a1574f33.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"67861f4658328c475597e540","avatarUrl":"/avatars/ff3d7b7912544cd0799d289e6c51db7a.svg","isPro":false,"fullname":"Seonghu Jeon","user":"SeonghuJeon","type":"user"},{"_id":"63198bf1615c77c25d63e9ab","avatarUrl":"/avatars/6d591f87366e9990fed3c221dafdfae0.svg","isPro":false,"fullname":"Yunsung Lee","user":"Maangeek","type":"user"},{"_id":"686fb8a66ea5d5fb0a4953a9","avatarUrl":"/avatars/b4aa187b82dd04a5a7ece3b922d86657.svg","isPro":false,"fullname":"Sunghwan Hong","user":"hongsunghwan","type":"user"},{"_id":"6752b6315281c3cae4b0783f","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/xmcyVEl2xBhk3G5_7dmpz.png","isPro":false,"fullname":"Paul Hyunbin Cho","user":"paulcho98","type":"user"},{"_id":"674e9fd3644d6056e583166c","avatarUrl":"/avatars/31dce6e8c1ff256372889c796c67e651.svg","isPro":false,"fullname":"Jinhyuk Jang","user":"JinhyukJang","type":"user"},{"_id":"668e24b961b6eff5a8ad2d34","avatarUrl":"/avatars/f3a4e2e766110e687010a702ed96d7af.svg","isPro":false,"fullname":"chloe","user":"chloe1929","type":"user"},{"_id":"66f3f780169b85adf963508d","avatarUrl":"/avatars/495ce69c478c332b5fafe897bf1ee80e.svg","isPro":false,"fullname":"Jaeyeong Kim","user":"jy9394","type":"user"},{"_id":"64cb5884d469fc2cf83bdd76","avatarUrl":"/avatars/10e63cf62d8200beef3e31846796e398.svg","isPro":false,"fullname":"JisooKim","user":"Jiiiiiisoo","type":"user"},{"_id":"6752ac9be0c39c0eaf6ba90d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/cbByUmYoPVUAr35MWQeVm.png","isPro":false,"fullname":"lee","user":"lshlsh","type":"user"},{"_id":"6752b5ebebb87145beedaecb","avatarUrl":"/avatars/1de059e88dad6fe070cb22ba96d32914.svg","isPro":false,"fullname":"Seungryong Kim","user":"seungryongkim","type":"user"},{"_id":"651277c2b6ffd31931db5290","avatarUrl":"/avatars/8495b84e8aed407da07908ee829e0510.svg","isPro":false,"fullname":"JihoPark","user":"jiho31","type":"user"},{"_id":"67136aa856c26d6294bc7dad","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/APivpGKA0OHK3EPV6AQqL.png","isPro":false,"fullname":"JungJae Lee","user":"Jerry112","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":2,"organization":{"_id":"63dd02d5f37111482523565a","name":"ETHZurich","fullname":"ETH Zürich","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1675428475378-63dcff68a8877129a1574f33.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2606/2606.17046.md","query":{}}">

Papers

arxiv:2606.17046

Geometric Action Model for Robot Policy Learning

Published on Jun 15

· Submitted by

Sunghwan Hong on Jun 16

#2 Paper of the day

ETH Zürich

Upvote

Authors:

Jisang Han ,

Seonghu Jeon ,

Honggyu An ,

Sunghwan Hong

Abstract

A geometric action model leverages pretrained geometric foundation models to enable language-conditioned manipulation policies with improved accuracy, robustness, and efficiency in 3D physical environments.

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

Generalist robot policies must follow user instructions while reasoning about how objects, cameras, and robot actions interact in the 3D physical world. Recent vision-language-action models (VLAs) and video world-action models (WAMs) inherit strong semantic or temporal priors from large-scale foundation models, but they still operate primarily on 2D image frames or 2D-derived latent spaces, leaving implicit the 3D geometry required for contact-rich manipulation. We propose the Geometric Action Model (GAM), a language-conditioned manipulation policy that directly repurposes a pretrained geometric foundation model (GFM) as a shared substrate for perception, temporal prediction, and action decoding. GAM splits the GFM at an intermediate layer: the shallow layers serve as an observation encoder, and a causal future predictor inserted at the split layer forecasts future latent tokens conditioned on language, proprioception, and action history. The predicted future tokens are then routed through the remaining GFM blocks for feature propagation and decoding, allowing a single backbone to produce both future geometry and actions. This design equips the GFM with language-conditioned temporal world modeling through minimal architectural modification while preserving its rich geometric priors. Across a broad suite of simulation and real-robot manipulation benchmarks, GAM is more accurate, more robust, faster, and lighter than current foundation-model-scale baselines.