Hugging Face Daily Papers · · 6 min read

LightNav-0: Eliciting VLM Spatial Intelligence for Generalist Embodied Navigation

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

Embodied navigation requires agents to translate heterogeneous goals and visual observations into actions across tasks, environments, and robot embodiments. Modern vision-language models (VLMs) already encode spatial priors for visual grounding, spatial reasoning, and pointing, but these capabilities are rarely elicited directly for robot control. Existing navigation systems instead rely on task- or embodiment-specific components, fragmenting perception, reasoning, and action while offering limited generalization. Here we present LightNav-0, a compact generalist embodied navigation model that elicits the spatial intelligence of a pretrained VLM and aligns it with navigation, without task-specific prediction heads. LightNav-0 represents diverse navigation tasks through a unified token interface: dual-channel pointing expresses task-, scene-, and embodiment-agnostic spatial intent, while a residual vector-quantized action tokenizer maps this intent to precise, embodiment-specific trajectories. Together with temporally aware visual history compression, ER mid-training, supervised fine-tuning, and reinforcement learning, this formulation supports instruction following, open-vocabulary object navigation, and visual tracking within a single model. The navigation training corpus spans 2K+ scenes and 4K+ hours of embodied navigation data. LightNav-ER, the embodied-reasoning checkpoint used to initialize LightNav-0, attains the highest complete-set average across 8 embodied-reasoning benchmarks, while LightNav-0 achieves state-of-the-art monocular success rates across all 10 public navigation simulation settings. Real-world evaluations further demonstrate zero-shot generalization across robot embodiments, diverse scenes, and static and dynamic targets. These results establish compact VLMs as a unified and transferable backbone for generalist embodied navigation.</p>\n","updatedAt":"2026-09-01T07:51:40.086Z","author":{"_id":"64b6e2abd3711dd619e476fc","avatarUrl":"/avatars/430f53c44d12c578a5128e42ada62527.svg","fullname":"Xinhang Liu","name":"xinhang0111","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.848848819732666},"editors":["xinhang0111"],"editorAvatarUrls":["/avatars/430f53c44d12c578a5128e42ada62527.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.30935","authors":[{"_id":"6a9681d4cd6ebc484732ef3c","name":"Shaoan Wang","hidden":false},{"_id":"6a9681d4cd6ebc484732ef3d","name":"Aocheng Luo","hidden":false},{"_id":"6a9681d4cd6ebc484732ef3e","name":"Fei Huang","hidden":false},{"_id":"6a9681d4cd6ebc484732ef3f","name":"Jingyi Xu","hidden":false},{"_id":"6a9681d4cd6ebc484732ef40","name":"Xiaoyang Wang","hidden":false},{"_id":"6a9681d4cd6ebc484732ef41","name":"Yueyu Wang","hidden":false},{"_id":"6a9681d4cd6ebc484732ef42","name":"Qianli Ma","hidden":false},{"_id":"6a9681d4cd6ebc484732ef43","name":"Fan Yang","hidden":false},{"_id":"6a9681d4cd6ebc484732ef44","name":"Ran Mei","hidden":false},{"_id":"6a9681d4cd6ebc484732ef45","name":"Jia Wei","hidden":false},{"_id":"6a9681d4cd6ebc484732ef46","name":"Jiangpeng Hu","hidden":false},{"_id":"6a9681d4cd6ebc484732ef47","name":"Xuhao Liu","hidden":false},{"_id":"6a9681d4cd6ebc484732ef48","name":"Hongming Chen","hidden":false},{"_id":"6a9681d4cd6ebc484732ef49","name":"Yuanbin Shao","hidden":false},{"_id":"6a9681d4cd6ebc484732ef4a","name":"Yiyang Lin","hidden":false},{"_id":"6a9681d4cd6ebc484732ef4b","name":"Ziliang Li","hidden":false},{"_id":"6a9681d4cd6ebc484732ef4c","name":"Liang Pan","hidden":false},{"_id":"6a9681d4cd6ebc484732ef4d","name":"Xinhang Liu","hidden":false},{"_id":"6a9681d4cd6ebc484732ef4e","name":"Yuntao Ma","hidden":false},{"_id":"6a9681d4cd6ebc484732ef4f","name":"Tingxiang Fan","hidden":false}],"publishedAt":"2026-08-31T00:00:00.000Z","submittedOnDailyAt":"2026-09-01T00:00:00.000Z","title":"LightNav-0: Eliciting VLM Spatial Intelligence for Generalist Embodied Navigation","submittedOnDailyBy":{"_id":"64b6e2abd3711dd619e476fc","avatarUrl":"/avatars/430f53c44d12c578a5128e42ada62527.svg","isPro":false,"fullname":"Xinhang Liu","user":"xinhang0111","type":"user","name":"xinhang0111"},"summary":"Embodied navigation requires agents to translate heterogeneous goals and visual observations into actions across tasks, environments, and robot embodiments. Modern vision-language models (VLMs) already encode spatial priors for visual grounding, spatial reasoning, and pointing, but these capabilities are rarely elicited directly for robot control. Existing navigation systems instead rely on task- or embodiment-specific components, fragmenting perception, reasoning, and action while offering limited generalization. Here we present LightNav-0, a compact generalist embodied navigation model that elicits the spatial intelligence of a pretrained VLM and aligns it with navigation, without task-specific prediction heads. LightNav-0 represents diverse navigation tasks through a unified token interface: dual-channel pointing expresses task-, scene-, and embodiment-agnostic spatial intent, while a residual vector-quantized action tokenizer maps this intent to precise, embodiment-specific trajectories. Together with temporally aware visual history compression, ER mid-training, supervised fine-tuning, and reinforcement learning, this formulation supports instruction following, open-vocabulary object navigation, and visual tracking within a single model. The navigation training corpus spans 2K+ scenes and 4K+ hours of embodied navigation data. LightNav-ER, the embodied-reasoning checkpoint used to initialize LightNav-0, attains the highest complete-set average across 8 embodied-reasoning benchmarks, while LightNav-0 achieves state-of-the-art monocular success rates across all 10 public navigation simulation settings. Real-world evaluations further demonstrate zero-shot generalization across robot embodiments, diverse scenes, and static and dynamic targets. These results establish compact VLMs as a unified and transferable backbone for generalist embodied navigation.","upvotes":23,"discussionId":"6a9681d5cd6ebc484732ef50","projectPage":"https://www.lightorigins.com/en/blog/lightnav-0","githubRepo":"https://github.com/lightorigins/LightNav-0","githubRepoAddedBy":"user","ai_summary":"LightNav-0 is a compact generalist navigation model that leverages a pretrained vision-language model’s spatial reasoning via unified pointing tokens and action tokenization to achieve state-of-the-art embodied navigation across diverse tasks and robots.","ai_keywords":["vision-language models","spatial priors","dual-channel pointing","residual vector-quantized action tokenizer","temporally aware visual history compression","ER mid-training","embodied reasoning","open-vocabulary object navigation","monocular success rates"],"ai_summary_model":"thinkingmachines/Inkling-Small","githubStars":180,"organization":{"_id":"6a69d90bc6b05159cdf05bd3","name":"LightOriginsHQ","fullname":"Light Origins","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/6a69bc220887c57325b805ac/30q8pyMpIFKfBosP82YFV.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"64b6e2abd3711dd619e476fc","avatarUrl":"/avatars/430f53c44d12c578a5128e42ada62527.svg","isPro":false,"fullname":"Xinhang Liu","user":"xinhang0111","type":"user"},{"_id":"674689c668844b51d9bde8a6","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/_NP586LyRFWnWOmKqpejD.jpeg","isPro":false,"fullname":"XiaoYoung","user":"xy-23","type":"user"},{"_id":"6630cc7e9ee8861dd0b9bdbd","avatarUrl":"/avatars/01da7d60cbb107d58329bbb80d924eb4.svg","isPro":false,"fullname":"Liang Pan","user":"lianganimation","type":"user"},{"_id":"6a0467a9242eaf4e1c28a7ca","avatarUrl":"/avatars/27e5ad5cde822bccb2ca65e466ee63a1.svg","isPro":false,"fullname":"Fan Yang","user":"lrlittley","type":"user"},{"_id":"665e88a7ddb96a4d7259ac65","avatarUrl":"/avatars/77acc8f2b2ed5eabd75f3a0a5c480cc9.svg","isPro":false,"fullname":"ace","user":"tuckerace","type":"user"},{"_id":"6516a5115b041d575d30b921","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/9zN09EW3ZnJQnbJvwIDpP.png","isPro":false,"fullname":"Yifan Yu","user":"MarkYu98","type":"user"},{"_id":"66a4e3f1dcf448e5a7a91e6b","avatarUrl":"/avatars/455f08f8e0f7e8055ed2f62a201ea4b1.svg","isPro":false,"fullname":"Shaoan Wang","user":"wsakobe","type":"user"},{"_id":"6768d5791d652acb0c6adfb8","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/muhfuUAqioTimaQ73fVVK.png","isPro":false,"fullname":"Tasbolat Taunyazov","user":"Tasbolat","type":"user"},{"_id":"66168d11123f393614f7c908","avatarUrl":"/avatars/754779a228cb2ff31d21eb1221c576de.svg","isPro":false,"fullname":"xiang huang","user":"xianghuang","type":"user"},{"_id":"6a43f0a854a7c7fac52f2538","avatarUrl":"/avatars/b910a8d21232e0dfa3b82195fa879970.svg","isPro":false,"fullname":"Aaron Humfrey","user":"Humfrey","type":"user"},{"_id":"6a226dd3b9a0c574d5c64296","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/_W37PWPhOACjRDR2ODwTv.png","isPro":false,"fullname":"xiaohui.huang","user":"HXH2ELSA","type":"user"},{"_id":"631e11fd2899cb403efdb775","avatarUrl":"/avatars/c19266fc97ddda817c79f0019732a2b0.svg","isPro":false,"fullname":"Y Ma","user":"articuno144","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"6a69d90bc6b05159cdf05bd3","name":"LightOriginsHQ","fullname":"Light Origins","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/6a69bc220887c57325b805ac/30q8pyMpIFKfBosP82YFV.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.30935.md","query":{}}">
Papers
arxiv:2608.30935

LightNav-0: Eliciting VLM Spatial Intelligence for Generalist Embodied Navigation

Published on Aug 31
· Submitted by
Xinhang Liu
on Sep 1
Authors:
,

Abstract

LightNav-0 is a compact generalist navigation model that leverages a pretrained vision-language model’s spatial reasoning via unified pointing tokens and action tokenization to achieve state-of-the-art embodied navigation across diverse tasks and robots.

Embodied navigation requires agents to translate heterogeneous goals and visual observations into actions across tasks, environments, and robot embodiments. Modern vision-language models (VLMs) already encode spatial priors for visual grounding, spatial reasoning, and pointing, but these capabilities are rarely elicited directly for robot control. Existing navigation systems instead rely on task- or embodiment-specific components, fragmenting perception, reasoning, and action while offering limited generalization. Here we present LightNav-0, a compact generalist embodied navigation model that elicits the spatial intelligence of a pretrained VLM and aligns it with navigation, without task-specific prediction heads. LightNav-0 represents diverse navigation tasks through a unified token interface: dual-channel pointing expresses task-, scene-, and embodiment-agnostic spatial intent, while a residual vector-quantized action tokenizer maps this intent to precise, embodiment-specific trajectories. Together with temporally aware visual history compression, ER mid-training, supervised fine-tuning, and reinforcement learning, this formulation supports instruction following, open-vocabulary object navigation, and visual tracking within a single model. The navigation training corpus spans 2K+ scenes and 4K+ hours of embodied navigation data. LightNav-ER, the embodied-reasoning checkpoint used to initialize LightNav-0, attains the highest complete-set average across 8 embodied-reasoning benchmarks, while LightNav-0 achieves state-of-the-art monocular success rates across all 10 public navigation simulation settings. Real-world evaluations further demonstrate zero-shot generalization across robot embodiments, diverse scenes, and static and dynamic targets. These results establish compact VLMs as a unified and transferable backbone for generalist embodied navigation.

Community

Paper submitter about 5 hours ago

Embodied navigation requires agents to translate heterogeneous goals and visual observations into actions across tasks, environments, and robot embodiments. Modern vision-language models (VLMs) already encode spatial priors for visual grounding, spatial reasoning, and pointing, but these capabilities are rarely elicited directly for robot control. Existing navigation systems instead rely on task- or embodiment-specific components, fragmenting perception, reasoning, and action while offering limited generalization. Here we present LightNav-0, a compact generalist embodied navigation model that elicits the spatial intelligence of a pretrained VLM and aligns it with navigation, without task-specific prediction heads. LightNav-0 represents diverse navigation tasks through a unified token interface: dual-channel pointing expresses task-, scene-, and embodiment-agnostic spatial intent, while a residual vector-quantized action tokenizer maps this intent to precise, embodiment-specific trajectories. Together with temporally aware visual history compression, ER mid-training, supervised fine-tuning, and reinforcement learning, this formulation supports instruction following, open-vocabulary object navigation, and visual tracking within a single model. The navigation training corpus spans 2K+ scenes and 4K+ hours of embodied navigation data. LightNav-ER, the embodied-reasoning checkpoint used to initialize LightNav-0, attains the highest complete-set average across 8 embodied-reasoning benchmarks, while LightNav-0 achieves state-of-the-art monocular success rates across all 10 public navigation simulation settings. Real-world evaluations further demonstrate zero-shot generalization across robot embodiments, diverse scenes, and static and dynamic targets. These results establish compact VLMs as a unified and transferable backbone for generalist embodied navigation.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images

· Sign up or log in to comment

Get this paper in your agent:

hf papers read 2608.30935
Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash

Models citing this paper

No model linking this paper

Cite arxiv.org/abs/2608.30935 in a model README.md to link it from this page.

Datasets citing this paper

No dataset linking this paper

Cite arxiv.org/abs/2608.30935 in a dataset README.md to link it from this page.

Spaces citing this paper

No Space linking this paper

Cite arxiv.org/abs/2608.30935 in a Space README.md to link it from this page.

Collections including this paper

No Collection including this paper

Add this paper to a collection 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.

More from Hugging Face Daily Papers