[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85026-en":3,"doc-seo-85026-105":29,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},85026,13056703019404,"Miles","https://ap-avatar.wpscdn.com/davatar_29158cc5080c5b710cf443261637dec0",8,"Research & Report","Scaling Mixture-of-Experts Video Pretraining for Embodied Intelligence","Video generative models for robots face a domain mismatch: existing systems optimize visual fidelity, creativity, and text alignment rather than physical correctness, controllability, and long-horizon consistency under interaction. LingBot-Video introduces a DiT-based video pretraining paradigm for embodied intelligence. It scales Mixture-of-Experts for sparse computation and capacity, builds a robot-augmented dataset with manipulation, navigation, and egocentric footage, and applies a multi-dimensional reward system enforcing physical rationality and task completion. Extensive evaluations assess both performance and efficiency as a video foundation model.","arXiv :2607 .07675v 1 [ cs .CV] 8 Jul 2026  \nScaling Mixture-of-Experts Video Pretraining for Embodied Intelligence  \nShuailei Ma∗ , Jiaqi Liao∗ , Xinyang Wang∗ , Jingjing Wang∗ , Chaoran Feng, Zijing Hu, Chong Bao,  \nZichen Xi, Yuqi Gan, Weisen Wang, Yanhong Zeng, Qin Zhao, Zifan Shi, Wei Wu, Hao Ouyang, Qiuyu Wang, Shangzhan Zhang, Jiahao Shao, Yipengjing Sun, Liangxiao Hu, Lunke Pan, Nan Xue,  \nKecheng Zheng, Yinghao Xu, Xing Zhu, Yujun Shen, Ka Leong Cheng†  \n∗Equal Contribution †Project Lead  \nDespite the recent promise in robot control, video generative models suffer from a domain mismatch due to their primary focus on content creation. For example, their design inherently prioritizes visual fidelity and creativity over computational efficiency and physical realism. In this work, we present LingBot-Video, a DiT-based video pretraining paradigm specifically tailored for embodied intelligence. From the architecture perspective, we adopt the Mixture-of-Experts (MoE), instead of dense, framework to achieve a better trade-off between modeling capacity and inference efficiency, and manage to scale it up from scratch. From the data perspective, we construct a data profiling engine that augments standard internet videos with extensive robot-oriented footage, encompassing manipulation, navigation, and egocentric perspectives, to equip the base model with an intrinsic understanding of actions and world dynamics. From the training perspective, we develop a multi-dimensional reward system to enforce the alignment regarding physical rationality and task completion, going beyond standard criteria such as aesthetics, prompt-following, and motion consistency. Comprehensive evaluations validate its performance and efficiency as a video foundation model. We contribute LingBot-Video as the inaugural large-scale, open-source MoE video foundation model to the community, in a pioneering effort to bridge digital creativity and physical actuation.  \nWebsite: [https://technology.robbyant.com/lingbot-video](https://technology.robbyant.com/lingbot-video)  \nGithub: [https://github.com/robbyant/lingbot-video](https://github.com/robbyant/lingbot-video)  \nCheckpoints: [https://huggingface.co/robbyant/lingbot-video](https://huggingface.co/robbyant/lingbot-video)  \n1 Introduction  \nBeyond their success in content creation, diffusion-based [7, 9, 36, 37, 81, 89, 105] and autoregressive [32, 101, 119] video models have demonstrated remarkable ability to synthesize temporally coherent and photorealistic sequences conditioned on text, images, and other control signals [10, 23, 31, 92, 96] . This capability has motivated a growing body of work that interprets video models as implicit simulators ofthe physical world, enabling their use in robotics [2, 52, 55], autonomous driving [77, 78], and interactive environments [10] . In this paradigm, video models serve not only as generative systems but also as predictive world models [4] that support planning, policy learning, and imagination-based control. However, translating these models from passive video generation to active embodied reasoning and intelligence remains an open challenge.  \nDespite their promise, a fundamental gap persists between video generation models and embodied intelligence requirements. Most video foundation models are optimized for perceptual quality—such as realism, aesthetics, and text alignment—rather than physical correctness or controllability. While these objectives yield visually compelling results, they do not explicitly enforce consistency with physical interaction constraints, such as contact stability, rigid-body dynamics, or long-horizon state consistency under intervention. This highlights a key tension: while internet-scale video provides rich visual diversity, it does not guarantee fidelity to the constraints of embodied interaction.  \nFigure 1. Samples of Text-to-Image and Text-to-Video tasks generated by LingBot-Video. LingBot-Video can produce images and videos with high ","cbCaifu6Fgg0OZa1","https://ap.wps.com/l/cbCaifu6Fgg0OZa1","pdf",36766475,1,51,"English","en",105,"# Introduction\n## Video foundation models and the domain gap\n## LingBot-Video: architecture, data, training objectives","[{\"question\":\"What problem does LingBot-Video aim to solve in video generation for robotics?\",\"answer\":\"It addresses the mismatch between video models optimized for perceptual quality and the physical correctness and controllability requirements of embodied intelligence under intervention.\"},{\"question\":\"How does LingBot-Video improve efficiency and scalability?\",\"answer\":\"It adopts a Mixture-of-Experts (MoE) design to enable sparse conditional computation, balancing modeling capacity and inference efficiency while scaling up from scratch.\"},{\"question\":\"What training data and objectives does LingBot-Video use?\",\"answer\":\"It constructs a robot-augmented pretraining corpus combining internet videos with robot manipulation, navigation, and egocentric footage, and employs a multi-dimensional reward system that enforces physical rationality and task completion beyond aesthetics and motion consistency.\"}]",1784200476,129,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":27},"scaling-mixture-of-experts-video-pretraining-for-embodied-intelligence","",{"@graph":35,"@context":85},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/scaling-mixture-of-experts-video-pretraining-for-embodied-intelligence/85026/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What problem does LingBot-Video aim to solve in video generation for robotics?","Question",{"text":75,"@type":76},"It addresses the mismatch between video models optimized for perceptual quality and the physical correctness and controllability requirements of embodied intelligence under intervention.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does LingBot-Video improve efficiency and scalability?",{"text":80,"@type":76},"It adopts a Mixture-of-Experts (MoE) design to enable sparse conditional computation, balancing modeling capacity and inference efficiency while scaling up from scratch.",{"name":82,"@type":73,"acceptedAnswer":83},"What training data and objectives does LingBot-Video use?",{"text":84,"@type":76},"It constructs a robot-augmented pretraining corpus combining internet videos with robot manipulation, navigation, and egocentric footage, and employs a multi-dimensional reward system that enforces physical rationality and task completion beyond aesthetics and motion consistency.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":45,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":45,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":45,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":45,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":45,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":45,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]