[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85379-en":3,"doc-seo-85379-105":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},85379,7971461741311,"Ophelia","https://ap-avatar.wpscdn.com/avatar/74000253aff267980c6?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779345379180704826",6,"Technology","Mixture of Frames Policy Multi-Frame Action Denoising for Bimanual Mobile Manipulation","Robotic manipulation is inherently multi-frame, with different coordinate frames making different action components easier to model. Existing diffusion-based visuomotor policies often commit to a single predefined action frame, forcing one denoiser to learn unnecessarily complex action distributions. Mixture of Frames Policy (MoF) performs synchronized denoising across multiple coordinate frames using a shared canonical diffusion state, frame-specialized experts, and fused noise predictions. A column-based 6D SE(3) rotation representation enables exact differentiable frame transforms during intermediate noisy states. MoF improves over oracle frame selection and standard Mixture-of-Experts baselines on nine simulated tasks and two real-world bimanual mobile manipulation tasks.","arXiv :2607 . 11884v1 [ cs .RO] 13 Jul 2026  \nMixture of Frames Policy: Multi-Frame Action Denoising for Bimanual Mobile Manipulation  \nDian Wang∗ Jisang Park∗ Xiaomeng Xu Han Zhang Shuran Song† Jeannette Bohg†  \nStanford University  \n[https://mofpo.github.io](https://mofpo.github.io)  \nFigure 1: Mixture of Frames Policy (MoF). The complexity of a manipulation action distribution depends strongly on the coordinate frame in which it is represented. The same relative motion can be compact and invariant in one frame (e.g. an end-effector frame), yet form a broader and harder-to-learn distribution in another. Conversely, some motions, such as keeping an object upright, are better represented in a base-aligned frame. Since no single frame is universally optimal, MoF trains frame-specialized experts and fuses their predictions in a shared canonical frame.  \nAbstract: Robotic manipulation is inherently multi-frame: local actions may be simple in an end-effector frame, while transport, upright-object handling, and whole-body coordination are better represented in a base-aligned frame. However, modern diffusion-based visuomotor policies typically commit to a single predefined action frame, forcing one denoiser to model action distributions that are often unnecessarily complex in that frame. We propose Mixture of Frames Policy (MoF), a diffusion policy that performs synchronized action denoising across multiple coordinate frames. MoF maintains a single canonical diffusion state, re-expresses it in several task-relevant frames, applies frame-specialized denoisers, and fuses their noise predictions back in the canonical frame. To make this possible for intermediate noisy diffusion states, we introduce a column-based 6D rotation representation within an SE(3) action parameterization that supports exact, differentiable frame transformations without requiring noisy rotations to lie on the SO(3) manifold. Across nine simulated bimanual manipulation tasks, we show  \n∗ Indicates equal contribution,† indicates equal advising.  \nthat the best action frame is task-dependent and that MoF improves over oracle frame selection and standard Mixture-of-Experts (MoE) baselines. We further evaluate MoF on two real-world bimanual mobile manipulation tasks, demonstrating that it outperforms all constituent single-frame baselines. Project homepage:  \n[https://mofpo.github.io](https://mofpo.github.io).  \nKeywords: Bimanual Mobile Manipulation, Visuomotor Policy Learning  \n1 Introduction  \nRobotic manipulation systems are inherently multi-frame. A bimanual mobile robot naturally operates with a base frame, left and right end-effector frames, camera frames, and other task-relevant coordinate frames defined by its kinematics and embodiment. These frames are not only alternative parameterizations of the same state, but can also make different parts of the same task easier to model. For example, the action of approaching and grasping a cup handle is more naturally expressed in the gripper frame: if the cup and gripper appear in the same relative configuration, the desired gripper frame action is invariant, even if the gripper-object pair is in different absolute poses relative to the base frame (Fig. 1 top) . In contrast, when the robot transports a filled cup, maintaining the cup upright and moving it coherently with the mobile base is more naturally expressed in the base frame.  \nThis frame choice matters because different tasks can be easier to learn when the action is represented in a specific frame. However, modern visuomotor policies such as Diffusion Policies [1] typically choose a single predefined coordinate frame to parameterize the action. Even when transformations among frames are provided as proprioceptive observations, the denoiser must still model the entire action distribution in one chosen frame. Moreover, the most useful frame might vary while a manipulation task is progressing. A single-frame policy therefore inherits whatever complexity the cho","cbCaitsQp2e8dumG","https://ap.wps.com/l/cbCaitsQp2e8dumG","pdf",8178597,1,20,"English","en",105,"# Introduction\n## Multi-frame nature of bimanual mobile manipulation\n## Limitation of single-frame diffusion policies\n## Mixture of Frames (MoF) approach\n## Contributions","[{\"question\":\"Why does representing actions in a single coordinate frame hurt diffusion-based visuomotor policies?\",\"answer\":\"Different tasks—and even different stages of the same task—can be easier to learn in different frames. A single-frame policy forces the denoiser to model the full action distribution in one parameterization, inheriting unnecessary complexity imposed by that chosen frame.\"},{\"question\":\"How does MoF perform multi-frame action denoising?\",\"answer\":\"MoF keeps one canonical diffusion state, re-expresses it in several task-relevant coordinate frames, runs frame-specialized expert denoisers, and fuses their noise predictions back into the canonical frame at each denoising step.\"},{\"question\":\"What rotation representation enables MoF’s frame transformations during noisy diffusion states?\",\"answer\":\"MoF introduces a column-based 6D rotation representation within an SE(3) action parameterization, supporting exact, differentiable frame transformations. This avoids requiring noisy rotations to lie on the SO(3) manifold.\"}]",1784202998,50,{"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},"mixture-of-frames-policy-multi-frame-action-denoising-for-bimanual-mobile-manipulation","",{"@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/technology/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/mixture-of-frames-policy-multi-frame-action-denoising-for-bimanual-mobile-manipulation/85379/",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},"Why does representing actions in a single coordinate frame hurt diffusion-based visuomotor policies?","Question",{"text":75,"@type":76},"Different tasks—and even different stages of the same task—can be easier to learn in different frames. A single-frame policy forces the denoiser to model the full action distribution in one parameterization, inheriting unnecessary complexity imposed by that chosen frame.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does MoF perform multi-frame action denoising?",{"text":80,"@type":76},"MoF keeps one canonical diffusion state, re-expresses it in several task-relevant coordinate frames, runs frame-specialized expert denoisers, and fuses their noise predictions back into the canonical frame at each denoising step.",{"name":82,"@type":73,"acceptedAnswer":83},"What rotation representation enables MoF’s frame transformations during noisy diffusion states?",{"text":84,"@type":76},"MoF introduces a column-based 6D rotation representation within an SE(3) action parameterization, supporting exact, differentiable frame transformations. This avoids requiring noisy rotations to lie on the SO(3) manifold.","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},[93,97,101,105,110,112,117,122,126,129,133],{"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":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":28,"slug":111},"technology",{"id":113,"doc_module":4,"doc_module_name":45,"category_name":114,"show_sort_weight":115,"slug":116},7,"Healthcare",40,"healthcare",{"id":118,"doc_module":4,"doc_module_name":45,"category_name":119,"show_sort_weight":120,"slug":121},8,"Research & Report",30,"research-report",{"id":123,"doc_module":4,"doc_module_name":45,"category_name":124,"show_sort_weight":21,"slug":125},9,"Religion & Spirituality","religion-spirituality",{"id":21,"doc_module":4,"doc_module_name":45,"category_name":127,"show_sort_weight":21,"slug":128},"World Cup","world-cup",{"id":130,"doc_module":4,"doc_module_name":45,"category_name":131,"show_sort_weight":130,"slug":132},10,"Lifestyle","lifestyle",{"id":134,"doc_module":4,"doc_module_name":45,"category_name":135,"show_sort_weight":106,"slug":136},19,"General","general"]