[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-81773-en":3,"doc-seo-81773-105":30,"detail-sidebar-cat-0-en-105":92},{"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":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},81773,549758252649,"Ivy","https://ap-avatar.wpscdn.com/avatar/8000253669c5317157?_k=1778319167496531819",8,"Research & Report","MuSix: Multi-scale Mixture of World Models for Embodied Agents in Evolving Environments","Embodied agents deployed in the real world need multiscale reasoning and continual knowledge adaptation under changing conditions. Mixture of Experts (MoE) is challenged by scale-agnostic routing that blocks targeted scale updates, and by a single uniform update policy that cannot match how quickly knowledge at each scale becomes outdated. MuSix addresses these issues with scale-aware world model mixture and evolution: two-stage scale routing using experiential distance and scale-dependent forgetting with gated inter-scale transfer. Experiments on EmbodiedBench and HAZARD show improvements over state-of-the-art baselines.","arXiv :2607 .00457v 1 [ cs .AI] 1 Jul 2026  \nMulti-scale Mixture of World Models for Embodied Agents in Evolving Environments  \nJinwoo Jang 1, Daniel J. Rho 1, Sihyung Yoon 1, Hyunsuk Cho 1, and  \nHonguk Woo 1 ,2  \n1 Sungkyunkwan University  \n{jinustar,danielrho,godboy3752,hscho5133,[hwoo}@skku.edu](hwoo}@skku.edu)  \n2 Corresponding author  \nAbstract. Embodied agents operating in the real world require multiscale reasoning and knowledge adaptation as conditions change. We identify two challenges in applying Mixture of Experts (MoE) to this setting:  \nrouting lacks an explicit notion of scale, preventing targeted updates at specific scales, and a uniform update policy cannot accommodate the different rates at which knowledge at each scale becomes outdated. We present MuSix, a framework that addresses both challenges through scale-aware world model mixture and evolution. A two-stage routing mechanism first maps experiential distance, a measure of situational novelty inspired by Construal Level Theory, to a weight over continuous scale space via a meta-router, then selects world models within the identified scale. For adaptation, scale-dependent forgetting rates allow low-scale knowledge to refresh rapidly while high-scale abstractions persist, and gated inter-scale transfer maintains coherence across the hierarchy. Experiments on EmbodiedBench and HAZARD show that MuSix improves over state-of-the-art baselines.  \nKeywords: Embodied AI · World model · Mixture of experts · Testtime training  \n1 Introduction  \nEmbodied agents powered by vision-language models (VLMs) have achieved significant progress in complex instruction following [7, 8, 11], yet real-world deployment demands multi-scale reasoning, from low-level physical dynamics to high-level abstract inference, while continuously adapting to non-stationary environments. Recent work has explored hierarchical world models that decompose predictions across scales [2,6,20], but these approaches do not address how world models should adapt when environmental conditions change over time.  \nMixture of Experts (MoE) [9] offers a natural substrate for these demands: its modular expert selection can accommodate qualitatively different knowledge types, while its selective activation enables targeted updates without disrupting other components, effectively mitigating catastrophic forgetting [12, 18] . However, conventional MoE is not scale-aware, introducing two limitations (Figure 1) . First, standard routing operates without an explicit notion of scale, so  \n2 Jang et al.  \nFig. 1: Explanation of mixture challenge and evolution challenge on conventional MoE  \nworld model selection is not tied to any identifiable scale, precluding test-time updates that target only the relevant scale (mixture challenge) . Second, a single uniform update policy cannot respect the fact that low-level knowledge about local dynamics changes frequently while high-level abstract rules remain relatively stable, preventing each scale from evolving at its own appropriate rate (evolution challenge) .  \nTo address these limitations, we propose MuSix, a multi-scale mixture of world models framework that enables embodied agents to dynamically mix and evolve world models at different scales. For the mixture challenge, we introduce two-stage scale-based routing that explicitly separates scale determination from world model selection, yielding transparent routing in which the identified scale directly determines how knowledge increments are distributed across groups. For the evolution challenge, we propose intra-and inter-scale knowledge adaptation mechanisms that allow each scale to evolve at its own characteristic rate while maintaining coherence across the hierarchy. To ground scale selection in a principled criterion, we draw on Construal Level Theory (CLT) [16], which posits that psychological distance governs the level of abstraction in human reasoning, and operationalize this principle through experiential distan","cbCaisGzVJJ2pqCZ","https://ap.wps.com/l/cbCaisGzVJJ2pqCZ","pdf",11556344,5,1,26,"English","en",105,"# Introduction\n# Related Works","[{\"question\":\"What problems does MuSix identify when applying Mixture of Experts to embodied agents?\",\"answer\":\"MuSix highlights two limitations: conventional MoE routing lacks an explicit notion of scale, and it uses a uniform update policy that cannot evolve each scale at its appropriate rate as knowledge becomes outdated differently over time.\"},{\"question\":\"How does MuSix determine which scale to use for routing?\",\"answer\":\"MuSix uses a two-stage routing mechanism where a meta-router maps experiential distance—a measure of situational novelty inspired by Construal Level Theory—onto a continuous scale space, then selects world models within the identified scale.\"},{\"question\":\"How does MuSix adapt knowledge over time across scales?\",\"answer\":\"MuSix applies scale-dependent forgetting rates so low-scale knowledge refreshes rapidly while high-scale abstractions persist, and it uses gated inter-scale transfer to maintain coherence across the hierarchy.\"}]",1784176060,66,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":87,"head_meta":89,"extra_data":91,"updated_unix":28},"musix-multi-scale-mixture-of-world-models-for-embodied-agents-in-evolving-environments","",{"@graph":36,"@context":86},[37,54,69],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":43,"position":53},"https://docshare.wps.com/document/musix-multi-scale-mixture-of-world-models-for-embodied-agents-in-evolving-environments/81773/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":24,"description":14,"dateModified":62,"datePublished":63,"encodingFormat":61,"isAccessibleForFree":64,"interactionStatistic":65},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":41,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-07-25","2026-07-16",true,{"@type":66,"interactionType":67,"userInteractionCount":20},"InteractionCounter",{"@type":68},"ViewAction",{"@type":70,"mainEntity":71},"FAQPage",[72,78,82],{"name":73,"@type":74,"acceptedAnswer":75},"What problems does MuSix identify when applying Mixture of Experts to embodied agents?","Question",{"text":76,"@type":77},"MuSix highlights two limitations: conventional MoE routing lacks an explicit notion of scale, and it uses a uniform update policy that cannot evolve each scale at its appropriate rate as knowledge becomes outdated differently over time.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"How does MuSix determine which scale to use for routing?",{"text":81,"@type":77},"MuSix uses a two-stage routing mechanism where a meta-router maps experiential distance—a measure of situational novelty inspired by Construal Level Theory—onto a continuous scale space, then selects world models within the identified scale.",{"name":83,"@type":74,"acceptedAnswer":84},"How does MuSix adapt knowledge over time across scales?",{"text":85,"@type":77},"MuSix applies scale-dependent forgetting rates so low-scale knowledge refreshes rapidly while high-scale abstractions persist, and it uses gated inter-scale transfer to maintain coherence across the hierarchy.","https://schema.org",{"og:url":52,"og:type":88,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":90,"canonical":52},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":93},[94,98,102,106,110,115,120,123,128,131,135],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":95,"show_sort_weight":96,"slug":97},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":99,"show_sort_weight":100,"slug":101},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":46,"category_name":103,"show_sort_weight":104,"slug":105},"Exam",70,"exam",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":20,"slug":138},19,"General","general"]