[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-86297-en":3,"doc-seo-86297-105":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},86297,13056703020460,"Valentina","https://ap-avatar.wpscdn.com/avatar/be000253dac470eee5d?_k=1778207105932848923",8,"Research & Report","From World Action Models to Embodied Brains: A Roadmap for Open-World Physical Intelligence","Artificial general intelligence requires agents that can learn, reason, and act in the physical world. World Action Models (WAMs) connect candidate interventions to predicted physical consequences, enabling consequence-aware decisions, yet progress fails to accumulate. The work surveys the evolution toward WAMs and organizes barriers into three coupled gaps involving model roles and representations, objectives and standardization, and system composition. It proposes a co-evolution roadmap centered on an embodied brain, supported by grounding harnesses, shared contracts, and closed-loop post-training.","arXiv :2607 . 11689v1 [ cs .RO] 13 Jul 2026  \nFrom World Action Models to Embodied Brains: A Roadmap for Open-World Physical Intelligence  \nYuanzhi Liang, Xufeng Zhan, Haibin Huang, Chi Zhang, Xuelong Li  \nArtificial general intelligence ultimately requires agents that can learn, reason, and act in the physical world. Action models, vision-language-action policies, and world models have made this ambition concrete. World Action Models (WAMs) are especially promising because they connect candidate interventions to predicted physical consequences, supporting consequence-aware decisions. Yet progress does not accumulate readily. Many action-centric systems bind outputs to action spaces and controllers. Predictive systems expose different variables; datasets, tasks, and runtimes use incompatible conventions. Scaling parameters or trajectories alone does not resolve these interfaces. We survey the evolution toward WAMs and organize these limitations into three coupled gaps in model roles and representations, objectives and standardization, and system composition. We then propose a co-evolution roadmap for scalable physical intelligence. At its center is the embodied brain, along-term model target for physical reasoning. It integrates multimodal context, compares possible interventions, and communicates an intended state transition or capability request instead of actuator commands. Current WAMs provide promising prototypes for the predictive capabilities this model may require without fixing its final architecture. A physical harness grounds the brain output, resolves tools and controllers, verifies execution, and records the resulting trace. Shared contracts make heterogeneous data, tasks, and components comparable while preserving embodiment details. Closed-loop post-training turns verified interaction into reusable experience. By separating general physical reasoning from local execution, the roadmap makes components independently testable, mutually compatible, and jointly improvable. The resulting physical-intelligence stack offers an actionable path toward adaptive, self-improving agents and more general intelligence grounded in real-world interaction.  \nCorresponding Author: Xuelong Li(xuelong [li@ieee.org](li@ieee.org))  \n1 Introduction  \nA long-standing ambition of artificial intelligence is to build general-purpose agents that can perceive, reason, act, and improve across open-ended tasks. Recent progress in large language models and multimodal foundation models has made this ambition increasingly concrete in digital domains. Models can now summarize knowledge, write code, use tools, follow instructions, and participate in multi-step workflows. Yet a purely digital notion of intelligence remains incomplete. Human intelligence is not only linguistic or symbolic; it is grounded in space, time, objects, forces, affordances, and consequences. A system that aspires to generalpurpose agency must eventually move beyond predicting text or images and acquire the ability to understand and intervene in the physical world. In this sense, physical intelligence is not a peripheral application of AGI, but a central test of whether intelligent systems can connect perception, prediction, decision making, and action under real-world constraints.  \nThis view is consistent with several influential perspectives on the future of AI. The “bitter lesson”emphasizes general methods that continue to benefit from computation and data (Sutton, 2019) . Recent discussions of an “era of experience” further argue that future agents should learn not only from static human-generated corpora, but also from their own interaction traces and consequences (Silver & Sutton, 2025) . In parallel, world-model and spatial-intelligence perspectives argue that intelligent behavior requires  \ninternal representations of the external world that support prediction, planning, and control (LeCun, 2022; Li, 2025) . Together, these perspectives point to a common requirement: ","cbCaimHEh0F6sfxN","https://ap.wps.com/l/cbCaimHEh0F6sfxN","pdf",1583562,6,1,31,"English","en",105,"# Introduction\n## From digital intelligence to physical agency\n## Future AI perspectives: bitter lesson and era of experience\n## Embodied AI and robot learning\n## The scaling problem across embodiments","[{\"question\":\"What are World Action Models (WAMs) designed to achieve?\",\"answer\":\"WAMs connect candidate interventions to predicted physical consequences, enabling consequence-aware decisions in physical environments.\"},{\"question\":\"Why does the document argue that progress toward WAMs does not accumulate easily?\",\"answer\":\"Because many action-centric systems tightly bind outputs to action spaces and controllers, while predictive systems use incompatible variables, dataset/task/runtime conventions, and interface mappings that scaling alone cannot reconcile.\"},{\"question\":\"How does the proposed roadmap support scalable physical intelligence?\",\"answer\":\"It centers an embodied brain for long-term physical reasoning, grounds outputs with a physical harness that resolves tools/controllers and verifies execution, uses shared contracts for comparability, and applies closed-loop post-training to turn verified interaction into reusable experience.\"}]",1784210288,78,{"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},"from-world-action-models-to-embodied-brains-a-roadmap-for-open-world-physical-intelligence","",{"@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/from-world-action-models-to-embodied-brains-a-roadmap-for-open-world-physical-intelligence/86297/",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","2026-07-27","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 are World Action Models (WAMs) designed to achieve?","Question",{"text":76,"@type":77},"WAMs connect candidate interventions to predicted physical consequences, enabling consequence-aware decisions in physical environments.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"Why does the document argue that progress toward WAMs does not accumulate easily?",{"text":81,"@type":77},"Because many action-centric systems tightly bind outputs to action spaces and controllers, while predictive systems use incompatible variables, dataset/task/runtime conventions, and interface mappings that scaling alone cannot reconcile.",{"name":83,"@type":74,"acceptedAnswer":84},"How does the proposed roadmap support scalable physical intelligence?",{"text":85,"@type":77},"It centers an embodied brain for long-term physical reasoning, grounds outputs with a physical harness that resolves tools/controllers and verifies execution, uses shared contracts for comparability, and applies closed-loop post-training to turn verified interaction into reusable experience.","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},[94,98,102,106,111,115,120,123,128,131,135],{"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":107,"doc_module":4,"doc_module_name":46,"category_name":108,"show_sort_weight":109,"slug":110},5,"Comic",60,"comic",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},"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":107,"slug":138},19,"General","general"]