[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-83090-en":3,"doc-seo-83090-105":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},83090,1099514067415,"Rowan","https://ap-avatar.wpscdn.com/avatar/100002539d78ffe74a7?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779092875211072502",6,"Technology","A Definition and Roadmap for World Models","World models are internal simulators that learn an environment’s structure and dynamics, enabling research spanning model-based reinforcement learning, video generation, and embodied robotics toward physical AI. The article addresses the lack of consensus on what world models fundamentally are, what they should predict, and how they should be built. It provides a scientific definition, analyzes key technical properties, and presents a staged roadmap for developing effective world models across architectures, training paradigms, and application domains.","arXiv :2607 .0640 1v 1 [ cs .AI ] 7 Jul 2026  \nA Definition and Roadmap for World Models  \nPhysical Intelligence Team, Shanghai AI Laboratory  \nAbstract  \nWorld models—internal simulators that learn the structure and dynamics of an environment—have become one of the most actively debated concepts in AI. From model-based reinforcement learning and video generation to embodied roboticsand ultimately, physical AI, researchers across AI subfields are building systems that they call “world models”, yet there is no consensus on what a world model fundamentally is, what it should predict, or how it should be built. This perspective article provides a scientific definition of world models, discussions of their key technical aspects, and a staged roadmap for developing effective world models.  \nContents  \n1 Introduction 2  \n2 What Is a World Model? 4  \n2.1 Definition and Major Properties ............................... 4  \n2.2 The Agent-Environment Loop ................................ 6  \n2.3 Understanding vs. Predicting: Two Views of the World Model ............. 7  \n2.4 A Functional Taxonomy: Renderers, Simulators, Planners ................ 8  \n2.5 A Two-Dimensional Taxonomy: Functions and Architectures .............. 9  \n2.6 Extending World Models Beyond Physical Environments ................ 10  \n3 Architectural Paradigms 12  \n3.1 Observation-Level Generative World Models ....................... 12  \n3.2 Latent-Space World Models ................................. 13  \n3.3 3D-Enhanced and Object-Centric World Models ..................... 14  \n3.4 The Unification Trend: Omnimodal World Models .................... 15  \n4 Training and Learning Paradigms 16  \n4.1 Self-Supervised and Generative Pretraining ........................ 16  \n4.2 Model-Based Reinforcement Learning ........................... 18  \n4.3 Policy Learning Inside World Models ........................... 21  \n4.4 Chain-of-Imagination: Reasoning Through World Models ................ 22  \n4.5 Physics-Informed and Constrained Learning ....................... 23  \n4.6 Counterfactual Reasoning .................................. 25  \n4.7 Long-Horizon and Hierarchical Planning ......................... 26  \n5 Application Domains 27  \n5.1 Robotics and Embodied AI .................................. 27  \n5.2 Scientific Discovery ...................................... 29  \n6 Open Challenges and Bottlenecks 30  \n6.1 Data Asymmetry ....................................... 31  \n6.2 Fidelity vs. Precision ..................................... 31  \n6.3 Compounding Prediction Errors .............................. 32  \n6.4 Sim-to-Real Transfer ..................................... 33  \n6.5 Evaluation and Benchmarks ................................. 33  \n6.6 Safety, Transparency, and Sustainability .......................... 34  \n7 Roadmap 36  \n7.1 Towards Unified Multimodal World Models ....................... 36  \n7.2 Towards a Unified Physical Representation ........................ 37  \n7.3 Foundation-Scale Interactive Simulators .......................... 38  \n8 Outlook: A Path to Physical AGI 38  \n1 Introduction  \nThe term “world model” has accumulated a diverse range of contextual usages that conceals as much as it reveals. As early as in 1943, Craik (1943) proposed that biological organisms survive  \nand thrive by holding “working models” of the physical world within their minds. Rather than relying exclusively on slow, trial-and-error physical interactions, these models enable the mind to simulate hypothetical scenarios, predict the outcomes of candidate actions, and precompute optimal strategies—this algorithmic instantiation represents the formal realization of what we today term the “world action model”. Craik’s hypothesis underpins some important AI architectures such as reinforcement learning agents and generative models (Sutton and Barto, 2018) . The concept has been invoked across computer vision, robotics, and generative modeling, often to describe very different systems","cbCaig6tio2aMDry","https://ap.wps.com/l/cbCaig6tio2aMDry","pdf",8964133,1,58,"English","en",105,"# Introduction\n# What Is a World Model?\n## Definition and Major Properties\n## The Agent-Environment Loop\n## Understanding vs. Predicting: Two Views of the World Model\n## A Functional Taxonomy: Renderers, Simulators, Planners\n## A Two-Dimensional Taxonomy: Functions and Architectures\n## Extending World Models Beyond Physical Environments\n# Architectural Paradigms\n## Observation-Level Generative World Models\n## Latent-Space World Models\n## 3D-Enhanced and Object-Centric World Models\n## The Unification Trend: Omnimodal World Models\n# Training and Learning Paradigms\n## Self-Supervised and Generative Pretraining\n## Model-Based Reinforcement Learning\n## Policy Learning Inside World Models\n## Chain-of-Imagination: Reasoning Through World Models\n## Physics-Informed and Constrained Learning\n## Counterfactual Reasoning\n## Long-Horizon and Hierarchical Planning\n# Application Domains\n## Robotics and Embodied AI\n## Scientific Discovery\n# Open Challenges and Bottlenecks\n## Data Asymmetry\n## Fidelity vs. Precision\n## Compounding Prediction Errors\n## Sim-to-Real Transfer\n## Evaluation and Benchmarks\n## Safety, Transparency, and Sustainability\n# Roadmap\n## Towards Unified Multimodal World Models\n## Towards a Unified Physical Representation\n## Foundation-Scale Interactive Simulators\n# Outlook: A Path to Physical AGI","[{\"question\":\"What problem does the article address about world models?\",\"answer\":\"It highlights that major AI subfields use “world models” differently, with no agreement on their fundamental definition, prediction targets, or construction principles.\"},{\"question\":\"How does the article frame the core idea behind world models?\",\"answer\":\"World models act as internal simulators that learn environment dynamics, letting agents simulate hypothetical outcomes of actions and support planning through predicted consequences.\"},{\"question\":\"What are key limitations and open challenges mentioned for building effective world models?\",\"answer\":\"The article discusses issues such as data asymmetry, the tradeoff between fidelity and precision, compounding prediction errors, sim-to-real transfer, and difficulties in evaluation, safety, transparency, and sustainability.\"}]",1784185148,146,{"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},"a-definition-and-roadmap-for-world-models","",{"@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/a-definition-and-roadmap-for-world-models/83090/",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 the article address about world models?","Question",{"text":75,"@type":76},"It highlights that major AI subfields use “world models” differently, with no agreement on their fundamental definition, prediction targets, or construction principles.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does the article frame the core idea behind world models?",{"text":80,"@type":76},"World models act as internal simulators that learn environment dynamics, letting agents simulate hypothetical outcomes of actions and support planning through predicted consequences.",{"name":82,"@type":73,"acceptedAnswer":83},"What are key limitations and open challenges mentioned for building effective world models?",{"text":84,"@type":76},"The article discusses issues such as data asymmetry, the tradeoff between fidelity and precision, compounding prediction errors, sim-to-real transfer, and difficulties in evaluation, safety, transparency, and sustainability.","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},[93,97,101,105,110,113,118,123,128,131,135],{"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":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":111,"slug":112},50,"technology",{"id":114,"doc_module":4,"doc_module_name":45,"category_name":115,"show_sort_weight":116,"slug":117},7,"Healthcare",40,"healthcare",{"id":119,"doc_module":4,"doc_module_name":45,"category_name":120,"show_sort_weight":121,"slug":122},8,"Research & Report",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"]