[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-81651-en":3,"doc-seo-81651-105":30,"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":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},81651,16904993612988,"Olivia Brown","https://ap-avatar.wpscdn.com/davatar_a8503ba1806abce46bf441b54a3ca4cd",8,"Research & Report","Embodied Multi-Agent Coordination by Aligning World Models Through Dialogue","Effective collaboration between embodied agents depends on communication that is grounded in each agent’s evolving understanding of the world, not only on acting in a shared environment. With partial observability, coordination without communication is provably difficult, but dialogue may align agents’ world models by sharing observations. This work extends PARTNR with a natural-language dialogue channel, then evaluates whether dialogue causes genuine world-model alignment using observation convergence, information novelty, and belief-sensitive messaging.","Embodied Multi-Agent Coordination by Aligning World Models Through  \nDialogue  \nVardhan Dongre & Dilek Hakkani-Tür  \nSiebel School of Computing & Data Science  \nUniversity of Illinois Urbana-Champaign  \n{vdongre2, [dilek}@illinois.edu](dilek}@illinois.edu)  \narXiv :2605 . 12920v3 [ cs .MA] 10 Jul 2026  \nAbstract  \nEffective collaboration between embodied agents requires more than acting in a shared environment; it demands communication grounded in each agent’s evolving understanding of the world. When agents can only partially observe their surroundings, coordination without communication is provably hard, but communication can, in principle, bridge this gap by allowing agents to share observations and align their world models. In this work, we examine whether LLM-based embodied agents actually realize the ability to communicate. We extend PARTNR, a benchmark for collaborative household robotics, with a natural-language dialogue channel that enables two agents with partial observability to communicate during task execution. To evaluate whether dialogue leads to genuine world-model alignment rather than superficial coordination, we propose a framework for measuring worldmodel alignment defined over per-agent world graphs: observation convergence (do private world models align over time?), information novelty (do messages convey what the partner lacks?), and belief-sensitive messaging (do agents model what their partner knows?) . Our experiments across three LLMs reveal that dialogue reduces action conflict by 41–93 percentage points but degrades task success relative to silent coordination. Using our metrics, we characterize the gap between superficial coordination and genuine world-model alignment, and identify where current models fall on this spectrum. Project Website  \n1 Introduction  \nWhen humans coordinate on a shared physical task, they naturally benefit from language. They negotiate responsibilities, share what they observe, flag unexpected obstacles, and revise their joint plan as new information arrives (Clark, 1996 ; Tomasello et al., 2005) . Every utterance carries an implicit model of the listener: what they already know, what  \nthey need to hear, and what they will do next. Effective collaboration is not about exchanging information; it is about using language to align world models so that two decentralized minds can plan as though they shared one. Central to this process is Theory of Mind (Premack and Woodruff, 1978) where each partner continuously models what the other knows, doesn’t know, what they intend, and uses language to close the gaps (Grosz and Kraus, 1996 ; Frank and Goodman, 2012) . Therefore, in problems grounded in a physical (or simulated) world, effective joint planning is not about exchanging arbitrary information; it is about aligning individual world models so that decentralized agents can plan as one.  \nThis intuition has a precise formal counterpart in joint planning with multiple decentralized agents. Decentralized partially observable Markov decision processes (Dec-POMDPs) with two or more agents are provably intractable (NEXP-complete)(Bernstein et al., 2002), a double-exponential jump over the single-agent case. But with an effective communication channel, the problem collapses towards centralized planning, where a single coordinator sees all observations (which is only PSPACEcomplete) 1 (Goldman and Zilberstein, 2004, 2008) . The reduction happens because a shared communication channel lets two agents’ partial observations be combined into a joint view; reducing the problem to its centralized form, where a single planner has access to both agents’ observations.  \nLarge language models now routinely serve as planners for embodied agents (Zhang and Soh, 2023 ; Mandi et al., 2024 ; Zu et al., 2025 ; Gonget al., 2024 ; Dongre et al., 2025), and on text-based benchmarks they reportedly approach human per-  \n1NEXP-complete problems require time exponential in an exponential function of input s","cbCaitf18PF0D89w","https://ap.wps.com/l/cbCaitf18PF0D89w","pdf",2443849,2,1,14,"English","en",105,"# Introduction\n## Contributions","[{\"question\":\"What problem does the document address about embodied multi-agent coordination?\",\"answer\":\"It addresses why effective collaboration requires language-based communication that aligns agents’ world models, especially under partial observability where coordination without communication is provably hard.\"},{\"question\":\"How does the work extend the PARTNR benchmark?\",\"answer\":\"It extends PARTNR with a natural-language dialogue channel so two embodied LLM agents with partial observability can exchange messages during task execution.\"},{\"question\":\"How does the paper measure whether dialogue leads to genuine world-model alignment?\",\"answer\":\"It proposes metrics based on per-agent world graphs, including observation convergence, information novelty, and belief-sensitive messaging to distinguish alignment from superficial coordination.\"}]",1784175175,35,{"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":86,"head_meta":88,"extra_data":90,"updated_unix":28},"embodied-multi-agent-coordination-by-aligning-world-models-through-dialogue","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,47,50],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":20},"https://docshare.wps.com/document/","Document",{"item":48,"name":12,"@type":43,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":43,"position":52},"https://docshare.wps.com/document/embodied-multi-agent-coordination-by-aligning-world-models-through-dialogue/81651/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-24","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 document address about embodied multi-agent coordination?","Question",{"text":75,"@type":76},"It addresses why effective collaboration requires language-based communication that aligns agents’ world models, especially under partial observability where coordination without communication is provably hard.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does the work extend the PARTNR benchmark?",{"text":80,"@type":76},"It extends PARTNR with a natural-language dialogue channel so two embodied LLM agents with partial observability can exchange messages during task execution.",{"name":82,"@type":73,"acceptedAnswer":83},"How does the paper measure whether dialogue leads to genuine world-model alignment?",{"text":84,"@type":76},"It proposes metrics based on per-agent world graphs, including observation convergence, information novelty, and belief-sensitive messaging to distinguish alignment from superficial coordination.","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":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"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":106,"slug":138},19,"General","general"]