[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-seo-657308-105":3,"detail-sidebar-cat-0-en-105":80,"doc-detail-657308-en":130},{"code":4,"msg":5,"data":6},0,"ok",{"site_id":7,"language":8,"slug":9,"title":10,"keywords":11,"description":12,"schema_data":13,"social_meta":73,"head_meta":75,"extra_data":77,"updated_unix":79},105,"en","generative-floor-plan-design-with-llms-via-reinforcement-learning-with-verifiable-rewards-abstract-and-introduction","Generative Floor Plan Design with LLMs via Reinforcement Learning with Verifiable Rewards - Abstract and Introduction","","An AI system for professional floor plan design must precisely control room dimensions and areas while respecting desired connectivity between rooms and maintaining functional and aesthetic quality. Existing generative approaches prioritize connectivity but fail to enforce numerical constraints. This work introduces text-based floor plan generation using an LLM fine-tuned on real plans and reinforcement learning with verifiable rewards to improve adherence to topological and numerical requirements while reducing invalid or overlapping layouts through constraint adherence metrics.",{"@graph":14,"@context":72},[15,34,55],{"@type":16,"itemListElement":17},"BreadcrumbList",[18,23,27,31],{"item":19,"name":20,"@type":21,"position":22},"https://docshare.wps.com","Home","ListItem",1,{"item":24,"name":25,"@type":21,"position":26},"https://docshare.wps.com/document/","Document",2,{"item":28,"name":29,"@type":21,"position":30},"https://docshare.wps.com/document/research-report/","Research & Report",3,{"item":32,"name":10,"@type":21,"position":33},"https://docshare.wps.com/document/generative-floor-plan-design-with-llms-via-reinforcement-learning-with-verifiable-rewards-abstract-and-introduction/657308/",4,{"url":32,"name":10,"@type":35,"image":36,"author":41,"headline":10,"publisher":44,"fileFormat":47,"inLanguage":8,"description":12,"dateModified":48,"datePublished":49,"encodingFormat":47,"isAccessibleForFree":50,"interactionStatistic":51},"DigitalDocument",{"url":37,"@type":38,"width":39,"height":40},"https://docshare.wps.com/thumbnails/generative-floor-plan-design-with-llms-via-reinforcement-learning-with-verifiable-rewards-abstract-and-introduction/657308.png","ImageObject",300,407,{"name":42,"@type":43},"Tirek","Person",{"url":19,"name":45,"@type":46},"DocShare","Organization","application/pdf","2026-10-30","2026-10-26",true,{"@type":52,"interactionType":53,"userInteractionCount":33},"InteractionCounter",{"@type":54},"ViewAction",{"@type":56,"mainEntity":57},"FAQPage",[58,64,68],{"name":59,"@type":60,"acceptedAnswer":61},"What problem does the approach address in generative floor plan design?","Question",{"text":62,"@type":63},"It targets the need to control both room connectivity and numerical constraints such as exact room sizing, which prior connectivity-focused methods do not reliably enforce.","Answer",{"name":65,"@type":60,"acceptedAnswer":66},"How does the method generate floor plans while enforcing constraints?",{"text":67,"@type":63},"It fine-tunes a large language model on real plans, then applies reinforcement learning with verifiable rewards to increase adherence to topological and numerical constraints while discouraging invalid or overlapping outputs.",{"name":69,"@type":60,"acceptedAnswer":70},"How are the generated plans evaluated?",{"text":71,"@type":63},"The document defines constraint adherence metrics that measure alignment between generated floor plans and user-defined constraints, and it reports improvements across Realism, Compatibility, and Diversity metrics.","https://schema.org",{"og:url":32,"og:type":74,"og:title":10,"og:site_name":45,"og:description":12},"article",{"robots":76,"canonical":32},"index,follow",{"doc_id":78,"site_id":7},657308,1792977263,{"code":4,"msg":81,"data":82},"success",[83,87,91,95,100,105,110,114,119,122,126],{"id":22,"doc_module":4,"doc_module_name":25,"category_name":84,"show_sort_weight":85,"slug":86},"Story & Novel",90,"story-novel",{"id":26,"doc_module":4,"doc_module_name":25,"category_name":88,"show_sort_weight":89,"slug":90},"Literature",80,"literature",{"id":33,"doc_module":4,"doc_module_name":25,"category_name":92,"show_sort_weight":93,"slug":94},"Exam",70,"exam",{"id":96,"doc_module":4,"doc_module_name":25,"category_name":97,"show_sort_weight":98,"slug":99},5,"Comic",60,"comic",{"id":101,"doc_module":4,"doc_module_name":25,"category_name":102,"show_sort_weight":103,"slug":104},6,"Technology",50,"technology",{"id":106,"doc_module":4,"doc_module_name":25,"category_name":107,"show_sort_weight":108,"slug":109},7,"Healthcare",40,"healthcare",{"id":111,"doc_module":4,"doc_module_name":25,"category_name":29,"show_sort_weight":112,"slug":113},8,30,"research-report",{"id":115,"doc_module":4,"doc_module_name":25,"category_name":116,"show_sort_weight":117,"slug":118},9,"Religion & Spirituality",20,"religion-spirituality",{"id":117,"doc_module":4,"doc_module_name":25,"category_name":120,"show_sort_weight":117,"slug":121},"World Cup","world-cup",{"id":123,"doc_module":4,"doc_module_name":25,"category_name":124,"show_sort_weight":123,"slug":125},10,"Lifestyle","lifestyle",{"id":127,"doc_module":4,"doc_module_name":25,"category_name":128,"show_sort_weight":96,"slug":129},19,"General","general",{"code":4,"msg":81,"data":131},{"doc_id":78,"user_id":132,"nickname":42,"user_avatar":133,"doc_module":4,"category_id":111,"category_name":29,"doc_title":10,"doc_description":12,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":33,"is_deleted":4,"is_public":22,"is_downloadable":22,"audit_status":22,"page_count":139,"language":140,"language_code":8,"site_id":7,"html_lang":8,"table_of_contents":141,"faqs":142,"seo_title":143,"seo_description":12,"update_tm":79,"read_time":108},5909896572756,"https://ap-avatar.wpscdn.com/davatar_155a257f0dc6eb9ab79c44ca47cae57d","Generative Floor Plan Design with LLMs via Reinforcement Learning with  \nVerifiable Rewards  \nLuis Lara1 , Aristides Milios1,2 , Zhi Hao Luo1,3 , Aditya Sharma1,3 , Ge Ya Luo1,2 , Christopher Beckham1 , Florian Golemo1 , Christopher Pal1,2,3,4  \n1Mila – Quebec AI Institute, 2Université de Montréal,  \n3Polytechnique Montréal, 4 Canada CIFAR AI Chair  \nCorrespondence: [luis.lara@mila.quebec](luis.lara@mila.quebec)  \narXiv :2605 . 14117v1 [ cs .CL] 13 May 2026  \nAbstract  \nAn AI system for professional floor plan design must precisely control room dimensions and areas while respecting the desired connectivity between rooms and maintaining functional and aesthetic quality. Existing generative approaches focus primarily on respecting the requested connectivity between rooms, but do not support generating floor plans that respect numerical constraints. We introduce a text-based floor plan generation approach that fine-tunes a large language model (LLM) on real plans and then applies reinforcement learning with verifiable rewards (RLVR) to improve adherence to topological and numerical constraints while discouraging invalid or overlapping outputs. Furthermore, we design a set of constraint adherence metrics to systematically measure how generated floor plans align with user-defined constraints. Our model generates floor plans that satisfy user-defined connectivity and numerical constraints and outperforms existing methods on Realism, Compatibility, and Diversity metrics. Across all tasks, our approach achieves at least a 94% relative reduction in Compatibility compared with existing methods.1 Our results demonstrate that LLMs can effectively handle constraints in this setting, suggesting broader applications for text-based generative modeling.  \n1 Introduction  \nGenerative models have emerged as powerful tools for accelerating design across different sectors. However, their widespread adoption depends on their ability to strike a balance between flexibility and precise control. This challenge is particularly pronounced in floor plan generation, where users often need to specify strict constraints, including exact room sizing and connectivity. To address this, we propose a model that handles user-defined  \n1Project code is available at [https://github.com/lud](https://github.com/lud)olara/floor-plan-rlvr  \nconstraints, and we introduce evaluation metrics that measure how well generated floor plans satisfy those constraints.  \nMost existing generative modeling techniques for floor plan design rely on top-down 2D renderings and use input graphs, referred to as bubble diagrams, to represent the spatial connectivity between rooms. However, such approaches have inherent drawbacks. Vision-based generative models, for instance, produce rasterized image-based outputs, making it challenging to access or modify them directly. Additionally, many models focus exclusively on adjacency constraints, disregarding room geometry entirely, thereby reducing the degree of control users can exert over the generated layouts.  \nTo overcome these limitations, we adopt a JSONbased representation for generative floor plan modeling that encodes room layouts as polygonal structures. Our method builds on prior vectorized floor plan approaches, such as House-GAN, HouseGAN++ (Nauata et al., 2020, 2021), and HouseDiffusion (Shabani et al., 2023), and leverages data from the RPLAN dataset (Wu et al., 2019) . This structured format facilitates greater control over spatial parameters, including room size and connectivity. To validate this approach, we fine-tune a large language model (LLM) to generate JSONencoded floor plans with spatial constraints specified in the prompt. See Figure 1 for an overview of the inference process.  \nThrough our experiments, we validate the effectiveness of our approach in maintaining high compliance with the specified constraints, as measured by our proposed constraint-adherence metrics. In the most complex setting, the eight-room task, ou","cbCaidVpDKbVDZ4H","https://ap.wps.com/l/cbCaidVpDKbVDZ4H","pdf",1280220,16,"English","# Introduction\n## Generative Models\n## Related Work","[{\"question\":\"What problem does the approach address in generative floor plan design?\",\"answer\":\"It targets the need to control both room connectivity and numerical constraints such as exact room sizing, which prior connectivity-focused methods do not reliably enforce.\"},{\"question\":\"How does the method generate floor plans while enforcing constraints?\",\"answer\":\"It fine-tunes a large language model on real plans, then applies reinforcement learning with verifiable rewards to increase adherence to topological and numerical constraints while discouraging invalid or overlapping outputs.\"},{\"question\":\"How are the generated plans evaluated?\",\"answer\":\"The document defines constraint adherence metrics that measure alignment between generated floor plans and user-defined constraints, and it reports improvements across Realism, Compatibility, and Diversity metrics.\"}]","Generative Floor Plan Design with LLMs via Reinforcement Learning with Verifiable Rewards - Abstract and Introduction | PDF"]