[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-82873-en":3,"doc-seo-82873-105":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},82873,8796095462418,"Noah","https://ap-avatar.wpscdn.com/avatar/80000253c1241d02b47?x-image-process=image/resize,m_fixed,w_180,h_180&k=1778826106357471780",8,"Research & Report","PRISM Personalized Robotic Dataset Generation via Image-based Scene and Motion Synthesis","Large-scale vision-language-action models improve robot policy learning but degrade when deployed in user-specific environments because target-aligned data are scarce. Teleoperation provides well-aligned demonstrations yet is costly and difficult to scale, while simulation scales poorly at matching real environments and producing task-specific trajectories. PRISM introduces an end-to-end pipeline that generates personalized robotic datasets from a single image and natural-language instruction. It synthesizes digital-cousin scenes and executable demonstrations without human teleoperation. Experiments on LIBERO and real manipulation tasks show higher success and robust sim-to-real generalization.","arXiv :2607 .04880v 1 [ cs .RO] 6 Jul 2026  \nPRISM: Personalized Robotic Dataset Generation via Image-based Scene and Motion Synthesis  \nDogyu Ko∗ , Haneul Kim∗ , Chanyoung Yeo∗ , Dowoon Lee, Taeho Park, Hyoseok Hwang†  \nKyung Hee University  \n{kodogyu, sky9893, ducksdud08, pauldoun, katehoya, [hyoseok](hyoseok}@khu.ac.kr)[}](hyoseok}@khu.ac.kr)[@khu.ac.kr](hyoseok}@khu.ac.kr)  \n∗Equal Contribution †Corresponding Author  \nFigure 1: PRISM generates personalized robotic datasets from a single image and a task instruction of the target environment. Scenes and trajectories are synthesized while preserving semantic and geometric structure, and datasets are constructed through structured randomization.  \nAbstract: Recent advances in large-scale pretrained vision-language-action models have improved robot policy learning, but directly deploying such policies in user-specific environments remains challenging due to limited generalization, which inevitably requires collecting a dataset tailored to the target environment.  \nTeleoperation yields well-aligned data but is costly and difficult to scale, whereas simulation scales easily but struggles to resemble the target environment and generate task-specific trajectories. To meet both simultaneously, we propose PRISM, an end-to-end pipeline that generates personalized robotic datasets from a single image and a natural-language instruction. PRISM constructs digital cousin scenes that are semantically and geometrically aligned with the user environment yet diverse at the instance level, and synthesizes executable demonstrations without human teleoperation. Extensive experiments show that policies trained on PRISMgenerated datasets outperform those trained on baseline-generated datasets on LIBERO and LIBERO-Plus, achieve up to 100% success rate on three real-world manipulation tasks, and maintain stronger performance when evaluated in environments that differ from those seen during training.  \nKeywords: Robotic data generation, Sim-to-Real transfer, VLA  \n1 Introduction  \nVision-Language-Action (VLA) models [1, 2, 3] have rapidly emerged as a leading paradigm for general-purpose robotic manipulation, adapting large vision-language foundation models to predict robot actions directly from images and natural-language instructions. Trained jointly on web-scale vision-language data and robot demonstrations across many robots, tasks, and environments [4, 5], modern VLAs acquire broad zero-shot competence over unseen objects, scenes, and instructions. Their success suggests a promising path toward general-purpose robotic manipulation.  \nYet this broad competence does not translate to a user’s deployment environment, where the performance of a pretrained VLA often degrades sharply on tasks it would otherwise handle reliably. The cause is structural, since a VLA’s competence is bounded above by the empirical distribution from which its training trajectories were drawn, and that distribution rarely covers any specific user’s environment densely enough for reliable execution [6] . Closing this gap therefore requires fine-tuning on data tailored to the target environment, and recent studies show that the resulting policy’s quality is governed mostly by how closely the training data match the deployment target [6, 7, 8] . The central question is therefore how to obtain such target-aligned data without prohibitive human effort.  \nExisting approaches to acquiring target-aligned data fall into three families. Teleoperation in the target environment, by direct robot control [4, 5] or with specialized hardware [9], yields data that faithfully reflect the target environment, achieving structural alignment with task supervision, since they are gathered by a human who performs the task in the deployment scene itself. Simulationbased approaches sidestep the cost of physical data collection and split into two further families. One line of work uses LLMs or VLMs to compose tasks and scenes from a large 3D asset lib","cbCaitk5tusb3gO0","https://ap.wps.com/l/cbCaitk5tusb3gO0","pdf",22115424,1,47,"English","en",105,"# Introduction\n## Problem: target deployment gap for pretrained VLA policies\n## Existing approaches: teleoperation and simulation families\n## Proposed solution: PRISM pipeline overview","[{\"question\":\"Why do pretrained vision-language-action policies perform poorly in user-specific environments?\",\"answer\":\"Performance degrades because the model’s capability is bounded by the distribution of trajectories seen during training, which rarely covers a specific user environment densely enough for reliable execution.\"},{\"question\":\"What limitation exists when using teleoperation to collect target-aligned data?\",\"answer\":\"Teleoperation yields data that faithfully reflect the target environment, but it is costly and labor-intensive, making it difficult to scale.\"},{\"question\":\"How does PRISM create personalized robotic datasets without human teleoperation?\",\"answer\":\"PRISM extracts object categories and geometry from a single target image, retrieves matched assets from a 3D asset library to compose digital-cousin scenes, then parses the natural-language instruction to synthesize executable demonstrations at scale.\"}]",1784183595,118,{"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},"prism-personalized-robotic-dataset-generation-via-image-based-scene-and-motion-synthesis","",{"@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/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/prism-personalized-robotic-dataset-generation-via-image-based-scene-and-motion-synthesis/82873/",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},"Why do pretrained vision-language-action policies perform poorly in user-specific environments?","Question",{"text":75,"@type":76},"Performance degrades because the model’s capability is bounded by the distribution of trajectories seen during training, which rarely covers a specific user environment densely enough for reliable execution.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"What limitation exists when using teleoperation to collect target-aligned data?",{"text":80,"@type":76},"Teleoperation yields data that faithfully reflect the target environment, but it is costly and labor-intensive, making it difficult to scale.",{"name":82,"@type":73,"acceptedAnswer":83},"How does PRISM create personalized robotic datasets without human teleoperation?",{"text":84,"@type":76},"PRISM extracts object categories and geometry from a single target image, retrieves matched assets from a 3D asset library to compose digital-cousin scenes, then parses the natural-language instruction to synthesize executable demonstrations at scale.","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},[93,97,101,105,110,115,120,123,128,131,135],{"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":111,"doc_module":4,"doc_module_name":45,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":45,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":121,"slug":122},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"]