[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-86118-en":3,"doc-seo-86118-105":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},86118,1374391974468,"Eden","https://ap-avatar.wpscdn.com/davatar_29158cc5080c5b710cf443261637dec0",8,"Research & Report","GraspGraphNet Graph-Structured Multi-Embodiment Dexterous Grasp Generation Slideshare 2607 11031","Dexterous grasp generation across different robot hands is difficult because hands differ in kinematic topology, actuation dimensions, and command spaces. GraspGraphNet proposes a topology-aware framework that models each hand as a URDF-derived kinematic graph and directly generates executable palm poses and joint configurations. It combines hierarchical object surface encoding, differentiable forward kinematics, and dynamic robot–object world-edge message passing. Conditional flow matching is applied in palm-pose and joint-state spaces to avoid post-processing optimization, inverse kinematics, and retargeting.","GraspGraphNet: Graph-Structured Multi-Embodiment Dexterous  \nGrasp Generation  \nYeonseo Lee 1 , Taeyeop Lee 1 , Hyosup Shin 1 , Guebin Hwang 1 , and Sungho Jo 1†  \narXiv :2607 . 11031v1 [ cs .RO] 13 Jul 2026  \nAbstract—Dexterous grasp generation across robot hands is challenging because hands differ in kinematic topology, actuation dimensions, and native command spaces. We introduce GraspGraphNet, a topology-aware grasp generation framework that represents each hand as a URDF-derived kinematic graph and directly generates executable palm poses and joint configurations. GraspGraphNet combines hierarchical object surface encoding, differentiable forward kinematics, and dynamic worldedge message passing to model evolving robot-object interactions. It applies conditional flow matching directly in executable palmpose and joint-state space, avoiding post-processing optimization, inverse kinematics, and retargeting. Using a shared model trained on Barrett Hand, Allegro Hand, and Shadow Hand, GraspGraphNet achieves an average success rate of 83.48% with 40 ms inference time per grasp on a 40-object benchmark. Without retraining, the same model achieves 72.70% success on controlled finger-removal variants, demonstrating robustness to hand-topology variations. These results suggest that graphstructured hand representations can effectively support dexterous grasp generation across robot hands with different kinematic structures. Project: [https://lysees.github.io/graspgraphnet-page](https://lysees.github.io/graspgraphnet-page)  \nI. INTRODUCTION  \nDEXTEROUS grasping is a fundamental capability for  \nrobotic manipulation, requiring a multi-fingered robot hand to coordinate high-dimensional joint motion while establishing physically meaningful contact with object surfaces. Recent learning-based methods [1]–[5] have made substantial progress in generating stable and diverse grasps for dexterous hands and complex object geometries. Despite this progress, scaling dexterous grasp generation across diverse robot hands remains challenging. Dexterous grasp generation must consider object geometry as well as hand-specific kinematics, link geometry, joint limits, and actuation spaces.  \nDexterous robot hands have their own physical characteristics, varying in the number of fingers, link geometry, joint arrangement, and kinematic tree. As a result, dexterous robot hands do not share a common configuration space. Both the hand representation and the joint command output are handspecific, so a representation designed for one embodiment is generally not directly applicable to another. This makes it difficult to design a single grasping model that supports multiple dexterous hands. These challenges highlight the need for a unified model that can operate across robot-hand embodiments with distinct kinematic structures and actuation spaces.  \nTo address this challenge, prior learning-based grasp generation methods [3], [6], [7] have explored transferable rep-  \n1 Korea Advanced Institute of Science and Technology (KAIST), Daejeon, South Korea.  \n†Corresponding author.  \nFig. 1. GraspGraphNet represents different dexterous hands as URDF-derived kinematic graphs and generates an executable grasp by evolving an open-hand state x(0) to a final grasp state x(K) with dynamic robot-object world edges.  \nresentations that reduce dependence on a specific robot configuration space. Some methods introduce transferable grasp cues, such as object contacts [6], [8], contact maps [3], [9], or contact-centric representations [5], [10], [11] . These representations are transferred to different hands through downstream optimization, retargeting, or hand-specific decision modules. Other methods instead explicitly model robot-object interaction. DRO-Grasp [7] represents dense robot-object distances, while TRO-Grasp [12] generates robot-object spatial transformations with graph diffusion. While such contact and interaction-level representations improve transfer across hands, t","cbCaimVmUbDnaG3o","https://ap.wps.com/l/cbCaimVmUbDnaG3o","pdf",3349047,6,1,9,"English","en",105,"# Introduction\n## Challenges in cross-hand dexterous grasp generation\n## Limits of existing transferable and interaction-based methods\n## GraspGraphNet approach and key ideas\n## URDF-based kinematic graph representation\n## Dynamic robot–object interaction modeling","[{\"question\":\"Why is cross-embodiment dexterous grasp generation challenging across different robot hands?\",\"answer\":\"Different hands have distinct kinematic topologies, actuation dimensions, and native command spaces, so a representation or output learned for one embodiment often does not transfer directly to another.\"},{\"question\":\"How does GraspGraphNet represent each robot hand?\",\"answer\":\"Each hand is represented as a URDF-derived articulated kinematic graph (link–joint graph), enabling information propagation along kinematic connections and supporting variable graph sizes and joint dimensions.\"},{\"question\":\"What does GraspGraphNet generate directly, and how does it reduce inference overhead?\",\"answer\":\"It directly outputs executable palm poses and actuated joint configurations using conditional flow matching in the executable state spaces, avoiding additional post-processing such as optimization, inverse kinematics, and retargeting.\"}]",1784208618,23,{"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},"graspgraphnet-graph-structured-multi-embodiment-dexterous-grasp-generation-slideshare-2607-11031","",{"@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/graspgraphnet-graph-structured-multi-embodiment-dexterous-grasp-generation-slideshare-2607-11031/86118/",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},"Why is cross-embodiment dexterous grasp generation challenging across different robot hands?","Question",{"text":76,"@type":77},"Different hands have distinct kinematic topologies, actuation dimensions, and native command spaces, so a representation or output learned for one embodiment often does not transfer directly to another.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"How does GraspGraphNet represent each robot hand?",{"text":81,"@type":77},"Each hand is represented as a URDF-derived articulated kinematic graph (link–joint graph), enabling information propagation along kinematic connections and supporting variable graph sizes and joint dimensions.",{"name":83,"@type":74,"acceptedAnswer":84},"What does GraspGraphNet generate directly, and how does it reduce inference overhead?",{"text":85,"@type":77},"It directly outputs executable palm poses and actuated joint configurations using conditional flow matching in the executable state spaces, avoiding additional post-processing such as optimization, inverse kinematics, and retargeting.","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},[94,98,102,106,111,115,120,123,127,130,134],{"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":22,"doc_module":4,"doc_module_name":46,"category_name":124,"show_sort_weight":125,"slug":126},"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":46,"category_name":128,"show_sort_weight":125,"slug":129},"World Cup","world-cup",{"id":131,"doc_module":4,"doc_module_name":46,"category_name":132,"show_sort_weight":131,"slug":133},10,"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":46,"category_name":136,"show_sort_weight":107,"slug":137},19,"General","general"]