[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-84211-en":3,"doc-seo-84211-105":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},84211,962075114101,"Seraphina","https://ap-avatar.wpscdn.com/avatar/e000253a75eb197efd?x-image-process=image/resize,m_fixed,w_180,h_180&k=1780044092746381165",8,"Research & Report","TouchWorld: A Predictive and Reactive Tactile Foundation Model for Dexterous Manipulation","Dexterous manipulation in everyday environments demands both anticipation and rapid correction as robots predict evolving contact and counter local errors from slip, misalignment, unstable grasps, or force mismatch. Vision and language aid semantic and geometric guidance but cannot reliably expose hidden contact states such as force, slip, and contact stability. TouchWorld introduces a hierarchical predictive-and-reactive tactile foundation model that uses touch for tactile world-model prediction and high-frequency residual refinement, improving local contact adaptation. Evaluated on six long-horizon, contact-rich tasks, it achieves 65.0% success in clean settings and 53.7% under human perturbations, surpassing the strongest baseline.","arXiv :2607 .07287v2 [ cs .RO] 9 Jul 2026  \nTouchWorld: A Predictive and Reactive Tactile Foundation Model for Dexterous Manipulation  \nJianyi Zhou1 ,2∗ , Feiyang Hong1 ,2∗ , Yunhao Li1 ,2∗ , Yicheng Zhao1 ,2 , Yongjue Cen1 ,2 , Zirui Liu1 ,2 , Jiakang Huang1 ,2 , Zirui Chen1 ,2 , Ruiyang Zhang1 ,2 , Weizhuo Zhu1 ,2 , Xuhua Song1 ,2 , Shuo Yang1 ,2✉  \n1 Harbin Institute of Technology, Shenzhen  \n2 PHANES AI  \n∗ Equal contribution, ✉ Corresponding author  \nAbstract  \nDexterous manipulation in everyday environments requires both anticipation and reaction: a robot must predict how contact should evolve while rapidly correcting local errors caused by slip, misalignment, unstable grasping, or force mismatch. Vision and language provide semantic and geometric guidance, but they cannot reliably reveal hidden contact states such as force, slip, and contact stability. Although tactile sensing exposes these physical cues, most existing policies treat touch as a low-frequency observation stream within a monolithic action model, coupling slow task reasoning, action generation, and fast contact feedback in a single loop. We introduce TouchWorld, a predictive-and-reactive tactile foundation model for dexterous manipulation. TouchWorld uses a hierarchical policy that separates vision-language subtask planning, tactile world-model prediction, visuo-tactile goal-conditioned action generation, and high-frequency tactile residual refinement. A High-Level Planning Layer produces executable subtasks and predicts tactile subgoals; a Visuo-Tactile Goal-Conditioned Policy generates nominal action chunks; anda Tactile-Conditioned Refinement Policy performs online residual correction using recent tactile and proprioceptive feedback. By using touch as both a predictive contact reference and a fast feedback signal, TouchWorld preserves the semantic generalization of vision-language-action policies while improving local contact adaptation. Across six long-horizon and contact-rich dexterous manipulation tasks, TouchWorld achieves 65.0% success in the clean setting and 53.7% success under human perturbations, outperforming the strongest baseline by 15.7 and 18.5 percentage points, respectively.  \nDate: July 10, 2026  \nCorrespondence: [shuoyang@hit.edu.cn](shuoyang@hit.edu.cn)  \nProject Page: [https://phanes-lab.github.io/TouchWorld-website/](https://phanes-lab.github.io/TouchWorld-website/)  \n1 Introduction  \nRobots operating in everyday environments must perform manipulation tasks that go beyond reaching and moving objects. Many daily skills, such as watering, power-plug insertion, cup insertion and tissue pulling, require the robot to anticipate how contact should evolve and to react when the actual contact deviates from expectation. While vision and language provide semantic and geometric guidance, they often cannot reveal whether a grasp is stable, an object is slipping, an insertion is aligned, or the applied force is appropriate. Tactile feedback provides direct access to these local contact states [6, 7 , 26 , 29 , 31 , 33], making it essential for robust contact-rich manipulation. This naturally introduces a multi-timescale control problem. Semantic task reasoning evolves slowly, visuo-tactile action generation operates at an intermediate action-chunk rate, and tactile feedback must support fast local correction when contact changes.  \nFigure 1 Conceptual overview of TouchWorld. The High-Level Planning Layer contains a Subtask Planner that produces executable subtasks and a Tactile World Model that predicts visual-tactile subgoals. A visuo-tactile goal-conditioned policy generates nominal action chunks, and a tactile-conditioned refinement policy refines the final action online using high-frequency tactile feedback.  \nDespite recent progress in vision-language-action policies [2, 4 , 8–12, 16 , 18 , 20 , 32], most existing systems still generate actions through a single monolithic model. Tactile observations, when available, are often appended as ad","cbCair1YpAQFqVrr","https://ap.wps.com/l/cbCair1YpAQFqVrr","pdf",2463571,1,17,"English","en",105,"# Abstract\n# Introduction\n## Multi-timescale contact control problem\n## Limits of monolithic vision-language-action policies\n## Proposed TouchWorld hierarchical model\n## High-level planning, goal-conditioned action, and tactile refinement","[{\"question\":\"What problem does TouchWorld address in dexterous manipulation?\",\"answer\":\"TouchWorld targets the need to both anticipate how contact should evolve and quickly react to local deviations such as slip, misalignment, unstable grasping, or force mismatch.\"},{\"question\":\"How does TouchWorld use tactile information differently from existing approaches?\",\"answer\":\"TouchWorld treats touch as both a predictive contact reference (for tactile world-model prediction and tactile subgoals) and a fast feedback signal (for high-frequency residual refinement inside the control loop).\"},{\"question\":\"What performance gains does TouchWorld report on benchmark tasks?\",\"answer\":\"Across six long-horizon, contact-rich tasks, TouchWorld achieves 65.0% success in clean settings and 53.7% under human perturbations, outperforming the strongest baseline by 15.7 and 18.5 percentage points respectively.\"}]",1784194011,43,{"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},"touchworld-a-predictive-and-reactive-tactile-foundation-model-for-dexterous-manipulation","",{"@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/touchworld-a-predictive-and-reactive-tactile-foundation-model-for-dexterous-manipulation/84211/",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 TouchWorld address in dexterous manipulation?","Question",{"text":75,"@type":76},"TouchWorld targets the need to both anticipate how contact should evolve and quickly react to local deviations such as slip, misalignment, unstable grasping, or force mismatch.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does TouchWorld use tactile information differently from existing approaches?",{"text":80,"@type":76},"TouchWorld treats touch as both a predictive contact reference (for tactile world-model prediction and tactile subgoals) and a fast feedback signal (for high-frequency residual refinement inside the control loop).",{"name":82,"@type":73,"acceptedAnswer":83},"What performance gains does TouchWorld report on benchmark tasks?",{"text":84,"@type":76},"Across six long-horizon, contact-rich tasks, TouchWorld achieves 65.0% success in clean settings and 53.7% under human perturbations, outperforming the strongest baseline by 15.7 and 18.5 percentage points respectively.","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"]