[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85588-en":3,"doc-seo-85588-105":29,"detail-sidebar-cat-0-en-105":90},{"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":11,"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},85588,1649267921044,"Ava Thompson","https://us-avatar.wpscdn.com/avatar/1800007509477c92dfb?_k=1782875107921204101",8,"Research & Report","Learning Tactile-Aware Quadrupedal Loco-Manipulation Policies","Quadrupedal loco-manipulation often relies on vision and proprioception, but contact-rich manipulation remains unreliable because those modalities cannot directly observe evolving interface interactions. Tactile sensing provides direct contact observability, yet scalable tactile-aware learning frameworks are limited. This paper proposes a hierarchical policy learning pipeline using tactile-conditioned visuotactile demonstrations and large-scale simulation reinforcement learning to obtain a transferable tactile-aware whole-body controller. Real-world contact-rich evaluations show an average 28.54% improvement over vision-only and visuotactile baselines.","Learning Tactile-Aware Quadrupedal Loco-Manipulation Policies  \nPokuang Zhou 1 , Yuhao Zhou 1 , Quan Khanh Luu 1 , Seungho Han 1 , Heng Zhang 1 ,2 , Binghao Huang3 , Yunzhu Li3 , Arash Ajoudani2 , Zhengtong Xu 1 ,†, and Yu She 1 ,†  \narXiv :2604 .27224v3 [ cs .RO] 12 Jul 2026  \nAbstract—Quadrupedal loco-manipulation is commonly built on visual perception and proprioception. Yet reliable contactrich manipulation remains difficult: vision and proprioception alone cannot resolve uncertain, evolving interactions with the environment. Tactile sensing offers direct contact observability, but scalable tactile-aware learning framework for quadrupedal loco-manipulation is still underexplored. In this paper, we present a tactile-aware loco-manipulation policy learning pipeline with a hierarchical structure. Our approach has two key components. First, we leverage real-world human demonstrations to train a tactile-conditioned visuotactile highlevel policy. This policy predicts not only end-effector trajectories for manipulation, but also the evolving tactile interaction cues that characterize how contact should develop over time. Second, we perform large-scale reinforcement learning in simulation to learn a tactile-aware whole-body control policy that tracks diverse commanded trajectories and tactile interaction cues, and transfers zero-shot to the real world. Together, these components enable coordinated locomotion and manipulation under contact-rich scenarios. We evaluate the system on realworld contact-rich tasks, including in-hand reorientation with insertion, valve tightening, and delicate object manipulation. Compared to vision-only and visuotactile baselines, our method improves performance by 28.54% on average across these tasks. The project website is available at [https://pokuangzhou.github](https://pokuangzhou.github). io/tactile-aware-quadrupedal-loco-manipulation/.  \nI. INTRODUCTION  \nLoco-manipulation based on quadrupedal mobile manipulators substantially expands a robot’s operational workspace and task repertoire, making these platforms promising for real-world deployment in inspection, maintenance, and field robotics applications [1]–[3] . Recent quadrupedal locomanipulation systems are commonly built on vision and proprioception, learning coordinated base–arm behaviors from data. However, reliable and generalizable contact-rich manipulation remains challenging: vision and proprioception lack direct access to the rich, precise contact information needed to infer evolving interaction dynamics at the interface, and vision is further degraded by occlusion and partial observability.  \nA growing body of evidence shows that tactile feedback substantially improves precision and robustness during physical interaction [4]–[6] . However, making quadrupedal loco-manipulation tactile-aware at scale remains challenging: tactile signals are high-dimensional and strongly coupled with whole-body dynamics, and contact-rich tasks require  \n1Purdue University, West Lafayette, IN, USA. 2Istituto Italiano di Tecnologia, Genoa, Italy. 3Columbia University, New York, NY, USA. †Equal advising. This work was supported by the National Science Foundation under Grants 2423068 and 2520136, and the United States Department of Agriculture under Grants 2023-67021-39072 and 2024-67021-42878 . This work used GPU resource from NSF ACCESS CIS 260072 .  \nHuman Demonstration  \nVisuo-Tactile Gripper  \nRobot Deployment  \nRobotic Arm  \nQuadruped Robot  \nVisuo-Tactile Gripper  \nFig. 1: We achieve fully autonomous real-world tactile-aware quadrupedal loco-manipulation by learning from demonstrations collected with a hand-held, tactile-instrumented UMI gripper. Our approach enables learning and deployment across tasks and objects, supporting contact-rich manipulation (e.g., in-hand reorientation with insertion and valve tightening) as well as damage-free handling of delicate items such as fragile chips and fruits.  \nreasoning over how interaction evolves over time w","cbCaich3ulcWCiqO","https://ap.wps.com/l/cbCaich3ulcWCiqO","pdf",2941530,2,1,"English","en",105,"# Introduction\n## Tactile-aware hierarchical learning framework\n## Contributions and evaluation","[{\"question\":\"Why are vision and proprioception insufficient for contact-rich quadrupedal loco-manipulation?\",\"answer\":\"Vision and proprioception lack direct access to precise contact information needed to infer how interactions evolve, and vision can degrade under occlusion and partial observability.\"},{\"question\":\"How does the proposed method use tactile sensing in its learning pipeline?\",\"answer\":\"It combines tactile-conditioned visuotactile high-level policy learning from human demonstrations with a low-level tactile-aware whole-body controller learned via large-scale reinforcement learning in simulation.\"},{\"question\":\"How is performance evaluated and how much improvement is reported?\",\"answer\":\"The system is tested on real-world contact-rich tasks such as in-hand reorientation with insertion, valve tightening, and delicate object manipulation, achieving an average 28.54% improvement over vision-only and visuotactile baselines.\"}]",1784204765,20,{"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":85,"head_meta":87,"extra_data":89,"updated_unix":27},"learning-tactile-aware-quadrupedal-loco-manipulation-policies","",{"@graph":35,"@context":84},[36,52,67],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,46,49],{"item":40,"name":41,"@type":42,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":20},"https://docshare.wps.com/document/","Document",{"item":47,"name":12,"@type":42,"position":48},"https://docshare.wps.com/document/research-report/",3,{"item":50,"name":13,"@type":42,"position":51},"https://docshare.wps.com/document/learning-tactile-aware-quadrupedal-loco-manipulation-policies/85588/",4,{"url":50,"name":13,"@type":53,"author":54,"headline":13,"publisher":56,"fileFormat":59,"inLanguage":23,"description":14,"dateModified":60,"datePublished":61,"encodingFormat":59,"isAccessibleForFree":62,"interactionStatistic":63},"DigitalDocument",{"name":9,"@type":55},"Person",{"url":40,"name":57,"@type":58},"DocShare","Organization","application/pdf","2026-07-25","2026-07-16",true,{"@type":64,"interactionType":65,"userInteractionCount":20},"InteractionCounter",{"@type":66},"ViewAction",{"@type":68,"mainEntity":69},"FAQPage",[70,76,80],{"name":71,"@type":72,"acceptedAnswer":73},"Why are vision and proprioception insufficient for contact-rich quadrupedal loco-manipulation?","Question",{"text":74,"@type":75},"Vision and proprioception lack direct access to precise contact information needed to infer how interactions evolve, and vision can degrade under occlusion and partial observability.","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"How does the proposed method use tactile sensing in its learning pipeline?",{"text":79,"@type":75},"It combines tactile-conditioned visuotactile high-level policy learning from human demonstrations with a low-level tactile-aware whole-body controller learned via large-scale reinforcement learning in simulation.",{"name":81,"@type":72,"acceptedAnswer":82},"How is performance evaluated and how much improvement is reported?",{"text":83,"@type":75},"The system is tested on real-world contact-rich tasks such as in-hand reorientation with insertion, valve tightening, and delicate object manipulation, achieving an average 28.54% improvement over vision-only and visuotactile baselines.","https://schema.org",{"og:url":50,"og:type":86,"og:title":13,"og:site_name":57,"og:description":14},"article",{"robots":88,"canonical":50},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":91},[92,96,100,104,109,114,119,122,126,129,133],{"id":21,"doc_module":4,"doc_module_name":45,"category_name":93,"show_sort_weight":94,"slug":95},"Story & Novel",90,"story-novel",{"id":20,"doc_module":4,"doc_module_name":45,"category_name":97,"show_sort_weight":98,"slug":99},"Literature",80,"literature",{"id":51,"doc_module":4,"doc_module_name":45,"category_name":101,"show_sort_weight":102,"slug":103},"Exam",70,"exam",{"id":105,"doc_module":4,"doc_module_name":45,"category_name":106,"show_sort_weight":107,"slug":108},5,"Comic",60,"comic",{"id":110,"doc_module":4,"doc_module_name":45,"category_name":111,"show_sort_weight":112,"slug":113},6,"Technology",50,"technology",{"id":115,"doc_module":4,"doc_module_name":45,"category_name":116,"show_sort_weight":117,"slug":118},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":45,"category_name":124,"show_sort_weight":28,"slug":125},9,"Religion & Spirituality","religion-spirituality",{"id":28,"doc_module":4,"doc_module_name":45,"category_name":127,"show_sort_weight":28,"slug":128},"World Cup","world-cup",{"id":130,"doc_module":4,"doc_module_name":45,"category_name":131,"show_sort_weight":130,"slug":132},10,"Lifestyle","lifestyle",{"id":134,"doc_module":4,"doc_module_name":45,"category_name":135,"show_sort_weight":105,"slug":136},19,"General","general"]