[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85075-en":3,"doc-seo-85075-105":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},85075,1099514067438,"River Wang","https://ap-avatar.wpscdn.com/avatar/100002539ee87300030?x-image-process=image/resize,m_fixed,w_180,h_180&k=1780474512215547542",8,"Research & Report","SkillPlug：用于机器人操作少样本自适应的无监督技能挖掘","Learning transferable visuomotor imitation policies that generalize across diverse robotic manipulation tasks and adapt rapidly from only a few demonstrations remains difficult because most end-to-end policies map observations directly to low-level actions, leaving reusable behavioral structure implicit and transfer data-inefficient. SkillPlug introduces a plug-in framework that mines a shared transferable skill library from raw multi-task demonstrations using self-supervised objectives, then adapts to unseen tasks by fine-tuning only lightweight router and action heads. Evaluations on two simulation benchmarks and a real robot show consistent improvements in multi-task performance and few-shot adaptation.","SkillPlug: Unsupervised Skill Mining for Few-Shot Adaptation in Robotic Manipulation  \nZi-han Ding 1 , Ziwei Wang 1∗  \narXiv :2607 .08354v 1 [ cs .RO] 9 Jul 2026  \nAbstract—Learning transferable visuomotor imitation policies that generalize across diverse manipulation tasks and adapt rapidly to new tasks from only a handful of demonstrations remains challenging. Most modern policies are trained end-toend to map observations directly to low-level actions, offering little explicit structure for reusing and recombining behaviors across tasks and making transfer data-inefficient under limited supervision. We propose SkillPlug, a plug-in framework that augments an existing visuomotor policy with a skill-conditioning module and mines a shared, transferable skill library from raw multi-task demonstrations. SkillPlug learns skills via selfsupervised objectives that promote compact, reusable, and nonredundant behavior-level primitives, forming a task-shared prior for compositional control. After skill mining, we keep the learned skills fixed and specialize to unseen tasks by fine-tuning only lightweight router and action head, enabling efficient adaptation without full end-to-end retraining. We evaluate SkillPlug on two simulation benchmarks and on a real robot, and observe that the mined transferable skills consistently improve both multi-task performance and few-shot adaptation. Overall, SkillPlug offers a scalable way to mine reusable skills that improve data-efficient generalization in robotic manipulation.  \nIndex Terms—Deep Learning in Grasping and Manipulation; AI-Enabled Robotics  \nI. INTRODUCTION  \nImitation-based visuomotor policies have driven rapid progress in robotic manipulation. Trained on large collections of demonstrations [1]–[3] and scaled with modern transformer backbones [4]–[10], a single policy can now solve a wide range of tasks. Despite these advances, most policies are still optimized as end-to-end mappings from observations to lowlevel actions, leaving reusable behavioral structure implicit. Asa result, while end-to-end policies can perform well in multitask settings, they often fail to uncover and reuse cross-task shared behavioral patterns, making it difficult to efficiently transfer to novel tasks with only a few demonstrations.  \nA natural remedy is to introduce an explicit skill abstraction as an intermediate layer. Existing imitation-learning-based approaches to skill learning broadly fall into two categories. One line relies on human annotation or VLM-based segmentation to define skills or phases (e.g., approach, grasp,  \nManuscript received: December 23, 2025; Revised: March 30, 2026; Accepted: May 17, 2026 .  \nThis paper was recommended for publication by Editor Markus Vincze upon evaluation of the Associate Editor and Reviewers comments.  \nThis work was supported by MoE AcRF Tier 2 (MOE-T2EP50125-0004) .  \n1Zi-han Ding and Ziwei Wang are with the School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore.  \n{zihan.ding, [ziwei.wang](ziwei.wang}@ntu.edu.sg)[}](ziwei.wang}@ntu.edu.sg)[@ntu.edu.sg](ziwei.wang}@ntu.edu.sg)[ ](ziwei.wang}@ntu.edu.sg)∗ Corresponding author: Ziwei Wang.  \nDigital Object Identifier (DOI): see top of this page.  \nFig. 1. Conventional imitation-based visuomotor policies often struggle to adapt to a new task from only a handful of demonstrations. SkillPlug augments the policy with a skill interactor and router, and mines a transferable skill library from multi-task demonstrations, enabling efficient reuse and recombination of learned skills for successful few-shot transfer.  \nplace), and then trains hierarchical controllers conditioned on these labels [11]–[16] . While such methods offer interpretable structure, they typically require extensive manual design for segmentation, skill definition, and phase-boundary specification. A second line learns skills directly from data, e.g., via trajectory clustering, learnable skill embeddings, or mixtur","cbCailf6VYJmF8J9","https://ap.wps.com/l/cbCailf6VYJmF8J9","pdf",4874626,4,1,"English","en",105,"# Introduction\n## Background and Motivation\n## Related Work on Skill Learning","[{\"question\":\"SkillPlug解决了什么核心问题？\",\"answer\":\"它解决的是：端到端视觉-运动模仿策略难以从少量演示中高效适应新任务，且跨任务可复用的行为结构难以显式复用，从而导致数据效率不足。\"},{\"question\":\"SkillPlug的关键方法是什么？\",\"answer\":\"SkillPlug在现有视觉-运动策略上以插件形式加入技能条件模块，并从多任务原始演示中通过自监督目标挖掘可迁移、可复用且非冗余的技能库。\"},{\"question\":\"适应未见任务时，模型需要怎么训练？\",\"answer\":\"在技能挖掘后，保持已学习技能固定，仅对未见任务微调轻量化的router和action head，从而避免全量端到端重训练并实现高效少样本适应。\"}]",1784200893,20,{"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},"skillplug-unsupervised-skill-mining-for-few-shot-adaptation-in-robotic-manipulation","",{"@graph":35,"@context":84},[36,52,67],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":21},"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":20},"https://docshare.wps.com/document/skillplug-unsupervised-skill-mining-for-few-shot-adaptation-in-robotic-manipulation/85075/",{"url":51,"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","2026-07-24","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},"SkillPlug解决了什么核心问题？","Question",{"text":74,"@type":75},"它解决的是：端到端视觉-运动模仿策略难以从少量演示中高效适应新任务，且跨任务可复用的行为结构难以显式复用，从而导致数据效率不足。","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"SkillPlug的关键方法是什么？",{"text":79,"@type":75},"SkillPlug在现有视觉-运动策略上以插件形式加入技能条件模块，并从多任务原始演示中通过自监督目标挖掘可迁移、可复用且非冗余的技能库。",{"name":81,"@type":72,"acceptedAnswer":82},"适应未见任务时，模型需要怎么训练？",{"text":83,"@type":75},"在技能挖掘后，保持已学习技能固定，仅对未见任务微调轻量化的router和action head，从而避免全量端到端重训练并实现高效少样本适应。","https://schema.org",{"og:url":51,"og:type":86,"og:title":13,"og:site_name":57,"og:description":14},"article",{"robots":88,"canonical":51},"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":46,"doc_module":4,"doc_module_name":45,"category_name":97,"show_sort_weight":98,"slug":99},"Literature",80,"literature",{"id":20,"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"]