[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-84115-en":3,"doc-seo-84115-105":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},84115,1099514067438,"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","SIEVE: Structure-Aware Data Selection for Imitation Learning with VLA Models","Vision-Language-Action (VLA) models rely on imitation learning from robot demonstration datasets, but larger datasets can still hurt performance due to redundancy, noise, and uneven task coverage. Existing selection techniques often score whole trajectories or isolated state-action pairs, overlooking reusable structures behind long-horizon behaviors. SIEVE treats demonstrations as compositions of reusable visuo-motor primitives and transition interfaces, then selects central medoid trajectories per composition pattern. Experiments across multiple benchmarks show consistent gains and efficient training with only 50% demonstrations and steps.","SIEVE: Structure-Aware Data Selection for Imitation Learning with VLA Models  \nChangti Wu 1,2 * , Bin Yu3,2 * , Zhaolong Shen4,2 * , Shijie Lian5,2 , Xiaopeng Lin6,8 , Cong Huang7 , Zhirui Zhang8 , Lei Zhang 1†, Kai Chen7,2,8†  \n1East China Normal University 2Zhongguancun Academy 3Harbin Institute of Technology 4Beihang University  \n5Huazhong University of Science and Technology 6The Hong Kong University of Science and Technology (Guangzhou)  \n7Zhongguancun Institute of Artificial Intelligence 8DeepCybo  \narXiv :2607 .06442v 1 [ cs .RO] 7 Jul 2026  \nAbstract  \nVision-Language-Action (VLA) models are typically trained by imitation learning on large-scale robot demonstration datasets, but more data does not necessarily yield better policies due to redundancy, noise, and uneven coverage. Existing data selection methods often assess demonstrations at either the trajectory or state-action level, missing the reusable structures that compose long-horizon behaviors. In this paper, we propose SIEVE, a structure-aware data selection method for VLA imitation learning. SIEVE views demonstrations as compositions of reusable primitives and transition interfaces. It first discovers visuo-motor primitives from segmented trajectories, then allocates selection budgets to composition patterns by maximizing reuse-aware structural exposure under diminishing returns. Finally, it selects medoid trajectories within each composition-pattern bucket to retain central, stable, and imitation-friendly demonstrations. Experiments across multiple datasets, benchmarks, and VLA models show that SIEVE consistently outperforms competitive data selection baselines. Notably, SIEVE can surpass fulldata training while using only 50% of demonstrations and 50% of training steps, suggesting that reusable structure, captured through primitives and transitions, is an important signal for efficient VLA imitation learning.  \nCode—[https://github.com/ChangtiWu/SIEVE](https://github.com/ChangtiWu/SIEVE)  \nIntroduction  \nVision-Language-Action (VLA) models have emerged asa scalable paradigm for robotic control, typically acquiring manipulation skills through imitation learning (IL) over large-scale demonstrations (Zitkovich et al. 2023; O’Neillet al. 2024; Kim et al. 2024; Black et al. 2024; Intelligence et al. 2025) . However, the rapid growth of robot demonstration datasets does not automatically translate into better policies. In practice, such datasets often contain substantial trajectory redundancy, noisy human demonstrations, suboptimal behaviors, and uneven task coverage (Sathyanarayan, Vantilborgh, and Abraham 2025; Belkhale, Cui, and Sadigh 2023; Lin et al. 2025a; Xing et al. 2025) . Training on unfiltered data can repeatedly expose the model to near-duplicate behaviors while also propagating inconsistent or low-quality  \n*These authors contributed equally  \n†Corresponding author  \nTraining Demonstrations Policy Execution  \nFigure 1: Motivation of SIEVE. Training demonstrations contain recurring primitive-transition patterns that can be viewed as reusable behavioral subprograms. Inspired by the MDL principle, SIEVE aims to select demonstrations that expose such reusable structures, enabling the policy to internalize shared subprograms.  \nsupervision. These issues make data selection an increasingly important problem for VLA imitation learning: given a large demonstration pool, we aim to retain a compact subset that is more beneficial for policy learning.  \nExisting data selection methods for imitation learning typically curate demonstrations by estimating sample utility at different granularities. One line relies on trajectory-level signals, such as trajectory-representation similarity for redundancy removal, demonstration reliability, or downstream task feedback (Dass et al. 2025; Chen et al. 2025; Zhang et al. 2025; Xu et al. 2026) . While these signals provide a global view of demonstration utility, they may collapse a long-horizon trajectory into a single score a","cbCaieYfiCe5PO8x","https://ap.wps.com/l/cbCaieYfiCe5PO8x","pdf",456318,1,10,"English","en",105,"# Introduction\n## Motivation and problem of redundancy/noise\n## Limitations of existing trajectory-level and state-action-level selection\n## SIEVE approach: primitives, transition interfaces, and reuse-aware budgeting\n## Selection via medoid trajectories per composition pattern\n## Experimental validation and efficiency results","[{\"question\":\"Why doesn’t adding more robot demonstrations always improve VLA imitation learning?\",\"answer\":\"More data can introduce redundancy, noise, and uneven coverage, which can repeatedly expose the model to near-duplicate or low-quality behaviors and reduce policy learning effectiveness.\"},{\"question\":\"What is the key idea behind SIEVE’s data selection?\",\"answer\":\"SIEVE treats demonstrations as reusable compositions, first discovering visuo-motor primitives from segmented trajectories and then selecting demonstrations that maximize reuse-aware structural exposure.\"},{\"question\":\"How does SIEVE decide which demonstrations to keep?\",\"answer\":\"SIEVE allocates a selection budget to composition patterns using diminishing-returns reuse criteria, then chooses medoid trajectories within each pattern bucket to retain central, stable, imitation-friendly examples.\"}]",1784192942,25,{"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},"sieve-structure-aware-data-selection-for-imitation-learning-with-vla-models","",{"@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/sieve-structure-aware-data-selection-for-imitation-learning-with-vla-models/84115/",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 doesn’t adding more robot demonstrations always improve VLA imitation learning?","Question",{"text":75,"@type":76},"More data can introduce redundancy, noise, and uneven coverage, which can repeatedly expose the model to near-duplicate or low-quality behaviors and reduce policy learning effectiveness.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"What is the key idea behind SIEVE’s data selection?",{"text":80,"@type":76},"SIEVE treats demonstrations as reusable compositions, first discovering visuo-motor primitives from segmented trajectories and then selecting demonstrations that maximize reuse-aware structural exposure.",{"name":82,"@type":73,"acceptedAnswer":83},"How does SIEVE decide which demonstrations to keep?",{"text":84,"@type":76},"SIEVE allocates a selection budget to composition patterns using diminishing-returns reuse criteria, then chooses medoid trajectories within each pattern bucket to retain central, stable, imitation-friendly examples.","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},[93,97,101,105,110,115,120,123,128,131,134],{"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":21,"doc_module":4,"doc_module_name":45,"category_name":132,"show_sort_weight":21,"slug":133},"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":45,"category_name":136,"show_sort_weight":106,"slug":137},19,"General","general"]