[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-84766-en":3,"doc-seo-84766-105":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},84766,4398048950312,"Violet","https://ap-avatar.wpscdn.com/avatar/400002538284de19e3c?_k=1778320343897328908",8,"Research & Report","CAC-VLA Context-Gated Action Conditioning for Vision-Language-Action Models","Vision-Language-Action (VLA) models enable generalist robot manipulation by using vision-language representations to condition continuous action generation, yet these representations are not explicitly optimized for action-level conditioning. Existing action-reasoning approaches add modules for action plans or action-space signals, often requiring separate action-generation frameworks. CAC-VLA introduces a context-gated latent-action interface trained natively inside the VLM, predicting coarse-to-fine latent actions from future segments and using a context gate to calibrate their influence on an expert controller. Experiments on LIBERO and LIBERO-Plus show 98.3% and 89.5% average success rates, respectively.","arXiv :2607 .048 16v 1 [ cs .RO] 6 Jul 2026  \nCAC-VLA: Context-Gated Action Conditioning for Vision-Language-Action Models  \nYifu Xiong1* , Wenhao Yu1*†, Jiaxuan Lin1 , Bojun Zou1 , Jiahao Li1 Lu Zhang2‡, Yanyong Zhang1 , Jianmin Ji1‡  \n1University of Science and Technology of China (USTC)  \n2Institute of Artificial Intelligence, Hefei Comprehensive National Science Center  \nAbstract: Vision-Language-Action (VLA) models have become a promising paradigm for generalist robot manipulation, where visual-language representations are used to condition continuous action generation. However, these representations are not explicitly optimized for action conditioning, leaving the action expert to bridge the gap between multimodal understanding and precise motor control. Recent action-reasoning methods introduce additional modules to generate explicit action plans or action-space reasoning signals, demonstrating the benefit of action-level guidance but often requiring separate action-generation frameworks. We propose CAC-VLA, a Context-Gated Action Conditioning framework that learns a lightweight latent-action interface directly within the VLM. Instead of generating executable trajectories, CAC-VLA trains the VLM to predict coarseto-fine latent actions, which are structured representations encoded from future action segments, and adaptively leverages them to condition the action expert via a context gate. This enables VLM-native action conditioning while calibrating the influence of latent-action guidance on expert action generation. Experiments on LIBERO and LIBERO-Plus demonstrate the effectiveness of CAC-VLA, achieving 98.3% average success rate on LIBERO and 89.5% on LIBERO-Plus, suggesting that context-gated latent-action conditioning is an effective interface for continuous expert control.  \nKeywords: VLA, Action Conditioning, Robotic Manipulation  \n1 Introduction  \nA robot that follows language instructions must not only understand what the task means, but also determine what physical actions should be executed. Recent Vision-Language-Action (VLA) models [1, 2, 3, 4, 5] have made significant progress toward generalist robot manipulation by coupling pretrained Vision-Language Models (VLMs) [6, 7] with action experts that generate continuous robot commands. These models benefit from the semantic knowledge and multimodal reasoning capabilities of large-scale VLMs, enabling promising generalization across diverse objects, scenes, and instructions. However, most VLA architectures still use VLM representations as the primary interface between multimodal understanding and action generation. This leaves the transformation from perception and language to control largely implicit: the action expert must infer both the task-level action structure and the fine-grained motor commands from representations that are not explicitly organized around actions.  \nA growing line of work introduces intermediate reasoning signals to make this transformation more structured. Language-based approaches decompose tasks into subgoals or textual plans [4, 8], while vision-based approaches synthesize goal images or future observations [9, 10, 3] . More recent action-reasoning methods further move the intermediate representation closer to the control space  \n* Equal contributions,† Project leader,‡ Corresponding authors  \nby generating action plans, reference trajectories, or action-space priors to guide downstream action prediction [11] . These works suggest that action generation can benefit from intermediate guidance that is more directly tied to the action space than generic visual-language representations. Nevertheless, many existing action-reasoning methods obtain such guidance through additional actiongeneration or action-reasoning modules. Moreover, the resulting action guidance is often treated asa fixed conditioning signal, although its reliability and usefulness may vary with the current scene, task phase, and action-generation state.  \nThis motivates","cbCaimxCT5ThCA6q","https://ap.wps.com/l/cbCaimxCT5ThCA6q","pdf",3095151,1,16,"English","en",105,"# Introduction","[{\"question\":\"What problem does CAC-VLA address in vision-language-action models?\",\"answer\":\"Most VLA architectures use vision-language representations implicitly for control, so the transformation from perception and language to fine-grained motor commands is not explicitly structured around actions. CAC-VLA tackles this by adding an explicit latent-action interface within the VLM.\"},{\"question\":\"How does CAC-VLA represent “latent actions” and use them for control?\",\"answer\":\"CAC-VLA trains the VLM to predict coarse-to-fine latent actions encoded from future action segments. These latent predictions serve as action-structured conditioning for an action expert, which still generates the final continuous robot commands.\"},{\"question\":\"What role does the context gate play in CAC-VLA?\",\"answer\":\"Latent actions may be incomplete, noisy, or misaligned with the current control requirement, and their usefulness can vary across task phases. The context gate adaptively calibrates how much the latent-action guidance influences expert action generation.\"}]",1784198110,40,{"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},"cac-vla-context-gated-action-conditioning-for-vision-language-action-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/cac-vla-context-gated-action-conditioning-for-vision-language-action-models/84766/",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 CAC-VLA address in vision-language-action models?","Question",{"text":75,"@type":76},"Most VLA architectures use vision-language representations implicitly for control, so the transformation from perception and language to fine-grained motor commands is not explicitly structured around actions. CAC-VLA tackles this by adding an explicit latent-action interface within the VLM.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does CAC-VLA represent “latent actions” and use them for control?",{"text":80,"@type":76},"CAC-VLA trains the VLM to predict coarse-to-fine latent actions encoded from future action segments. These latent predictions serve as action-structured conditioning for an action expert, which still generates the final continuous robot commands.",{"name":82,"@type":73,"acceptedAnswer":83},"What role does the context gate play in CAC-VLA?",{"text":84,"@type":76},"Latent actions may be incomplete, noisy, or misaligned with the current control requirement, and their usefulness can vary across task phases. The context gate adaptively calibrates how much the latent-action guidance influences expert action generation.","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},[93,97,101,105,110,115,119,122,127,130,134],{"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":28,"slug":118},7,"Healthcare","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":125,"slug":126},9,"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":45,"category_name":128,"show_sort_weight":125,"slug":129},"World Cup","world-cup",{"id":131,"doc_module":4,"doc_module_name":45,"category_name":132,"show_sort_weight":131,"slug":133},10,"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":45,"category_name":136,"show_sort_weight":106,"slug":137},19,"General","general"]