[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85890-en":3,"doc-seo-85890-105":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},85890,687197207639,"Asher","https://ap-avatar.wpscdn.com/davatar_a8503ba1806abce46bf441b54a3ca4cd",8,"Research & Report","PrismAD Decoupled Planning via Semantic Mixture of Planners for End to End Autonomous Driving","PrismAD presents a decoupled end-to-end autonomous driving framework built on a Semantic Mixture-of-Planners. Prior planners aggregate heterogeneous scene tokens into a single coupled representation, limiting factor-specific reasoning for agent interaction, road geometry, and driving intention. PrismAD partitions tokens into interaction, geometry, and intent groups and assigns them to independent planning experts with shared architecture but separate parameters. A semantics-aware router adaptively aggregates expert predictions for motion prediction and ego planning, using sparse top-K activation with noisy gating to improve robustness and reduce computation. Experiments on nuScenes open-loop and NeuroNCAP closed-loop show competitive results.","PrismAD: Decoupled Planning via Semantic Mixture-of-Planners for  \nEnd-to-End Autonomous Driving  \nKang Ding 1 ,2 ,3 , Zhigui Lin5 , Hongsong Wang4 , Jie Gui3 , Qi Liu 1 ,2 ,4 , Zhe Wang6 , Luqi Tang6 , and Lei He 1 ,2†  \narXiv :2607 . 10336v1 [ cs .RO] 11 Jul 2026  \nAbstract—This letter presents PrismAD, a decoupled endto-end autonomous driving framework based on a Semantic Mixture-of-Planners. Existing planners usually aggregate heterogeneous scene tokens into a coupled representation space, forcing a single planning branch to jointly model agent interaction, road geometry, and driving intention. Such coupling may weaken factor-specific reasoning and obscure the contribution of different planning cues. To address this limitation, PrismAD partitions scene tokens into interaction, geometry, and intent groups, and assigns them to independent planning experts with the same architecture but separate parameters. Each expert learns a specialized motion-planning representation, while a semantics-aware router adaptively aggregates expert predictions with separate routing weights for motion prediction and ego planning. Sparse top-K activation with noisy gating is further introduced to improve routing robustness and reduce unnecessary expert computation. Extensive experiments on the nuScenes open-loop dataset and NeuroNCAP closed-loop benchmark demonstrate that PrismAD exhibits competitive performance. Our code will be released soon.  \nI. INTRODUCTION  \nEnd-to-end autonomous driving has become an important research direction because it directly maps sensory observations to planning decisions and reduces the hand-crafted dependencies in traditional modular pipelines. Recent planningoriented methods have made steady progress by integrating perception, prediction, mapping, and planning into unified neural frameworks [1], [2], [3] . In particular, query-based and sparse-representation planners have shown promising efficiency by representing traffic scenes with compact object, map, and ego tokens. However, planning remains challenging because a safe ego trajectory is jointly determined by multiple heterogeneous factors, including dynamic interactions  \n*This work is supported by National Key R&D Program of China 2024YFB2505500 and Guangxi Science and Technology Major Program (No.AA24206054)  \n†Corresponding Author: Lei He. E-mail: [helei2023@tsinghua.edu.cn](helei2023@tsinghua.edu.cn)  \n1 School of Vehicle and Mobility, Tsinghua University, Beijing 100084, China.  \n2 State Key Laboratory of Intelligent Green Vehicle and Mobility, Tsinghua University, Beijing 100084, China.  \n3 School of Cyberspace Security, Southeast University, Nanjing 210096, China.  \n4 School of Computer Science and Engineering, Southeast University, Nanjing 210096, China.  \n5 School of Automotive Engineering, Wuhan University of Technology, Hubei Key Laboratory of Advanced Technology for Automotive Components, Hubei Collaborative Innovation Center for Automotive Components Technology, and Hubei Research Center for New Energy & Intelligent Connected Vehicles, Wuhan 430070, China.  \n6 SAIC GM Wuling Automobile Co., Ltd., Guangxi Laboratory of New Energy Automobile, and Guangxi Key Laboratory of Automobile Artificial Intelligence, Liuzhou 545000, China.  \nFig. 1. Comparison of our approach with existing methods. (a) Traditional end-to-end Paradigm. (b) Classical MoE-based Paradigm. (c) Our proposed Paradigm.  \nwith surrounding agents, local road geometry, and high-level navigation intention. Most existing end-to-end planners[4],[5], [6], [7] aggregate these heterogeneous scene tokens into a single coupled representation space. Although this unified design simplifies the model architecture, it forces one planner to simultaneously reason about factors with different semantic roles. For example, surrounding-agent tokens mainly support interaction reasoning and collision avoidance, map tokens provide geometric constraints from lanes and road boundaries, and command-related tok","cbCaic3xiBZopKWh","https://ap.wps.com/l/cbCaic3xiBZopKWh","pdf",10867529,4,1,"English","en",105,"# Introduction\n## Problem with coupled planners\n## Need for expert specialization and semantic routing\n# Proposed Approach\n## Token partitioning into interaction, geometry, and intent\n## Semantic mixture-of-planners and routing mechanism\n## Sparse top-K activation and noisy gating\n# Experiments\n## nuScenes open-loop evaluation\n## NeuroNCAP closed-loop benchmark","[{\"question\":\"What limitation do existing end-to-end autonomous driving planners face?\",\"answer\":\"They typically aggregate heterogeneous scene tokens into a coupled representation space, forcing one planning branch to jointly model interaction, geometry, and intention. This can weaken factor-specific reasoning and obscure the contribution of different planning cues.\"},{\"question\":\"How does PrismAD improve over coupled planning?\",\"answer\":\"PrismAD partitions scene tokens into interaction, geometry, and intent groups and routes them to independent planning experts. Each expert learns a specialized motion-planning representation with separate parameters for each semantic factor.\"},{\"question\":\"What mechanisms does PrismAD use to make routing robust and efficient?\",\"answer\":\"A semantics-aware router aggregates expert predictions with separate routing weights for motion prediction and ego planning. It also introduces sparse top-K activation with noisy gating to improve routing robustness while reducing unnecessary expert computation.\"}]",1784206971,20,{"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},"prismad-decoupled-planning-via-semantic-mixture-of-planners-for-end-to-end-autonomous-driving","",{"@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/prismad-decoupled-planning-via-semantic-mixture-of-planners-for-end-to-end-autonomous-driving/85890/",{"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","2026-07-26","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},"What limitation do existing end-to-end autonomous driving planners face?","Question",{"text":74,"@type":75},"They typically aggregate heterogeneous scene tokens into a coupled representation space, forcing one planning branch to jointly model interaction, geometry, and intention. This can weaken factor-specific reasoning and obscure the contribution of different planning cues.","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"How does PrismAD improve over coupled planning?",{"text":79,"@type":75},"PrismAD partitions scene tokens into interaction, geometry, and intent groups and routes them to independent planning experts. Each expert learns a specialized motion-planning representation with separate parameters for each semantic factor.",{"name":81,"@type":72,"acceptedAnswer":82},"What mechanisms does PrismAD use to make routing robust and efficient?",{"text":83,"@type":75},"A semantics-aware router aggregates expert predictions with separate routing weights for motion prediction and ego planning. It also introduces sparse top-K activation with noisy gating to improve routing robustness while reducing unnecessary expert computation.","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"]