[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-seo-502071-105":3,"detail-sidebar-cat-0-en-105":81,"doc-detail-502071-en":131},{"code":4,"msg":5,"data":6},0,"ok",{"site_id":7,"language":8,"slug":9,"title":10,"keywords":11,"description":12,"schema_data":13,"social_meta":74,"head_meta":76,"extra_data":78,"updated_unix":80},105,"en","curricuvlm-towards-safe-autonomous-driving-via-personalized-safety-critical-curriculum-learning-with-vision-language-models","CurricuVLM: Towards Safe Autonomous Driving via Personalized Safety-Critical Curriculum Learning with Vision-Language Models","","Ensuring safety in autonomous driving systems remains a critical challenge, especially when encountering rare but potentially catastrophic safety-critical scenarios. Existing scenario-generation research has limited impact on policy learning, and adaptive training curricula that respond to an AV’s evolving behavior and bottlenecks are still underexplored. CurricuVLM proposes a Vision-Language Model–driven framework that analyzes agent behavior, pinpoints weaknesses, and dynamically generates personalized training scenarios from narrative descriptions. Experiments on the Waymo Open Motion Dataset show strong gains in navigation success, driving efficiency, and safety metrics, and the method integrates with reinforcement learning algorithms.",{"@graph":14,"@context":73},[15,34,56],{"@type":16,"itemListElement":17},"BreadcrumbList",[18,23,27,31],{"item":19,"name":20,"@type":21,"position":22},"https://docshare.wps.com","Home","ListItem",1,{"item":24,"name":25,"@type":21,"position":26},"https://docshare.wps.com/document/","Document",2,{"item":28,"name":29,"@type":21,"position":30},"https://docshare.wps.com/document/research-report/","Research & Report",3,{"item":32,"name":10,"@type":21,"position":33},"https://docshare.wps.com/document/curricuvlm-towards-safe-autonomous-driving-via-personalized-safety-critical-curriculum-learning-with-vision-language-models/502071/",4,{"url":32,"name":10,"@type":35,"image":36,"author":41,"headline":10,"publisher":44,"fileFormat":47,"inLanguage":8,"description":12,"dateModified":48,"datePublished":49,"encodingFormat":47,"isAccessibleForFree":50,"interactionStatistic":51},"DigitalDocument",{"url":37,"@type":38,"width":39,"height":40},"https://docshare.wps.com/thumbnails/curricuvlm-towards-safe-autonomous-driving-via-personalized-safety-critical-curriculum-learning-with-vision-language-models/502071.png","ImageObject",300,407,{"name":42,"@type":43},"Aditya","Person",{"url":19,"name":45,"@type":46},"DocShare","Organization","application/pdf","2026-10-17","2026-10-03",true,{"@type":52,"interactionType":53,"userInteractionCount":55},"InteractionCounter",{"@type":54},"ViewAction",11,{"@type":57,"mainEntity":58},"FAQPage",[59,65,69],{"name":60,"@type":61,"acceptedAnswer":62},"What problem does CurricuVLM address in autonomous driving safety?","Question",{"text":63,"@type":64},"It targets the challenge of improving safety policy learning for rare but potentially catastrophic safety-critical scenarios and filling the gap in personalized curricula that adapt to an AV’s evolving behavior and bottlenecks.","Answer",{"name":66,"@type":61,"acceptedAnswer":67},"How does CurricuVLM use vision-language models?",{"text":68,"@type":64},"CurricuVLM leverages VLM multimodal understanding to analyze agent behavior, identify performance weaknesses, reason over unsafe driving situations described as narratives, and generate tailored training scenarios.",{"name":70,"@type":61,"acceptedAnswer":71},"What evidence supports CurricuVLM’s effectiveness?",{"text":72,"@type":64},"Extensive experiments on the Waymo Open Motion Dataset show it outperforms state-of-the-art baselines on both regular and safety-critical scenarios, improving navigation success, driving efficiency, and safety metrics.","https://schema.org",{"og:url":32,"og:type":75,"og:title":10,"og:site_name":45,"og:description":12},"article",{"robots":77,"canonical":32},"index,follow",{"doc_id":79,"site_id":7},502071,1791359752,{"code":4,"msg":82,"data":83},"success",[84,88,92,96,101,106,111,115,120,123,127],{"id":22,"doc_module":4,"doc_module_name":25,"category_name":85,"show_sort_weight":86,"slug":87},"Story & Novel",90,"story-novel",{"id":26,"doc_module":4,"doc_module_name":25,"category_name":89,"show_sort_weight":90,"slug":91},"Literature",80,"literature",{"id":33,"doc_module":4,"doc_module_name":25,"category_name":93,"show_sort_weight":94,"slug":95},"Exam",70,"exam",{"id":97,"doc_module":4,"doc_module_name":25,"category_name":98,"show_sort_weight":99,"slug":100},5,"Comic",60,"comic",{"id":102,"doc_module":4,"doc_module_name":25,"category_name":103,"show_sort_weight":104,"slug":105},6,"Technology",50,"technology",{"id":107,"doc_module":4,"doc_module_name":25,"category_name":108,"show_sort_weight":109,"slug":110},7,"Healthcare",40,"healthcare",{"id":112,"doc_module":4,"doc_module_name":25,"category_name":29,"show_sort_weight":113,"slug":114},8,30,"research-report",{"id":116,"doc_module":4,"doc_module_name":25,"category_name":117,"show_sort_weight":118,"slug":119},9,"Religion & Spirituality",20,"religion-spirituality",{"id":118,"doc_module":4,"doc_module_name":25,"category_name":121,"show_sort_weight":118,"slug":122},"World Cup","world-cup",{"id":124,"doc_module":4,"doc_module_name":25,"category_name":125,"show_sort_weight":124,"slug":126},10,"Lifestyle","lifestyle",{"id":128,"doc_module":4,"doc_module_name":25,"category_name":129,"show_sort_weight":97,"slug":130},19,"General","general",{"code":4,"msg":82,"data":132},{"doc_id":79,"user_id":133,"nickname":42,"user_avatar":134,"doc_module":4,"category_id":112,"category_name":29,"doc_title":10,"doc_description":12,"doc_content":135,"file_id":136,"file_url":137,"file_type":138,"file_size":139,"view_count":55,"is_deleted":4,"is_public":22,"is_downloadable":22,"audit_status":22,"page_count":140,"language":141,"language_code":8,"site_id":7,"html_lang":8,"table_of_contents":142,"faqs":143,"seo_title":144,"seo_description":12,"update_tm":145,"read_time":146},962085564549,"https://ap-avatar.wpscdn.com/davatar_085a072bc5b1113ac321206ff7593b45","arXiv :2502 . 15119v1 [ cs .RO] 21 Feb 2025  \nCurricuVLM: Towards Safe Autonomous Driving via Personalized Safety-Critical Curriculum Learning with Vision-Language Models  \nZihao Sheng†a , Zilin Huang†a , Yansong Qub , Yue Lengc , Sruthi Bhavanamc , Sikai Chena,∗  \na Department of Civil and Environmental Engineering, University of Wisconsin-Madison, Madison, WI 53706, USA bLyles School of Civil and Construction Engineering, Purdue University, West Lafayette, IN 47907, USA  \nc Google, Sunnyvale, CA 94089, USA  \nAbstract  \nEnsuring safety in autonomous driving systems remains a critical challenge, particularly in handling rare but potentially catastrophic safety-critical scenarios. While existing research has explored generating safety-critical scenarios for autonomous vehicle (AV) testing, there is limited work on effectively incorporating these scenarios into policy learning to enhance safety. Furthermore, developing training curricula that adapt to an AV’s evolving behavioral patterns and performance bottlenecks remains largely unexplored. To address these challenges, we propose CurricuVLM, a novel framework that leverages Vision-Language Models (VLMs) to enable personalized curriculum learning for autonomous driving agents. Our approach uniquely exploits VLMs’ multimodal understanding capabilities to analyze agent behavior, identify performance weaknesses, and dynamically generate tailored training scenarios for curriculum adaptation. Through comprehensive analysis of unsafe driving situations with narrative descriptions, CurricuVLM performs in-depth reasoning to evaluate the AV’s capabilities and identify critical behavioral patterns. The framework then synthesizes customized training scenarios targeting these identified limitations, enabling effective and personalized curriculum learning. Extensive experiments on the Waymo Open Motion Dataset show that CurricuVLM outperforms state-of-the-art baselines across both regular and safety-critical scenarios, achieving superior performance in terms of navigation success, driving efficiency, and safety metrics. Further analysis reveals that CurricuVLM serves as a general approach that can be integrated with various RL algorithms to enhance autonomous driving systems. The code and demo video are available at: [https://zihaosheng.github.io/CurricuVLM/](https://zihaosheng.github.io/CurricuVLM/) .  \nKeywords: Autonomous Driving, Vision-Language Models, Safety-Critical Systems, Curriculum Learning, Reinforcement Learning  \n1. Introduction  \nThe advancement of artificial intelligence and sensor technologies has enabled autonomous vehicles (AVs) to achieve remarkable progress over the past decades, particularly in perception, decision-making, and control capabilities (Di and Shi, 2021; Wu et al., 2024; Sheng et al., 2024c; Huang et al., 2024a) . These advancements have facilitated the transition of autonomous driving from theoretical research to preliminary commercial deployment, with leading technology companies like Waymo and Cruise initiating pilot autonomous taxi services in selected cities across the United States (Kusano et al., 2024; California Department of Motor Vehicles, 2023) . While AVs promise to revolutionize transportation with improved mobility and reduced emissions, their widespread deployment remains constrained by persistent safety-related challenges (Ding et al., 2023; Cao et al., 2022; Huang et al., 2024c; He et al., 2024) . Real-world traffic environments are inherently complex, presenting a “long tail” of rare but critical scenarios, such as sudden pedestrian crossings or abrupt loss of control by nearby vehicles. These scenarios, while infrequent, often carry significant consequences and present substantial challenges. Due to their urgency and safety-critical nature, instantaneous decision-making is demanded, where subtle misjudgments could lead to fatal outcomes. Even human drivers require extensive training to handle them reliably. In light of these challenges, e","cbCairLVv9V5uu0m","https://ap.wps.com/l/cbCairLVv9V5uu0m","pdf",2188916,22,"English","# Introduction\n## Safety-critical scenarios in autonomous driving\n## Literature overview: policy-level vs scene-level approaches","[{\"question\":\"What problem does CurricuVLM address in autonomous driving safety?\",\"answer\":\"It targets the challenge of improving safety policy learning for rare but potentially catastrophic safety-critical scenarios and filling the gap in personalized curricula that adapt to an AV’s evolving behavior and bottlenecks.\"},{\"question\":\"How does CurricuVLM use vision-language models?\",\"answer\":\"CurricuVLM leverages VLM multimodal understanding to analyze agent behavior, identify performance weaknesses, reason over unsafe driving situations described as narratives, and generate tailored training scenarios.\"},{\"question\":\"What evidence supports CurricuVLM’s effectiveness?\",\"answer\":\"Extensive experiments on the Waymo Open Motion Dataset show it outperforms state-of-the-art baselines on both regular and safety-critical scenarios, improving navigation success, driving efficiency, and safety metrics.\"}]","CurricuVLM: Towards Safe Autonomous Driving via Personalized Safety-Critical Curriculum Learning with Vision-Language Models | PDF",1791028540,55]