[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85785-en":3,"doc-seo-85785-105":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":4,"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},85785,8796095462418,"Noah","https://ap-avatar.wpscdn.com/avatar/80000253c1241d02b47?x-image-process=image/resize,m_fixed,w_180,h_180&k=1778826106357471780",8,"Research & Report","Runtime Safety Filtering for Learned Small UAS Separation Policies under GNSS Degradation","Learning-based separation assurance for small Unmanned Aircraft Systems (sUAS) using reinforcement learning can achieve near-zero collision rates in high-density simulations, but relies on accurate GNSS-derived position and velocity. In urban settings, multipath, blockage, and interference degrade navigation integrity, creating unsafe misinterpretations of traffic state. This work evaluates two runtime safety-filtering architectures under adversarial GNSS degradation: action filtering versus observation filtering. Both compute a worst-case traffic state under bounded uncertainty; observation filtering reduces near mid-air collisions by 90% and remains robust to control barrier tradeoffs.","Runtime Safety Filtering for Learned Small UAS Separation Policies under GNSS Degradation  \nAlex Zongo Peng Wei  \nDepartment of Mechanical and Aerospace Engineering  \nGeorge Washington University  \nWashington, D.C., USA  \n{a. zongo, [pwei](pwei}@gwu.edu)[}](pwei}@gwu.edu)[@gwu.edu](pwei}@gwu.edu) .  \narXiv :2607 . 10014v1 [ cs .RO] 10 Jul 2026  \nAbstract—Learning-based separation assurance, notably reinforcement learning, for small Unmanned Aircraft Systems (sUAS) has demonstrated strong performance in simulation, achieving near-zero collision rates while maintaining traffic throughput in high-density scenarios. However, these learned control policies assume accurate position and velocity information derived from Global Navigation Satellite Systems (GNSS); an assumption that fails in urban environments where multipath propagation, signal blockage, and intentional interference routinely degrade navigation integrity. Under such conditions, when deploying a learned aircraft separation assurance policy, a fundamental architectural question arises: should runtime safety mechanisms filter the policy’s actions to satisfy some constraints, or filter its observation input to present a conservative view of the traffic state? This work evaluates both approaches for multi-agent sUAS separation under adversarial GNSS degradation. Both architectures share a common first step: estimating a worst-case traffic state consistent with some bounded observation uncertainty. The approaches differ thereafter. Action filtering constrains policy outputs to satisfy hand-designed safety conditions, implemented via discretetime control barrier functions, and evaluated at the worst-case state. Observation filtering, on the other hand, presents the worstcase state to the policy as a corrected input, allowing the policy to determine its own response. Experimental evaluation reveals that action filtering provides negligible safety improvement, while observation filtering reduces near mid-air collisions by 90% and remains robust to the control barrier function’s tradeoff between separation distance and closing rate. These results suggest that, for policies with learned safety behaviors, preserving the policy’s decision authority outperforms overriding its actions with handdesigned control filtering.  \nIndex Terms—UAS separation assurance, GNSS Degradation, runtime safety, observation filtering, reinforcement learning, control barrier functions  \nI. INTRODUCTION  \nThe integration of small unmanned aircraft systems (sUAS) into low-altitude urban airspace presents a fundamental challenge: tactical separation assurance must scale beyond humansupervised operations while maintaining safety standards that aviation demands. Under the UAS Traffic Management (UTM) paradigm [1], [2], high-density traffic flows require decentralized deconfliction, where each aircraft makes autonomous speed or heading adjustments to maintain safe separation from nearby traffic. Multi-agent reinforcement learning (MARL) has emerged as a promising approach to this challenge, with recent work demonstrating separation policies that achieve  \nnear-zero collision rates while maintaining efficient traffic flow in simulation [3]–[6] . These policies learn sophisticated deconfliction behaviors through millions of training interactions, developing responses/strategies to multi-aircraft encounter scenarios that would be difficult to hand-design.  \nA critical assumption underlies this success: policies receive accurate aircraft state information derived from Global Navigation Satellite Systems (GNSS) . In urban environments, this assumption often fails. Multipath propagation from building surfaces introduces position errors exceeding tens of meters [8], [9] . Signal blockage in urban canyons creates coverage gaps. Intentional interference, whether it is jamming or spoofing, poses an increasing threat to unmanned operations [10]–[13] . When a separation control policy receives degraded observations, it","cbCaidfy06uuE3Oi","https://ap.wps.com/l/cbCaidfy06uuE3Oi","pdf",615185,1,9,"English","en",105,"# Introduction\n## Problem and motivation\n## Runtime safety filtering architectures\n## Contributions","[{\"question\":\"Why do learned sUAS separation policies struggle under GNSS degradation?\",\"answer\":\"They assume accurate aircraft state from GNSS. In urban environments, multipath, blockage, and interference can distort position and velocity, leading the policy to misread encounter geometry and reduce separation safety performance.\"},{\"question\":\"What are the two runtime safety filtering approaches compared in the work?\",\"answer\":\"The paper compares action filtering, which constrains policy outputs using safety conditions evaluated at a worst-case state, and observation filtering, which corrects policy inputs by presenting a worst-case state so the policy decides its own response.\"},{\"question\":\"Which approach improves safety more and what result is reported?\",\"answer\":\"Observation filtering reduces near mid-air collisions by 90%, while action filtering provides negligible safety improvement even though both use the same worst-case state estimation.\"}]",1784206272,23,{"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},"runtime-safety-filtering-for-learned-small-uas-separation-policies-under-gnss-degradation","",{"@graph":35,"@context":84},[36,53,67],{"@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/runtime-safety-filtering-for-learned-small-uas-separation-policies-under-gnss-degradation/85785/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":61,"encodingFormat":60,"isAccessibleForFree":62,"interactionStatistic":63},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-16",true,{"@type":64,"interactionType":65,"userInteractionCount":4},"InteractionCounter",{"@type":66},"ViewAction",{"@type":68,"mainEntity":69},"FAQPage",[70,76,80],{"name":71,"@type":72,"acceptedAnswer":73},"Why do learned sUAS separation policies struggle under GNSS degradation?","Question",{"text":74,"@type":75},"They assume accurate aircraft state from GNSS. In urban environments, multipath, blockage, and interference can distort position and velocity, leading the policy to misread encounter geometry and reduce separation safety performance.","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"What are the two runtime safety filtering approaches compared in the work?",{"text":79,"@type":75},"The paper compares action filtering, which constrains policy outputs using safety conditions evaluated at a worst-case state, and observation filtering, which corrects policy inputs by presenting a worst-case state so the policy decides its own response.",{"name":81,"@type":72,"acceptedAnswer":82},"Which approach improves safety more and what result is reported?",{"text":83,"@type":75},"Observation filtering reduces near mid-air collisions by 90%, while action filtering provides negligible safety improvement even though both use the same worst-case state estimation.","https://schema.org",{"og:url":51,"og:type":86,"og:title":13,"og:site_name":58,"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":20,"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":52,"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":21,"doc_module":4,"doc_module_name":45,"category_name":123,"show_sort_weight":124,"slug":125},"Religion & Spirituality",20,"religion-spirituality",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":127,"show_sort_weight":124,"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"]