[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-84870-en":3,"doc-seo-84870-105":28,"detail-sidebar-cat-0-en-105":89},{"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":11,"language":21,"language_code":22,"site_id":23,"html_lang":22,"table_of_contents":24,"faqs":25,"seo_title":13,"seo_description":14,"update_tm":26,"read_time":27},84870,8796095461564,"Liam","https://ap-avatar.wpscdn.com/davatar_155a257f0dc6eb9ab79c44ca47cae57d",8,"Research & Report","Delay-Aware Active Triangulation with Uncertainty-Driven Multi-Agent Reinforcement Learning for Counter-UAS","Multi-agent active visual triangulation improves precise 3D localization of aerial targets in Counter-UAS by coordinating mobile observers and controllable cameras. Existing approaches assume instantaneous state feedback, overlooking cumulative latency from detection, communication, and decision propagation. A delay-aware, uncertainty-driven multi-agent reinforcement learning framework is introduced using a DecPOMDP with AoI-augmented observations, demonstrating a 10.6 percentage-point validity gain. Controlled comparisons show perception-consistent rewards improve aggregate accuracy and reduce track losses, revealing a stability–robustness tradeoff. Multisource covariance propagation reduces RMSE up to 2.8× versus angular-only modeling, and MAPPO results reach 0.547 ± 0.217 m RMSE with 78.1% validity.","Delay-Aware Active Triangulation with Uncertainty-Driven Multi-Agent Reinforcement Learning for Counter-UAS  \nSeungwook Lee 1 , David Hyunchul Shim 1  \narXiv :2607 .05957v 1 [ cs .RO] 7 Jul 2026  \nAbstract—Multi-agent active visual triangulation enables precise 3D localization of aerial targets by coordinating mobile observers with controllable cameras. However, existing methods assume instantaneous state feedback, ignoring cumulative latency from detection, communication, and decision propagation. We present a delay-aware, uncertainty-driven multiagent reinforcement learning framework for target localization in Counter-UAS applications. Our contributions are: (1) a DecPOMDP formulation with Age-of-Information (AoI) augmented observations enabling delay-aware coordination—AoI improves triangulation validity by 10.6 percentage points; (2) a controlled comparison showing that perception-consistent rewards outperform privileged clean-state rewards (0 .547 m vs. 0.633m RMSE, 27% fewer track losses)—both policies are trained through identical observation noise but differ in what they are optimized for, producing a stability–robustness tradeoff; and (3) multisource analytical covariance propagation incorporating pixel, pose, gimbal, and intrinsics uncertainties—restricting to angular noise alone causes 2.8-fold RMSE degradation. Experiments with MAPPO in 4096 parallel environments achieve 0.547 ± 0.217m RMSE with 78. 1% triangulation validity, while MLP policies achieve near-zero validity (0 .7%), confirming recurrent memory as essential for delay compensation.  \nI. INTRODUCTION  \nThe proliferation of small unmanned aerial systems (UAS) has created urgent demand for Counter-UAS (C-UAS) technologies capable of detecting, tracking, and neutralizing aerial threats [1], [2] . Regardless of the neutralization method, effective C-UAS response depends on precise, continuous 3D target localization. Vision-based tracking offers passive operation, rich semantics, and scalable deployment on small platforms.  \nMulti-agent active triangulation—where mobile observers dynamically reposition to optimize measurement geometry—provides instantaneous 3D position estimates without the convergence delays of bearing-only filtering based methods [3] . Modern platforms further enable controllable sensors: gimbal-stabilized cameras with optical zoom decouples the viewing direction from platform motion and the distance from the subject.  \nDespite these capabilities, real-world multi-agent systems face a critical bottleneck: system delays. Cumulative latency from frame acquisition, object detection, inter-agent communication, and control propagation can range from tens to thousands of milliseconds. These delays are asymmetric: ego-state estimates benefit from high-rate onboard sensors and computations, whereas inter-agent observations traverse  \n1 S. Lee and D. H. Shim are with the Korea Advanced Institute of Science and Technology (KAIST), Daejeon, Korea. Email: {seungwook1024, [hcshim](hcshim}@kaist.ac.kr)[}](hcshim}@kaist.ac.kr)[@kaist.ac.kr](hcshim}@kaist.ac.kr)  \nthe full delay pipeline. Prior perception-aware methods are predominantly single-agent [4] and do not encounter this multi-agent delay structure.  \nGavin et al. [5] demonstrated multi-agent RL for triangulation with analytical covariance-based rewards and real world flight tests. We extend their framework to incorporate realistic deployment conditions: stochastic communication delays with perspective asymetric staleness, controllable gimbalzoom sensors with field-of-view constraints, and multisource uncertainty propagation. Our three contributions are:  \n1) Delay-aware Dec-POMDP formulation for active triangulation: We show that multi-agent active triangulation requires explicit delay modeling under realistic communication conditions. Formulating the problem as a Dec-POMDP with AoI-augmented observations [6], [7] and recurrent policies, AoI improves triangulation validity by 10.6 %p, while MLP po","cbCainI5bmUlmYhA","https://ap.wps.com/l/cbCainI5bmUlmYhA","pdf",1224197,1,"English","en",105,"# Introduction\n## Counter-UAS and active perception\n## Multi-agent active triangulation\n## System delays and the need for delay-aware learning\n# Related Work\n## C-UAS and active perception\n# Method Contributions\n## Delay-aware Dec-POMDP with AoI-augmented observations\n## Reward design: privileged vs perception-consistent\n## Multi-source analytical covariance propagation","[{\"question\":\"Why is delay modeling necessary for multi-agent active triangulation in Counter-UAS?\",\"answer\":\"Real systems accumulate latency across frame acquisition, detection, inter-agent communication, and control propagation, creating asymmetric delays that break assumptions of instantaneous feedback. Delay modeling captures this structure explicitly for coordination and learning.\"},{\"question\":\"What does the proposed Dec-POMDP with AoI-augmented observations achieve?\",\"answer\":\"It incorporates Age-of-Information into observations so agents coordinate under staleness-aware delay dynamics. AoI improves triangulation validity by 10.6 percentage points compared with setups without it.\"},{\"question\":\"How do perception-consistent rewards compare with privileged clean-state rewards?\",\"answer\":\"Both are trained under identical observation noise but optimize different reward targets. Perception-consistent rewards improve aggregate metrics (e.g., lower RMSE) and reduce track losses, while privileged rewards provide more temporally stable coordination, indicating a stability–robustness tradeoff.\"}]",1784198929,20,{"code":4,"msg":29,"data":30},"ok",{"site_id":23,"language":22,"slug":31,"title":13,"keywords":32,"description":14,"schema_data":33,"social_meta":84,"head_meta":86,"extra_data":88,"updated_unix":26},"delay-aware-active-triangulation-with-uncertainty-driven-multi-agent-reinforcement-learning-for-counter-uas","",{"@graph":34,"@context":83},[35,52,66],{"@type":36,"itemListElement":37},"BreadcrumbList",[38,42,46,49],{"item":39,"name":40,"@type":41,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":43,"name":44,"@type":41,"position":45},"https://docshare.wps.com/document/","Document",2,{"item":47,"name":12,"@type":41,"position":48},"https://docshare.wps.com/document/research-report/",3,{"item":50,"name":13,"@type":41,"position":51},"https://docshare.wps.com/document/delay-aware-active-triangulation-with-uncertainty-driven-multi-agent-reinforcement-learning-for-counter-uas/84870/",4,{"url":50,"name":13,"@type":53,"author":54,"headline":13,"publisher":56,"fileFormat":59,"inLanguage":22,"description":14,"dateModified":60,"datePublished":60,"encodingFormat":59,"isAccessibleForFree":61,"interactionStatistic":62},"DigitalDocument",{"name":9,"@type":55},"Person",{"url":39,"name":57,"@type":58},"DocShare","Organization","application/pdf","2026-07-16",true,{"@type":63,"interactionType":64,"userInteractionCount":4},"InteractionCounter",{"@type":65},"ViewAction",{"@type":67,"mainEntity":68},"FAQPage",[69,75,79],{"name":70,"@type":71,"acceptedAnswer":72},"Why is delay modeling necessary for multi-agent active triangulation in Counter-UAS?","Question",{"text":73,"@type":74},"Real systems accumulate latency across frame acquisition, detection, inter-agent communication, and control propagation, creating asymmetric delays that break assumptions of instantaneous feedback. Delay modeling captures this structure explicitly for coordination and learning.","Answer",{"name":76,"@type":71,"acceptedAnswer":77},"What does the proposed Dec-POMDP with AoI-augmented observations achieve?",{"text":78,"@type":74},"It incorporates Age-of-Information into observations so agents coordinate under staleness-aware delay dynamics. AoI improves triangulation validity by 10.6 percentage points compared with setups without it.",{"name":80,"@type":71,"acceptedAnswer":81},"How do perception-consistent rewards compare with privileged clean-state rewards?",{"text":82,"@type":74},"Both are trained under identical observation noise but optimize different reward targets. Perception-consistent rewards improve aggregate metrics (e.g., lower RMSE) and reduce track losses, while privileged rewards provide more temporally stable coordination, indicating a stability–robustness tradeoff.","https://schema.org",{"og:url":50,"og:type":85,"og:title":13,"og:site_name":57,"og:description":14},"article",{"robots":87,"canonical":50},"index,follow",{"doc_id":7,"site_id":23},{"code":4,"msg":5,"data":90},[91,95,99,103,108,113,118,121,125,128,132],{"id":20,"doc_module":4,"doc_module_name":44,"category_name":92,"show_sort_weight":93,"slug":94},"Story & Novel",90,"story-novel",{"id":45,"doc_module":4,"doc_module_name":44,"category_name":96,"show_sort_weight":97,"slug":98},"Literature",80,"literature",{"id":51,"doc_module":4,"doc_module_name":44,"category_name":100,"show_sort_weight":101,"slug":102},"Exam",70,"exam",{"id":104,"doc_module":4,"doc_module_name":44,"category_name":105,"show_sort_weight":106,"slug":107},5,"Comic",60,"comic",{"id":109,"doc_module":4,"doc_module_name":44,"category_name":110,"show_sort_weight":111,"slug":112},6,"Technology",50,"technology",{"id":114,"doc_module":4,"doc_module_name":44,"category_name":115,"show_sort_weight":116,"slug":117},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":44,"category_name":12,"show_sort_weight":119,"slug":120},30,"research-report",{"id":122,"doc_module":4,"doc_module_name":44,"category_name":123,"show_sort_weight":27,"slug":124},9,"Religion & Spirituality","religion-spirituality",{"id":27,"doc_module":4,"doc_module_name":44,"category_name":126,"show_sort_weight":27,"slug":127},"World Cup","world-cup",{"id":129,"doc_module":4,"doc_module_name":44,"category_name":130,"show_sort_weight":129,"slug":131},10,"Lifestyle","lifestyle",{"id":133,"doc_module":4,"doc_module_name":44,"category_name":134,"show_sort_weight":104,"slug":135},19,"General","general"]