Implementation of monocular visual SLAM with ARCog-NET for aerial robot swarm indoor mapping

This paper presents a real-time distributed framework for Unmanned Aerial Vehicle (UAV) swarms operating in GPS-denied indoor environments, built on the Aerial Robot Cognitive Network Architecture (ARCog-NET) cognitive architecture and enhanced with monocular visual Simultaneous Localization and Mapping (SLAM). ARCog-NET employs a multi-layered Edge-Fog-Cloud (EFC) hierarchy that supports decentralized decision-making and adaptive coordination, including dynamic path optimization. Through a novel formulation, each UAV jointly estimates its own trajectory and contributes to a shared 3D reconstruction of the environment by exchanging matched visual landmarks across the network. The system dynamically adapts navigation paths in response to operational events using reinforcement learning guided by trajectory coverage metrics and historical decision weights. A full deployment with six DJI Ryze Tello UAVs was conducted in a controlled indoor lab environment, demonstrating autonomous swarm navigation and collaborative 3D mapping. Performance was evaluated through metrics such as trajectory error, point cloud fidelity, decision convergence, and knowledge reuse rate. Results confirm that the proposed method enables scalable, autonomous SLAM and planning capabilities in real-world UAV networks, highlighting the cognitive synergy between navigation and perception in distributed aerial robotics.