Dronuum: A smart and energy efficient drone application within the Computing Continuum

This paper presents Dronuum, a drone-based application designed within the Computing Continuum to enable efficient, real-time wildfire detection and environmental monitoring. Addressing the inherent limitations of Unmanned Aerial Vehicles (UAVs), notably constrained computational resources, energy capacity, and intermittent connectivity, the proposed system leverages a distributed, microservice-oriented architecture spanning Edge (drone) and Cloud environments. By integrating principles of Osmotic Computing and leveraging Liquid Computing (LIQO)-enabled multi-cluster orchestration, Dronuum can migrate application components across heterogeneous resources in response to mission requirements. The system decomposes the wildfire detection pipeline into modular services, including image acquisition, preprocessing, inference, and alerting. A lightweight YOLOv8n-based classifier is employed for fire detection. Experimental evaluation, conducted on a testbed combining Raspberry Pi Edge nodes and Cloud Virtual Machines (VMs), demonstrates the effectiveness of the proposed approach. Results indicate that classification accuracy is independent of the deployment scenario. Energy measurements confirm that LIQO-based offloading reduces the per-image energy cost from 6.85mWh to 4.58mWh, enabling up to 49.6% more images per battery charge, while microservice migration incurs a service downtime below (Formula presented) .