A team has coupled an aerial perception system based on the YOLOv8 neural network with a reinforcement-learning drone route planner, aiming to coordinate supply delivery in emergency scenarios. The proposal is presented as a closed-loop proof of concept: visual detection feeds real-time decisions on vehicle assignment and replanning.

The work addresses an open problem in humanitarian logistics: how to keep terrain perceptions and delivery decisions consistent when conditions change during an operation. Conventional systems tend to separate computer vision from route planning, which forces reliance on precomputed maps and hinders adaptation to obstacles or unforeseen demand.

In the proposed architecture, YOLOv8 identifies and classifies relevant elements in images captured by the drones, while a reinforcement-learning agent adjusts routes and resource allocation based on that information. The closed loop allows visual updates to directly modify the planning policy without human intervention.

The piece is described as a proof of concept, so its results focus on the feasibility of the coupling rather than large-scale validation. The authors note that integrating calibrated perception and adaptive planning could reduce response times in relief operations, although trials under real conditions will be needed to quantify its advantage over existing methods.