AirEye: UAV-Based Intelligent DRL Mobile Target Visitation

Abstract

From traffic monitoring to livestock tracking, and military reconnaissance to marine discovery, unmanned aerial vehicles (UAVs) are indispensable. Its dependence on a battery for power supply limits the flight time to visit all planned locations. Consequently, target visitation needs to be smart and minimize the mechanical energy. We propose to develop the AirEye UAV-based smart platform that can perform target visitation in the shortest time possible without knowing targets' exact locations, but with a known probabilistic distribution. We show how to integrate a UAV with the proper hardware to control it and execute commands from an on-ground command and control station. A pre-built machine learning model was modified to detect and identify targets, along with a reinforcement learning (RL) model to autonomously navigate the drone and ensure that all targets are visited while consuming minimal energy. We propose a drone energy model that can be used to estimate the total energy consumed by the drone in a complex scenario. We then use this energy model to compare the total energy consumed by the proposed RL-based technique in comparison with two other heuristic strategies, namely, random, and zigzag motion.