Approximation of an Autonomous Navigation System in UAVs Based on Reinforcement Learning

This study aimed to identify the UAV features that most significantly impact energy consumption using Machine Learning (ML) models. A UAV simulation was developed in CoppeliaSim to model UAV features and analyze them with ML models. The simulated UAV followed predefined paths while varying its mass to simulate irrigation scenarios. Although a wind environment was designed, data collection was limited to no-wind conditions due to time constraints. Key features captured included flight time, traveled distance, X and Y traveled distance, mean velocity, and X and Y mean velocity. Energy consumption at each waypoint was calculated using methodologies that converted exerted force to RPM. Despite simulator limitations requiring thrust modeling through exerted force, the energy values obtained were consistent with expected magnitudes. Multiple ML models were implemented, with the Bagged Trees model showing the best performance (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> = 0.7286), identifying travel time as the most influential factor, followed by mean velocity. These findings enhance the understanding of UAV energy consumption during agricultural irrigation. Future work should incorporate the wind environment for more realistic simulations and use simulators that allow UAV control through RPM and torque for more accurate energy calculations.