Improved Real-Time UAV Positioning Using Extended Kalman Filter and Machine Learning

This paper discusses the integration of satellite and inertial navigation systems for Unmanned Aerial Vehicles (UAVs) to enhance positioning accuracy, particularly in urban environments where satellite signals are often blocked. Since standalone inertial navigation systems accumulate errors over time, fusing satellite and inertial data using Kalman filters— specifically, the Extended Kalman Filter (EKF)—is a common approach for UAV navigation. To further improve accuracy, this study combines EKF with machine learning models to refine the sensor data outputs. The results show that machine learning significantly enhances position estimates in complex and dynamic flight scenarios. Various machine learning algorithms—such as linear regression, K-Nearest Neighbors (KNN), decision trees, and random forests—were compared based on performance metrics like Mean Absolute Error (MAE) and R². Among these, Random Forest outperformed the others, demonstrating strong fitting abilities, excellent generalization, and effective handling of complex data relationships. This method offers a promising solution for precise, real-time UAV navigation in urban areas.