Система автономного керування дроном для перегонів на основі згорткових нейронних мереж

Abstract

This paper presents the development of an autonomous control system for a racing drone based on convolutional neural networks (CNNs). Modern drone racing requires high-speed decision-making, precise navigation, and robustness in dynamic environments, where traditional control approaches often become insufficient. The proposed system utilizes onboard visual data processed by CNNs to detect track elements, estimate the drone’s position, and generate control commands in real time. Special attention is given to end-to-end learning and hybrid approaches that combine deep learning with classical control algorithms. The advantages of CNN-based methods, including fast visual processing and adaptability, are discussed alongside challenges such as computational complexity and sim-to-real transfer. The results demonstrate that the integration of convolutional neural networks significantly improves the performance and autonomy of racing drones.