Design and Implementation of Machine Learning Algorithm in UAV Intelligent Control

• Published in: 2025 International Conference on Electrical Drives, Power Electronics & Engineering (EDPEE) pp. 879 - 884
• Main Authors: Yan, Yan, Quan, Shenming
• Format: Conference Proceeding
• Language: English
• Published: IEEE 26.03.2025
• Subjects: Autonomous aerial vehicles; Collision avoidance; Convolutional neural networks; Decision making; Deep Q-Network; Deep Reinforcement Learning; Intelligent control; Machine Learning Algorithm; Machine learning algorithms; Robustness; Safety; Security; Training; UAV
• DOI: 10.1109/EDPEE65754.2025.00159

With the rapid development of sensor technology, communication technology and computing power, UAV is developing from simple remote control operation to highly autonomous and intelligent direction. In this study, an algorithm framework based on Deep Reinforcement Learning (DRL) is proposed. By learning the control strategy directly from the original sensor data, the UAV's environmental awareness, decision-making efficiency and flight safety are enhanced. In this study, the intelligent control problem of UAV is modeled as markov decision processes (MDP), and the deep Q network (DQN) is used as the core model, and the convolutional neural network (CNN) is used to process high-dimensional input and output the Q value of the action. Through experience playback and target network technology, the training process is effectively stabilized. The simulation test results show that the autonomous navigation and obstacle avoidance ability of UAV control system based on DRL in complex environment is significantly better than that of traditional methods, and the average task completion time is reduced by about 30%, and the collision rate is reduced by nearly 80%. In addition, DRL method also shows superior obstacle avoidance ability in low light or complex meteorological conditions. This study not only provides technical support for UAV intelligent control, but also provides a new direction for future research on multi-UAV cooperative control, intelligent adaptation in complex environment and robustness and security of the algorithm.