SOFTWARE-HARDWARE COMPLEX FOR AUTOMATED VEHICLE CONTROL USING NEURAL NETWORK MODELING AND GENETIC ALGORITHMS

Abstract

The article presents a hardware-software complex for providing autopiloting using neural network modeling and genetic algorithms. The complex uses the YOLO v6 library for real-time object recognition and for further decision-making regarding the road situation. The authors propose the use of a multilayer perceptron, the training of which was optimized using a genetic algorithm, which significantly reduced the duration of its training. The developed system uses DirectX 11 to transmit images from any camera, which allows integrating various visual data sources. The main scientific result is that an improved approach to automatic vehicle control using a multilayer perceptron has been proposed, the accelerated training of which is carried out using a genetic algorithm, which allows a significant increase in system performance. One of the key advantages of the approach proposed by the authors is the ability to adapt to changes. As the software-hardware complex for automated driving of a car collects more data about the environment, the genetic algorithm can update the weights of the neural network, increasing the accuracy of decisions and the safety of the system. The collected data is used to further improve the system based on new generations of GA.
The study shows that the combination of genetic algorithms and MLP allows to significantly accelerate learning and increase the accuracy of decision-making in autopilot systems. The use of YOLO v6 for real-time object detection provides high performance and adaptability of the system to changing conditions.