Intelligent Control of Automated Driving Systems in Vehicles Using Neural Networks and Genetic Algorithm

Abstract

Evolutionary neural networks are a form of machine learning that uses genetic algorithms to train a neural network by estimating neural network weights, optimization of learning rules, as well as optimization of neural network architecture, thus obtaining more accurate and less human-dependent neural networks.

 This research uses a genetic algorithm as an optimal way to choose the weights of the neural network in order to obtain the optimal steering angle for a self-driving car. Each car represents a different chromosome in the generation, in other words a unique set of neural network weights, which are evaluated and transmitted to the next generation by the fitness function (fitness score). The fitness function is the distance traveled by the vehicle along the path without colliding with the boundary of the path, so that the vehicle that stayed the longest period on the path is the best vehicle. The trained car was tested on several methods that differ from the training path, and according to different speeds, and the practical results proved the efficiency of the algorithm used to reach an appropriate steering angle.