Development of Algorithms and Game Mechanics of a Computer Game Using Machine Learning Methods

The object of development is the algorithm and game mechanics of a computer game using machine learning methods. The purpose of this work is to develop algorithms and game mechanics for controlling the movement of a car along a track using a genetic algorithm for training a neural network with the ability to save and load the obtained weights during training. As part of the final qualification work, a genetic algorithm and a neural network were developed using the C ++ programming language without extraneous libraries. The built-in UE4 vehicle class has been modified to scan the area around the vehicle. Upon completion of all the preparatory work, several tests were carried out to determine the optimal parameters and methods for training the neural network. After the completion of testing, the optimal parameters of the algorithm were selected and the final training of the neural network was carried out for 200 generations. The result was a generation capable of competing on the track with the average player. The developed program that implements the game mechanics of driving a car can be integrated into any project of a similar subject, based on the Unreal Engine 4 game engine. The volume of the final qualifying work is 52 pages, the number of illustrations is 26, the number of formulas is 8, the number of sources used is 32.