Application of Deep Learning to Generate Auto Player Mode in Car Based Game

This paper sets out a framework for constructing deep learning models for autonomous driving in Forza Horizon 5. By playing a game for few hours, we focus on a merging model of AlexNet and Deep Reinforcement Learning known as AlexNet-DRL. The data are useful for research and include ambient characteristics, model choices, and vehicle telemetry. Time efficiency, collision rates, and perception accuracy are all evaluated as well. Comparatively, the performance analysis of autonomous driving in object recognition, collision avoidance, and real-time decision making can be done in comparison to AlexNet-DRL. Although AlexNet-DRL shows promise, it only has 92.5% perceptual accuracy, and humans have 95%. The human player had a collision rate of 1.5%, while the model had 3.4%. Results contribute to the development of models on DRL for autonomous navigation that can be more advanced than humans.

I. Introduction

Recently, autonomous driving has grown significantly, with the potential to transform transportation and mobility [1]. Deep learning, especially convolutional neural networks like AlexNet, may help build autonomous driving systems. The automotive industry has advanced autonomous driving technology from driver assistance to completely autonomous vehicles [2]. Deep learning, especially convolutional neural networks, has helped autonomous driving by processing and interpreting complicated, highdimensional sensor input. This study uses the AlexNet-DRL paradigm to create an auto player mode for self-driving cars in games [3] [4] [5]. The popular racing game Forza Horizon 5 is used to test the direct perception technique in a complicated and dynamic setting. This game's realistic graphics and numerous driving settings are ideal for testing the model's ability to navigate complex virtual landscapes.