Steering Angle Prediction for Self-Driving Car Using Proposed CNN Architecture

Automobiles offer significant advantages in terms of convenience, time efficiency, privacy, and cargo capacity. However, they also present challenges such as high maintenance costs, parking difficulties, and, most notably, reliance on human drivers, whose decision-making can be compromised by various factors. This often leads to road accidents, with human error contributing to over half of these incidents. To resolve this issue, this study proposes creating a deep learning-based model that adjusts the steering angle when a vehicle deviates from its lane. With the advantages of convolutional neural networks (CNNs), the proposed model aims to predict and correct steering angles in real-time, thereby enhancing vehicle control and reducing the risk of accidents. As shown by the experimental results, the proposed model secured an accuracy of 82% in predicting steering angles during testing, demonstrating its effectiveness in learning, and generalizing steering angle adjustments. The operational performance of the model will be assessed through comparison with the predicted and actual steering angles, which will be visualized through simulated driving video data. Therefore, the primary objective is to minimize human error in driving, thereby ensuring safer and more reliable transportation solutions.