A self-driving car, also known as an autonomous or automated vehicle, is designed to navigate complex environments autonomously. This study utilizes deep learning, enhancing vehicle autonomy and safety. Implementing Convolutional Neural Networks (CNNs) for visual perception and integrating sensor fusion techniques, the system gains a robust understanding of the environment, adapting dynamically to road conditions and unexpected obstacles. The system architecture is divided into four primary modules: car vision, sensor fusion, steering prediction, and ROS integration. The car vision module leverages CNNs for real-time lane detection, obstacle recognition, and traffic sign classification. Sensor fusion merges data from LIDAR, radar, and ultrasonic sensors, providing a 360-degree environmental view and precise object localization. The neural network-based steering prediction module continuously adjusts the vehicle's steering based on live driving data. Lastly, ROS integration ensures seamless communication among subsystems, supporting real-time decision-making. Tested in simulated environments, this structured approach aims to push autonomous vehicles towards full autonomy across diverse road networks, significantly enhancing operational safety and efficiency. The implementation showcases the potential for advanced autonomous systems to navigate with increased independence, marking a step forward in the evolution of self-driving technology.
Kamal et al. (Mon,) studied this question.