DQN-Augmented V2X Communication with AoI-Based Data Freshness for Safer Autonomous Driving

Autonomous vehicles (AVs) struggle to achieve accurate perception, rapid decision-making, and reliable real-time communication in complex urban environments. Challenges such as incorrectly sensing its environment, delayed or incorrect decision-making, and unreliable communication with other vehicles and infrastructure can reduce AV performance. To address this problem, we propose a comprehensive framework to improve AV operations through simulations and deep learning. We created simulations in Webots, focusing on vehicle-to-everything (V2X) communication, including vehicle-to-vehicle (V2V) and vehicle-to-pedestrian (V2P) interactions. The simulation environment features ten vehicles, each equipped with multiple sensors, including LiDAR, camera, GPS, and gyroscope. We tested scenarios such as obstacle detection, managing sudden pedestrian crossings, and effective vehicle-to-vehicle (V2V) communications. We integrate a rule-based system with the Deep Q Network (DQN) to enhance the decision-making process. We also included Age of Information (AoI) to ensure that vehicles take action on the freshest data sent. The results show improved safety and efficiency of AVs in complex urban environments.