Network Load Reduction Using Variational Auto-Encoder for Connected and Automated Vehicles

To address the growing challenges of road safety, efficiency, and environmental sustainability, Cooperative, Connected, and Automated Mobility (CCAM) leverages advanced communication and automation technologies across road networks. In Europe, Cooperative Intelligent Transportation Systems (C-ITS) facilitate communication among vehicles, infrastructure, and other entities, enhancing situational awareness and safety. Cooperative Awareness Messages (CAMs) provide continuous status information-such as vehicle position, speed, and direction-to nearby vehicles and infrastructure. Vehicles send CAMs to other vehicles at high frequencies (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$1-10 \text{Hz}$</tex>), which, especially in dense networks, risks overloading communication channels. This may degrade network performance and potentially reduce the effectiveness of road safety applications. In this paper, we propose a new deep learning-based communication mechanism to reduce the risk of channel overload in C-ITS. Our approach enables nearby vehicles to form a temporary trust group valid for a specified period. During this period, vehicles initially send CAMs at a frequency of 1 Hz. Then, each vehicle predicts the subsequent messages of its neighbors using a variational autoencoder. Next, vehicles periodically verify the predicted CAMs by sending actual CAMs. This process ensures prediction reliability and reduces channel load by allowing vehicles to decrease CAM transmissions. We validated our approach in a simulation environment (using Omnet++, Sumo, and Artery) demonstrating its effectiveness in maintaining road safety, while reducing communication overhead.