Redefining Multimodal Microscopic Traffic Modeling with Information‐Theoretic Influence Areas and Self‐Supervised Learning

The concept of the influence area is fundamental to developing accurate traffic models, with significant implications for traffic simulation and Automated Vehicle (AV) collective perception by improving information transmission in Connected Cooperative and Automated Mobility (CCAM). In this work, we investigate the existence and define the extent of the influence area of various vehicle types in urban road networks using an information-theory-based approach. We develop an enhanced self-supervised model based on a LSTM-Transformer Autoencoder to replicate the car-following behavior of all vehicle types. The approach is trained and evaluated on high-resolution trajectory data obtained using Unmanned Aerial Vehicles (UAV-drones) from the center of Athens, Greece. The results regarding the influence area indicate that its extent is approximately 50 m for conventional passenger cars, slightly lower for motorcycles, and higher for buses and heavy vehicles. The proposed traffic model, exploiting influence area-specific input, achieves remarkably enhanced performance, indicating the suitability of the modelling approach and the benefits of considering the area of influence as input.