Spatio-Temporal AI Modeling for Urban Traffic Calibration - A SUMO-Based Approach

This conference talk explores innovative AI solutions for urban traffic management challenges, focusing on spatio-temporal learning techniques for traffic calibration. Learn about the development of an AI engine that utilizes digital twin scenarios powered by microscopic simulations to capture detailed vehicle behaviors including interactions, lane changes, and driver dynamics. The presentation introduces the Dynamic Spatio-Temporal Graph Attention Network (DSTGAT), a sophisticated hybrid model combining multi-head Graph Attention Networks (GATv2) with Long Short-Term Memory (LSTM) networks to process urban traffic data represented as graphs. Discover how this approach exploits joint spatio-temporal relationships by analyzing sequential traffic snapshots, with GATv2 layers extracting spatial embeddings and LSTM aggregating these patterns over time to predict future traffic flows in real-time. The talk, presented by Pablo Manglano-Redondo and co-authored with Alvaro Paricio-Garcia and Miguel A. Lopez-Carmona, demonstrates how the AI engine's iterative feedback loop continuously calibrates traffic control parameters, showing promising results in reducing congestion and improving urban mobility through SUMO-based implementation.