🚗 Advancements in Traffic Simulation based on cm-level Real-World Data We would like to share insights from a recent paper that explores modeling car-following behavior using #leveLXData's high-quality naturalistic traffic data. This research is crucial for improving #microscopic #traffic #simulations by enhancing accuracy and realism. For those in transportation planning, traffic engineering, and urban development, this study offers tools to optimize traffic flow and enhance road safety. By leveraging real-world data, we can gain a better understanding of driver behavior and develop more effective infrastructure solutions. Consider how leveLXData's vast trajectory database might be integrated into your projects for improved outcomes. #TrafficSimulation #UrbanPlanning #DroneData #ItelligentTransport
🤔Why do drivers still behave so differently, even in nearly identical traffic situations? I’m excited to share that our paper, “When Context Is Not Enough: Modeling Unexplained Variability in Car-Following Behavior,” is allocated to Transportation Research Part B, and will be presented in a lectern session at ISTTT 2026, widely regarded as the premier conference in traffic flow theory (only 36 papers were selected for podium presentation). This work reflects a question that has motivated much of my PhD: observable traffic context alone cannot fully explain human driving behavior. Even under similar spacing, speed, and relative speed, drivers show substantial differences shaped by heterogeneity, memory, and latent variability. In this paper, we develop a stochastic framework that learns context-dependent uncertainty and temporal correlations, leading to more realistic simulations and better-calibrated uncertainty on the leveLXData by fka HighD dataset. Beyond model performance, I find the behavioral insights especially interesting: drivers appear more cautious at tight headways, more reactive in critical spacing regimes, and more stable and anticipatory at wider gaps. The results also suggest that behavioral heterogeneity is itself context-dependent. I’m especially grateful that this will be my second ISTTT paper during my PhD :) I’m very thankful to my co-authors Zhengbing He and Cathy Wu, and especially my advisor, Lijun Sun, for their support. I’m looking forward to the discussions at ISTTT 2026. See you in Munich!