Calibrating heterogeneous car-following models for human drivers in oscillatory traffic conditions

Accurately modeling the realistic and unstable traffic dynamics of human-driven traffic flow is crucial to being able to to understand how traffic dynamics evolve, and how new agents such as autonomous vehicles might influence traffic flow stability. This work is motivated by a recent dataset that allows us to calibrate accurate models, specifically in conditions when traffic waves arise. Three microscopic carfollowing models are calibrated using a microscopic vehicle trajectory dataset that is collected with the intent of capturing oscillatory driving conditions. For each model, five traffic flow metrics are constructed to compare the flow-level characteristics of the simulated traffic with experimental data. The results show that the optimal velocity-follow the leader (OV-FTL) model and the optimal velocity relative velocity model (OVRV) model are both able to reproduce the traffic flow oscillations, while the intelligent driver model (IDM) model requires substantially more noise in each driver's speed profile to exhibit the same waves.