Characterizing Heterogeneous Car-Following Behaviors of Human Drivers in Mixed Traffic

With an increasing number of autonomous vehicles (AVs) being tested and deployed on public roads, the sharing of the road between human-driven vehicles (HVs) and AVs will become inevitable. However, there is a lack of research on the interactions between HVs and AVs, particularly in terms of how HVs behave differently when following AVs compared to following HVs, and how this potential difference affects traffic safety and stability. To fill this gap, we utilized the Lyft dataset to investigate the variances between HV-following-AV and HV-following-HV scenarios. Hierarchical clustering was used to classify the driving styles of the following vehicles in all following events. Then, driving volatility measures and surrogate safety measures, i.e., speed volatility and reciprocal time-to-collision (reTTC) value, were used to compare the outcomes of different car-following events. The clustering results showed that human drivers exhibited four different car-following styles in the mixed traffic and when following AVs and HVs, drivers may exhibit different following styles. Compared to the leading HV, the leading AV resulted in a smaller TTC (i.e., a larger reTTC), regardless of the driving styles of the following HVs. When driving at low speeds, drivers had lower speed volatility when following AVs, indicating more stable traffic; while at high speed, when drivers were not sensitive to changes in the speed of the vehicle ahead, the leading AVs led to larger speed volatility. Future AV control algorithms should take the styles of the following HV into consideration to better optimize traffic stability and improve the safety of mixed traffic.