Data-Driven Adaptive Automated Driving Model in Mixed Traffic

Abstract

The interplay between Connected Automated Vehicles (CAVs) and Human-driven Vehicles (HDVs) in mixed traffic environments is often presumed to influence the behavior of the other, and the dynamic impacts of such interplay on traffic flows is a critical aspect that is absent in most existing studies. This study employs a data-driven optimization approach to model the driving behavior of Connected Automated Vehicles (CAV) in mixed traffic and investigates the impact of CAVs on overall traffic performance. Specifically, considering a scenario of a gradual increase in the penetration of CAVs in the conventional traffic stream, currently dominated by Human-driven vehicles (HDV), four possible car-following configurations are identified where a CAV has to behave differently. Regarding such configurations, existing car-following and lane-changing models of CAVs are tuned using a Lipschitzian optimization algorithm and a local search method with data obtained from the WAYMO Open Dataset. The developed driving model of CAVs is used to simulate mixed traffic on a freeway section attached to an on-ramp, which often induces traffic bottlenecks. Under varying market penetration of CAVs, traffic performances, including travel time, throughput, and string stability, are compared with conventional traffic. The findings suggest significant improvements at a network level, for example, by delaying and dampening shockwaves. However, on an individual level, CAVs feel hindered by the slower-moving HDVs.