An unsupervised learning framework for detecting adaptive cruise control operated vehicles in a vehicle trajectory data

Abstract

The traffic dynamics are expected to change with the widespread utilization of advanced driver assistance systems (ADAS). Currently, simulation tools are adopted to capture the impacts of ADAS technologies on traffic dynamics. Real-world data collection of different ADAS technologies is required to support realistic modeling of these technologies in simulation tools. Vehicle trajectories are one of the cornerstones of modern traffic flow theory with applications in driver behavior studies and automated vehicle (AV) research. Unfortunately, the current trajectory datasets fail to provide any information on the utilization of ADAS technologies. This study proposes collecting and using a new trajectory dataset that contains multiple instances of probe vehicles using adaptive cruise control (ACC) to identify ACC-type behavior across the entire trajectory dataset. Since the trajectory data is not labeled based on ACC utilization, clustering is an excellent approach to arrange similar trajectories in the dataset into the same group. Using this dataset combined with clustering, this study identifies the vehicle trajectories with similar dynamics to the vehicles using ACC.