Automated Driving Control in Mixed Traffic Flow Using V2V Communication

Abstract

Automated vehicles (AVs) are expected to transform the future of intelligent transportation systems. To enhance the feasibility of integrating AVs with human-driven vehicles (HDVs), AV technology needs to accurately assess driving situations and operate within human expectations. Car-following (CF) and lane-changing (LC) are key models for improving AV driving control. However, most CF and LC models were developed separately and tested on a few vehicle samples, without considering their impacts on traffic flow. Therefore, this study develops an automated driving control model that simultaneously generates CF and LC decisions. The model utilizes vehicle-to-vehicle (V2V) technology that incorporates thirty-four features from the subject vehicle and six surrounding vehicles. A hybrid deep learning model is proposed using LSTM with two parallel structures for each maneuver. By applying the NGSIM dataset, the proposed model outperformed all other models with the accuracy of matching ratios of 70.5%, 96.1%, and 98.4% at acceleration, speed, and position levels, respectively. Traffic flow simulations were conducted, validating that vehicles perform driving behaviors within human expectations and the simulations are reliable for describing the actual traffic flow. For comfort and safe AV driving, jerk, or change rate of acceleration, is restricted within a certain range. With the AV rate increasing, traffic congestion was mitigated with speed increasing from 47.5 to 74.9 km/h and density decreasing from 36.8 to 25.4 veh/km. However, traffic safety remained unstable unless the rate reached 100%. This study contributes to the development of V2V-based AV driving control in mixed traffic scenarios.