Abstract
Human drivers can have diverse car-following behaviors when interacting with connected and automated vehicles (CAVs) and other human-driven vehicles in mixed traffic where many human-driven vehicles and a limited number of CAVs frequently interact and share the road. In this study, Inverse Reinforcement Learning (IRL) is used to model unique car-following behaviors of different human drivers when interacting with the CAV and another human-driven vehicle by using their driving demonstrations collected from in-field driving tests. The learned driver behavior model is shown that the personalized driving behaviors accurately and consistently can be characterized when following the different types of preceding vehicles in a variety of traffic situations. Furthermore, the energy efficiency of different human-driven vehicles when interacting with the CAV and the human-driven vehicle is investigated with the heterogeneous characteristics of drivers’ behaviors, considering driving behaviors have significant influences on vehicle fuel economy. A detailed analysis reveals the significant fuel-saving benefits of the CAV to the following human-driven vehicles during the car-following scenario and the extent of such benefits varies among tested human drivers owing to their intrinsic preferences and perception of CAV. These findings suggest that human-CAV interactions can be effectively leveraged to improve the energy efficiency of mixed traffic.
Highlights
In recent years, the number of vehicles registered in the United States has shown a considerable increase [1], and CO2 emissions and the amount of fuel consumption have elevated substantially
The results show that connected and automated vehicles (CAVs) with eco-approach can help the following human-driven vehicle to achieve a 6% fuel economy improvement
A driver behavior model based on inverse reinforcement learning is designed to capture the driving behaviors of the different following human drivers when interacting with another human-driven vehicle and a CAV from the demonstration
Summary
The number of vehicles registered in the United States has shown a considerable increase [1], and CO2 emissions and the amount of fuel consumption have elevated substantially. It is crucial to investigate the differences and impacts of the human drivers’ behaviors on fuel economy when interacting with a preceding vehicle, whether it is another human-driven vehicle or a CAV. In our previous work [22], the IRL approach is used to mimic the observed driving style of different drivers with a cost function that captures the car-following preferences of an individual driver, such as driving comfort and relative distance to the preceding human-driven vehicle. A driver behavior model based on inverse reinforcement learning is designed to capture the driving behaviors of the different following human drivers when interacting with another human-driven vehicle and a CAV from the demonstration. The objective is to recover a unique cost function from the driver’s demonstration, which consists of human-driven vehicle trajectories such as speed and position, that can best explain the driving behaviors, assuming that the. The trajectory set is divided into the trajectory segments s1, s2, . . . , sN , and the following steps of the algorithm are applied for the driver behavior learning process
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
