Eco-Driving With Partial Wireless Charging Lane at Signalized Intersection: A Reinforcement Learning Approach

Abstract

Consumer electronics such as advanced GPS, vehicular sensors, inertial measurement units (IMUs), and wireless modules integrate vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) within internet of thing (IoT), enabling connected autonomous electric vehicles (CAEVs) to optimize energy optimization through eco-driving. In scenarios with traffic light intersections and partial wireless charging lanes (WCL), an eco-driving algorithm must consider net and gross energy consumption, safety, and traffic efficiency. We introduced a deep reinforcement learning (DRL) based eco-driving control approach, employing a twin-delayed deep deterministic policy gradient (TD3) agent for real-time acceleration planning. This approach uses reward functions for acceleration, velocity, safety, and efficiency, incorporating a dynamic velocity range model which not only enables the vehicle to smoothly pass the signalized intersections but also uses partial WCL efficiently and time-adaptively while ensuring traffic efficiency in diverse traffic scenarios. Tested in Simulation of Urban Mobility (SUMO) across various intersections with partial WCL, our method significantly lowered net and gross energy consumption by up to 44.01% and 17.19%, respectively, compared to conventional driving, while adhering to traffic and safety norms.