Optimal Control of Electric Vehicles in Dynamic Wireless Charging Lanes Using Deep Q-Networks

The increasing adoption of electric vehicles (EVs) presents new challenges in optimizing energy consumption, charging efficiency, and travel time. Dynamic Wireless Charging Lanes (DWCLs) offer a promising solution by allowing EVs to charge while driving, reducing dependency on static charging stations and minimizing downtime. However, optimizing the decision-making process for when and at what speed a vehicle should enter and exit a DWCL while maintaining an efficient driving strategy remains a complex problem. This work proposes a Deep Q-Network (DQN)-based reinforcement learning (RL) approach to dynamically control lane-switching and speed adjustments in a multi-lane road environment. The RL agent is designed to balance energy efficiency, travel time minimization, and charging maximization, while also considering real-world constraints such as road topology, accurate energy consumption modeling, and vehicle alignment with charging coils. By leveraging CARLA as a high-fidelity simulation environment, we implement and train a policy that autonomously decides lane-switching, speed adaptation, and charging strategies. The proposed framework not only improves the charging efficiency of EVs but also ensures smooth lane transitions while maintaining energy-optimal driving behavior. Experimental results show that the RL agent effectively learns the optimal use of DWCL, minimizing unnecessary lane switches and ensuring a highly efficient, and real-time deployable EV management system.