Drone-Assisted Lane Change Maneuver using Reinforcement Learning with Dynamic Reward Function

This paper provides a Lane Change Assistance (LCA) platform that communicates with Unmanned Aerial Vehi-cles (UAV). The proposed platform is based on a reinforcement learning technique, where a Deep Q-Network (DQN) is trained to make lane change decisions. The reward function of the DQN agent considers safety, comfort and efficiency perspectives. Specifically, the safety reward, based on the road vehicular density, is adapted dynamically by the drone during the training phase. Performance analysis proves that the proposed platform improves the total travel time while reducing the collision rate and responding to urgent lane changes in a timely manner.