Vehicle Energy-Saving Motion Planning Based on Deep Q-Network

Abstract

Energy-saving driving of vehicles has become the general trend. This paper will take driverless vehicle as the research object, and use the algorithm of deep reinforcement learning to carry out energy-saving motion planning. First, a virtual driving simulation scene of driverless vehicle that can be applied with deep reinforcement learning is built. Secondly, the vehicle longitudinal DQN algorithm is designed. Aiming at the energy-saving driving behavior in the vehicle-following scenario, the proposed DQN controller is verified when the speed of preceding vehicle is fixed and variable. Last but not least, the problem of energy-saving passing through traffic light intersections is explored, mainly on two typical conditions in passing traffic lights: quickly pass and glide through. The vehicle based on DQN algorithm can pass the above two traffic light intersections without stopping midway and violating traffic rules. In the above experiments, compared with the comparison experiment, the average instantaneous energy consumption of the ego vehicle based on the DQN algorithm is reduced.