Efficient self-driving control for lead vehicle following in a mixed traffic environment

Abstract

Autonomous control for vehicles is rapidly developed since it not only improves the mixed traffic flow but also decreases the traffic accidents and the energy consumption. Especially self-driving technology has been attention since it progressively plays an important role in prompting comport driving condition as well as safety. In addition, eco-driving strategy is considered as the energy efficient driving technology of electric vehicles (EVs) since the eco-driving can increase the fuel economy of the limited power capacity of battery. This research introduces an energy efficient EVs control strategy with Model predictive control (MPC) and deep neural networks (DNN). MPC and DNN are combined to solve the optimal problem that is to minimize battery consumption and maximize the energy efficiency during self-driving. The estimated vehicle state information by DNN and the ratio of power consumption at the previous state are fed into the next state information of the MPC controller. Thus, the controller can optimize the energy consumption ratio with respect to the state values such as the vehicle state and the battery state of charge. The hybrid control algorithm was evaluated using PGDrive which is a virtual vehicle driving environment, and simulation results showed that the MPC controller with DNN can efficiently reduce the battery consumption and increase the energy efficient up to 1.7%.

Keywords

• Eco-driving
• Self-driving
• Electric vehicles
• Learning predictive control
• Deep reinforcement learning
• Energy consumption