Energy-Efficient Autonomous Vehicle Control Using Reinforcement Learning and Interactive Traffic Simulations

Connected and autonomous vehicles are expected to improve mobility and transportation, as well as to provide energy efficiency benefits. The integration of safety and energy efficiency aspects is challenging as there are certain tradeoffs between them, and also because the assessment of these attributes requires different time horizons. This paper illustrates the development of a controller for highway driving that, through reinforcement learning, can simultaneously address requirements of safety, comfort, performance and energy efficiency for battery electric vehicles. The training process of the decision policy exploits traffic simulations that are capable of representing the interactive behavior of vehicles in traffic based on game theory. Results indicate the potential for improved energy efficiency by adding powertrain-related states in the decision policy and by suitably defining the reward function.