A Reinforcement Learning Approach for Traffic Control

Intelligent traffic control is a key tool to achieve and to realize resource-efficient and sustainable mobility solutions. In this contribution, we study a promising data-based control approach, reinforcement learning (RL), and its applicability to traffic flow problems in a virtual environment. We model different traffic networks using the microscopic traffic simulation software SUMO. RL-methods are used to teach controllers, so called RL agents, to guide certain vehicles or to control a traffic light system. The agents obtain real-time information from other vehicles and learn to improve the traffic flow by repetitive observation and algorithmic optimization. As controller models, we consider both simple linear models and non-linear radial basis function networks. The latter allow to include prior knowledge from the training data and a two-step training procedure leading to an efficient controller training.