Reinforcement Learning for Multi-Agent Systems with an Application to Distributed Predictive Cruise Control

In this paper, we propose a reinforcement learning (RL) approach to the coordinated control of multi-agent systems with a special emphasis on intelligent transportation systems. As a result, for a network of autonomous vehicles, a novel distributed predictive cruise control (PCC) algorithm based on RL is proposed to reduce idle time and maintain an adjustable speed depending on the traffic signals. Under the proposed distributed PCC law, given the signal phase and timing (SPaT) message from upcoming traffic intersections, autonomous vehicles can cross intersections without stopping. The effectiveness of the proposed approach has been validated through Paramics microscopic traffic simulations by choosing a scenario in Statesboro, Georgia. For different traffic demands, the travel time and fuel consumption rate of vehicles are compared between non-PCC and PCC algorithms. Microscopic traffic simulation results show that the proposed PCC algorithm is able to reduce both fuel consumption rate and travel time.