RLAgentBasedTraffic​Control

Traffic congestion is always a daunting problem that affects people's daily life across the world. The objective of this work is to develop an intelligent traffic signal management to improve traffic performance, including alleviating traffic congestion, reducing waiting times, improving the throughput of a road network, and so on. Traditionally, traffic signal control typically formulates signal timing as an optimization problem. In this work, reinforcement learning (RL) techniques have been investigated to tackle traffic signal control problems through trial-and-error interaction with the environment. Comparing with traditional approaches, RL techniques relax the assumption about the traffic and do not necessitate creating a traffic model. Instead, it is a more human-based approach that can learn through trial-and-error search. The results from this work demonstrate the convergence and generalization performance of the RL approach as well as a significant improvement in terms of less waiting time, higher speed, collision avoidance, and higher throughput.