Traffic Signal Control for An Isolated Intersection Using Reinforcement Learning

Recent advances in reinforcement learning techniques have shown promising results in solving complex control problems with high dimensional state and action spaces. Inspired by the success, we show that two advanced reinforcement learning algorithms, Q-Learning and Q-Learning with function approximation, can predict the traffic signal timing plan for an isolated intersection. To deal with the large state cardinality of the input traffic state, we propose a discrete state representation and a finite set of actions for a sample case. The proposed algorithm helps to converge the reward function, average queue in early episodes. Our methods show promising results for an intersection traffic control simulated using SUMO micro simulator.