Deep Q-Network Model for Intelligent Traffic Light

The problem of controlling traffic lights at inter-sections has begun to be regulated using artificial intelligence (AI) or machine learning. One of the most widely implemented methods is by using Q-Learning. However, this method has some limitations, such as the inflexibility of the number of states and the need for appropriate policies in order for the system to be converged. For these reasons, in this paper, we propose a method to control the traffic light cycle using Deep Q-Network (DQN), a combination of Reinforcement Learning (RL) and Neural Networks (NN). The proposed method can eliminate the state limitation problem which becomes a problem for Reinforcement Learning. This proposed method is compared to a conventional traffic light control system in their queue length performance.