Search

Intelligent traffic light control is one of the modern approaches to solve traffic congestion, where reinforcement learning is a widely used method. Conventionally, reinforcement learning is used to determine whether to change the current phase (or choose a traffic phase) after each small interval. One major drawback of these approaches is that it makes the current traffic light phase duration uncertain before the current phase terminates. Directly determining the duration of the traffic light phase can effectively avoid this shortcoming. An adaptive traffic light timing system is proposed in this paper which can directly control the phase duration. In the proposed system, the Q-learning algorithm is employed and the action space is defined as all possible phase durations. In addition, the reward function is redefined to guide the agent to balance more traffic metrics, and the state is redefined to reduce the state space. Finally, the proposed system is evaluated by equal, unequal, and complex traffic scenarios. Results show that the proposed system has a better performance compared with other methods in controlling traffic lights, even on complex traffic scenarios. [ABSTRACT FROM AUTHOR]

Reinforcement Learning (RL) is an effective method for adaptive traffic signals control. As one type of RL, the teacher-student framework has been found helpful in improving the model performance for different application fields (such as robot control, game, hybrid intelligence), but it is rarely applied for traffic control due to that the hyper-parameters and the number of state-action pairs experienced are difficult to determine. In this work, the teacher-student framework is used for traffic signal control, where only a single reward function is designed to guide the student agent and by using this method the number of hyper-parameters and the model complexity are reduced. Specifically, the teacher agent uses an importance function to evaluate and guide the student, where the importance function combines with environment reward to form a synthetic reward for the student agent. Experimental results under different traffic environments show that the proposed method achieves the expected performance enhancement and is better than most of the state-of-the-art RL-based traffic signal control methods. [ABSTRACT FROM AUTHOR]