Intelligent traffic light control using collaborative Q-Learning algorithms

In most large cities with high population densities, there are many problems that arise due to traffic jam as it happens in metropolitan cities in Indonesia. There are many factors that contribute to traffic jam or traffic congestion, for instance high traffic density, inadequate road capacity, bad driving behaviors, and ineffective traffic light setting. From those factors, traffic light setting problem is what we try to address. If the traffic light sequence and duration of green light can be set based on the condition at the intersections at that time (adaptive) of course, this will help reduce congestion. Basically, the main problem in traffic light control is having to make right decisions sequentially. One method that matches the characteristics of these problem is a Reinforcement Learning (RL). With this background and the results of previous studies, the author has developed a simulation application about Intelligent Traffic Light Control with Collaborative Q-Learning Algorithms methods. The purpose of this research is to optimize the waiting time at traffic light control based on method of collaborative Q-Learning that can be used as a reference model for the solution of traffic congestion in real world. Based on test results, it can be concluded that the Collaborative Q-Learning Algorithms is the best traffic light control algorithms among the other method tested with a waiting time is 54.67 seconds. The simulation process is using Green Light District Simulator. The test results showed that the best parameters for Collaborative Q-Learning algorithms method are the learning rate is 1 and the discount factor is 0.8, and, the immediate reward method is results of the subtraction of the waiting time before, and the waiting time at the moment.