Reinforcement Learning for Traffic Light Control

What is this project about?

Traffic signal controllers generally work on electro-mechanical controllers that use dial timers with fixed, signalized intersection time plans. These systems are inefficient in dealing with different scenarios encountered at crossroads. In this project, I propose a reinforcement learning method for controlling traffic lights in a wide range of scenarios. We created a crossroad environment whose variables can be altered to create multiple settings. The Q-learning algorithm was used to train the RL agent.