Traffic light controllers are often unable to handle traffic congestion. Existing systems are based on a theoretical controller or base that is responsible for changing traffic lights based on traffic conditions. The goal is to reduce vehicle delays during normal traffic conditions and increase vehicle productivity during congestion. However, current traffic light controllers cannot achieve such variable goals, and a human controller can only manage a few intersections.

Reinforcement learning (RL) is adopted to solve this problem, but RL usually works in a static environment, and traffic environments are not static.

// –>