Congestion Estimation at Junction Using Reinforcement Learning

Traffic jam in city intersections is a big problem because of growing vehicle numbers and static traffic control systems limitations. Usual fixed-time traffic lights can’t match changing traffic conditions, causing problems such as longer wait times, unequal flow of cars and pollution of the environment. To solve these issues, this project studies how to use State-Action-Reward-State-Action algorithm - a type of learning method that encourages improvements over time - for making optimization in controlling traffic lights more flexible and adjustable. This project puts in place a system based on State-Action-Reward-State-Action in an imitation environment using Python’s Pygame library. This is to replicate a four-way intersection. Important aspects being researched include the learning rate, discount factor and exploration rate that control how adaptable and performing the algorithm can be. The function of this system involves adjusting timings for traffic lights according to real- time patterns of congestion with the goal of reducing queue sizes as well as delay time for vehicles on average. Metrics like total throughput, waiting times of vehicles, and lengths of queues are monitored for assessing system effectiveness to guarantee productive learning and adaptability Key Words: Traffic jam, City Intersection , Vehicle Number, congestion, Delay Time, Pollution.