BrotherResearchers from Aston University have developed an AI program that controls traffic lights according to real-time data for smooth traffic flow.
Simulate highway traffic using AI-controlled traffic lights. Image: Aston University
Aston University’s AI system reads images from the camera in real time and adjusts lights to keep traffic flowing and reduce congestion. The system uses deep reinforcement learning, where the program recognizes when it’s not doing well and tries another way or continues to improve. In testing, the system outperformed any other method that relies heavily on the operator to change the headlights.
In 2019, it is estimated that traffic congestion in urban areas in the UK costs residents about 115 hours, the amount of wasted fuel and lost income amounts to $1,090 per year. The main cause of congestion is the timing of the lamp signal not being matched. The researchers built a state-of-the-art traffic model called Traffic 3D to teach their program to handle a variety of traffic and weather situations. When tested on a real intersection, the system adapts itself to the intersections despite being trained purely on simulation. As a result, it can be effective in a real-life context.
Dr. Maria Chli, a computer science researcher at Aston University, explains that he and his colleagues set up the simulation as a traffic control game. The program receives a “reward” for taking the car through an intersection. Each time the car has to wait or there is a traffic jam, the program will be deducted from the reward. The team is only involved in controlling the reward system.
Currently, the main form of automatic traffic lights used at intersections depends on the magnetic induction circuit that recognizes passing vehicles. The program counts the number of vehicles and processes the data. Because the Aston University team’s AI “sees” high traffic before the car goes through the lights and makes a decision, it can react more quickly.
The program can be set to observe any actual or simulated intersection and start learning automatically. The reward system can be manipulated to incentivize the program to help emergency vehicles pass faster. The researchers hope to begin testing the program on the road this year.
An Khang (According to Phys.org)
at Blogtuan.info – Source: vnexpress.net – Read the original article here
