Science

No more traffic jams thanks to the traffic light system controlled by artificial intelligence

⇧ [VIDÉO] You may also like this affiliate content (after ads)

In the near future, traffic lights may be able to decide for themselves whether you have the right to pass or not. Scientists at Ashton University have just developed an artificial intelligence (AI)-based system that can determine whether a traffic light should turn green or not based on real-time traffic.

“It is estimated that in 2019 traffic jams in UK urban areas cost the average UK resident about 115 hours a year – and £894 in wasted fuel and lost income,” Ashton University said in a press release. details this innovation. “One of the main causes of traffic congestion is a mismatch in traffic light schedules.”

Therefore, the research team decided to tackle this problem and think about a more efficient traffic light system that will make traffic smoother by taking into account real-time data. To do this, they relied on artificial intelligence, or rather on a program equipped with an automatic learning system. A success, they say, as early tests were able to demonstrate that their program smoothed motion more efficiently than any other automatic system already in existence. Their findings were published by the university and presented at the Autonomous Agents and Multi-Agent Systems conference on May 11, 2022.

The principle of their technology is relatively simple, as AI detects traffic density based on the image from the video camera and adjusts the traffic lights accordingly. Depending on what she can discern through the video, she makes a decision and, above all, can learn from her mistakes in order to improve her scores as she passes the tests.

Dealing with surprises

As training, the scientists had a simulation program called Traffic 3D. It offers different road and weather conditions… Ambulances have even been implemented to test the behavior of the AI ​​towards them. With each success, the system received an acknowledgment, a kind of “reward” a bit like when you want to train an animal.

The team “set it up as a traffic management game. The program receives a “reward” when it drives a car through an intersection. Every time a car has to wait where there is a congestion, there is a negative “reward”. In fact, there is no input from us; we just control the reward system,” says Maria Chli, senior lecturer in computer science at Ashton University.

If machine learning is favored, it is because it allows the system to adapt to situations even if they have not been specifically tested or programmed in advance. For example, the team put in a physical barrier to create an unusual bottleneck, and the program was still able to adapt. “As long as there is a causal relationship, the computer will eventually figure out what the relationship is. It’s an extremely powerful system,” says George Vogiatzis, professor of computer science at Aston University.

When the AI ​​was tested on a real road, it managed to adapt. Therefore, researchers believe that it can be used in many real-life situations. Traffic light automation systems already exist. However, according to the scientists involved in this study, they usually rely on a detection system that is triggered when a car passes through a magnetic induction loop. So this allows traffic lights to adapt to traffic, but not really predict it, as AI can do based directly on visual data.

The researchers hope that this year they will be able to start testing their system in real conditions. They say that it can be placed at any intersection so that it begins to learn “on its own.” In addition, the deployment will not necessarily be costly as traffic cameras are already installed in many cities such as London, even giving access to the public in a designated area.

Aston University.

Back to top button