AI traffic light system aims to end congestion

Long queues at traffic lights are a thing of the past thanks to a new artificial intelligence system developed by researchers at Aston University.

The system – the first of its kind – reads live camera images and adjusts the lights to compensate, keep traffic flowing and reduce congestion.

The system uses deep reinforcement learning, where a program understands when things aren’t going well and tries a different course of action — or continues to improve as it progresses.

In testing, the system significantly outperformed all other methods, which typically rely on manually designed phase transitions.

In 2019, it was estimated that congestion in the UK’s urban areas results in the average UK resident wasting around 115 hours of time each year – and £894 wasted on fuel and lost income. An important cause of congestion is insufficient timing of traffic lights.

The researchers built a state-of-the-art photorealistic traffic simulator, Traffic 3D, to train their program and teach it to deal with various traffic and weather scenarios. When the system was tested on a real intersection, it then adapted to real traffic interchanges, despite being fully trained on simulations. It can therefore be effective in many real world situations.

dr. Maria Chli, reader in Computer Science at Aston University, explains: “We set this up as a traffic control game. The program gets a ‘reward’ when it gets a car through a fork. Every time a car has to wait or there is a breakdown there’s a negative reward. There’s really no input from us, we’re just checking the reward system.”

Currently, the main form of traffic light automation used at intersections relies on magnetic induction loops; a wire lies on the road and registers cars driving over it. The program counts that and then responds to the data. Because the AI ​​created by the team at Aston University “sees” the high volume of traffic before the cars have passed through the light and then makes a decision, it reacts faster and can react faster.

Dr George Vogiatzis, a senior lecturer in computer science at Aston University, said: “The reason we have based this program on learned behavior is that it can understand situations that it has not explicitly experienced before. We tested this with a physical obstacle that causes congestion, rather than traffic light phasing, and the system still performed well. As long as there is a causal connection, the computer will eventually figure out what that connection is. It is an intensely powerful system.

The program can be set to watch any traffic junction – real or simulated – and will start learning autonomously. The reward system can be manipulated, for example to encourage the program to allow emergency vehicles to pass quickly. But the program always learns itself, rather than being programmed with specific instructions.

The researchers hope to test their system on real roads this year.

Reference: Garg D, Chli M, Vogiatzis G. Fully autonomous vision-based traffic signal control: from simulation to reality. In: ACM; 2022. Accessed 12 May 2022.

This article has been republished based on the following materials. Note: Material may have been edited for length and content. For more information, please contact the said source.

Leave a Comment