Traffic jams will be a distant memory thanks to AI traffic lights, developed by researchers at Aston University

Long queues at traffic lights could be a thing of the past, thanks to a new artificial intelligence system developed by researchers at Aston University. Inefficient traffic light control is one of the main causes of congestion in urban road networks. Dynamically changing traffic conditions and real-time traffic state estimation are fundamental challenges that limit the ability of the existing signaling infrastructure to provide individualized real-time signal control.

This first-of-its-kind system reads live camera images and adjusts the lights to compensate, helping to smooth traffic and reduce congestion. Researchers are using deep reinforcement learning (DRL) to address these challenges. Due to the economic and security constraints associated with training such agents in the real world, a practical approach is to do so in simulation prior to deployment. Domain randomization is an effective technique to bridge the reality gap and ensure efficient transfer of agents trained in simulation to the real world.

In testing, the system far outperformed all other methods, which usually rely on manually designed phase transitions. In 2019, it was estimated that traffic congestion in UK urban areas results in around 115 hours of lost time – and 894 of wasted fuel and lost income – every year for the average UK resident. A major cause of congestion is improper timing of traffic lights.

The researchers built a state-of-the-art photorealistic traffic simulator, Traffic 3D, to train their program, teaching it to handle different traffic and weather scenarios. When the system was tested on a real intersection, it then adapted to real intersections, although it was fully trained on simulations. So it could be effective in many real-world settings.

A fully autonomous, vision-based DRL agent that performs adaptive signal control in complex, imprecise and dynamic traffic environments has been developed. The agent uses live visual data (i.e. a stream of real-time RGB sequences) from an intersection to extensively perceive the traffic environment and then act upon it.

Using domain randomization, researchers examine the agent’s generalization abilities under varying traffic conditions, both in the simulation and in the real world. In a diverse validation set, independent of the training data, the traffic control agent reliably adapted to new traffic situations and demonstrated a positive transfer to real intersections never seen before, despite having t trained entirely in simulation.

Dr Maria Chli, a computer science lecturer at Aston University, explains: We set it up as a traffic control game. The program receives a “reward” when it drives a car through a crossroads. Every time a car has to wait or there is a traffic jam, there is a negative reward. In fact, we do not intervene; we just control the reward system.

Currently, the main form of traffic light automation used at intersections is based on magnetic induction loops; a wire is placed on the road and records the cars passing over it. The program counts them and then reacts to these data. As the AI ​​created by the team at Aston University sees a high volume of traffic before the cars have passed through the lights and makes its decision then, it is more responsive and can react faster .

Aston University computer science lecturer Dr George Vogiatzis said: “The reason we based this program on learned behaviors is that it can understand situations it hasn’t explicitly experienced. previously. We tested it with a physical obstacle causing the congestion, rather than the phasing of traffic lights, and the system still performed well. As long as there is a causal link, eventually the computer will figure it out. It is an extremely powerful system.

A netizen who goes by the name AKLmfreak says: Humans can literally cause a traffic jam on 6 lanes of a perfectly straight highway with no intersections or ramps. I appreciate that deep machine learning is trying to help, but we need deep cranial learning where I live. That’s what I thought. Public transport and infrastructure built with non-motorized transport in mind are the real answers to traffic problems. And we have the ability to implement these solutions now.

Netizen says the technology won’t do what the media headlines say: I’m not saying this smart traffic light technology is bad (although there are arguments to be made for that), but it does won’t do what the headlines want us to believe at all and it’s not the right tool for the job in the first place.

The program can be configured to visualize any crossroads, real or simulated, and will start learning on its own. The reward system can be manipulated, for example to encourage the program to let emergency vehicles through quickly. But the program always learns by itself, rather than being programmed with specific instructions. The researchers hope to start testing their system on real roads this year.

Source : Universit d’Aston

And you?

What is your opinion on the subject?

See as well :

The Dutch scandal is a warning for Europe on the risks linked to the use of algorithms, the tax administration has ruined thousands of lives with an algorithm

Clearview AI, the controversial facial recognition specialist, has agreed to curb its sales in the United States, after a resounding lawsuit

AI Models Remain Racist, Even With More Balanced Training, NIST Report Says

New startup says its AI can analyze human motions and come up with solutions to better manipulate them to hit sales targets