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 major causes of traffic congestion in the urban road network. Dynamically changing traffic conditions and real-time traffic state estimates are fundamental challenges that limit the ability of existing signaling infrastructure to provide unique real-time signal control.
The first such system reads live camera images and adjusts the lighting to compensate, which helps to smooth traffic and reduce traffic congestion. Researchers are using deep reinforcement education (DRL) to address this challenge. Due to the economic and security constraints associated with training such agents in the real world, a practical approach is to do so in simulations before deployment. Domain randomization is an effective strategy to bridge the gap between realities and ensure efficient transfer of agents trained in simulation to the real world.
In testing, the system surpassed all other methods, usually relying on manually designed phase transitions. In 2019, it was estimated that approximately 115 hours of time wasted as a result of traffic jams in urban areas of the UK – and 894 wasted fuel and lost income – for the average UK resident each year. Improper timing of traffic lights is a major cause of traffic jams.
Researchers have created a state-of-the-art photorealistic traffic simulator, Traffic 3D, to train their program to teach them how to handle different traffic and weather conditions. When the system was tested at a real intersection, it was adapted to real intersections, although it was fully trained on simulation. So it can be effective in many real-world settings.
A fully autonomous, vision-oriented DRL agent that performs adaptive signal control in complex, obscure and dynamic traffic environments. Agents use live visual data (i.e. a stream of real-time RGB sequences) from an intersection to better understand the traffic environment and then work on it.
Using domain randomization, researchers tested the generalizability of agents under different traffic conditions, both in simulation and in the real world. In a varied validity set, free from training data, traffic control agents have reliably adapted to new traffic situations and, despite not being fully trained in simulation, have demonstrated a positive shift to real intersections never seen before.
Dr. Maria Chholi, a computer science lecturer at Aston University, explains: We set it up as a traffic control game. The program gets a “reward” when it drives a car through a road. Every time you have to wait for a car or there is a traffic jam, there is a negative reward. In fact we do not interfere; 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 car passing over it. The program counts them and then responds to these data. Since AI cars created by the Aston University team see a lot of traffic before passing through the light and then decide, it can respond more responsibly and quickly.
Dr George Vogiatzis, a computer science lecturer at Aston University, said: “The reason we build this program on learning behavior is because it understands situations that it has never experienced before. Has done and the system still performs well.As long as there is a causal link, the computer will finally get it out.This is a very powerful system.
A netizen named AKLmfreak says: People can literally create a traffic jam in the 6 lanes of a completely straight highway without an intersection or ramp. I’m grateful that deep machine learning is trying to help, but we need deep cranial learning wherever I live. That’s what I thought. Public transport and infrastructure, built with non-motorized transportation in mind, are the real answer to the traffic problem. And we now have the ability to implement this solution.
Netizens say technology won’t do what media headlines say: I’m not saying this smart traffic light technology is bad (although it has to be argued), but headlines won’t do what we want them to believe at all and it’s right to work in the first place. Not a tool.
The program can be configured to imagine any crossroads, real or simulated and will start learning on its own. The rewards system can be manipulated, for example by encouraging the program to allow emergency vehicles to pass quickly. But the program always learns by itself, but is programmed with specific instructions. Researchers hope to begin testing their systems on real roads this year.
Source: Aston University
What is your opinion about it?
A warning to Europe about the risks associated with the use of the Dutch scandal algorithm, the tax administration has destroyed thousands of lives with an algorithm
Clearview AI, the controversial facial recognition expert, has agreed to block its sale in the United States following a lawsuit.
AI models remain racist, even with more balanced training, the NIST report says
The new startup says its AI can analyze people’s movements and come up with solutions to better manage them to hit sales targets.