Sci tech daily pointed out that in 2014 alone, Americans spent 6.9 billion hours commuting In addition, the additional gasoline consumption caused by traffic congestion is also as high as 19 gallons on average - equivalent to a loss of $160billion in time and fuel per year More specifically, in many large cities in the United States, the average driver may waste more than 100 hours of commuting time every year.
(from Aston University)
Referring to the typical workplace, the time wasted on the road is enough for commuters to take two and a half weeks' leave every year. The good news is that many researchers are trying to reduce traffic congestion - whether it is the development of autonomous vehicle or the introduction of AI traffic lights.
The 21st aamas Artificial intelligence signal lamp technology from Aston University research team included in the conference papers. It can be seen that the system can scan video clips in real time to adjust the light control signal to reduce congestion / keep the road unobstructed.
Deepeka Garg, Maria chli and George vogiatzis explained in the article entitled "full autonomous, vision based traffic signal control: from simulation to reality":
● with the help of deep reinforcement learning, AI software can distinguish when the road traffic is blocked, and try a new method to improve the signal lights, or continue to improve when progress is made.
● the performance of this AI system during the test is unparalleled, far beyond the changes usually made by relying on manual design.
● in addition, the insufficient release time of signal lights is a major cause of road congestion during peak periods.
In order to find the optimal solution to solve traffic congestion, researchers at Aston University first constructed a set of traffic 3D traffic simulator scenes with fidelity similar to photos.
By teaching AI system to deal with different Jiatong and weather conditions, deep reinforcement learning can make this system adapt quickly when facing the real intersection picture.
Maria chli, a computer science major at the University, explained:
We set it as a traffic signal control game. When the program lets the car pass through an intersection, it can get appropriate 'rewards'.
When the car has to wait or the road is congested, the system will receive a negative evaluation. But in fact, we do not have any input, but only reward and punish the control system.
As a reference, the traffic signal automatic control system that has been put into use at present mostly relies on the magnetic induction circuit on the road to count the passing vehicles and make some responses accordingly.
However, the AI system newly created by the research team of Aston University can quickly "predict" higher traffic flow and make decisions before the car passes the red light intersection.
Dr George vogiatzis, senior lecturer in computer science at Aston University, added:
The reason why this program is based on behavioral learning is that it can understand situations that have not been clearly experienced in the past.
We have passed a physical obstacle that causes congestion, rather than the phase of traffic lights, to verify the test, and the results show that the performance of this system is still good.
Obviously, as long as there is a causal relationship between various factors, the computer will eventually understand the powerful management, which is also the extraordinary strength of this AI traffic signal control system.
For example, the program can be set to view any traffic intersection in a certain area, and then start autonomous learning without using specific instructions to program. In addition, if necessary, the traffic management unit can also give priority to the rapid passage of emergency rescue vehicles.
Finally, if all goes well, the researchers hope they can test the system on the real road from this year.