For the system to work, only 10% of the vehicles are connected to the route optimization network.

Road traffic is undoubtedly the area where human irrationality is most damaging. We are building more and more traffic lanes around cities, bearing in mind a totally erroneous assumption: on the road, individuals behave in such a way as to encourage regular and efficient driving and to deter dangerous and anarchic driving . Thus, attempts are made to circulate unbelievable quantities of cars in a limited space, at a speed high enough for the traffic to be fluid.

Obviously, it does not work. Why ? Because human behavior in no way resembles the ideal behavior that is modeled to predict the evolution of traffic. People stuck in the corks start and accelerate abruptly as soon as the car that precedes them advances three meters, a rainy weather favors nervousness and unwarranted slowdowns, and finally, everyone presses the brake pedal.

Because there are prudent drivers, unwary drivers, impulsive persons, softheads and heads-up specialists in driving, there is really no "typical driving" that would make predictions Accurate information on the state of traffic in a time t. Consequences: most metropolises are so congested that one would gain to go to work in hang-gliding.

In recent years, urban planners, engineers and other prospective urban prospects have imagined that the emergence of autonomous cars will solve all the traffic problems of the world. If we could get rid of the irrational behavior of humans behind the wheel by delegating car steering to an AI, then we could seriously streamline road traffic. The problem is that all these jovial experts had not foreseen that the public would be reluctant to entrust its destiny to an intelligent system. They had also not foreseen that in 2017, the horizon of the fully functional autonomous car would still be very distant.

While waiting for this blessed day, computer scientists at Nanyang Technology University have developed a new intelligent routing algorithm that reduces the frequency of "spontaneous traffic jams" - the accumulations of cars caused by the convergence of several traffic lanes . It is an extremely fast distributed computing system that meets all the requirements of a real-time traffic management system. The work of the researchers is described in the April issue of IEEE Transactions on Emerging Topics in Computational Intelligence.

Routing (or route selection) is a traditional branch of computing. Here, it is more precisely to rely on the theory of graphs, a domain that used algorithms long before the first electronic computer was built. The problem is that routing algorithms can be extremely complex, even when they have to manage only a small number of possible actors and paths. Real road traffic requires calculations involving a large number of roads, cars and other agents (such as traffic lights).

The algorithm of Nanyang researchers starts from the assumption that from a certain density of circulation, the problems will emerge on their own. At one point or another, a driver will inevitably take a bad initiative that will disrupt the traffic until it breaks down. This technical term means that for a certain period of time, on a given route segment, the outgoing traffic will be less than the incoming traffic.

"We assume that we already have a good traffic allocation model and that the probability of a traffic disruption is greater than zero (meaning that at some point there will necessarily be a slowdown or a traffic jam). Our goal is to redirect the flow of traffic so that the probability of traffic breakdown is reduced to the maximum, "write Hongliang Guo and his colleagues. In other words, "our objective is to maximize the probability that none of the roads in the system will be subject to traffic congestion. "

This objective is that of the algorithm, which is materialized in the form of a very simple equation. From there, it was enough to build an efficient machine learning system that allows to analyze the traffic in a time t, to add the additional traffic load that could affect the network at any moment and then to calculate the probability Of a break at each of the nodes of the network (or intersections). Add to this a hint of linear algebra and you get an ideal route calculation system.

Guo and co. have succeeded in developing the mathematical optimization tools that allow this kind of real-time calculation. They were able to demonstrate the effectiveness of their algorithm in simulations, and are currently working on improving their BMW partnership system, which provides them with a large database of its shared fleet of Munich.

This technology could be implemented very quickly. In addition, it is enough to optimize the itinerary of 10% of the vehicles of the network to fluidify the traffic, which means that most motorists will be able to continue to drive like branches, in all serenity. Great, is not it?

Source: Motherboard, by Micheal Byrne, March 28th, 2017