Pages

With advancements in AI, Signalized intersections are going to be less chaotic and more organized for Motorists, says MIT Researchers

Every time we are in a rush to cross the traffic signal as soon as it is green because none of us like to wait so long until the signal turns green again. This rush can become a cause of serious accidents and sometimes leads to traffic violations with heavy fines. As much as it’s tiring for us to wait at traffic signals, and it consumes our time along with fuel, it also damages the environment. To put this problem to an end, policymakers and motorists are looking out for a much more relevant and reliable alternative, even if it is very small. A team of investigators at the Massachusetts Institute of Technology (MIT) believes that they have got a solution to this problem.

They put effort into looking for many ways to make sure that drivers do not have to wait at the traffic junction until the signal turns green. Instead of waiting so long for the signal to turn green, what if they reach the traffic junction exactly at the time when the signal is green. Despite the fact that it is hard for a human driver to calculate the precise timing of the signal turning green and arrive at that very moment, it can be very consistently achieved by an autonomous car, a car that does not require a driver, and works on Artificial Intelligence.

A vehicle that works on AI can be regulated accordingly, its speed can be set in such a way that the vehicle arrives at the traffic signal at the exact time when it turns green, and it doesn’t have to wait for the signal to turn green.

The team had published a study previously, on the electronic preprints archive arXiv.org, in which they validated a machine learning approach, which is going to learn to direct a fleet of self-driving vehicles in such a manner that runs a smooth flow of traffic. The team of investigators was headed by a postgraduate student, Vindula Jayawardana, and the team reported that this machine learning approach requires less petrol usage and helps reduce emissions, however, it aids to increase the speed of an automobile, hence decreasing our travel time.

One complication here is that the investigators want this approach to minimize petrol usage along with decreasing the duration of the journey, but for reducing emissions they do not require the car to completely come to halt or just slow down. Hence, it can be difficult for the learning agent to cater to both the incompatible objectives.

To cope with this problem, these investigators came up with an alternative solution called reward shaping, through which the learning agent was provided with the information that was not possible for it to learn by itself. They gave penalties to the system, each time the automobile would come to halt in order to make it learn not to repeat this mistake ever in the future.

Through the help of a traffic simulation model that had only one junction, they tried out their control algorithm once it was completely ready. As soon as the car comes near the junction, the system would not stop the car at once, as it would have in stop-and-go traffic.

As compared to self-driving cars, many of the autonomous cars passed in just one phase of the green signal, and indeed this approach led to greater fuel savings and lesser emissions.

Therefore, if every automobile on the road is self-moving, with no human input required, and has a built-in system, fuel consumption can be decreased by 18 percent and along with that, emission of Carbon dioxide can also be reduced by 25 percent. Moreover, it can also improve the average travel speed of a vehicle by 20 percent. To the least, if only 2 percent of the automobiles are self-moving, they can contribute to the reduction of a minimum of 50 percent of the fuel and emissions altogether.


Read next: Big Tech Is Pushing Carbon Removal to Solve Climate Change, Here’s Why That Won’t Work

No comments: