Achieving fluid traffic in cities is essential to improve the quality of life of its citizens. And the Massachusetts Institute of Technology works with artificial intelligence to achieve it in the future.
Traffic lights at intersections were designed to organize traffic, but sometimes impatience or haste leads to a traffic jam. this ends slowing down traffic even more, generating greater environmental and noise pollution.
Recently, we told you about a Ford experiment with which the American brand tries to speed up the movements of emergency vehicles in large cities.
Now we know another of a similar cut, although even more ambitious. This is an investigation initiated by MIT engineers whose objective is stop waiting at red lights and, incidentally, with traffic jams at intersections.
Artificial intelligence and autonomous cars
Of course, this project is aimed at its long-term application, since the objective is to implement it in autonomous vehicles. The plan is to ensure that, through a learning process with artificial intelligence, this type of vehicle don’t have to wait standing still at red lights.
To do this, MIT researchers demonstrate a machine learning approach that can learn to control a fleet of autonomous vehicles as they approach and travel through a signalized intersection in a way that keeps traffic flowing.
“A single intervention with a 20 to 25% reduction in fuel or emissions is really incredible”
This, according to the members of the study, generates several advantages. On the one hand, fuel consumption and emissions are reduced. On the other hand, it improves the average speed of the vehicle, thus shortening the time needed for each journey.
Simulations show that the technique performs best if all cars are autonomous, but even if only 25% use its control algorithm, substantial fuel and emissions benefits are still generated.
“This is a really interesting field to intervene in”says lead author Cathy Wu, Gilbert W. Winslow Assistant Professor of Professional Development in the Department of Civil and Environmental Engineering and a member of the Institute for Data, Systems, and Society (IDSS) and the Laboratory for Information and Decision Systems (LIDS). ).
“Being stuck in an intersection doesn’t make anyone’s life better. With many other climate change interventions, a difference in quality of life is expected, so there is a barrier to entry. In this case, the barrier is much lower»Wu points out.
complex intersections
In large cities, on many occasions it is not as simple as crossing an intersection of two streets, since there are many variables that complicate multiple circulation points. Several lanes, signs, number of vehicles, speed, pedestrians, cyclists, scooters… there are many elements to take into account.
To deal with these complexities, the MIT Traffic Improvement Project takes a different approach than usual: a model-free technique known as deep reinforcement learning.
Reinforcement learning is a trial and error method in which the control algorithm learns to make a sequence of decisions. You are rewarded when you find a good sequence. With deep reinforcement learning, the algorithm takes advantage of the assumptions learned by a neural network to find shortcuts to good sequences, even if there are billions of possibilities.
This is useful for solving a long-term problem like traffic in cities. To do this, «the control algorithm must issue more than 500 acceleration instructions to a vehicle for an extended period of time,” explains Wu.
“And we have to get the sequence right before we know we’ve done a good job of mitigating emissions and getting to the intersection at a good speed,” he adds. But there is one more challenge: the researchers want the system to learn a strategy that reduces fuel consumption and limits the impact on travel time. These goals can be contradictory.
“To reduce travel time, we want the car to go fast, but to reduce emissions, we want the car to slow down or not move at all. Those competitive rewards can be very confusing for the learning agent.”Wu admits.
While it is challenging to solve this problem in all its complexity, the researchers employed a workaround using a technique known as reward modeling. With the bounty setup, they give the system domain knowledge that it can’t learn on its own. In this case, they penalized the system every time the vehicle came to a complete stop, so that it would learn to avoid that action.
Óscar Magro compares two heavyweights in the SUV market: the Mazda CX-5 and the Volkswagen Tiguan.
Once they had developed an effective control algorithm, they evaluated it using a single intersection traffic simulation platform. The control algorithm is applied to a fleet of connected autonomous vehicles, which you can contact the next traffic lights to receive information about signal phase and time and observe your immediate surroundings. The control algorithm tells each vehicle how to speed up and slow down.
His system did not generate intermittent traffic as vehicles approached the intersection. In the simulations, more cars made it through a single green phase, which outperformed a model simulating human drivers.
“A single intervention with a 20 to 25% reduction in fuel or emissions is really incredible. But what I find interesting, and was really hoping to see, is this non-linear scaling. If we only control 25% of the vehicles, that gives us 50% of the benefits in terms of emissions and fuel reduction. That means we don’t have to wait until we have 100% autonomous vehicles to reap the benefits of this approach.”
“The objective of this work is to move the needle in sustainable mobility. We also want to dream, but these systems are huge monsters of inertia. Identify intervention points that are small changes to the system but have a significant impact is something that gets me up in the morning»he concludes.
Source: MIT