A Blog by Jonathan Low


Jul 14, 2018

Can Artificial Intelligence Eliminate Traffic Congestion?

While better information is helpful, there are usually only so many routes available, especially for commuters or anyone headed to a specific destination.

For systems to really make a difference they are going to have to address a much broader issue: how to get drivers to switch to other transportation modes in order to reduce demand for the increasingly limited urban traffic space available. JL

Henry Williams reports in the Wall Street Journal :

(AI) has reduced waiting times at traffic lights by as much as 42%. Car makers are betting on short-distance communications between traffic-management systems, and between cars. Drivers are able to receive real-time information about the intersections they’re approaching, including how many seconds are left until the light turns green.“In the future it will suggest a speed” at which to drive to hit green lights along a route
If you drive a car, you’ve probably found yourself waiting at a red light while the intersection sits empty. Artificial intelligence could make that—and other frustrating inefficiencies of city traffic—a thing of the past.
For a sense of what the technology is capable of, consider that some research suggests artificial intelligence could allow networked autonomous vehicles to safely make their way around cities without any traffic lights at all if it weren’t for the presence of human-driven cars and pedestrians, says Dorsa Sadigh, a professor at Stanford University who specializes in the interaction between autonomous vehicles and humans. Humans aren’t going away anytime soon, of course, so neither are traffic lights. But researchers are taking steps toward a future where smart traffic lights and internet-connected cars can make getting around town smoother for both drivers and pedestrians—as well as provide other benefits, such as giving priority to public transit or emergency vehicles and reducing auto emissions.
Progress in Pittsburgh
For AI to do its potential magic, the first thing that’s needed is data. Lots of it. So several startups are connecting hundreds of sensors at traffic lights to understand why congestion is happening and learn how to manage it in real time.
For instance, Rapid Flow Technologies, which began as a Carnegie Mellon University research project, is testing its Surtrac traffic-management system in the East Liberty neighborhood in Pittsburgh.
Straddling a major arterial route and home to a Target store, the neighborhood has long been an area of heavy congestion as commuters, shoppers and local residents clog the roads.
“Traffic patterns changed so much over the course of the day that [the traffic signals] didn’t really work all that well” in keeping traffic moving, says Greg Barlow, a Rapid Flow co-founder.
Traditional traffic signals commonly change on a fixed schedule. Some are coordinated with those at the next intersection. More-advanced traffic lights can even sense when a car is waiting at the light and adjust the timing. But for the most part that’s as futuristic as it gets.
In East Liberty, Rapid Flow’s technology deployed at intersections allows coordination among all the lights where it has been installed—for example, allowing a light to stay green longer to clear traffic at a particular intersection.
“We have communication between intersections,” says Mr. Barlow. “It lets an intersection plan based on what it can see with its own sensors and what its [neighbors] can see with upstream sensors.”
The Surtrac system has reduced waiting times at traffic lights in the area by as much as 42%, Mr. Barlow says. That not only gets people to their destinations quicker, it also helps reduce auto emissions because cars are spending less time on the road.
Rapid Flow is working on a feature that would allow drivers to share their planned routes with the network, so that information could be used to adjust the timing of lights and possibly cut waiting times even further.
Because the project in Pittsburgh is in only a small part of the city right now, there were early issues with traffic backing up as cars moved from the area with the new AI-enhanced lights into areas without the technology. But AI managed to solve that, too, by recognizing the congestion on the fringes of the system and taking it into account in changing the lights under its control.
Rapid Flow is expanding, with a deployment of sensors across 24 intersections in Atlanta and other deployments in the Northeast—at three intersections in Portland, Maine, with nine more coming later this year, and at two intersections in Needham, Mass.
Predicting patterns
A startup called Vivacity Labs is taking a different approach in the town of Milton Keynes, in England. It is focusing on gathering data on traffic patterns with custom-made sensors installed at traffic lights throughout the town, with the aim of eventually using the system to provide predictive traffic information and guidance to drivers. Later still, controlling traffic lights would come into play.
The sensors don’t simply gather information; each is a powerful computer attached to a camera, capable of analyzing the traffic it can see at its intersection.
For years, traditional mapping apps, like Google Maps and Waze, have been capturing traffic information by crowdsourcing their data, monitoring the speed of individual users, getting accident reports from users, and then relaying those conditions to other users.
But by the time drivers plan a route through a city, or as they move along that route, the traffic information their app depends on can already be out of date.
Vivacity instead uses its sensors at intersections to gather traffic information that is continually sent back to a central computer. The systemwide data can be analyzed not only to recognize current traffic conditions but also to predict how traffic patterns will develop. Eventually, it should be able to direct drivers “based not on how busy the road is now, or how busy it was a few minutes ago, but how busy it will be when you get there,” says Peter Mildon, chief operating officer of Vivacity, which is based in London.
For example, the sensor at one intersection might know that five cars are waiting at a light and there’s congestion in the intersection itself. The next might be sensing a slowdown in traffic and some long trucks waiting to go through the intersection. All of this, along with information from all the other sensors in the system, is fed into a central AI algorithm that can project what traffic will be, say, five, 10 or 15 minutes in the future—or even hours later. The company hopes to eventually use the technology to control traffic signals for improved traffic flow.
Among other projects, Vivacity also has sensors deployed in the city of Cambridge, England, where it is trying to predict when lines for parking lots will start backing up and lead to gridlock.
Where to next?
Officials in Pittsburgh see the potential for Rapid Flow’s Surtrac system to optimize not just vehicle traffic but the movement of people around the city on mass transit.
“What if we want to really emphasize person throughput rather than vehicle throughput?” says Karina Ricks, Pittsburgh’s director of mobility and infrastructure. “What if we were able to tell the signal that not only is there a 30-person bus, but there is a 30-person bus with one person in it—the driver—or a 30-person bus with 40 people in it? That can get into the algorithm” to get the most people to their destinations as quickly as possible. The city is working toward this goal with Rapid Flow.
Meanwhile, car makers are betting on the technology developing further to include short-distance communications between traffic-management systems like these and cars, and between cars themselves.
Volkswagen’s Audi brand began putting AI-driven traffic technology in its 2017 models. Drivers in certain cities—including Las Vegas and —are able to receive real-time information about the intersections they’re approaching, including how many seconds are left until the light turns green.
Anupam Malhotra, Audi of America’s director of connected vehicles and data, is excited about how the technology could develop. “In the future it will suggest a speed” at which to drive to hit green lights along a route, he says.


Post a Comment