The MARLIN concept uses artificial intelligence and game theory to make traffic management smarter.
Smart traffic signals could bypass human controls to make better decisions on when to change
Few things are more annoying than being stuck at a red light that seems to go on forever. And gridlock is bad for the economy, in Canada and around the world; it's estimated that the cost of traffic jams in southern Ontario alone, due to lost time and productivity, is as high as $6-billion a year.
Could the problem be alleviated by changing how traffic flow is managed at intersections? Samah El-Tantawy and Baher Abdulhai think smart traffic lights could be the answer.
El-Tantawy, a University of Toronto engineering PhD student, and Dr. Abdulhai, her professor, have designed a traffic light system that uses decentralized artificial intelligence − with a bit of game theory thrown in − to unlock the grid.
The system is called MARLIN, which stands for the Multi-Agent Reinforcement Learning Integrated Network of adaptive traffic signal controllers. El-Tantawy developed the system after contemplating the snarling traffic jams in her hometown, Cairo, and realizing that Toronto is crawling in the same unpleasant direction toward constant gridlock.
There are different types of traffic lights around the world, she explains, and MARLIN is really the next generation. "The first are fixed lights, where the timing is signaled [to change] at set intervals."
More sophisticated systems operate on synchronized timers, where the lights along a major artery change in relation to each other. There are also some intersections that use activated signal controls, with sensors under the pavement that detect when a car is sitting at a red light for a while, while nobody is coming the other way.
While these systems are a step up from unsynchronized signals, none are as efficient or cost-effective − or as smart − as MARLIN, El-Tantawy says. "The existing systems are based on what's happening at other intersections downstream." Centralized traffic management systems that operate in big cities such as Toronto depend on having control management centres that collect all the data to one headquarters, which needs to be monitored by staff.
"This requires a massive communications centre. It's expensive and there's a single point of failure," she explains. MARLIN's adaptive signals use a form of analysis known as the Markov Decision Process: it involves a situation (in this case, traffic) interacting with a decision maker — in this case, not the drivers but the intersections themselves.
Using a combination of cameras, computers and sensors, each intersection makes its own "decisions" about the best timing to keep the traffic moving most smoothly in all directions. It adapts to changing conditions.
Unlike synchronized lights, which all respond in a row from a central command, a MARLIN intersection can constantly adjust its timing, by the second, and the more adjustments it makes, the more it learns. "It's as if each intersection has a brain," Ms. El-Tantawy says. It's a form of reinforcement learning − just as you won't touch a hot burner twice, the intersections learn from the data they gather as the cars roll by … or when they don't.
It also utilizes the concepts of game theory. To use a sporting analogy, hockey players know where their teammates are on the ice and constantly adjust where to skate, when to pass, when to look for the puck; they make decisions based on the team's need to win as well as their own will to make a good play.
MARLIN is not the only adaptive traffic system in the world, but it is catching the attention of politicians in the Greater Toronto Area, where regional traffic jams have gone from being a headache to a threat to economic growth and peoples' quality of life. So far the system has been computer-modeled at U of T and researchers are looking for a few good intersections in the region where they can put the system to the test.
In addition to saving money by shortening travel times and increasing productivity, adaptive traffic is also good for the environment, and it may be cheaper to install and run. Studies show that while it costs $80,000 to $100,000 to set up a typical traffic light intersection with a central command offsite, an adaptive system could be installed for as little as $20,000 to $30,000, El-Tantawy says.
Dr. Abdulhai says a $200,000 MARLIN system can move traffic as efficiently as an $8-million connected traffic light network. Researchers are already thinking beyond MARLIN, adds El-Tantawy: "In the future it would be even more accurate if the intersections could get information directly from the vehicles."
While smarter intersections should improve traffic flow, Benjamin Dachis, Senior Policy Analyst at the C.D. Howe Institute − who has written about the cost of congestion to regional economies − calls it a "stop-gap, stop-light solution."
He says, "When you have a short-term solution that makes it easier to get around, it's going to help in the next year or two, but what happens two or three years from now?"
He says the only solution that will work in the long term "is putting a price on roads that reflects the demand on traffic. It's politically difficult but when people see it works they like it."
For more innovation insights, visit www.gereports.ca
This content was produced by The Globe and Mail's advertising department, in consultation with GE. The Globe's editorial department was not involved in its creation.