A U of T academic has found a way of getting traffic lights to "talk" to one another.
Samah El-Tantawy, a University of Toronto postdoctoral fellow, has invented something she thinks could revolutionize the way vehicle traffic flows in cities. It’s called the Multi-Agent Reinforcement Learning for Integrated Network of Adaptive Traffic Signal Controllers, or MARLIN-ATSC for short.
The system isn’t in use yet, but it has been tested on a simulator with real data from Toronto’s busiest intersections. It “thinks” on its own, and takes into account oncoming traffic from multiple intersections, and from all directions. Traffic lights connected to MARLIN-ATSC are able to interact, sort of like members of a sports team. In theory, it’s a big improvement over the current system of managing traffic signals, which in most cases relies on simple timers and sensors.
MARLIN-ATSC, which El-Tantawy has been working on with experts and researchers from U of T since 2004, shows promise. In simulations, it’s able to reduce delays per vehicle at intersections by 40 per cent.
El-Tantawy’s supervisor, Professor Baher Abdulhai, is the director of the Toronto Intelligent Transportation Systems Centre and Testbed, a type of laboratory for traffic technology. He says, “In 2000, I started a simple proposition, a research question. Can traffic lights learn, or can machine-learning in general be used to control or manage traffic?”
Almost a decade later, El-Tantawy has shown him it can be done.
“She added to the dimension of how to make traffic lights independent, intelligent agents that can communicate with each other using game theory, where essentially agents or players negotiate. For a system to be able to discover that on its own, that’s the novelty here.”
Anything capable of speeding up the flow of traffic could be useful in Toronto, where road congestion is an ongoing source of commuter pain. According to the 2011 National Household Survey, daily commutes in Toronto average 32.8 minutes, which is longer than the average commute in any other Canadian city.
In Toronto, there are 1800 lights that operate on predetermined traffic patterns and 350 that rely on real data to determine timing. The latter type of light has been used for the past 20 years and is nearing the end of its useful life, according to Myles Currie, the director of the city’s traffic management centre.
Although the City is considering ways of improving the coordination of some of its traffic signals, there are currently no plans to implement MARLIN-ATSC on Toronto’s streets. Currie wouldn’t comment on El-Tantawy’s system, because it hasn’t been tested with real drivers and cars.
El-Tantawy says her system would not only improve traffic flow, but, in the long run, would also be less expensive for the City. She believes it would cost between $20,000 and $30,000 to install MARLIN-ATSC at an intersection, with additional expenses for installing the software and other components. Plus, the system doesn’t rely on a pricey communications network. Nor does it demand highly skilled operators or safety staff.
Currie says the City’s current systems cost between $125,000 and $150,000 to install. They also require extensive in-house training for Currie’s staff, which is made up mostly of traffic engineering graduates.
El-Tantawy says she’s in the process of negotiating with a partner in the traffic-control industry, in the hopes of bringing her system to market.
“We’re forming a start-up company with them and my supervisor,” she says. “Hopefully it works and we can have our first field testing and commercialize it from there.”