U of T student designs smarter traffic lights

Here’s something we could all use less of: gridlock—a political lightening rod and increasing limit on daily routines in Toronto, traffic congestion eats up our time, not to mention reserves of patience and good humour.

Now, one U of T student thinks she’s found a way to help tame congestion, by getting the lights at individual intersections to communicate directly with one another.

Samah El-Tantawy was inspired by the awful state of the roads both here in Toronto and in Cairo, where she grew up. Her traffic-management system formed the core of her graduate work (El-Tantawy earned her PhD in civil engineering in 2012), and is based on innovations in artificial intelligence research.

Right now, El-Tantawy explains, there are three types of traffic-management systems operating in Toronto:

  • Set times for light changes, based on prior calculations using historical records; these are optimized, but don’t adapt to the circumstances of any given moment.
  • Actuated controls: detectors under the pavement which send calls to traffic lights, so those lights can change based on immediate conditions. The shortcoming with these is that they are operating “as if blind,” El-Tantaway says. Since they only have inputs from vehicles in one direction, they don’t work based on the state of the intersection or road network as a whole.
  • Adaptive controls that are optimized in real time, based on traffic approaching an intersection; this system exists at about 300 intersections in Toronto. The main limitation with this system is that it works via a centralized command system, and thus requires a substantial communications network. (Any failure in that centralized system has, correspondingly, a huge impact on the whole network.)

The system El-Tantawy has developed is based on individualized intersection control, and comes with lower capital costs and risks of interruption compared to the adaptive control system. As she explains it, “Each intersection sends and receives information from its neighbours, and each of the neighbours does this in a cascading fashion.” Essentially, the lights at each intersection communicate with the ones at the connecting intersections, and this allows the lights at each intersection to change based on what those neighbouring lights are doing.

Unlike scheduled cascading traffic lights (where you hit a series of greens in a row if traffic conditions allow you to pace yourself just right), this system includes real-time responses to changing traffic conditions.

“Each one decides for itself,” El-Tantawy says. “But it considers what decisions what might be taken by the neighbours by having a model for each neighbour, and that model is built based on receiving information every second. They are actually deciding simultaneously.”

According to El-Tantawy’s simulation models, her traffic management system—called Multi-agent Reinforcement Learning for Integrated Network of Adaptive Traffic Signal Controllers (or MARLIN-ATSC)—can reduce delays by up to 40 per cent, and yield a 15- to 25-per-cent savings in travel time. It can also have environmental knock-off effects—up to a 30-per-cent reduction in CO2 emissions, since vehicles are spending less time on the road and travelling more efficiently when they do.

City of Toronto staff are aware of El-Tantawy’s work, and she’s hoping it will eventually be implemented in some intersections here. She needs to conduct field tests first, however, and is currently looking for quieter areas suitable for pilot projects next summer.

This article first appeared on Yonge Street.

Top Stories

Top Stories

Most Watched Today