Back to Bigger Picture

BIGGER PICTURE Public & everyday services

Traffic lights that tell the next junction what is coming

At nine Pittsburgh junctions, a planning system repeatedly reshuffled green times around actual approaching traffic. Its small but tangible target was the time people spent stopped between places.

AI-generated illustration: Neighbouring traffic signals coordinating vehicle flows.
Neighbouring traffic signals coordinating vehicle flows. Conceptual illustration; not a photograph or an exact depiction of the real project.AI-generated illustration of the idea, not a photograph of the actual people, equipment or event.
Original pilot
Nine intersections in 2012
Travel time in field comparison
−25.79%
Waiting time in field comparison
−40.64%

How the idea works

  1. Observe approaching traffic

    Each junction estimates its incoming vehicle flow.

  2. Schedule and inform neighbours

    Green-time plans include expected downstream arrivals.

  3. Update as traffic changes

    Controllers repeatedly adjust the local schedule.

A signal plan that keeps changing

SURTRAC treats each junction as a scheduling problem. Approaching vehicles compete for limited green time; the system repeatedly chooses an allocation based on the current inflow. In the original design, an intersection could update its plan as often as once a second. The intention is to respond to the traffic actually arriving, rather than rely only on a fixed timetable.

Sources: [1]

The next junction gets advance notice

Each controller shares its expected outgoing traffic with neighbouring junctions. Those neighbours can prepare for vehicles that have not yet reached their sensors. Coordination emerges from these local exchanges, while each junction still balances its own competing approaches. This is the central AI idea: connected planning agents rather than one rigid citywide sequence.

Sources: [1]

Drive the same routes and compare

The 2012 East Liberty pilot covered nine intersections. Researchers drove fixed routes before and after deployment, using GPS traces to compare travel and waiting. Their volume-weighted results reported 25.79% less travel time and 40.64% less waiting. A university student project later examined the practical commercialization and policy issues surrounding this deployment, beyond the technical demonstration.

Sources: [1][2]

A local result worth testing elsewhere

This was a before-and-after field comparison, not a randomized citywide experiment. Its reported emissions improvement was calculated, not directly measured from exhaust. Other streets, pedestrian needs and traffic patterns require their own evaluation. The durable insight is operational: improving the timing and coordination of existing infrastructure can produce a noticeable benefit without adding a new road.

Sources: [1][2]

The original sources

  1. Carnegie Mellon University research team / TRB 2013SURTRAC: Scalable Urban Traffic Control
  2. Carnegie Mellon University course projectSmart Traffic Light: Non-Market Strategy Analysis (student project)

What to explore next

Treasury says AI helped recover $1 billion lost to cheque fraudA robot chemist discovered when less light works better
Explore all 99 applications