Hate Red Lights? How AI is Working to Optimize Traffic Intersections

    By: Illah Nourbakhsh, ROBO Global Strategic Advisor & Professor at Carnegie Mellon University


    We all have that one troubling traffic light in town that undoubtedly turns red the moment we pull up to it. Who is the mastermind working against us and our daily drive home? The answer is simple: probably no one or nothing, which begs for a much-needed update, don’t you agree? In many ways, the roadway intersection is a perfect representation of organized chaos in our lives, serving the essential purpose of keeping people safe, but also causing us all irritation behind the wheel.

    Our current traffic system has become the status quo over the past century but could it be that a more optimized, data-driven approach is possible? Could AI come to the rescue at red lights?

    That is the very question researchers here at Carnegie Mellon University set out to solve. Starting years ago, my colleagues first began their efforts to study urban intersections in Pittsburgh. They initially focused on developing camera systems and artificial intelligence–based computer-vision software that could accurately observe traffic conditions at an intersection: How many cars, bicycles, and pedestrians are waiting to cross, every minute of the day and in every direction of flow?

    Observations like this turned into large-scale data analytics problems where my colleagues discovered something startling. Two forms of inefficiency were abundant at traffic lights:


    1. Commuters’ time was consistently wasted, with great potential time savings if only traffic lights could be made responsive to true traffic conditions in real-time. 
    2. Carbon output resulting from idling cars and trucks comprised a significant portion of the commute budget, so saving time would also result in real environmental benefits.


    Next came a city-university partnership that would make real-world algorithms possible. The City of Pittsburgh allowed researchers to not only observe but also to control traffic signals at test intersections. The programmers began fashioning AI algorithms that would minimize the wait times of everyone—cars, bicycles, and pedestrians alike—a literal win-win in the world of optimization.

    This kernel of an idea became a product, Surtrac, and a spinoff company, Rapid Flow, that matured the technology across numerous cities and demonstrated quantitative savings across the board: 


    • 25% reduction in travel times for local commuters, including pedestrians
    • 40% reduction in time spent waiting at red lights
    • 20% reduction in vehicle emissions due to a significant reduction in the amount of stopping and accelerating.


    Thus, AI optimization techniques are solving a problem that people could not have solved on their own by taking real-time data into account across distributed intersection control systems and optimizing every criteria without having to compromise one user group for the benefit of another. These techniques have reduced overall waste in a way that improves the experience of every user group. 

    Complexity is part of our engineered world—from elevator control in high-rise buildings to medical diagnostics systems to air traffic control. In every case, the newest AI algorithms give us the prospect of experiencing something new through the elimination of waste and inefficiencies that benefit the world—not just users, but also the passive environment around us, by reducing greenhouse gas emissions and energy usage.

    ROBO Global’s constituent companies represent the innovators that are needed to achieve this kind of network optimization. Companies like Cognex, Hexagon, Optex, and Omron provide the computer vision and infrastructure sensors needed to evaluate human behavior in complex networks. These sensing systems, in conjunction with computational optimization innovations, will blaze a more efficient trail (or in this case, road), on which each of us will be provided with more value.

    Back to All Entries