Traffic control: how Google wants to make traffic lights smarter with AI

Google's Green Light project, which is based on machine learning and artificial intelligence and with which the internet company models traffic patterns and suggests optimized traffic light circuits, has produced initial results as part of test implementations. Traffic on some busy streets in Seattle is flowing a little more smoothly thanks to the technology, reports Scientific American. A spokeswoman for the US West Coast city's transportation department told Scientific American that the experience has been positive. Green Light has provided "concrete, actionable recommendations", drawn attention to bottlenecks in the transportation system and confirmed known bottlenecks.

Google launched its Green Light pilot program in Seattle and a dozen other cities in autumn 2023. These include Hamburg as well as some notoriously congested cities such as Rio de Janeiro in Brazil and Calcutta in India. During these test runs, local traffic engineers use the system's suggestions to adjust traffic light circuits. The company's aim with the initiative is to reduce waiting times at traffic lights, improve traffic flow on busy main roads and intersections and ultimately reduce greenhouse gas emissions. According to Google, $(LEhttps://www.gstatic.com/gumdrop/sustainability/google-2024-environmental-report.pdf:erste figures indicate a potential of up to 30 percent fewer stops at traffic lights and 10 percent less CO2 emissions.

At the heart of Green Light is an AI-based model of each intersection. This takes into account its structure, traffic patterns such as driving and stopping periods, traffic light planning and the way in which traffic and signal systems interact. Based on this model, the system develops recommendations that can be passed on to city planners via a special interface. A major advantage of the project is that it does not require expensive, permanently installed sensors and does not need to be constantly monitored on site. Instead, existing traffic data from Google Maps is collected from vehicles and smartphone users, who ultimately act as "mobile sensors".

Some green light proposals are not a direct hit

Henry Liu, Director of the Institute for Transportation Research at the University of Michigan, takes a more cautious view of the technology. According to him, the time required at junctions could be reduced by 20 to 30 percent with Green Light in Birmingham, for example. "It all depends on the basis of comparison," the civil and environmental engineer explained to Scientific American. In Birmingham, for example, there are only fixed traffic light times. These are based on car counts that have not been updated for some time. The climate argument should also be treated with caution: According to official figures, only around two percent of all traffic-related emissions in the USA are caused by waiting cars and traffic jams. Driving at higher speeds consumes significantly more fuel than a vehicle stopping for a red light.

Green Light does not yet take into account aggravating factors such as crossing bus and bicycle lanes, streetcars and busy crosswalks. Nevertheless, the tips do not always hit the mark. In Seattle, the planners withdrew a change to a set of traffic lights because it ultimately did not bring any benefits. In Manchester, another pilot city, traffic engineers repeatedly decided to deliberately ignore Google's recommendations. According to them, traffic lights there are sometimes deliberately set so that buses have priority or commuters in residential areas have to allow more time. The AI approach of minimizing the number of stops at junctions is counterproductive here. In Germany, AI traffic lights in Essenbach and Hamm, for example, which work with a system from Yunex Traffic, are still quite annoying for drivers.

(nie)