How AI could help reduce your time waiting at a red light

Artificial intelligence powered traffic lights are being trialled, but will it ease traffic congestion? Artificial intelligence (AI) already has a foothold in many industries across Australia, but it could soon help speed up your commute. Currently, when you pull up at an intersection, your car is detected by loops under the road surface, and the traffic lights cycle through phases before turning green. In Australia, this phase-based system has been commonplace since the 1950s, and while reliable, researchers say it may no longer be optimal. It may be the latest "buzzword", but Queensland University of Technology Professor Ashish Bhaskar says AI could help improve our understanding of the movement of traffic at an intersection and use that information to reduce delays. In an Australian first, a Queensland council is set to trial the technology at a single intersection later this year. Professor Bhaskar says cameras have the potential to improve visibility at an intersection and AI can leverage that "smartness" to provide a better input to the algorithm, leading to better traffic control. That means the system will have greater awareness of the mix of cars, buses, pedestrians and cyclists at an intersection, and will act accordingly. "Now you have more knowledge, you have more visibility of what is happening at the intersection," he says. "Instead of focusing on being reactive to what's happening, you can be more proactive, looking at the network-wide level or the corridor level and then make decisions at different intersections." Unlike the phase-based system, the new technology does not run on a fixed cycle, with the combinations instead determined based on demand. Professor Bhaskar says the algorithm can be fine-tuned to suit the area's needs, for example, to prioritise pedestrians in the CBD or cars in high traffic areas. The $170,000 trial at Petrie in Queensland's Moreton Bay will start later this year and is expected to run until early 2029, with a low-traffic intersection isolated from the main transport network deliberately chosen. The City of Moreton Bay says a camera will identify objects, which will be fed back, before the controller and AI determine the best sequence based on the detection and algorithm. The council says an evaluation of the intersection before and after the trial will help determine its success. The system, by Austrian-based company SWARCO, is already operating in Denmark and in cities across Europe, and the company says councils across Australia have expressed interest. In the US city of Pittsburgh, the AI-based system SURTRAC, by researchers from Carnegie Mellon University, was found in 2019 to have reduced the average travel time by 25 per cent, with cars spending up to 40 per cent less time idling. Looking beyond a single intersection University of Sydney transport engineering associate professor Mohsen Ramezani has welcomed the move in Moreton Bay, but with caution. He says he does not expect massive gains to be made at an individual intersection, with the "real challenge" being effectively managing traffic at closely spaced intersections in CBD's. While he admits he might sound pessimistic, he supports councils taking on trials and being "open-minded" to how to manage transport. He says if AI or a new approach could help, it would be a "game changer". According to Dr Ramezani, the biggest area for improvement is in the algorithm rather than detection. "We can equip an intersection with heaps of sensors," he says. "But it's the decision-making … that's the smartness that still needs to be improved." Later this year, a two-year $15 million trial will begin in Brisbane. Councillor Ryan Murphy says it will go beyond "making one intersection smarter" with AI to be introduced "into the traffic management system itself". "By coordinating entire corridors, we can deliver more reliable journeys and keep Brisbane moving as our city continues to grow," he said. The council says the project is in the final stages of procurement, with the roads to be part of the trial and further detail on how AI will be incorporated yet to be determined.