Traffic lights at intersections are managed by simple computers that assign the right of way to the nonconflicting direction. However, studies looking at travel times in urban areas have shown that delays caused by intersections make up 12-55% of daily commute travel, which could be reduced if the operation of these controllers were made more efficient.
A team of researchers led by Dr. Guni Sharon has developed a self-learning system that utilizes machine learning to improve the coordination of vehicles passing through intersections.
Recent studies have shown learning algorithms can be used to optimize the controller’s signal. This strategy enables controllers to make a series of decisions and learn what actions improve its operation in the real world.
Sharon noted that these optimized controllers could not be used in practical applications because the underlying operation that controls how it processes data uses deep neural networks (DNNs).
Despite how powerful they are, DNNs are unpredictable and inconsistent in their decision-making, which makes understanding why they take certain actions as opposed to others a cumbersome process for traffic engineers. This, in turn, makes DNNs difficult to regulate and learn the different policies.
Using a simulation of an actual intersection, the team found that their approach was particularly effective in optimizing their interpretable controller, resulting in up to a 19.4% reduction in vehicle delays in comparison to commonly deployed signal controllers.
Other contributors to this research include Dr. Josiah P. Hanna, research associate in the School of Informatics at the University of Edinburgh, and James Ault, doctoral student in the Pi Star Lab at Texas A&M.