Simulation-Driven Multi-objective Evolution for Traffic Light Optimization

The constant growth of vehicles circulating in urban environments poses a number of challenges in terms of city planning and traffic regulation. A key aspect that affects the safety and efficiency of urban traffic is the configuration of traffic lights and junctions. Here, we propose a general framework, based on a realistic urban traffic simulator, SUMO, to aid city planners to optimize traffic lights, based on a customized version of NSGA-II. We show how different metrics -such as number of accidents, average speed of vehicles, and number of traffic jams- can be taken into account in a multi-objective fashion to obtain a number of Pareto-optimal light configurations. Our experiments, conducted on two city scenarios in Italy and different combinations of fitness functions, demonstrate the validity of this approach and show how evolutionary optimization is an effective tool for traffic light optimization.