Traffic Light Optimization for Vehicles and Pedestrians through Evolution Strategies

The optimization of urban traffic lights is a relevant problem. With the increasing occupation of urban pathways comes mobility deterioration: increasing delays, traffic jams and other consequential losses. Its relevance led to several proposals on traffic light optimization; the majority of them only consider vehicular traffic, to the detriment of pedestrians. Nonetheless, the longer pedestrians wait to cross, the riskier their behavior becomes, since they become more impatient. We tackle this problem through the optimization of traffic lights considering the average delays of both pedestrians and vehicles, by using microscopic traffic simulations. The problem is modeled on the basis of reference works of the area, and solved by an Evolution Strategy (ES). Several constraint handling methods are compared, including one proposed in this work, Two-Level Ranking (TLR), that aims to quickly find feasible solutions, which is important for real-time execution. The ES was able to find solutions that keep the pedestrian delays within the limits given by related work. Furthermore, in the evaluated scenario, a solution that satisfies the constraints is found, in average, at approximately 18.6 seconds with TLR, which is shorter than what other methods yield, allowing real-time operation.