Reinforcement learning based adaptive traffic signal control method invariant to the configuration of the traffic lights

The paper presents an adaptive traffic signal control method that is invariant to its configuration. The proposed method uses a single neural network model to control traffic lights of different configurations, i.e. having a different set of phases and number of controlled lanes. Experimental studies on the effectiveness of the proposed method were carried out in a specialized traffic flow modeling environment, SUMO (Simulation of Urban MObility). The results obtained confirmed the operability and efficiency of the proposed method.