A learning technique for deploying self-tuning traffic control systems

Currently, a considerable amount of human effort and time is spent for initialization or calibration of operational traffic control systems. Typically, this optimization (fine-tuning) procedure is conducted manually, via trial-and-error, relying on expertise and human judgment and does not always lead to a desirable outcome. This paper presents a new learning/adaptive algorithm that enables automatic fine-tuning of general traffic control systems. The efficiency and online feasibility of the algorithm is investigated through extensive simulation experiments. The fine-tuning problem of three mutually-interacting control modules - each one with its distinct design parameters - of an urban traffic signal control strategy is thoroughly investigated. Simulation results indicate that the learning algorithm can provide efficient automatic fine-tuning, guaranteeing safe and convergent behavior.