Application of Reinforcement Learning as a Tool of Adaptive Traffic Signal Control on Isolated Intersections

Abstract—For a long time it was believed that the systems responding to real time traffic would enable significant benefits. However, numerous limitations have appeared such as the existence of the models with high level of detail precision, the uncertainty in predicting future traffic flows, the difficulty in arrival time estimation, the lack of self-adjusting mechanism. The difficulties in optimising the signal control strategy have initiated new researches. The results highlight the artificial intelligence methods as a possible solution. These systems are characterized with the ability to accumulate and use knowledge, set a problem, learn, process, conclude, solve the problem and exchange knowledge. The research presented in this paper proposes an adaptive signal control performed by a control agent able to adapt to an optimal policy by learning from the environment. The goal to be achieved is minimization of the delays in the system. First, the problem of reinforcement learning has been set. The first computation results of the Q-learning application for adaptive traffic signal control are presented. It is concluded that the results obtained are in favor of the adaptive signal control strategy compared to the fixed and actuated signal control.