Q-learning

In the article, modern approaches to the development of Intelligent Transportation Systems (ITS) aimed at optimizing urban traffic are analyzed. Special attention is paid to model-free reinforcement learning algorithms (Q-Learning and Deep Q-Learning) used for controlling traffic lights in dynamic road traffic conditions. Simulation results in the SUMO environment have proven that implementing such algorithms significantly reduces intersection queues and increases the capacity of the transportation network.

The model of matrix stochastic game for decision-making in the conditions of uncertainty is developed. The method of Q-learning for stochastic game solving with a priori unknown gains matrices is offered. The formulation of a game problem is executed. The Markovian recurrent method and algorithm for the game solving are described. Results of computer modelling of stochastic game with Q-learning are received and analysed.