Road traffic optimisation based on a learning approach

摘要

Road traffic management is an important issue for authorities, road operators and users. Many researchers have worked since many years in order to propose smart solutions. Some of them have been adopted by road operators and applied on road signalling or on traffic light management. During the last decade, cooperative intelligent transport systems (C-ITS) have emerged and are being deployed. They permit on one hand to increase the number of connected vehicles and on another hand to deploy road side units (RSU) along roads and junctions. RSUs together with central road operator servers compose the road infrastructure used for infrastructure to vehicle communications (I2V). RSUs disseminate continuously messages about road safety and traffic flows to vehicles. Vehicles and RSUs are able to measure, to compute, and to disseminate relevant information. In this study, we add to these actors the ability to learn road behaviours in order to be able to provide the optimal path in terms of travel time for vehicles seeking to reach their destination. The solution is based on a reinforcement learning technique, in particular the Q-learning, that is used to learn the best action to take into account according to various situations. The transit delay from one location to another is used to evaluate the rewards. The simulation results confirm that the proposed Q-learning approach outperforms the existing greedy algorithm with better performances in terms of a transit delay.