A REINFORCEMENT LEARNING APPROACH TO VEHICLE PATH OPTIMIZATION IN URBAN ENVIRONMENTS

Road traffic management in metropolitan cities and urban areas, in general, is an important component of Intelligent Transportation Systems (ITS). With the increasing number of world population and vehicles, a dramatic increase in road traffic is expected to put pressure on the transportation infrastructure. Therefore, there is a pressing need to devise new ways to optimize the traffic flow in order to accommodate the growing needs of transportation systems. This work proposes to use an Artificial Intelligent (AI) method based on reinforcement learning techniques for computing near-optimal vehicle itineraries applied to Vehicular Ad-hoc Networks (VANETs). These itineraries are optimized based on the vehicle’s travel distance, travel time, and traffic road congestion. The problem of traffic density is formulated as a Markov Decision Process (MDP). In particular, this work introduces a new reward function that takes into account the traffic congestion when learning about the vehicle’s best action (best turn) to take in different situations. To learn the effect of this approach, the work investigated different learning algorithms such as Q-Learning and SARSA in conjunction with two exploration strategies: (a) e-greedy and (b) Softmax. A comparative performance study of these methods is presented to determine the most effective solution that enables the vehicles to find a fast and reliable path. Simulation experiments illustrate the effectiveness of proposed methods in computing optimal itineraries allowing vehicles to avoid traffic congestion while maintaining reasonable travel times and distances.