Problème de routage des véhicules avec l’apprentissage par renforcement

Abstract

Abstract : Reinforcement learning constitutes a powerful method for solving the Vehicle Routing Problem, with interesting applications in logistics and transportation. In the context of this thesis, we examine the methodologies and approaches used to optimize delivery missions using reinforcement learning-based Vehicle Routing Problem (VRP) agents. We begin by providing a comprehensive study of the VRP, along with different research approaches and methodologies. Then, by reviewing previous articles on the subject, we evaluate their advantages and disadvantages, thereby paving the way for new research opportunities to enhance the efficiency and generalization of these methods in real vehicle routing scenarios. Our study demonstrates that reinforcement learning, notably through the actor-critic algorithm we have chosen, offers a promising approach to solving the VRP. By combining reinforcement learning with techniques such as the use of neural networks and advanced optimization algorithms. This approach proves to be particularly effective in addressing the geographical specificities of the Algerian context, as this VRP will be used for the wilayas of Algeria.