Research on Vehicle Dispatch Problem Based on Kuhn-Munkres and Reinforcement Learning Algorithm

With the development of artificial intelligence and 5G communication technology, autonomous vehicles are gradually becoming more achievable. Autonomous vehicles are used in urban transportation to provide taxi service, which effectively reduces labor cost and realizes intelligent transportation systems. The vehicle system combined with 5G technology can quickly obtain traffic information, which provides a decision basis for the vehicle dispatching. Thus, it is necessary to develop an efficient way to distribute and allocate these vehicles to maximize the potential income for the system. This paper studies the vehicle dispatching based on the travel data from the 2016 New York City Green Taxi data and propose two dispatching methods. First, we consider the dispatching problem as a maximum weight value matching problem. Then a distance-based dispatching method is proposed with the goal of minimizing the waiting time of passengers by using the Kuhn and Munkres (KM) algorithm. Finally, we formulate the decision of vehicle dispatch with a Markov Decision Process (MDP) and introduce a Reinforcement Learning (RL)-based dispatching method, which combines RL algorithm and KM algorithm to solve the dispatching problem with the goal of maximizing long-term revenue of divers. In the experiment, KM algorithm is compared with the full permutation algorithm to prove the effectiveness of KM algorithm. The performance of the distance-based dispatching method and RL-based dispatching method are presented in a small-scale dispatching and a large-scale dispatching. Experiment results show that the total revenue of vehicles is improved by about 20% by using RL-based dispatching method, compared to dispatching method based on the distance. Thus, RL-based dispatching method is more effective in a dispatching platform. It could be used by future public autonomous vehicle companies to achieve the fulfill the need of maximizing the potential income for the system.