Optimizing Mobility on Demand Systems: Multiagent Reinforcement Learning Approaches to Order Assignment and Vehicle Guidance

This dissertation explores ways to improve Mobility on Demand (MoD) systems, which are services like ride-sharing and autonomous taxi systems. The main goal is to make these services more efficient and reliable, benefiting both passengers and drivers by better matching the number of available vehicles with the number of people needing rides. For ride-sharing services, a new method called T-Balance helps match riders with drivers and guides empty taxis to areas where more people need rides. This reduces wait times for passengers and increases earnings for drivers. Another method, called GRL-HM, looks at how riders and drivers behave to further improve the system’s efficiency and fairness. For autonomous taxi services, the dissertation introduces an advanced technique called Pr-DDQN to better plan routes for empty vehicles. This technique performs better than traditional methods, leading to higher customer satisfaction and shorter times for vehicles to get to passengers. A new framework is also proposed to improve how data is processed and how the system learns, making autonomous taxi services more efficient and scalable. Overall, this research develops smarter ways to manage rides and vehicles in MoD systems, helping create more sustainable and user-friendly urban transportation solutions.