ORCIBS Seminar- Alireza Khani

In most mid-size American cities, fixed-route transit has limited geographic or temporal coverage, making it less competitive with personal vehicles. Transit users may have to take a long and inconvenient walk from the last alighting stop to their destination. This issue is known as transit last mile access problem and has been shown to be a significant factor in discouraging travelers from riding transit. An emerging solution for the transit last mile access problem is mobility on demand systems with autonomous vehicles (AMoD). In an AMoD system, managing the fleets of vehicles, commonly known as “dispatching” and “rebalancing” is crucial to ensure that vehicles are distributed properly to meet customer demands. This seminar presents an optimal control approach to optimize vehicle dispatching and rebalancing in an AMoD system integrated with fixed-route transit service. A linear discrete-time dynamical model is developed for the AMoD system and equilibrium points are derived analytically. A model predictive control (MPC) framework is applied to find the optimal dispatching and rebalancing policy in real time. Numerical results from a real-world application show significant improvement in the quality of service to customers compared to the state-of-the-art algorithms in the literature. As an alternative approach, a model-free, data-efficient Q-learning-based algorithm is presented to solve the H-infinity control of the AMoD system. An adaptive optimal controller is designed, and the parameters of the action and critic networks are learned online without the knowledge of the system dynamics.