A Double-Layered MPC of Autonomous Mobility-on-Demand Systems

This paper presents a double-layered model predictive control (MPC) approach to control an Autonomous Mobilityon-Demand (AMoD) System whereby fleets of self-driving vehicles provide on-demand service. Specifically, we present the upper layer computing the dispatching strategy of the AMoD system with a time-expanded network flow model firstly. Then routes for rebalancing vehicles are optimized according to current congestion situation in the lower layer. To improve the scalability in the lower layer, we utilize an adaptive piecewise linearization method to transform the original optimization problem into a convex linear program while reserving the feature of Bureau of Public Roads (BPR) function. Finally, the performance of the double-layered MPC is compared with a single-layered MPC. Simulation results show that the proposed algorithm can ease road congestion while ensuring operational performance.