Distributed routing and rebalancing optimization for autonomous mobility-on-demand system

Abstract

This thesis addresses decision-making for networked autonomous vehicles in mobility on demand (MoD) systems. We introduce an optimization formulation, Decentralized Dial-A-Ride Problem (DDARP), that simultaneously accounts for the dispatching of vehicles in response to existing service requests and the routing vehicles for assigned requests. Additionally, we extend the D-DARP for ride-sharing and transfer problems, Decentralized Dial-A-Ride Problem with transfer (D-DARPT). The DARP comprises optimization to assign transportation requests and planning of individual vehicle routes. In the case of self-optimizing on-demand mobility, a dispatcher does not know the exact cost of the vehicle to perform each request. So we decompose DARP into D-DARP with sub-problems of decision-makers in the decentralized AMoD system. We develop an alternating direction method of multipliers (ADMM) based decomposition method to solve this optimization problem in a distributed manner effectively. The ADMM-based framework enables (i) decomposing the DARP into a dispatching problem and vehicle sub-problems with a subset of assigning decisions based on an ADMM

and (ii) optimizing individual vehicles to ensure the convergence to the feasible solution of the global system by an individual heuristic algorithm. The coordination algorithm is based on a variable neighborhood search and adaptive penalty selection strategy, accelerating the ADMM considering the binary decision. Numerical examples demonstrate the efficacy and the benefits of the optimization model in the context of real-world autonomous mobility-on-demand systems. Additionally, these optimization algorithms provide a hierarchical solution structure that aids in solving the autonomous mobility-on-demand system decision-making problems.