Data-Driven Model Predictive Control of Autonomous Mobility-on-Demand Systems

Abstract

The goal of this paper is to present an end-to-end, data-driven framework to control Autonomous Mobility-on-Demand systems (AMoD, i.e. fleets of self-driving vehicles). We first model the AMoD system using a time-expanded network, and present a formulation that computes the optimal rebalancing strategy (i.e., preemptive repositioning) and the minimum feasible fleet size for a given travel demand. Then, we adapt this formulation to devise a Model Predictive Control (MPC) algorithm that leverages short-term demand forecasts based on historical data to compute rebalancing strategies. Using simulations based on real customer data from DiDi Chuxing, we test the end-to-end performance of this controller with a stateof-the-art LSTM neural network to predict customer demand: we show that this approach scales very well for large systems (indeed, the computational complexity of the MPC algorithm does not depend on the number of customers and of vehicles in the system) and outperforms state-of-the-art rebalancing strategies by reducing the mean customer wait time by up to to 89.6 %.

Stochastic Model Predictive Control for Autonomous Mobility on Demand

“…Third, we demonstrate through experiments that the proposed algorithm outperforms the aforementioned deterministic counterparts when the demand distribution has significant variance. In particular, on the same DiDi Chuxing dataset, our controller yields a 62.3 percent reduction in customer waiting time compared to the work presented in [6].…”

“…In this section, we first present a stochastic, time-varying network flow model for AMoD systems that will serve as the basis for our control algorithms. Unlike in [6], the model does not assume perfect information about the future, instead it assumes that customer travel demand follows an underlying distribution, which we may estimate from historical and recent data. Then, we present the optimization problem of interest: how to minimize vehicle movements while satisfying as much travel demand as possible.…”

“…However, [3] suffers from computational complexity as the fleet size grows, and [3], [6] do not account for uncertainty on the forecasts. While the authors of [6] show impressive results in their experiments, it can be shown that the difference between the stochastic optimum and the certainty equivalent one can be arbitrarily large. To address stochasticity of demand, [7] proposes a distributionally robust approach leveraging semidefinite programming.…”

Enhanced Mobility With Connectivity and Automation: A Review of Shared Autonomous Vehicle Systems

“…Topic Reference Optimization Fleet sizing [38], [73], [92], [93], [98], [115], [130] Vehicle routing / trip assignment [81], [98], [106], [120], [123], [125], [127], [131]- [136] Vehicle rebalancing / relocation…”

Smart Charging Benefits in Autonomous Mobility on Demand Systems

“…On the modeling side, queueing theoretical models capture the stochasticity of the customers [8], while network flow models efficiently optimize the fleet control in a static setting [9]. Owing to their simplicity, network flow based formulations are commonly used for algorithmic control of routing and rebalancing in a receding-horizon fashion [10], and to control congestion effects [11]. In addition to AMoD technology, the transportation sector is looking to increase utilization of electric vehicles (EVs) for consumers, companies, and fleet operations.…”