Ride-pooling Electric Autonomous Mobility-on-Demand: Joint Optimization of Operations and Fleet and Infrastructure Design

TL;DR

This work tackles designing and operating Electric Autonomous Mobility-on-Demand (E-AMoD) systems with ride-pooling in dense urban settings, aiming to minimize fleet energy consumption and vehicle-hours traveled while optimally locating charging infrastructure. It develops a time-invariant, multi-layer network flow model that tracks vehicle state-of-charge (SoC) and charging actions, and embeds this in a mixed-integer linear program (MILP) that jointly optimizes ride-pooling, routing, and infrastructure siting; a pruning step yields iso-energy arcs to keep the problem tractable. Ride-pooling is incorporated by transforming the original demand into an equivalent set of travel requests $\mathcal{M}'$ under constraints $\bar{t}$, $\bar{\delta}$, and capacity $K$, solved efficiently via a Knapsack-like algorithm to maintain polynomial-time performance. Case studies in Manhattan show that optimal charging infrastructure siting provides about a 1% energy improvement over heuristics, while ride-pooling can reduce total energy by up to 45%, with vehicle design effects depending on demand and waiting-time tolerance; these results highlight the framework’s potential to support energy-efficient planning for large-scale E-AMoD deployments.

Abstract

This paper presents a modeling and design optimization framework for an Electric Autonomous Mobility-on-Demand system that allows for ride-pooling, i.e., multiple users can be transported at the same time towards a similar direction to decrease vehicle hours traveled by the fleet at the cost of additional waiting time and delays caused by detours. In particular, we first devise a multi-layer time-invariant network flow model that jointly captures the position and state of charge of the vehicles. Second, we frame the time-optimal operational problem of the fleet, including charging and ride-pooling decisions as a mixed-integer linear program, whereby we jointly optimize the placement of the charging infrastructure. Finally, we perform a case-study using Manhattan taxi-data. Our results indicate that jointly optimizing the charging infrastructure placement allows to decrease overall energy consumption of the fleet and vehicle hours traveled by approximately 1% compared to an heuristic placement. Most significantly, ride-pooling can decrease such costs considerably more, and up to 45%. Finally, we investigate the impact of the vehicle choice on the energy consumption of the fleet, comparing a lightweight two-seater with a heavier four-seater, whereby our results show that the former and latter designs are most convenient for low- and high-demand areas, respectively.

Figures (10)

• Figure 1: Multi-layer network schematically representing an E-AMoD system. Each layer corresponds to a battery state of charge (SoC). The top layer represents the original network where each node, called geo-node, represents a geographical location. The curved geo-arcs connect every layer to the same geographical location. As a car travels, it drives through the solid arcs and decrease its SoC. The yellow arcs are placed at charging stations and capture charging activities.
• Figure 2: Original road network of Manhattan (grey). Pruned synthetic network (colored) with equal (energy) weights.
• Figure 3: Illustrative example of no ride-pooling and ride-pooling with two vehicles; dashed arrows indicate vehicles traveling without users on-board. Figure taken from PaparellaPedrosoEtAl2024b.
• Figure 4: The rebalancing and overall energy consumption for the whole fleet leveraging an optimal approach and an heuristic method, betweenness centrality, to do the siting of the charging infrastructure for Manhattan, NYC. Results are both with and without ride-pooling scenario, with a maximum of two users per vehicle at the same time, $K=2$. The results are normalized w.r.t. the overall energy usage of the no ride-pooling scenario with heuristic placement.
• Figure 5: Improvement of overall daily energy usage of the ride-pooling scenario w.r.t. no pooling, for a varying waiting time and delay $\bar{t}$ and $\bar{\delta}$, respectively. Seat capacity of the vehicles $K=2$. Overall number of demands equal to $14$ thousand per hour. All charging stations have the same power available.