On Accessibility Fairness in Intermodal Autonomous Mobility-on-Demand Systems

TL;DR

This paper addresses the gap between traditional cost-focused metrics and transportation justice by defining and optimizing accessibility unfairness in Intermodal Autonomous Mobility-on-Demand (I-AMoD) systems. It builds on a network-flow framework to compute minimum-time operations and casts a linear program for minimum-accessibility-unfairness, validated on a real-world Eindhoven case using OpenStreetMap, GTFS data, and ALBATROSS-generated demand; under a fleet cap of $N_{ ext{cars,max}}=4000$ and a travel-time threshold of $T_{ ext{max}}=20$ min, accessibility fairness can be improved with only a small increase in average travel time. The results show substantial reductions in unfairness metrics at the o-d-pair and path levels, and reveal a divergence between path-based and flow-based fairness, prompting further discussion on the proper fairness definition in dynamic settings. Overall, the work demonstrates that incorporating fairness into I-AMoD control is practically feasible and highlights directions for richer models incorporating multi-modal benefits, turn-taking mechanisms, and transdisciplinary considerations.

Abstract

Research on the operation of mobility systems so far has mostly focused on minimizing cost-centered metrics such as average travel time, distance driven, and operational costs. Whilst capturing economic indicators, such metrics do not account for transportation justice aspects. In this paper, we present an optimization model to plan the operation of Intermodal Autonomous Mobility-on-Demand (I-AMoD) systems, where self-driving vehicles provide on-demand mobility jointly with public transit and active modes, with the goal to minimize the accessibility unfairness experienced by the population. Specifically, we first leverage a previously developed network flow model to compute the I-AMoD system operation in a minimum-time manner. Second, we formally define accessibility unfairness, and use it to frame the minimum-accessibility-unfairness problem and cast it as a linear program. We showcase our framework for a real-world case-study in the city of Eindhoven, NL. Our results show that it is possible to reach an operation that is on average fully fair at the cost of a slight travel time increase compared to a minimum-travel-time solution. Thereby we observe that the accessibility fairness of individual paths is, on average, worse than the average values obtained from flows, setting the stage for a discussion on the definition of accessibility fairness itself.

Figures (6)

• Figure 2: Comparison of time-based modal share as a function of the average travel time of user-weighted o-d-pairs for the objectives of minimum travel time (top) and minimum accessibility unfairness (bottom).
• Figure 3: Time-based modal share difference as a function of the average travel time of user-weighted o-d-pairs.
• Figure 4: Comparison of time-based modal share as a function of the average travel time of user-weighted paths computed with Algorithm \ref{['alg:paths']} for the different objectives of minimum travel time (top) and minimum-unfairness-accessibility (bottom).
• Figure 5: Time-based modal share difference as a function of the average travel time of user-weighted paths.
• Figure 6: Average accessibility unfairness levels in the PC4 regions $u_r$ of Eindhoven, NL, for the minimum-travel-time (left) and minimum-accessibility-unfairness operation (right).