Citation

Espinosa Mireles de Villafranca A, Roncoli C. Transp. Res. Interdiscip. Persp. 2024; 25: e101042.

Copyright

(Copyright © 2024, Elsevier Publishing)

DOI

10.1016/j.trip.2024.101042

PMID

unavailable

Abstract

We present a computational study of network ensembles with two types of coexisting vehicle classes: an altruistically routing vehicle (ARV) class - potentially automated vehicles that are routed to reduce total system travel time - and a selfishly routing vehicle (SRV) class, corresponding to human-driven vehicles. We investigate the performance of these networks when some links are reserved for exclusive use by the ARVs. The goal of these interventions is to avoid or mitigate the detrimental effects of the SRVs on the costs of the ARVs. We formulate the problem as a bi-level network design problem, where the upper level deals with optimising the choice of ARV-exclusive links minimising the statistical dispersion of used-route costs, while the lower level finds the corresponding traffic equilibrium under static traffic assignment conditions. We tackle the ARV-exclusive link selection with a genetic algorithm, where the fitness of solutions is based on the dispersion of the costs of routes used by ARVs. The mixed equilibrium is found by solving a multi-class static traffic assignment problem, with constraints on the SRV flows on the ARV-exclusive links. SRVs attempt to minimise their personal travel time, whilst ARVs attempt to drive the flows to system optimal. Our approach is effective in reducing the per-vehicle travel cost of the ARVs to below that of the SRVs, making altruistic routing a more attractive option on average. Our results are consistent across networks with different structures and demand levels.

Keywords

Altruistic routing; Automated vehicles; Genetic algorithm; Multiclass equilibrium; Network design; Traffic management