Ride-Sharing Networks with Mixed Autonomy

TL;DR

This paper models and analyzes the integration of autonomous vehicles into ride-sharing networks, demonstrating how mixed fleets can optimize profits and providing computational methods for equilibrium analysis.

Contribution

It introduces a novel mixed autonomy ride-sharing model, formulates a nonconvex profit optimization, and offers a convex approximation for efficient computation.

Findings

Mixed autonomy can increase platform profits.

Optimal vehicle mix depends on autonomous vehicle costs.

A convex problem formulation enables efficient profit computation.

Abstract

We consider ride-sharing networks served byhuman-driven vehicles and autonomous vehicles. First, wepropose a novel model for ride-sharing in this mixed autonomysetting for a multi-location network in which the platformsets prices for riders, compensation for drivers, and operatesautonomous vehicles for a fixed price. Then we study thepossible benefits, in the form of increased profits, to the ride-sharing platform that are possible by introducing autonomousvehicles. We first establish a nonconvex optimization problemcharacterizing the optimal profits for a network operatingat a steady-state equilibrium and then propose a convexproblem with the same optimal profits that allows for efficientcomputation. Next, we study the relative mix of autonomous andhuman-driven vehicles that results at equilibrium for variouscosts of operation for autonomous vehicles. In particular, weshow that there is a regime for which the platform will chooseto mix autonomous and human-driven vehicles in order tooptimize profits. Our results provide insights into how suchride-sharing platforms might choose to integrate autonomousvehicles into their fleet.