Rebalancing shared mobility-on-demand systems: A reinforcement learning approach

Wen, Jian; Zhao, Jinhua; Jaillet, Patrick

doi:10.1109/itsc.2017.8317908

Cited by 108 publications

(81 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Both of these deal with a matching market where the agents are constantly moving around and hence it is important to characterize the flow patterns of these agents. A number of works study the prediction aspects as well as the dynamics of flow patterns in both bike-sharing (e.g., [27,12]) as well as in ridesharing (e.g., [35,34,17,36,29]) platforms.…”

Section: Assumptions and Related Workmentioning

confidence: 99%

Mix and Match: Markov Chains and Mixing Times for Matching in Rideshare

Curry

Dickerson

Sankararaman

et al. 2019

Web and Internet Economics

View full text Add to dashboard Cite

Rideshare platforms such as Uber and Lyft dynamically dispatch drivers to match riders' requests. We model the dispatching process in rideshare as a Markov chain that takes into account the geographic mobility of both drivers and riders over time. Prior work explores dispatch policies in the limit of such Markov chains; we characterize when this limit assumption is valid, under a variety of natural dispatch policies. We give explicit bounds on convergence in general, and exact (including constants) convergence rates for special cases. Then, on simulated and real transit data, we show that our bounds characterize convergence rates-even when the necessary theoretical assumptions are relaxed. Additionally these policies compare well against a standard reinforcement learning algorithm which optimizes for profit without any convergence properties. IntroductionRideshare firms such as Uber, Lyft, and Didi Chuxing dynamically match riders to drivers via an online, digital platform. Riders request a driver through an online portal or mobile app; a driver is matched by the platform to a rider based on geographic proximity, driver preferences, pricing, and other factors. The rideshare driver then picks up the rider at her request location, transfers her to her destination, and reenters the platform to be matched again-albeit at a new geographic location. Part of the larger sharing economy, rideshare firms are increasingly competitive against traditional taxi services due to their ease of use, lower pricing, and immediacy of service [14].Matching riders to rideshare drivers is nontrivial. While the core process is a form of the well-studied online matching problem [23], current models developed in the EconCS, AI, and Operations Research communities [28,11,5] do not completely capture the mobile aspects of both the drivers and riders. Drivers and riders are agents who move about a constrained space (e.g., city streets), becoming (in)active periodically due to the matching process. When a platform receives a request, it must make a near-real-time dispatch decision amongst nearby drivers who are available at the current time. The platform's goal is to maximize an objective (e.g., revenue or throughput) by servicing requests in an online fashion, subject to various realworld constraints and challenges like setting prices, predicting supply and demand, fairness considerations, competing with other firms, and so on [8,17,4].In this paper, we study the dynamics of the nascent rideshare market under different dispatch strategies. Recent work uses Markov chains to model complex ride-sharing dynamics in a closed-world system-that is, a system with a fixed total supply of cars [28,4,5]. These assume the Markov chains reach their stationary distributions quickly, and thus all prior work optimizes for dispatch strategies in the limit. As a complement, the present paper characterizes-theoretically and empirically-when that limit assumption is valid under a variety of natural strategies.

show abstract

Section: Assumptions and Related Workmentioning

confidence: 99%

Mix and Match: Markov Chains and Mixing Times for Matching in Rideshare

Curry

Dickerson

Sankararaman

et al. 2019

Web and Internet Economics

View full text Add to dashboard Cite

show abstract

“…However, numerous operational issues remain to be addressed, vehicle rebalancing being one of main open issues in one-way car sharing systems [2], [3]. Due to inherent differences in patterns of people movement at different times of the day, vehicles tend to accumulate in certain areas.…”

Section: Introductionmentioning

confidence: 99%

“…Simulations show that ability of vehicles to autonomously rebalance and predictively reposition to high-demand locations reduces both customer waiting time and overall required fleet size [4]. However, research remains to be done on strategies and algorithms for rebalancing; current approaches are centralized, consider the full network and are therefore computationally intensive, and assume all cars belong to a single fleet, i.e., are controlled by a central entity [3]. In addition, most current rebalancing approaches do not consider ride-sharing [3]; vehicles can either be occupied or unoccupied, and only unoccupied ones are considered as available.…”

Section: Introductionmentioning

confidence: 99%

“…However, research remains to be done on strategies and algorithms for rebalancing; current approaches are centralized, consider the full network and are therefore computationally intensive, and assume all cars belong to a single fleet, i.e., are controlled by a central entity [3]. In addition, most current rebalancing approaches do not consider ride-sharing [3]; vehicles can either be occupied or unoccupied, and only unoccupied ones are considered as available. With the introduction of ride-sharing, partially occupied vehicles also need to be taken into account in the overall strategy, as they are able to potentially serve further requests along their travel route to the destination, i.e., effectively being rebalanced to the areas they are travelling through.…”

Section: Introductionmentioning

confidence: 99%

“…[6] propose a centralized RL process to learn re-allocation patterns for the whole fleet. In the decentralized approach described in [3], each vehicle uses Deep Q-learning to learn whether to take a rebalancing action during regularly scheduled rebalancing windows. A similar approach was developed in [7], where an individual autonomous taxi learns the most likely passengers pick up points.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

SAMoD: Shared Autonomous Mobility-on-Demand using Decentralized Reinforcement Learning

Guériau

Dusparić

2018

2018 21st International Conference on Intelligent Transportation Systems (ITSC)

View full text Add to dashboard Cite

Shared mobility-on-demand systems can improve the efficiency of urban mobility through reduced vehicle ownership and parking demand. However, some issues in their implementations remain open, most notably the issue of rebalancing non-occupied vehicles to meet geographically uneven demand, as is, for example, the case during the rush hour. This is somewhat alleviated by the prospect of autonomous mobility-on-demand systems, where autonomous vehicles can relocate themselves; however, the proposed relocation strategies are still centralized and assume all vehicles are a part of the same fleet. Furthermore, ride-sharing is not considered, which also has an impact on rebalancing, as already occupied vehicles can also potentially be available to serve new requests simultaneously. In this paper we propose a reinforcement learning-based decentralized approach to vehicle relocation as well as ride request assignment in shared mobility-on-demand systems. Each vehicle autonomously learns its behaviour, which includes both rebalancing and selecting which requests to serve, based on its local current and observed historical demand. We evaluate the approach using data on taxi use in New York City, first serving a single request by a vehicle at a time, and then introduce ride-sharing to evaluate its impact on the learnt rebalancing and assignment behaviour.

show abstract

Carpool services for ride‐sharing platforms: Price and welfare implications

Wang

Zhang

2021

Naval Research Logistics

View full text Add to dashboard Cite

There has been rapid growth in on‐demand ride‐hailing platforms that serve as an intermediary to match individual service providers (drivers) with consumer demand (riders). Several major players of this market have introduced carpool services that allow passengers heading toward the same direction to share a ride at a discounted fare. In this article, we develop an analytical model to study the pricing issues of ride‐sharing platforms in the presence of carpool services, and their economical and social implications. We show that the carpool service should be provided when its quality and/or the pooling efficiency is high. When the platform finds it optimal to offer the carpool service option, the platform achieves a larger market coverage and the riders are able to enjoy more affordable rides without compromising on service quality. Our analysis reveals that the provision of carpool services benefits the platform and the riders in general, but may hurt the drivers.

show abstract

Rebalancing shared mobility-on-demand systems: A reinforcement learning approach

Cited by 108 publications

References 14 publications

Mix and Match: Markov Chains and Mixing Times for Matching in Rideshare

Mix and Match: Markov Chains and Mixing Times for Matching in Rideshare

SAMoD: Shared Autonomous Mobility-on-Demand using Decentralized Reinforcement Learning

Carpool services for ride‐sharing platforms: Price and welfare implications

Contact Info

Product

Resources

About