Sitemap
A list of all the posts and pages found on the site. For you robots out there, there is an XML version available for digesting as well.
Pages
Posts
Future Blog Post
Published:
This post will show up by default. To disable scheduling of future posts, edit config.yml and set future: false.
Blog Post number 4
Published:
This is a sample blog post. Lorem ipsum I can’t remember the rest of lorem ipsum and don’t have an internet connection right now. Testing testing testing this blog post. Blog posts are cool.
Blog Post number 3
Published:
This is a sample blog post. Lorem ipsum I can’t remember the rest of lorem ipsum and don’t have an internet connection right now. Testing testing testing this blog post. Blog posts are cool.
Blog Post number 2
Published:
This is a sample blog post. Lorem ipsum I can’t remember the rest of lorem ipsum and don’t have an internet connection right now. Testing testing testing this blog post. Blog posts are cool.
Blog Post number 1
Published:
This is a sample blog post. Lorem ipsum I can’t remember the rest of lorem ipsum and don’t have an internet connection right now. Testing testing testing this blog post. Blog posts are cool.
portfolio
Portfolio item number 1
Short description of portfolio item number 1
Portfolio item number 2
Short description of portfolio item number 2
publications
A customized deep neural network approach to investigate travel mode choice with interpretable utility information
Published in Journal of Advanced Transportation, 2020 [Link]
Discrete choice modeling of travel modes is an essential part of traffic planning and management. Thus far, this field has been dominated by multinomial logit (MNL) models with a linear utility specification. However, deep neural networks (DNNs), owing to their powerful capacity of nonlinear fitting, are now rapidly replacing these models. This is because, by using DNNs, mode choice can be assimilated with the classification problems within the machine learning community. This article proposes a newly designed DNN framework for traffic mode choice in the style of two hidden layers. First, a local-connected layer automatically extracts an effective utility specification from the available data, and then, a fully connected layer augments the feature representation. Validated by a practical city-wide multimodal traffic dataset in Beijing, our model significantly outperforms the random utility models and simple fully connected neural network in terms of the prediction accuracy. Besides the comparison of the predictive power, we also present the interpretability of the proposed model.
Online operations strategies for automated multistory parking facilities
Published in Transportation Research Part E, 2021 [Link]
Parking at megacities has become a major problem that is garnering increasing attention. The fundamental cause of the parking problem is the imbalance of demand and supply in core areas, where parking demand is high but parking provision is limited owing to exorbitant land prices. The idea of multistory parking facilities is proposed to serve larger parking demands with fewer land possessions. The newly developed automated multistory parking facilities are able to pick-up and place cars on different stories automatically. This paper proposes online operations method (OOM) of automated multistory parking facilities in response to intensive parking demands to reduce customers’ waiting time. The proposed online optimization model is composed of two tiers: in the first tier, a reinforcement learning framework is adopted to determine parking spot selections for incoming parking demands, and the second tier executes the plan acquired from the first tier by optimizing the action sequences of the automated elevator. Numerical experiments with multiple demand patterns are conducted to verify the proposed methodology. The results show that the learned strategy distinguishes from common practice in that it shows non-greedy patterns for some time during the day, and achieves significant improvements in various cases.
Designing transit-oriented multi-modal transportation systems considering travelers’ choices
Published in Transportation Research Part B, 2022 [Link]
One goal of future transit-oriented transportation systems is to promote door-to-door mobility for travelers by integrating different public transportation modes into a whole. We propose a mathematical design framework for such a transit-oriented multi-modal transportation system from a societal aspect considering three categories of public transportation modes, i.e., general on-demand modes, local on-demand modes, and fixed-schedule modes. A system-state equilibrium is brought up to describe travelers’ rational travel choices and their reverse effects on agency service levels using the continuous approximation method, and a centralized system designer then manages to achieve a system-beneficial outcome with the minimum cost. To solve the design problem, we prove that the transportation system reaches a unique equilibrium when decision variables are given. By this discovery, we construct a global search framework based on the DIRECT algorithm to solve the optimal design. In analyzing the problem property, we rigorously prove that in the designed systems, the bus service as a fixed-schedule mode is absent from the design scheme under the cases with sufficiently low demands, and the design problem thus reduces to a one-dimensional line search. The ride-hailing service as a general on-demand mode is similarly proved to be excluded when the demand is sufficiently high. In this context, the approximate design parameters of the bus service and the total system cost are developed analytically. The local on-demand mode, bike-sharing service, as an option of bus feeders, is proved to be efficient under a realistic setting. Extensive numerical examples provide evidence verifying the preceding analyses and indicating the behavior of travelers and agencies. Further sensitivity tests show that the subway is favorable for intensive demands and autonomous vehicles may promote the ride-hailing industry. For completeness, an immediate application of the proposed framework in generalized cases validates the model reliability.
The online integration of competitive on-demand service markets
Published in Manufacturing & Service Operations Management (Under Revision), 2024 [Link]
Problem definition: As platforms for on-demand services flourish, a novel online integration trend rises to promote service efficiency. In view of pioneers like the ride-hailing industry, this paper explores the formation and impacts of online integration, where the third-party integrator establishes an integrated platform (IP) by delivering matching orders for participating service providers (SPs) and implements a commission policy on SPs’ actual completed revenue share. Methodology: We employ a three-stage game-theoretic approach to formulate the IP establishment process as a non-linear mixed-integer mathematical problem. By inspecting SPs’ feasibility and the integrator’s profitability, we derive analytical integration results in various typical markets, specify their impacts on the engaged parties, and propose regulating policies. Results: We have found that the integration outcome is primarily shaped by four interactive featured forces and obtained insightful findings. First, within the same IP, profit redistribution among differently sized SPs hinders larger SPs’ from integration as their comparative advantage diminishes. Second, excessive integration may inadequately stimulate customer demand and transform competitors into collaborators, thus reducing SPs’ interest in integration. Moreover, the integrator also circumvents excessive integration scale in the trade-off with commission rates for higher earnings. These findings deepen our understanding of contrasting IP situations in diverse real-world markets. For example, all minor SPs are possibly integrated if an extremely dominant SP exists like in China ride-hailing services, and the integration of two major SPs is less likely if they are divergent in size alike in the U.S. Further regarding the impacts, we find that the profit extraction can eclipse the integration benefits, impairing the entire industry particularly when worker supply is sufficient or balanced. Managerial implications: Integrators should make subtle decisions on market intervention and integration scale in consideration of market fragmentation, demand elasticity, supply levels, etc. Regulators can adopt a ceiling-commission policy towards integrators to safeguard the industry welfare.
Modeling and analyzing ride-hailing market equilibrium considering drivers’ multi-homing choice behavior
Published in Transportation Research Part E, 2025 [Link]
The success of ride-hailing services has engendered the rise of multiple service platforms in a competitive market. To this end, drivers are widely observed delivering services on more than one platform simultaneously to maximize their profits, referred to as the “multi-homing” behavior. Drivers’ spontaneous multi-homing has the potential to mitigate market fragmentation but may also intensify their competition for passengers across platforms. This study aims to address the impacts brought by multi-homing drivers on the stakeholders of ride-hailing services. We employ a game-theoretical framework to model the choice of freelance drivers on whether and how to multi-home, as well as the decision of passengers on which platform to attend in an asymmetric market with loyal drivers. On account of the multi-homing behavior, we specifically identify the effective supply of platforms dictating the matching efficiency behind the seeming blossom of supply. On this basis, we compare the equilibrated market under drivers’ multi-homing behavior with a monopoly market and another equilibrated market having drivers single-homing on one platform. Our finding suggests that compared to a monopoly, multi-homing alleviates the wild-goose chase (WGC) dilemma and benefits all stakeholders in supply shortages due to its weaker matching efficiency, resulting in a rational demand level and more available vehicles. In contrast to the benchmark where drivers single-home, multi-homing enhances overall system performance due to the enlarged pool of vacant vehicles for passengers. Specifically, it strengthens the supply advantage of dominant platforms in more balanced markets, while facilitating the expansion of weaker platforms in highly asymmetric markets with a moderate freelance fleet. Extensive numerical results derived from real-world datasets indicate that multi-homing becomes a viable strategy for freelance vehicles in markets with limited supply, and multi-homing contributes to the growth of less influential platforms in practice.
Should micro electric vehicles be embraced? A perspective from equilibrium analysis
Published in Accepted to Transportation Research Part E, 2025 [Link]
The success of ride-hailing services has engendered the rise of multiple service platforms in a competitive market. To this end, drivers are widely observed delivering services on more than one platform simultaneously to maximize their profits, referred to as the “multi-homing” behavior. Drivers’ spontaneous multi-homing has the potential to mitigate market fragmentation but may also intensify their competition for passengers across platforms. This study aims to address the impacts brought by multi-homing drivers on the stakeholders of ride-hailing services. We employ a game-theoretical framework to model the choice of freelance drivers on whether and how to multi-home, as well as the decision of passengers on which platform to attend in an asymmetric market with loyal drivers. On account of the multi-homing behavior, we specifically identify the effective supply of platforms dictating the matching efficiency behind the seeming blossom of supply. On this basis, we compare the equilibrated market under drivers’ multi-homing behavior with a monopoly market and another equilibrated market having drivers single-homing on one platform. Our finding suggests that compared to a monopoly, multi-homing alleviates the wild-goose chase (WGC) dilemma and benefits all stakeholders in supply shortages due to its weaker matching efficiency, resulting in a rational demand level and more available vehicles. In contrast to the benchmark where drivers single-home, multi-homing enhances overall system performance due to the enlarged pool of vacant vehicles for passengers. Specifically, it strengthens the supply advantage of dominant platforms in more balanced markets, while facilitating the expansion of weaker platforms in highly asymmetric markets with a moderate freelance fleet. Extensive numerical results derived from real-world datasets indicate that multi-homing becomes a viable strategy for freelance vehicles in markets with limited supply, and multi-homing contributes to the growth of less influential platforms in practice.
Network-based pick-up and drop-off location optimization for shared autonomous vehicles
Published in Urban Informatics, 2025 [Link]
Shared autonomous vehicles (SAVs) are reshaping the mobility system, with ride-hailing services being regarded as a pioneering application market. Despite its benefits, SAVs inevitably exacerbate the competition for limited curb space, which is already a scarce resource. Designated pick-up and drop-off sites, if reasonably planned, present a promising solution to this challenge. In this paper, we propose a novel network-based data-driven approach to optimize PUDO locations for SAVs. Network kernel density estimation (NKDE) is used to measure the PUDO demand along the road network based on historical ride-hailing data. The results are fed into a maximal coverage location problem (MCLP) model to optimize the spatial distribution of PUDO sites by maximizing the demand they can cover. The spatial optimization model is used both prescriptively and descriptively, assessing transport policies and enhancing the operations of SAVs. Results show that our model outperforms the baseline models. We also reveal that a suitable PUDO site density is about 20/10 km2, with denser sites in bustling urban centre and sparser sites in suburbs. The framework is highly generalizable and reproducible and can be extended to future studies.
Efficiency and interventions of strategic driver relocation for ride-hailing platforms
Published in Production and Operations Management, 2025 [Link]
Idle drivers’ spontaneous relocation, when provided with sufficient information, has the potential to mitigate the supply–demand imbalance in ride-hailing services. As this potential remains ambiguous, this study aims to address two fundamental questions and derive managerial insights: Q1: To what extent can drivers’ spontaneous relocation resolve the supply-demand imbalance? Q2: How can subsidies be designed to induce the platform optimum? We propose a leader-follower game-theoretic framework to investigate the dynamic relocation game on arbitrary service networks. The platform, as the leader, designs relocation subsidies to achieve service-oriented or interest-oriented goals, while idle drivers, as followers, decide relocation and compete for individual returns in a multi-stage context. Despite the inherent problem complexity, we investigate the existence, uniqueness, transformation, and solution of the driver equilibrium, providing insights into these critical questions. First, the platform should avoid issuing subsidies purely for relocation purposes when drivers’ actions align with platform goals, such as (1) when the supply and demand are sufficiently imbalanced in volume, eliminating drivers’ gambling behavior, and (2) when commission rates are low—a plausible case in practice—due to the tension between the platform and drivers. Second, we present the potential of the platform intervention in other cases, providing theoretical references for the platform expectations. Notably, due to the extra responsibility gap, the subsidy impact on completed trips exceeds that on platform profits, with the latter at most doubling. Third, we investigate the imbalance in demand that values the platform’s intervention. While a certain spatial imbalance or a demand surge in peak hours emphasizes subsidies for platform goals, the spatial and temporal imbalance can possibly complement each other, inducing coupled demand that directly achieves platform-optimal goals. These results comprehensively answer Q1 along with practical suggestions. Fourth, for Q2, we confirm its feasibility and provide actionable subsidy configurations that induce platform-optimal goals to aid platform management. Numerical examples demonstrate that sparse network structures can lead to larger realized gaps. Additionally, drivers with appropriately random behavior can enhance platform performance, suggesting careful information disclosure by platforms.
Area-weighted resource allocation for humanitarian logistics: An enhanced proximal policy optimization approach
Published in 2025 13th International Conference on Traffic and Logistic Engineering (ICTLE), 2025 [Link]
This paper addresses the critical challenge of resource allocation in humanitarian logistics during disaster response. We develop a comprehensive framework that optimizes the distribution of limited relief resources across multiple affected areas while balancing efficiency, effectiveness, and equity objectives. The novel contribution is the introduction of area weights into the optimization model, enabling prioritization based on region-specific factors such as population density, infrastructure importance, and proportion of vulnerable populations. We formulate the problem as a Markov Decision Process and propose an enhanced Proximal Policy Optimization (PPO) algorithm with an Actor-Critic architecture to solve the resulting nonlinear optimization model. Our approach focuses specifically on the critical 72-hour “golden rescue period” after disasters occur, when resource allocation decisions have the most significant impact on saving lives. The PPO algorithm efficiently navigates the complex trade-offs between competing objectives while accounting for capacity constraints and dynamic state transitions over multiple time periods. Numerical experiments demonstrate that our approach achieves significant improvements in convergence speed, solution quality, and stability across various problem scales. The model effectively adapts to different objective priorities and maintains robust performance under demand uncertainty, making it well-suited for complex, dynamic humanitarian operations. This research provides a practical solution for resource allocation decisions in disaster response scenarios, contributing to the development of more effective humanitarian logistics systems that can better serve affected populations during crises.
Optimized parking lot locations for shared autonomous vehicles: A perspective from minimum route distance
Published in Accepted to Journal of Transport Geography, 2025
Shared autonomous vehicles (SAVs) are revolutionizing urban mobility systems, necessitating significant updates to transportation infrastructure, particularly parking facilities, due to their unique driverless nature and shared operational mode. For SAVs, parking lots not only function as depots for idle vehicles but also serve as origins and destinations of services, thereby impacting their service levels and fleet operational efficiency. Along with the emission reduction initiatives, the strategic planning of parking lot locations for SAVs is a critical issue. In this paper, we propose a two-step optimization framework to locate SAV parking lots. The first step employs a vehicle-trip assignment strategy to explore SAV operations, addressing the high computational costs of simulation methods and the limited problem size of traditional operations research approaches. The second step extends the classical capacitated p-median problem (CPMP) model to a capacitated flow termini median model (CFTMM) for location selection, incorporating dual-node demand to better depict reality. Using Chengdu as a case study, we demonstrate the effectiveness of our framework and gain some actionable insights for urban policymakers and transport network companies. It shows that the CFTMM demonstrates superior performance compared to the two benchmarks by effectively minimizing the empty driving costs to and from parking lots. Moreover, 6,500 SAVs can meet all travel demand, with over 75\% of SAVs operating during the day and gradually finishing their work after 8 PM. The optimized parking lot location distribution highlights the importance of retaining some parking spaces in the city center and balancing the parking demand of urban and suburban areas. The extended capacitated flow termini median model with dynamics (CFTMM-D) yields comparable results, further confirming the effectiveness and efficiency of CFTMM.
Optimal parking planning for shared autonomous vehicles: A two-stage stochastic programming approach
Published in Under Review, 2025 [Link]
The successful deployment of shared autonomous vehicle systems relies on well-designed parking lots, which serve as both storage depots and dispatch hubs. A key challenge is the coordination between parking planning and fleet operations, exacerbated by varying land prices and demand uncertainty. To address these issues, we develop a two-stage stochastic programming model. The first stage determines the parking lot location, capacity, and initial SAV allocation, considering land prices. The second stage optimizes fleet operations through a time-space network to track vehicle movements. The model is solved using the sample average approximation method and validated through a case study in Chengdu, China. Results demonstrate that our approach consistently outperforms benchmark methods, highlighting the importance of an optimizationbased decision support for SAV parking planning. Further analysis reveals that as the fleet size increases, parking lots increasingly function as intermediate stops during operation, substantially reducing unnecessary vehicle movements and associated energy consumption.
Coordinating the port-carrier supply chain: Optimal design of volume-commitment discount contracts
Published in Under Review, 2025 [Link]
Ports operate as capacity-constrained service platforms that face substantial demand volatility and congestion risk from multiple heterogeneous carriers. In such case, a delicate design of contracts between the ports and carriers can foster their ability to cope with the inefficiency. This study develops a contract-based coordination framework that enables risk sharing and efficiency improvement in port-carrier cargo transfer. Specifically, we focus on volumecommitment discount contracts, which refer to the port offering a price incentive in exchange for the carriers’ binding commitment to a specific cargo volume, enhancing resource planning and operational efficiency. We first demonstrate that simple volume-commitment discount contracts, while Pareto-improving compared to no contract, fail to achieve full coordination due to double marginalization. To address this inefficiency, we propose a coupled volume-commitment discount and shortage-subsidy contract that simultaneously induces system-optimal commitment, mitigates congestion externalities, and flexibly allocates surplus among participants. Analytically, we establish that the coupled contract can fully coordinate the decentralized system and recover the system-optimal outcome under both single-carrier and multi-carrier settings with correlated stochastic demand. Comparative statics analyses further reveal that the port strategically designs differentiated contract terms considering carrier heterogeneity, offering more favorable discounts to carriers with less shortage-replenishment cost. Moreover, the port offers deeper discounts as a risk premium to carriers with higher demand uncertainty to manage systemic risks. This research contributes to the literature on supply chain coordination and maritime logistics by formalizing a risk-sharing contract that internalizes congestion externalities and enhances the operational resilience of port-carrier supply chain under uncertainty.
Coordination of competitive robots and workers in on-demand logistics services: Filling in the price of equity
Published in Working Paper, 2026
The emergence of robotic technologies threatens to replace human workers in on-demand logistics services, such as autonomous vehicles in ride-hailing and drones in delivery. In platforms currently operating with all-human workforces, we identify a notable inefficiency⏤termed the “price of equity” (POE)⏤stemming from indiscriminate order dispatching that disregards differences in order profitability. Therefore, robot threats can possibly be resolved along with the POE if the platform strategically deploys robots to handle less profitable orders, such as those involving high pickup difficulties or on-road congestion, and reserves workers for more valuable assignments in the mixed fleets. To explore this potential, we begin with developing an analytical model to estimate the POE, comparing the all-inclusive strategy and selective order admission based on order profitability. Real-world data reveal a maximum potential loss of 24% in platform revenue during demand peaks. Motivated by this inefficiency, we propose a threshold matching strategy for a platform in charge of both human workers and robots, while ensuring uniform service levels for all customers. This operational approach shows the potential to safeguard workers’ welfare amid robot penetration, and we substantiate that with an appropriate number of robots, the price of equity in workers’ earnings can be fully restored. Concerning the fleet size decision of robots, we consider strict public regulations in the transition stage where strategic platforms have to protect workers’ welfare. Relative to the maximum allowable fleet that mandates no harm to workers, platforms may deem robot adoption profitable only when the technology is sufficiently mature (i.e., unit robot costs are sufficiently low). Through a cross-period coordination scheme combined with a subsidy that taxes part of workers’ surplus from robotaxi adoption to fund robot procurement, the maximum coordinating robot procurement cost for socially beneficial adoption doubles from approximately 150 to 300 thousand CNY per vehicle. Under such favorable per-robot costs, platforms spontaneously introduce robots, enhancing customer service levels without adverse effects on workers’ welfare.
talks
Talk 1 on Relevant Topic in Your Field
Published:
This is a description of your talk, which is a markdown file that can be all markdown-ified like any other post. Yay markdown!
Conference Proceeding talk 3 on Relevant Topic in Your Field
Published:
This is a description of your conference proceedings talk, note the different field in type. You can put anything in this field.
teaching
Teaching experience 1
Undergraduate course, University 1, Department, 2014
This is a description of a teaching experience. You can use markdown like any other post.
Teaching experience 2
Workshop, University 1, Department, 2015
This is a description of a teaching experience. You can use markdown like any other post.