Optimizing personalized menus and incentives to increase driver autonomy in ridesharing and crowdsourced delivery platforms with stochastic driver behavior

Thumbnail Image
Horner, Hannah
Issue Date
Electronic thesis
Research Projects
Organizational Units
Journal Issue
Alternative Title
Peer-to-peer logistics platforms have become increasingly popular in recent years for performing last mile delivery, ridesharing, and more. In general, current platforms have the suppliers i.e. the drivers either sift through and select from a large number of requests or are assigned a single request that they may or may not be able to reject. In this dissertation we offer an alternative framework, that provides drivers with a small but personalized menu of requests to choose from. This creates a Stackelberg game, in which the platform leads by deciding what menu of requests to send to each driver, and the drivers follow by selecting which request(s) to accept from their received menus. Determining optimal menus, menu size, and request overlaps is complex as the platform has limited knowledge of drivers' request preferences. Exploiting the problem structure when drivers signal willingness to fulfill each request, we reformulate our problem as an equivalent single-level Mixed Integer Linear Program (MILP) and apply the Sample Average Approximation (SAA) method. Computational tests recommend a training sample size for inputted SAA scenarios and a test sample size for completing performance analysis. Our stochastic optimization approach performs better than current approaches, as well as deterministic optimization alternatives. A simplified formulation ignoring `unhappy drivers' who accept requests but are not matched is shown to produce similar objective values with a fraction of the runtime. A ridesharing case study of the Chicago Regional transportation network provides insights for a platform wanting to provide driver autonomy via menu creation. The proposed methods achieved high demand performance as long as the drivers are well compensated (e.g., even when drivers are allowed to reject requests, on average over 90% of requests are fulfilled when 80% of the fare goes to drivers; this drops to below 60% when only 40% of the fare goes to drivers). Thus, neither the platform nor the drivers benefit from low driver compensation due to its resulting low driver participation and thus low request fulfillment. Finally, for the cases tested, a maximum menu size of 5 is recommended as it produces good quality platform solutions without requiring much driver selection time. Stochastic driver responses, independently accepting or rejecting each request in their menus, endogenously depend on the offered compensation for each request and the driver's effort required to fulfill the request (e.g. extra driving time). Therefore, in this dissertation, we also create and solve an optimization model to simultaneously determine personalized menus and incentives to offer drivers. We exploit variable properties to circumvent nonlinear variable relationships, formulating the model as a linear integer program. Stochastic driver responses are modeled as a sample of variable and fixed scenarios. An imposed premium counterbalances solution overfitting. Solution methods decompose and iterate, improving performance of computational experiments that use request/driver trip information from the Chicago Regional transportation network. Our approach outperforms alternative methods and our first approach that has no incentives by strategically using personalized incentives to prioritize promising matches and to increase drivers’ willingness to accept requests. This benefits both customers and drivers: the average driver income is increased by 4.1% compared to the menu-only model, and 96.6% of requests are matched (4.1% higher than the menu-only method). Higher incentives are offered when drivers are more likely to accept, while fewer incentives and menu slots are reserved for driver-request pairs less likely to be accepted. We design a third approach to examine the tradeoffs between the potential performance gains with the inclusion of incentives, and ensuring a fair experience for drivers. We use the same methodology and experiment data as our second framework producing menus and incentives, with added constraints that enforce one of three fairness types. These constraints require that nearby drivers receive the same compensation for the same request (driver proximity fairness), that drivers closer to the request receive higher compensation as they incur a shorter customer wait time (closer higher fairness), or that all compensation offers for a request are the same (all equal fairness). Computational experiments illustrate that personalized incentives even under such fairness constraints can still benefit the platform. Several of the properties of the fairness-constrained solutions, namely, match rate, profit, and driver income, are in between those of the two extremes found in the no-incentives solutions and the incentives-without-fairness-constraints solutions, while also providing a certain level of incentive fairness. Compared to the remaining fairness settings, solutions from driver proximity fairness with a low distance threshold value (the method with the fewest fairness constraints) and from our method with no fairness constraints have fewer compensation offers that have a nonzero incentive, but have the highest incentive offer average.
August 2021
School of Science
Full Citation
Rensselaer Polytechnic Institute, Troy, NY
Terms of Use
PubMed ID