Crowdsourcing package delivery is an emerging business model, inspired by the “Sharing Economy”, which utilizes people’s existing travel for last-mile packages’ deliveries. While crowdsourcing has the potential of answering the growing expectations for faster and more efficient package delivery, there are several distinct architectures in which these systems can be deployed and operationalized.
Our aim at this study is to compare the two architectures that were prominent in the literature as architectures for crowdsourcing package delivery. The first, the one-hop architecture, assumes packages can be transferred between origin and destination stop points by one carrier, i.e. one-hop. The second, the multi-hop architecture, allows carriers to transfer packages to their destination by leaving them at stop points, which are intermediate storage units, until the next carrier picks them up. Although past studies have suggested these architectures, we are the first to compare between the two using actual data of large-scale population sample.
We used large real-world dataset of anonymized locations of ~1.8M distinct mobile users over a one-month period as a proxy to drivers’ movements. In addition, to model the demand for delivery requests in the multi-hop hub architecture, we used official data of the number of packages ordered from out of Israel at different Israeli cities.
In this talk we will present the preprocessing steps applied on the mobility dataset, such as outlier trajectory detection, stop point identification and stop points extraction from trajectories. We will then discuss the two models we used to evaluate performance measures of the system. The first model is based on a routing graph formulation that models in a high-level granularity the potential handoffs of packages. The second model, in which we combine trajectory aggregated data and demand data into a maximum-flow network formulation, helps us to evaluate the number of packages that can flow in such systems.
Our findings point to the effectiveness of the multi-hop architecture, but also highlight some of its limitations, with regard to geographical areas and delivery performance. We also provide a model that predicts performance measures given different participation proportions of the population in the service. Finally, we discuss how data-based simulation can be used to investigate crowdsourcing model before they are deployed.