Crowdshipping with Anticipated Crowdshipper Availability: An Approximate Dynamic Programming Approach
Files
Sadiki_40941900_2025.pdf
Open access - Adobe PDF
- 1.71 MB
Details
- Supervisors
- Faculty
- Degree label
- Abstract
- Crowdshipping is a collaborative delivery method in which private individuals, known as ‘crowdshippers’, deliver parcels during their personal journeys in exchange for a remuneration based on the detours made. This delivery method has the potential to be an interesting way of reducing delivery costs and environmental impact. However, it raises significant challenges in terms of the dynamic crowdshippers-to-parcels assignment, in a context where present assignment decisions affect future ones and future arrivals of parcels and crowdshippers are uncertain. This is all the more pronounced in peer-to-peer crowdshipping systems, which are characterised by highly heterogeneous journeys and no restrictions on origin or destination. This master's thesis is based on the Approximate Dynamic Programming (ADP) approach developed by Innocente and Tancrez (2025) in which the crowdshipping platform uses an ADP algorithm to dynamically assign parcels to crowdshippers. While the existing literature generally considers that crowdshippers declare themselves available at the exact moment of their journey, we explore a model in which crowdshippers announce their availability in advance. This anticipated information is modelled through a parameter epsilon, which represents the number of periods between the announcement of availability and the actual time of the journey. Our contributions consist of a comparative analysis of the ADP approach and the myopic one, integrating this anticipation parameter. We determine how the relative efficiency of the ADP approach varies according to several configurations of the problem: the delivery window, the problem size, the crowdshipper-to-parcel arrivals ratio, the cost structure as well as an alternative scenario introducing uncertainty about parcel arrivals. The results of 210 experiments show that the introduction of anticipation reduces the comparative advantage of the ADP approach over the myopic one. Indeed, as the information horizon extends, the capacity of the ADP approach to postpone assignments in hope of better opportunities becomes less relevant. In addition, we note that this reduction is accentuated when the system initially has large optimisation possibilities: a high crowdshipper-to-parcel arrivals ratio, a larger delivery window and a cost structure making delivery via crowdshipping attractive. Finally, fluctuations in parcel arrivals have a limited impact on the relative efficiency of the ADP approach, because parcels with the same characteristics at a decision period t, are aggregated and considered as a single parcel that can be delivered by a crowdshipper.