Efficient Assignment in Peer-to-Peer Crowdshipping: An Approximate Dynamic Programming Approach

Crowdshipping is a collaborative delivery system that outsources delivery tasks to ordinary citizens that act as non-professional couriers, with the potential to reduce delivery costs and improve sustainability. The crowdshippers deliver shipments by taking a short detour from their planned trips in return for a compensation fee. The dynamic assignment of crowdshippers to parcels over time is challenging as the arrival of crowdshippers and parcels is stochastic and their availability is dynamically revealed and as present assignments affect future ones. Peer-to-peer crowdshipping, which encompasses all types of crowdshipper journeys and parcel deliveries, with no restrictions on origin or destination to specific locations, is subject to a high level of uncertainty. This work presents an approximate dynamic programming approach based on value function approximation for the dynamic assignment problem of peer-to-peer crowdshipping platforms. The approach is based on the offline adaptive approximation of parcel values and provides non-myopic behavior while only solving a sequence of assignment problems no larger than in a myopic approach. Through numerical results, we demonstrate our methodology’s effectiveness as, compared to a myopic benchmark, it increases the cost savings achieved via crowdshipping and reduces crowdshipper detours. Our analysis highlights the parameters affecting the relevance of implementing a non-myopic assignment method.