Crowdsourced last mile delivery: Collaborative workforce assignment

In this paper, we propose a last mile delivery selection model using crowdsourced workers that optimizes the trade-off between cost, time, and workers' performance. Most of the current methods utilize either greedy worker–task assignments or a task-by-task basis selection to reach a sufficient worker–task assignment. However, a better trade-off between the distance traveled and delivery time can be further obtained by considering the quality of performance on the tasks as a whole rather than treating tasks individually. As a solution, we present a novel framework for last mile delivery which separates the routing and assignment aspects of the problem and solves the assignment problem by maximizing the overall quality of the delivery. The Quality of Service (QoS) is defined as a non-linear function of the number of allocated tasks, distance traveled, timeliness of the delivery, workers' reputation, and confidence in delivery completion. In the first step, the delivery tasks to be shipped from a single warehouse are clustered using k-medoids. The set of tasks in each cluster are to be delivered by the same worker. The shipping provider will send a truck to handover the corresponding parcels to each worker. Accordingly, the shortest route for the truck is computed using Tabu search, where the handover points to the potential workers are the centroids of the clusters. Tabu search is also used to compute the potential workers' routes from the handover point through all the tasks in the cluster. Finally, genetic algorithm is used to effectively solve the assignment problem where each worker is assigned to several neighboring tasks. The performance of the proposed assignment mechanism is evaluated and compared to greedy solutions with respect to the QoS as well as its components. The results show that the proposed algorithm achieves 100% task allocation ratio while outperforming greedy selections in terms of QoS. Moreover, it is able to increase confidence in task completion by 20.3% on average and prevent delays to the schedule of the truck.