Parcel consolidation approach and routing algorithm for last-mile delivery by unmanned aerial vehicles

The last-mile delivery by unmanned aerial vehicles (UAVs), emerging from the online grocery retailing industry, has attracted much attention and interest from the scientific and industrial communities. In reality, the online retailing platform receives grocery orders and promises ultrafast delivery service to the customers via a two-echelon logistics and distribution network. The uncertain arrivals of the orders have a nontrivial negative effect on the performance of the whole delivery network, which has not been well studied in the literature. This paper investigates the parcel consolidation policy and the UAVs routes for the last-mile delivery from a transshipment site in a distribution network to the final customers. First, this study proposes a nonmyopia consolidation policy considering the upcoming grocery parcels with random arrival times to minimize the total delivery cost including the possible penalty cost due to the delivery delay to the customer. To make the problem tractable in the real-time computational setting, an approximation method based on Bayesian estimation is proposed to reduce a large number of random arrival scenarios of upcoming parcels to an expected scenario. The problem is then approached with a deterministic model under the expected scenario. Subsequently, a discrete particle swarm optimization (DPSO) algorithm is developed to solve the model. In the algorithm, a novel decoding method is designed to evaluate the particles with respect to the constraints of the load and battery capacities of a UAV and a neighborhood search based on reinforcement learning is developed to improve the quality of the particles in the searching process. The experimental results based on a case study validate the performance of the proposed parcel consolidation approach and the UAV routing algorithm. Some findings are given based on the variations of benchmarks with different distributions of customer locations. The approaches and insights in this paper could be used as a reference for last-mile delivery by UAVs.