Modeling the Last Mile: A Large-Scale Agent-Based Simulation Framework for Parcel Delivery

The growing volume of parcel deliveries challenges urban infrastructure. While numerous simulation studies have explored logistical strategies, a gap in large-scale, comprehensive, and cross-regional studies assessing parcel delivery traffic’s impact across different spatial area types has been identified. To address the gap in the literature, an agent-based freight simulation tool focusing on parcel deliveries was developed in this doctoral research project using the MATSim simulation framework. Unlike other studies that depend on smaller or hypothetical datasets, the approach benefits from using a comprehensive dataset as the base for the simulation framework. Thus, the actual parcel demand could be used to establish a calibrated baseline scenario. As a first step, a general agent-based transport simulation model of the Hanover Region was set up to represent the transportation infrastructure and traffic dynamics accurately. The model was extended to encompass parcel delivery traffic, leveraging the MATSim freight extension. A comprehensive parcel demand input dataset was extrapolated. Geospatial analysis was incorporated into the framework, identifying eight distinct spatial types. This integration allowed for a quantitative assessment of the impacts of delivery traffic across these areas. The complex Vehicle Routing Problem was effectively addressed, minimizing the computational workload. The model was utilized to evaluate the spatial impact of parcel Last-Mile Delivery (LMD) of the baseline scenario, with variability being uncovered across the area types. To understand how new logistic concepts affect urban and rural areas, two distinct strategies were integrated into the simulation framework as case studies: In the first case study, an extended Collection Point Delivery (CPD) or parcel locker strategy was explored. Introducing Collection Points (CP) enhanced the rate of successful first-time deliveries. Compared to the baseline scenario, the results demonstrated a reduction of traveled kilometers by up to 10 % and a decrease in CO2 emissions by up to 13 %. Shared CPs, used by different logistics service providers, generally yield better efficiencies at higher network penetrations, whereas dedicated lockers are more effective at lower rates. This observation stems from the segmentation of routes, initially only minimally simplified by the introduction of shared CPs. Compared to rural areas, deliveries to CPs are particularly efficient when the lockers are located in densely populated areas with high parcel demand in both scenarios. In the second case study, Consolidated Last-Mile Delivery (CLMD) strategies were evaluated, which involve a single provider managing all parcel deliveries within a defined area. Two CLMD scenarios were tested: one focusing on consolidation in low-density rural areas alongside conventional operations and a fully consolidated scenario across the entire region. The simulation demonstrated improvements in operational efficiency and environmental impact, with a decrease of up to 59 % in total kilometers driven by delivery vehicles and a 44 % reduction in CO2 emissions compared to the baseline scenario. Yet, despite these advances, the overall impact of these strategies remains modest compared to total traffic emissions. Therefore, it is questionable whether implementing the concept and associated market transformation is feasible. However, local consolidated solutions for inefficient delivery areas might present a more feasible option for implementation. In conclusion, to accurately represent the transportation infrastructure and traffic dynamics, a parcel LMD simulation model was used to assess new logistical concepts incorporating different spatial area types. The findings of this thesis emphasize the importance of expanding research beyond merely urban areas. More than half of the daily CO2 emissions from delivery vehicles are generated in urban peripheries and rural areas – areas that previous research has overlooked. The effectiveness of the analyzed logistics concepts varies depending on the spatial conditions in which they are implemented. Thus, the impact of other spatial area types on logistics operations must be acknowledged and integrated into implementing logistics solutions.