PrimeTime: A Case Study of An Autonomous Last-Mile Parcel Delivery Network in New Jersey

Abstract

In the past two decades, we have seen significant breakthroughs in machine learning and artificial intelligence that have propelled the development and deployment of autonomous transportation entities. Notably, these entities can operate in their respective physical environment without needing additional, costly infrastructure and can interact with humans and other objects in their environment. This technology has fundamentally impacted nearly every business sector, especially the retail sector, and noteworthy examples include successful demonstrations of self-driving cars on public roads, widespread deployments of fulfillment robots in warehouses and brick-and-mortar stores, and experimentation of drones for last-mile parcel deliveries. As automation continues to disrupt many business models, especially those that have been traditionally more labor-intensive, the last-mile of parcel deliveries is relatively underappreciated in the automation space. As the descriptor implies, the last-mile of parcel deliveries is the final and generally the shortest leg of a parcel’s journey to its destination. However, its short distance belies the complex, multi-dimensional problems faced by logistics companies and parcel couriers alike. However, the successful application of autonomous technologies to the last-mile has significant upside potential for both senders and receivers in various dimensions, including those of time, geography, and financial costs. The goals of this thesis are to show the potential of synthesizing demand-based residential parcel data, provide a quantitative, multi-dimensional analysis of autonomous last-mile residential deliveries, and catalyze further development and adoption of autonomous technologies in the lastmile segment of supply chain logistics in spite of the inertia of this complex problem. Despite the vast amount of data that exists regarding the billions of packages delivered annually in the United States, much of it remains in secrecy, held by parcel couriers and e-commerce businesses. Nonetheless, this thesis attempts to accurately recreate this information by synthesizing demand-based parcel data using alternative data sources, such as household travel surveys and Census data. From our generated residential parcel data, we conduct a last-mile parcel delivery study in the context of New Jersey. Specifically, we compare the traditional, or current, human-based last-mile market for residential parcel deliveries with an alternative autonomous last-mile market and explore the multi-dimensional differences for various couriers and different delivery vehicle sizes. Additionally, this thesis proposes a more optimal autonomous last-mile market under a universal autonomous delivery fleet. In doing so, we propose a framework for autonomous fleets that is more optimally designed to leverage the spatial, financial, and temporal advantages and constraints of autonomous delivery vehicles rather than those of today’s human-operated vehicles. The problems that this thesis attempts to address are certainly nontrivial as the multidimensional benefits of automating last-mile deliveries will far outweigh the costs and have widespread implications for consumers and businesses alike. Furthermore, it’s clear that an automated last-mile for residential parcel deliveries is inevitable. However, current technological, social, and political roadblocks, as well as logistical complexities of the last-mile, have hindered significant progress in this area. Nonetheless, this thesis provides a novel analysis of the pros and cons of automating the last-mile of parcel deliveries to underscore the vast potential in this underappreciated area. Additionally, we propose an alternative framework for today’s last-mile delivery network that is designed with the constraints and advantages of autonomous delivery fleets in mind.