System Design and Operations of Package Delivery with Crowdshipping and Advanced Air Mobility

This dissertation presents methodological studies on designing, modeling, and evaluating of package delivery systems that employ crowdshipping- and Advanced Air Mobility (AAM) inspired electric Vertical Takeoff and Landing (eVTOL) aircraft-based delivery. This involves developing optimization models and scalable solution methods for crowdshipping and eVTOL-based package delivery systems for large-scale real-world applications as a substitute for traditional truck-based delivery systems. Three studies are presented where the first study focuses on crowdshipping for last-mile delivery, and the second and third studies focus on eVTOL-based middle-mile to last-mile delivery. The first study proposes a Deep Reinforcement Learning (DRL)-based approach to the dynamic on-demand crowdshipping-based last-mile delivery problem in which requests constantly arrive in a crowdshipping system for pickup and delivery. The pickup and delivery of requests are handled by crowdsourcees, who are everyday individuals participating in the crowdshipping system. They allocate their varying and limited time and carrying capacity to perform crowdshipping tasks. In exchange, they receive compensation from the delivery service provider, who regularly assigns requests to crowdsourcees throughout the day to minimize shipping expenses. To train DRL agents, algorithms based on Heuristics-embedded Deep Q-Networks are employed, incorporating both double and dueling structures. To tackle the hard constraints pertaining to crowdsourcee and request time windows, three new constraint-handling strategies are proposed and integrated into the DRL training and testing. An extensive numerical experiment is conducted to demonstrate the superiority and scalability of the DRL-based approach over other traditional methods while keeping a very small and acceptable optimality gap. The second study proposes an innovative package delivery system using eVTOL aircraft as a substitute for trucks to carry freight from a warehouse to vertiports that are close to the final destinations of the requests in a metro region. In using eVTOLs for delivery, a particular concern is the associated noise that will generate negative impacts on neighborhoods surrounding the vertiports where eVTOLs take off and land. A generalizable method is proposed to estimate the community noise impact. The estimated community noise impact is integrated into an integer programming model to seek the optimal eVTOL schedules and vertiport choice while meeting package delivery demand, with the objective of simultaneously minimizing shipping cost and community noise impact. To solve this bi-objective problem, a customized solution method which builds on and extends the non-dominated sorting genetic algorithm is developed. The implementation of the model with the solution method is demonstrated in a case study of package delivery to the north suburbs of the Chicago metro region. Due to the spatial and temporal variation of the population in neighborhoods surrounding vertiports, minimizing the total community noise impact for large-scale eVTOL operations in close proximity to households may result in a disproportionate distribution of community noise impact among different communities. To mitigate this, the third study of this dissertation formulates an inter-community noise equity metric considering the number of flights scheduled at each vertiport and the percentage of the population affected by eVTOL noise depending on the eVTOL arrival time. This inter-community noise equity metric is designed to be used as a regulatory measure to enforce the delivery service provider to schedule its flights maintaining an equitable distribution of service. The mixed integer programming model proposed in this study considers that the eVTOLs are able to visit multiple vertiports in one route provided that the flying range and carrying capacity permit while maintaining a prespecified inter-community noise equity. A hybrid algorithmic solution method is proposed combining set covering optimization, local search heuristics, and adaptive large neighborhood search algorithms to solve the problem that can produce scalable results with a very small optimality gap. The proposed method is tested for a large-scale case study of the north suburb of the Chicago metro region.