Digital-twins and machine learning-assisted stable, energy-aware unmanned aerial and ground vehicles delivery in blockchain-enabled crowdsourcing framework

Rising demand in last-mile delivery has heightened sustainability, cost, and delay concerns in traditional logistics. Unmanned Aerial Vehicles (UAVs) offer a transformative solution to ground vehicles due to their efficiency, low emission, and ability to avoid traffic. However, realizing their potential in crowdsourced delivery faces fundamental barriers: limited flight time due to highly unpredictable energy consumption that varies significantly with flight environment and payload. Existing solutions, such as simplified payload-based energy models, in-transit recharging, or pairing UAVs with trucks, often overlook the effect of flight environment uncertainty and fluctuating energy demands. These approaches are typically tailored for structured delivery networks, making them unsuitable for crowdsourced systems. Moreover, the absence of transparent, accountable frameworks severely limits adoption for high-value perishable deliveries and platform scalability. This paper presents a comprehensive energy-aware, secure, and transparent framework that fundamentally addresses these industry-critical challenges through the integration of UAVs and ground vehicles within a crowdsourced last-mile delivery system. It features a Random Forest-based energy consumption model that considers the UAV's and flight environment data, as well as its payload. Additionally, it employs a delivery-success probability aware Gale-Shapely game-based task allocation mechanism to maximize Quality of Service (QoS) and Quality of Delivery (QoD). Digital twins are proposed and modeled for transparent and real-time package storage and status monitoring. Experiments show that the trained energy consumption model achieves a mean absolute error (MAE) of 1.66. The allocation evaluation results highlight that the proposed system improves the QoS by at least 37%, delivery success by at least 45%, worker reputation and payoff by at least 32% and 22%, respectively, compared to the benchmarks.