Dynamic UAV trajectory optimisation for parcel delivery with integrated third-party risk mitigation

Last-mile delivery by Uncrewed Aerial Vehicles (UAVs) is an emerging solution for urban parcel transport, offering a sustainable and cost-effective alternative to traditional ground-based logistics. However, these operations introduce significant Third-Party Risk (TPR), necessitating comprehensive safety assessments. Evaluating UAV TPR in urban environments is particularly challenging due to the complexity of such settings, where factors like population density, urban geometry, and UAV behaviour must be carefully considered. Recent studies have developed analytical models for UAV risk assessment, but these often rely on oversimplified assumptions. For instance, many models focus primarily on total power loss scenarios while neglecting other critical failure modes, such as sensor malfunctions or flight controller errors. Additionally, existing models frequently overlook the impact of urban geometry in crash simulations, relying instead on constant shelter factors. This oversight can lead to inaccurate risk estimations, as buildings, roads, and other structures significantly influence collision probabilities and third-party exposure. To address these limitations, this study introduces a novel UAV failure simulation framework that realistically models aircraft responses to the most common UAV hardware failures. A UAV simulator is employed to generate probabilistic failure trajectories, which are then used to assess third-party exposure within a detailed 3D digital urban environment. This TPR assessment is integrated into a multi-objective optimisation framework based on a k-Non Aligned Nearest Neighbours (k-NANN) approach. Compared to traditional grid-based path planning methods, this approach generates safer and more efficient trajectories, reducing operational risk by up to 30% and flight time by 20%.