Published Paper Details:

Unmanned Aerial Vehicle (UAV) delivery is moving from isolated pilot trials toward routine use in logistics, healthcare, and disaster-response operations, driven by parallel advances in Simultaneous Localization and Mapping (SLAM), real-time path optimization, and Explainable Artificial Intelligence (XAI). Because these three research threads have largely developed within separate communities - robotics, operations research, and machine-learning interpretability, respectively - the literature that connects them for the specific case of autonomous UAV delivery remains fragmented. This paper addresses that gap with a structured review that surveys and synthesizes 40 verifiable peer-reviewed and archival sources published between 2005 and 2025. The literature is organized into four technology areas: localization and mapping, path planning and optimization, perception and obstacle avoidance, and explainable decision-making. Representative techniques within each area are compared using structured tables covering their operating principles, strengths, and limitations. Building on this synthesis, the paper abstracts a generalized layered system architecture and operational workflow that captures common design patterns across the reviewed systems, and benchmarks five representative real-world UAV delivery deployments. A critical analysis distills six recurring research gaps, including GPS-denied localization robustness, the opacity of AI-driven flight decisions, energy-constrained operation, and multi-UAV coordination at scale. The review closes by outlining future research directions - including edge and federated learning for onboard inference, standardized XAI reporting suitable for airworthiness review, and swarm-level explainability - intended to guide subsequent system-level research on AI-driven autonomous UAV delivery.