Published Paper Details:

Abstract--This paper presents a design, implementation and experimental evaluation of an AI-Driven Autonomous Unmanned Aerial Vehicle (UAV) Delivery System. The system uses Simultaneous Localization and Mapping (SLAM)-based navigation and real-time path optimization. The proposed system combines Deep Reinforcement Learning, Convolutional Neural Networks and an Explainable AI decision module. This enables autonomous parcel delivery in complex urban environments. Our system uses a custom hexacopter drone platform with LiDAR, stereo cameras, IMU, barometric pressure sensors and ultrasonic proximity detectors. The SLAM pipeline uses an ORB-SLAM3 and LiDAR SLAM fusion approach. This achieves sub-10 cm localization accuracy. The real-time path optimization engine executes a Modified A* algorithm. It also uses D* Lite re-planning and DRL-guided adjustments. This operates at 50 Hz update frequency. The Explainable AI module generates natural-language explanations of every drone decision. This includes route changes, altitude adjustments, obstacle detections and emergency protocols. The system is fully auditable and trustworthy for compliance. Experimental results across 847 real-world flight tests demonstrate a delivery success rate of 97.3%. The mean path planning latency is 18.7 ms. The obstacle avoidance reaction time is 23 ms. There is an end-to-end energy efficiency improvement of 34% over baseline systems.