Last-mile Delivery by an Autonomous Drone Through a Balcony in Urban Environments

Last-mile delivery by a drone in urban environments with high-rise buildings is a challenging task due to the presence of severe GPS errors and a high risk of collisions. This paper proposes the technologies required for a drone to perform last-mile deliveries autonomously in urban environments. First, we recognize urban obstacles simulating high-rise buildings using 2D LiDAR and construct an occupancy grid map to avoid obstacles as the shortest path through the center point between obstacles. Next, we detect a balcony window look-alike marker using a deep learning model and then estimate the position and attitude of the origin-symmetric marker. Based on the estimated pose of the marker, the drone is guided to enter the balcony area and drop the cargo on the balcony. We also design a device and mechanism that fixes and drops cargo via lock and release operations, which is attached to the bottom of a small drone. Finally, we perform precision landing using a cascaded ArUco marker created from two ArUco markers with different IDs. We validate each of the proposed technologies in a realistic testing environment.