Drones in Warehouses: Advancements in Autonomous Inventory A Comparative Analysis of Vision-Based Techniques and Fiducial Marker Families for Indoor Drone Navigation and Relocation = Drones in Warehouses: Advancements in Autonomous Inventory A Comparative Analysis of Vision-Based Techniques and Fiducial Marker Families for Indoor Drone Navigation and Relocation

Autonomous navigation is one of the most interesting and complicated challenges of recent years: in particular, imagine the technical and application complexity of autonomously flying a drone within an indoor environment, where fixed and moving obstacles are present... Nowadays, there is a continuous attempt to automate and optimize any kind of process, in order to facilitate work from humans, and where possible, make machines do the most repetitive, alienating, time-consuming, error pruning and dangerous jobs. So that humans can focus on activities that enhance their intellectual value and business can grow faster and faster, by investing more and more in autonomous technology solutions. Thanks to today's cutting-edge technologies, drones can therefore be used to facilitate inventory and logistics operations in traditional warehouses, to carry out these operations in a safer, more efficient, and completely autonomous way. The state of the art regarding these issues is studied, and the main techniques and technologies that can be used to locate and navigate autonomously within an indoor environment are briefly explained. Solutions in which a standalone drone is used will be compared with others where the drone is supported by an AMR. After a study of the main indoor localization techniques such as UWB, wheel odometry, Lidar odometry etc; the main vision-based techniques are discussed more in-depth: VIO and fiducial marker-based relocalization, which often work in a complementary way with each other. Finally, the results of an extensive study of 2 families of fiducial markers are presented: Aruco vs April-Tag. The main differences in accuracy and performance using different metrics. This is done by comparing measurements obtained from 4 different cameras: two with a standard aperture, one wide-angle camera, and a fisheye camera.