A sample application on drone cargo transportation using explainable artificial intelligence methods

Along with the technological advancements, the use of unmanned aerial vehicles (UAVs) and the diversity of their utilization areas are increasing. In today's world, they are highly preferred in civil and commercial fields, as well as in the field of military. Artificial intelligence applications have been integrated into UAVs with the expansion in their fields of use. Cargo transportation is one of the major areas in which UAVs are currently used and are expected to be used significantly more in the future. The UAV with artificial intelligence is designed to be able to detect obstacles and adjust its route by avoiding obstructions during transportation from a certain starting point to the target point. Explainable artificial intelligence was used to comprehend the logic behind the decisions it made while encountering the obstacle. In the thesis study, an environment was created in ROS and Gazebo simulations through the use of AR.Drone. In order for UAV to recognize the environment and reach the target point from the starting point without being blocked by obstacles, mapping and path planning were performed. The mapping process was done with the OctoMap library with three-dimensional occupancy grid approach. Regarding the obstacle detection, a convolutional neural network (CNN), a deep learning model, was used. LIME and Grad-CAM architecture was chosen as the explainable model and observations were made to determine which region in the images contributed to the prediction.