Development of autonomous unmanned aerial systems in indoor environments

The need for humanoid tasks to be performed by robots has led to autonomous mobile robot applications that make their own decisions and perform a task accordingly, just like humans. An autonomous robot must know the geometric structure of its environment, position itself accordingly, and finally, based on this information, it must create a movement trajectory towards the task point. In order to develop autonomous robots sharing the same environment with humans, there are Odometry and Simultaneous Localization and Mapping (SLAM) studies to localize the mobile robot with different sensors in indoor environments where the Global Positioning System (GPS) is insufficient. In the last decade, researchers have focused on Visual SLAM (VO) and Visual SLAM (VSLAM) methods performed with cost-effective monocular cameras due to improvements in processor speed. In addition to cameras, VISLAM and VIO solutions, which include low-cost Inertial Measurement Unit (IMU) sensors, have recently been preferred because of their contribution to localization. Solutions so far often include traditional geometric-based solutions. Because the good representation of the real complex world is very difficult with these classical methods, reliable results are often not obtained and they are also very dependent on manually adjusted parameters. Therefore, nowadays, traditional solutions are replaced by Artificial Intelligence-based solutions in terms of adaptability to different environments and ease of application. In the light of the above-mentioned information, this thesis proposes three different applications for the development of an autonomous Unmanned Aerial Vehicle (UAV) in GPS-denied indoor environments. The first application offers a deep learning-based hybrid architecture and visual and IMU information-based work to predict the position of a UAV moving indoors. The second application implements a different artificial intelligence-based VIO application with a different fusion technique, which transforms the IMU information into an image and successfully estimates the angle information as well as the position of the UAV. The last application proposes a new path planning method for a UAV with known position information in a three-dimensional environment. Moreover, in addition to a new path planning method, an optimization and Artificial Intelligence-based application has been developed for the proposed method, resulting in real-time path planning. All three applications offer new methods for the development of an autonomous UAV in the indoor environment. All methods showed the performance to outperform most previous studies. In addition, the applications realized are capable of working in real-time systems.