Advanced guidance and navigation of small UAVs under GPS-denied environment with experimental validations

Small vertical take-off and landing (VTOL) Unmanned Aerial Vehicles (UAVs) are playing more and more important role in recent years for a wide range of applications including wireless communication, topographic mapping, disaster relief, and military action. The size of the UAVs becomes smaller to enable challenging applications such as indoor exploration. With the reduced size of the structure, battery, and on-board sensors, the payload is constrained. As a result, the quality of the on-board sensors and the on-board computation ability of small UAVs are limited. Under these conditions, to complete complicated tasks, research on small VTOL UAVs are required in many aspects. This thesis mainly focuses on four aspects: 1) Object recognition and tracking for vessel deck landings; 2) Simultaneous localization and mapping (SLAM); 3) Wind disturbance estimation; and 4) UAV platform design. Autonomous landing of a quadrotor UAV on a vessel deck is challenging due to the special sea environment. In this thesis, an on-board monocular vision-based solution that provides a quadrotor with the capability to autonomously track and land on a vessel deck platform with simulated high sea state conditions is presented. The whole landing process includes two stages: approaching from a long range and landing after hovering above the landing platform. Only on-board sensors are used in both stages, without external information input. This thesis use Parrot AR.Drone as the experimental quadrotor platform, and a self-designed vessel deck emulator is constructed to evaluate the effectiveness of the proposed vessel deck landing solution. Experimental results demonstrate the accuracy and robustness of the developed landing algorithms. Second, this thesis presents a novel solution to on-board sensor only feature based monocular 3-D SLAM algorithm for Small UAVs. The proposed SLAM algorithm can navigate autonomously in previously unknown environments without external positioning aid by only using one camera and an IMU. Main contribution of this work is the proposed method can generate absolute scale map without map initialization and optimization methods. These advantages make the presented SLAM algorithm more applicable to use than other existing methods for small UAVs. Applying the proposed method to a low-cost quadrotor, experimental results demonstrate that our system can generate map correctly with absolute scale. At the same time, the position estimation accuracy of the camera is also improved compared to using inertial navigation only. Wind speed estimation for VTOL UAVs is challenging due to the low accuracy of airspeed sensors, which can be severely affected by the rotor's down wash effect. Different from the traditional aerodynamic modeling solutions, in this thesis, a K Nearest Neighborhood learning based method which does not require the details of the aerodynamic information is presented. The proposed method includes two stages: an off-line training stage and an on-line wind estimation stage. Only flight data is used for the on-line estimation stage, without direct airspeed measurements. This thesis uses Parrot AR.Drone as the testing quadrotor, and a commercial fan is used to generate wind disturbance. Experimental results demonstrate the accuracy and robustness of the developed wind estimation algorithms under hovering conditions. For VTOL UAVs hardware system design, this thesis provides a quadrotor solution which can support most of the research of small VTOL UAVs. The quadrotor's structure is made by carbon fiber, which provides large freedom of weight for on-board computers and sensors. For the on-board computer, a high-level processor and a low-level processor work together as a two-level processor system. The low-level processor is responsible for sensor data acquiring, synchronization, and motor speed controls. Furthermore, the low-level processor also monitors the manual control signal during the flight and can switch the control source to manual control under emergency, which makes the platform safe to operate. For the high-level processor, an Ubuntu operation system is equipped. At the same time, Robot Operating System (ROS) is also embedded to access on-line available ROS packages easily. Researches can also make their own packages which contain their own algorithms. Another function for high-level processor is to communicate with users in real-time through wireless communication. The on-board sensor system includes: 1) a downward looking camera, 2) an Inertial Measurement Unit (IMU) which contains a 3-axis gyro and a 3-axis accelerometer, 3) a 3-axis compass, and 4) a downward looking ultrasonic sensor. These sensors provide rich information. Combined with the powerful two-level processor system, the whole system can help many important UAV research.Ph.D.Includes bibliographical reference