Unmanned air vehicles (UAVs) are attracting increased attention as their envelope of suitable tasks expands to include activities such as perimeter tracking, search and rescue assistance, surveillance and reconnaissance. The simplified goal of many of these tasks is to image an object for tracking or information-gathering purposes. The ability to determine the inertial location of a visible, ground-based object without requiring a priori knowledge of its exact location would therefore prove beneficial. This thesis discusses a method of localizing a ground-based object when imaged from a fixed-wing UAV. Using the target's pixel location in an image, with measurements of UAV position, attitude and camera pose angles, the target is localized in world coordinates. This thesis also presents a study of possible error sources and localization sensitivities to each source. From this study, an accuracy within 15.5 m of actual target location is expected. Also, several methods of filtering are presented, which allow for effective noise reduction. Finally, filtered hardware results are presented that verify these expectations by localizing a target from a fixed-wing UAV using on-board vision to within 10.9 meters.

Unmanned aerial vehicles are a salient solution for rapid deployment in disaster relief, search and rescue, and warfare operations. In these scenarios, the agility, maneuverability and speed of the UAV are vital components towards saving human lives, successfully completing a mission, or stopping dangerous threats. Hence, a high speed, highly agile, and small footprint unmanned aerial vehicle capable of carrying minimal payloads would be the best suited design for completing the desired task. This thesis presents the design, analysis, and realization of a dual-nacelle tiltrotor unmanned aerial vehicle. The design of the dual-nacelle tiltrotor aerial vehicle utilizes two propellers for thrust with the ability to rotate the propellers about the sagittal plane to provide thrust vectoring. The dual-nacelle thrust vectoring of the aerial vehicle provides a slimmer profile, a smaller hover footprint, and allows for rapid aggressive maneuvers while maintaining a desired speed to quickly navigate through cluttered environments. The dynamic model of the dual-nacelle tiltrotor design was derived using the Newton-Euler method and a nonlinear PD controller was developed for spatial trajectory tracking. The dynamic model and nonlinear PD controller were implemented in Matlab Simulink using SimMechanics. The simulation verified the ability of the controlled tiltrotor to track a helical trajectory. To study the scalability of the design, two prototypes were developed: a micro scale tiltrotor prototype, 50mm wide and weighing 30g, and a large scale tiltrotor prototype, 0.5m wide and weighing 2.8kg. The micro scale tiltrotor has a 1.6:1 thrust to weight ratio with an estimated flight time of 6 mins in hover. The large scale tiltrotor has a 2.3:1 thrust to weight ratio with an estimated flight time of 4 mins in hover. A detailed realization of the tiltrotor prototypes is provided with discussions on mechanical design, fabrication, hardware selection, and software implementation. Both tiltrotor prototypes successfully demonstrated hovering, altitude, and yaw maneuvering while tethered and remotely controlled. The developed prototypes provide a framework for further research and development of control strategies for the aggressive maneuvering of underactuated tiltrotor aerial vehicles.

<p>Project Solaris is a student research project with the goal to build a solar powered Unmanned Aerial Vehicle. This study is one in a set of studies that make up the initial phase of project Solaris. The main objective of this report is to investigate earlier solar powered airplanes as well as evaluate (or explore) potential future niche markets where solar powered UAVs could excel.A presentation of earlier solar powered airplanes will give an overall understanding of how solar powered airplanes have evolved and also provide information about the goals and ambitions behind the projects.Potential applications such as power line inspection and algal bloom observation will be described and a list of specifications for each application will be presented.</p> / Solaris

<p>The final thesis named Tracking in Sensor Networks was created by the company Instrument control Sweden. The thesis work is to create a plug-in to an already developed software pro- gram, SkyView. Via the software program one should be able to control an IP-camera from AXIS, model 215 PTZ, with the appropriate protocol over a network.  An Application Programming Interface, API, was used to easily control the camera through SkyView. The interface is called VAPIX and is used to all cameras from AXIS. VAPIX sup- ports Hyper Text Transfer Protocol, HTTP, and Real Time Streaming Protocol, RTSP.  The video was sent through the camera with RTSP, thus an investigation resulted in that the protocol was best suited at streaming video. Steering commands to the camera was sent via HTTP. A small program was written, which was then integrated to SkyView. Features to be able to control the camera with the keyboard, mouse and a joystick were implemented. Zoom, bright- ness, focus and some more features were also added. The problem with the delay should be dealt with as soon as possible.</p>

Project Solaris is a student research project with the goal to build a solar powered Unmanned Aerial Vehicle. This study is one in a set of studies that make up the initial phase of project Solaris. The main objective of this report is to investigate earlier solar powered airplanes as well as evaluate (or explore) potential future niche markets where solar powered UAVs could excel.A presentation of earlier solar powered airplanes will give an overall understanding of how solar powered airplanes have evolved and also provide information about the goals and ambitions behind the projects.Potential applications such as power line inspection and algal bloom observation will be described and a list of specifications for each application will be presented. / Solaris

The problem of stand-off tracking of a moving target using a quadrotor unmanned aerial vehicle (UAV) based on vision-sensing is investigated. A PID (Proportional-Integral-Derivative) controller is implemented for attitude stabilization of the quadrotor. An LQG-based (Linear-Quadratic-Gaussian) control law is designed and implemented for position control of the quadrotor for a moving target tracking task. A novel vision-based estimation algorithm is developed, enabling estimation of quadrotor’s position, altitude and yaw relative to the target based on limited information about the target. Two image processing algorithms are implemented and compared for the task of feature detection and feature tracking in a series of images. Image processing algorithms are integrated with quadrotor control and experiments are performed to validate proposed control and estimation approaches.

The problem of stand-off tracking of a moving target using a quadrotor unmanned aerial vehicle (UAV) based on vision-sensing is investigated. A PID (Proportional-Integral-Derivative) controller is implemented for attitude stabilization of the quadrotor. An LQG-based (Linear-Quadratic-Gaussian) control law is designed and implemented for position control of the quadrotor for a moving target tracking task. A novel vision-based estimation algorithm is developed, enabling estimation of quadrotor’s position, altitude and yaw relative to the target based on limited information about the target. Two image processing algorithms are implemented and compared for the task of feature detection and feature tracking in a series of images. Image processing algorithms are integrated with quadrotor control and experiments are performed to validate proposed control and estimation approaches.

A methodology is presented in this work for intelligent motion planning in an uncertain environment using a non-local sensor, like a radar sensor, that allows the sensing of the environment non-locally. This methodology is applied to an unmanned helicopter navigating a cluttered urban environment. It is shown that the problem of motion planning in a uncertain environment, under certain assumptions, can be posed as the adaptive optimal control of an uncertain Markov Decision Process, characterized by a known, control dependent system, and an unknown, control independent environment. The strategy for motion planning then reduces to computing the control policy based on the current estimate of the environment, also known as the "certainty equivalence principle" in the adaptive control literature. The methodology allows the inclusion of a non-local sensor into the problem formulation, which significantly accelerates the convergence of the estimation and planning algorithms. Further, the motion planning and estimation problems possess special structure which can be exploited to reduce the computational burden of the associated algorithms significately. As a result of the methodology developed for motion planning in this thesis, an unmanned helicopter is able to navigate through a partially known model of the Texas A&M campus.

Many modern mobile applications, such as Unmanned Aerial Vehicles (UAVs), require sophisticated processing capability with low power consumption in a small form factor. UAVs, for example, may require a platform capable of controlling a camera, performing digital signal processing techniques on the pictures to detect faces or motion, and guiding the vehicle based on decisions made from the processed data. Additionally, since the vehicle is mobile and aerial, its effectiveness is heavily dependent on the size and power consumption of the platform. In this report, we explore this set of requirements and how well they are met with a Texas Instruments OMAP SoC on a BeagleBoard. Specifically, we report on the computational performance and power drawn by the OMAP General Purpose Processor (GPP) when performing a facial detection algorithm with OpenCV. We also analyze the performance enhancement possible by offloading the facial detection algorithm to the OMAP DSP coprocessor. In summary we find that the Beagleboard would be an appropriate platform for a simpler UAV capable of pre-processing still images taken every few seconds, but not for processing video data real-time. We conclude by describing other applications that are suitable for the Beagleboard. / text

Les principaux travaux développés dans cette thèse traitent de la commande, du diagnostic et de la tolérance aux fautes utilisant le concept de la commande sans modèle récemment proposé dans la littérature. Les méthodes développées sont testées en temps réel et validées sur un quadrotor. Dans un premier temps, le modèle du quadrotor est présenté et analysé dans l'objectif de concevoir et implanter une loi de commande robuste aux perturbations et aux incertitudes de modèle. Pour cela, le concept de commande sans modèle est utilisé. Ce concept est basé sur la détermination d'une loi de commande calculée à partir de modèles ultra-locaux ajoutée à la commande calculée par des régulateurs classiques. Une technique utilisant le régulateur linéaire quadratique ainsi que la technique non linéaire de backstepping avec intégrateur augmentées de ce concept de commande sans modèle ont été proposées. Ces deux approches ont été testées et validées en les appliquant en temps-réel au drone considéré. Une étude comparative des lois de commandes avec et sans l'utilisation du concept de commande sans modèle a été proposée mettant en évidence les avantages de l'utilisation de ce concept. Le diagnostic de défauts est une étape importante pour la commande tolérante aux fautes. Les techniques de diagnostic développées dans cette thèse sont basées sur la génération de résidus issus de la comparaison entre les mesures réelles et les estimations de ces mesures obtenues par des estimateurs. La génération de résidus dépend de la qualité du modèle utilisé ainsi que des perturbations qui peuvent conduire à des fausses alarmes ou des non détections. Une nouvelle technique "d’estimation intelligente" inspirée du concept de commande sans modèle a été proposée et implantée afin de rendre la génération de résidus insensible aux incertitudes de modèle et aux perturbations et ainsi améliorer la décision quant à la présence de défauts. Deux "estimateurs intelligents" ont été proposés en rajoutant à l'observateur d'état classique et l'observateur de Thau un concept inspiré de la technique de commande sans modèle. L'estimation des sorties du système obtenue par l'estimateur intelligent est utilisée afin d'estimer l'amplitude des défauts d'actionneurs et de capteurs. L'estimation des défauts d'actionneurs est basée sur des modèles ultra-locaux. Quant aux défauts de capteurs, un algorithme structure a été proposé pour estimer leur amplitude en utilisant les résidus générés à partir de cet estimateur. Les résultats de détection et estimation de défauts ainsi obtenus sont ensuite utilisés pour compenser l'effet des défauts sur les performances du drone. La commande tolérante aux fautes mise en œuvre permet de modifier la loi de commande par rapport à l'estimation de l'amplitude du défaut d'actionneur, alors que lorsqu'un défaut de capteur est détecté et estimé, la trajectoire de référence est régénérée afin de compenser l'effet du défaut et maintenir le drone sur la trajectoire initialement définie. Les algorithmes proposés ont été implantés et testés sur un quadrotor Qball-X4. Les résultats de vol en temps-réel ont été analysés et ont permis de valider les techniques de commande et de tolérance aux fautes de capteurs et d'actionneurs. Des vidéos illustrant différentes expérimentations sont disponibles en ligne / The main objectives of this thesis consist of developing Control, Fault Detection and Diagnosis (FDD) and Fault-Tolerant Control (FTC) techniques based on a the Model-Free (MF) concept recently introduced in the literature. The proposed approaches are implemented, tested and validated on a quadrotor platform. The first step of this work consisted of the modelling of the quadrotor, and then analyzing, designing and implementing new robust control strategies based on the Model-Free Control (MFC) technique recently developed in the literature. The MFC algorithm helps compensating for disturbances and model uncertainties. The advantage of this recent concept is in the simplicity of the design of the controller by adding a control law using ultra-local models to the classical control techniques. To test and validate this new approach, the Linear-Quadratic-Regulator (LQR) and the Nonlinear-Integral-Backstepping (NIB) controllers have been considered by integrating the MFC concept to design a (LQR-MFC) and a (NIB-MFC), respectively. Both algorithms are validated through analytical and experimental procedures and the robustness checked and compared with respect to the initial controllers in the presence of disturbances and model uncertainties.The FDD is a very important step towards the development of FTC techniques. The FDD approach developed in this thesis is based on the residual generation between the measured outputs and the estimated outputs obtained using observers/estimators. Residuals are expected to be close to zero in the fault free case and deviate from zero in the presence of a fault or failure. However, as the residuals are generated using models, they highly depend on the quality of the model used and on the presence of disturbances which may lead to false alarms or to non-detections. A novel “intelligent estimator” inspired from the MF concept has been developed and used in order to improve the residual generation and the fault diagnostic. Two intelligent estimators have been designed by integrating the MF scheme with the state and Thau observers for Multi-Input-Multi-Output (MIMO) systems, where the intelligent Output-Estimator (iOE) represents the integration between the MF technique with the state observer, and the intelligent Thau Output-Estimator (iTOE) represents the augmentation of the MF technique with the Thau Observer. The estimation of the system outputs obtained using the iOE are then used to estimate the actuator and sensor faults. The estimation of the actuator faults is improved by using the ultra-local models. A structured algorithm is then developed and implemented in order to estimate sensor fault magnitudes using the residuals generated by the intelligent estimator. The results obtained from the fault detection and estimation are then used to compensate for the fault effect on the flight control performance. The implemented fault-tolerant control technique compensates for the actuator faults by adjusting the control law based on the fault estimation. In case a sensor fault is detected and estimated, the desired path is regenerated according to the estimated fault magnitude in order to compensate for the fault effect.The proposed algorithms are implemented and tested on the Qball-X4 quadrotor. The results of the real-time flight tests validate the proposed techniques and compensate for sensor and actuator faults. Footages of the flight tests are available online.