This thesis proposes two acoustic localisation techniques that are accurate in Non-Line-of-Sight (NLoS) conditions and system implementation of the proposed techniques. Such conditions can cause positive bias errors, namely NLoS errors, in the measured Time-of-Arrivals (ToAs) of first-arrival signals received by microphones and thus reduce the positioning accuracy. The primary issue of the thesis is to precisely estimate and correct the NLoS errors by modelling the received first-arrival signals. The first proposed technique uses multiple on-ground microphones to locate a sound source. The proposed technique approximately estimates and corrects the NLoS errors based on an initial guess of the sound source position and a map. The localisation is then achieved by iteratively correcting the ToAs and updating the sound source location. The strength of the proposed technique is that its accuracy is not notably affected by small or known obstacles. The proposed technique is implemented into two localisation systems of controlled and uncontrolled sound sources. The performance of the proposed technique is investigated by its comparison with three other time-based localisation techniques in series of experiments and simulations, showing at least 10% improvement by the proposed technique under various background noise levels. The second proposed technique localises a sound source using a single on-ground microphone subject to an assumption of a single diffraction in the first-arrival signal. To predict the angular and radial coordinates of the sound source relative to the diffraction point, a new magnitude delay frequency profile is proposed. The profile can be estimated by applying the uniform geometrical theory of diffraction and be extracted from measured signals using a derived formulation. Similar to the first technique, the second proposed technique estimates the measured delay of the first-arrival signal for computing the radial coordinate. The angular coordinate is then obtained by matching the estimated and measured profiles at the measured delay. A key achievement of the second proposed technique is enabling NLoS localisation using only one microphone without any time-consuming pre-measurement. This technique is implemented into a localisation system of a controlled sound source and validated experimentally with three different sound sources and under two background noise levels.

Micro aerial vehicle development and atmospheric flight on Mars are areas that require research in very low Reynolds number flight. Facilities for studying these problems are not widely available. The upper atmosphere of the Earth, approximately 100,000 feet AGL, is readily available and closely resembles the atmosphere on Mars, in both temperature and density. This low density also allows normal size test geometry with a very low Reynolds number. This solves a problem in micro aerial vehicle development; it can be very difficult to manufacture instrumented test apparatus in the small sizes required for conventional testing. This thesis documents the design, construction, and testing of a glider designed to be released from a weather balloon at 100,000 feet AGL and operate in this environment, collecting airfoil and aircraft performance data. The challenges of designing a vehicle to operate in a low Reynolds number, low temperature environment are addressed.

"Micro aerial vehicles (MAV) are small remotely piloted or autonomous aircraft. Wingspans of MAVs can be as small as six inches to allow MAV’s to avoid detection during reconnaissance missions. Improving the aerodynamic efficiency of MAV’s by increasing the lift to drag ratio could lead to increased MAV range and endurance or future decreases in aircraft size. In this project, biologically inspired flight is used as a framework to improve MAV performance since MAV’s operate in a similar flight regime to birds. A novel wind tunnel apparatus was constructed that allows the planform shape of a MAV wing to be easily altered. The scale-model wing mimics a bird wing by using variable feather lengths to vary the wing planform shape. Genetic algorithms that use natural selection as an optimization process were applied to establish successive populations of candidate wing shapes. These wing shapes were tested in the wind tunnel where wings with higher fitness values were allowed to ‘breed’ and create a next generation of wings. After numerous generations were tested an acceptably strong solution was found that yielded a lift to drag ratio of 3.28. This planform was a non conventional planform that further emphasized the ability of a genetic algorithm to find a novel solution to a complex problem. Performance of the best planform was compared to previously published data for conventional MAV planform shapes. Results of this comparison show that while the highest lift to drag ratio found from the genetic algorithm is lower than published data, inabilities of the test wing to accurately represent a flat plate Zimmerman planform and limitations of the test setup can account for these discrepancies."

The thesis focuses on special space--time computational techniquesintroduced recently for computational aerodynamics modeling of flapping wings of an actual locust. These techniques complement the Deforming-Spatial-Domain/Stabilized Space--Time (DSD/SST) formulation, which is the core computational technique. The DSD/SST formulation was developed for flows with moving interfaces, and the version used in the computations is "DST/SST-VMST," which is the space--time version of the residual-based variational multiscale (VMS) method. The special space--time techniques are based on using NURBS basis functions for the temporal representation of the motion of the locust wings. The motion data is extracted from the high-speed video recordings of a locust in a wind tunnel. In addition, temporal NURBS basis functions are used in representation of the motion of the volume meshes computed and also in remeshing. These ingredients provide an accurate and e fficient way of dealing with the wind tunnel data and the mesh. The thesis includes a detailed study on how the spatial and temporal resolutions influence the quality of the numerical solution.

Small scale unmanned aircraft, such as quadrotors, that are quickly emerging as versatile tools for a wide range of applications including search and rescue, hazardous environment exploration, or just shooting great video, are known as micro air vehicles (MAVs). However, for millimeter scale vehicles with weights under 10 grams, conventional flight technologies become greatly inefficient and instead inspiration is drawn from biology. Flapping wing MAVs (FWMAVs) have been created based on insects and hummingbirds in an effort to emulate their extreme agility and ability to hover in place. FWMAVs possess unique capabilities in terms of maneuverability, small size, and ability to operate in dynamic environments that make them particularly well suited for environmental monitoring and swarm applications such as artificial crop pollination. Despite their advantages, significant challenges in fabrication, power, and control must be overcome in order to make FWMAVs a reliable platform. Current designs suffer from high mechanical complexity and often rely on off-board power, sensing, and control, which compromises their autonomy and limits practical applications. The goal of my research is to develop a simple FWMAV design that provides high efficiency and controllability. An efficient, simple, and controllable vehicle design is developed utilizing the principles of resonance, emulation of biological flight control, and under-actuation. A highly efficient, resonant actuator is achieved by attaching a spring in parallel to the output shaft of a commercial geared DC micro-motor. This actuator directly drives the wings of the vehicle, allowing them to be controlled precisely and independently. This direct control strategy emulates biology and differs from other FWMAV designs that utilize complicated transmissions to generate flapping from rotary motor output. Direct control of the wings allows for emulation of biological wing kinematics, resulting in control based on wing motion alone. Furthermore, under-actuation is employed to mimic the rotational motion of insect wings. A rotational joint is added between the motor and wing membrane such that the wing rotates passively in response to aerodynamic forces that are generated as the wing is driven. This design is realized in several stages, initial prototyping, simulation and development of the actuator and wings, then finally a control system is developed. First the system was modeled and improved experimentally in order to achieve lift off. Improvements to the actuator were realized through component variation and custom fabrication increasing torque and power density by 161.1% and 666.8% respectively compared to the gearmotor alone and increased the resonant operating frequency of the vehicle from 4 Hz to 23 Hz. Advances in wing fabrication allowed for flexible wings that increased translational lift production by 35.3%, aerodynamic efficiency by 41.3%, and the effective lift coefficient by 63.7% with dynamic twisting. A robust control architecture was then developed iteratively based on a date driven system model in order to increase flight time from 1 second (10 wing strokes) to over 10 seconds (230 wing strokes). The resulting design improves lift to weight by 166%, allowing for a payload capacity of approximately 8.7 g and offers the potential for fully autonomous operation with all necessary components included on-board. A thermal model for micro-motors was developed and tuned to accurately predict an upper limit of system operation of 41 seconds as well as to optimize a heatsink that increases operating time by 102.4%.

This thesis investigates the possibility of using reinforcement learning (RL) techniques to create a flight controller for a quadrotor Micro Aerial Vehicle (MAV). A capable flight control system is a core requirement of any unmanned aerial vehicle. The challenging and diverse applications in which MAVs are destined to be used, mean that considerable time and effort need to be put into designing and commissioning suitable flight controllers. It is proposed that reinforcement learning, a subset of machine learning, could be used to address some of the practical difficulties. While much research has delved into RL in unmanned aerial vehicle applications, this work has tended to ignore low level motion control, or been concerned only in off-line learning regimes. This thesis addresses an area in which accessible information is scarce: the performance of RL when used for on-policy motion control. Trying out a candidate algorithm on a real MAV is a simple but expensive proposition. In place of such an approach, this research details the development of a suitable simulator environment, in which a prototype controller might be evaluated. Then inquiry then proposes a possible RL-based control system, utilising the Q-learning algorithm, with an adaptive RBF-network providing function approximation. The operation of this prototypical control system is then tested in detail, to determine both the absolute level of performance which can be expected, and the effect which tuning critical parameters of the algorithm has on the functioning of the controller. Performance is compared against a conventional PID controller to maximise the usability of the results by a wide audience. Testing considers behaviour in the presence of disturbances, and run-time changes in plant dynamics. Results show that given sufficient learning opportunity, a RL-based control system performs as well as a simple PID controller. However, unstable behaviour during learning is an issue for future analysis. Additionally, preliminary testing is performed to evaluate the feasibility of implementing RL algorithms in an embedded computing environment, as a general requirement for a MAV flight controller. Whilst the algorithm runs successfully in an embedded context, observation reveals further development would be necessary to reduce computation time to a level where a controller was able to update sufficiently quickly for a real-time motion control application. In summary, the study provides a critical assessment of the feasibility of using RL algorithms for motion control tasks, such as MAV flight control. Advantages which merit interest are exposed, though practical considerations suggest at this stage, that such a control system is not a realistic proposition. There is a discussion of avenues which may uncover possibilities to surmount these challenges. This investigation will prove useful for engineers interested in the opportunities which reinforcement learning techniques represent.

This thesis addresses flight control of the perch landing maneuver for micro-aerial vehicles. A longitudinal flight model is constructed for a pigeon-sized aircraft. In addition to a standard elevator control surface, wing-rotation also considered as a non-standard actuator for increasing low-speed aerodynamic braking. Optimal state and control trajectories for the perch landing maneuver are computed using commercial software. A neighboring optimal control law is then developed and implemented in a set of flight simulations. Simulations are run with both a quasisteady and an unsteady aerodynamic model. The effectiveness of wing rotation and of the neighboring optimal control law is discussed, as is the importance of unsteady aerodynamics during the maneuver. Wing rotation was found to be minimally effective in this case, but it showed potential to be more effective in further research. The unsteady aerodynamic model has significant influence over the success or failure of the maneuver.

M.Phil. / This work proposes a fast local image feature detector and descriptor that is im- plementable on a GPU. The BFROST feature detector is the first published GPU implementation of the popular FAST detector. A simple but novel method of feature orientation estimation which can be calculated in constant time is proposed. The robustness and reliability of the orientation estimation is validated against rotation invariant descriptors such as SIFT and SURF. Furthermore, the BFROST feature descriptor is robust to noise, scalable, rotation invariant, fast to compute in parallel and maintains low memory usage. It is demonstrated that BFROST is usable in real-time applications such as vision-based localization and mapping of images captured from micro aerial platforms.

Submitted by Kamila Costa (kamilavasconceloscosta@gmail.com) on 2015-06-15T20:50:21Z No. of bitstreams: 1 Dissertacao-Adriano B dos S Frutuoso.pdf: 1799620 bytes, checksum: acda5b957dcaee1690413438f4a99e8e (MD5) / Approved for entry into archive by Divisão de Documentação/BC Biblioteca Central (ddbc@ufam.edu.br) on 2015-06-16T14:58:57Z (GMT) No. of bitstreams: 1 Dissertacao-Adriano B dos S Frutuoso.pdf: 1799620 bytes, checksum: acda5b957dcaee1690413438f4a99e8e (MD5) / Approved for entry into archive by Divisão de Documentação/BC Biblioteca Central (ddbc@ufam.edu.br) on 2015-06-16T15:00:11Z (GMT) No. of bitstreams: 1 Dissertacao-Adriano B dos S Frutuoso.pdf: 1799620 bytes, checksum: acda5b957dcaee1690413438f4a99e8e (MD5) / Made available in DSpace on 2015-06-16T15:00:11Z (GMT). No. of bitstreams: 1 Dissertacao-Adriano B dos S Frutuoso.pdf: 1799620 bytes, checksum: acda5b957dcaee1690413438f4a99e8e (MD5) Previous issue date: 2015-02-02 / FAPEAM - Fundação de Amparo à Pesquisa do Estado do Amazonas / This work approaches the angles stability control of a quadrotor. The vehicle parts (mechanical structure and electronic devices), dynamic modeling, controllers design methodology and the experimental results are also presented. Attitude controllers were designed using a PD/H2 discrete control structure, which the gain of H2 part was performed by solving a convex optimization problem, described in linear matrix inequalities form. The experiments indicates that angles dynamic responses can be changed by setting the gain of H2, without modifying the gains of the proportional-derivative part. The results of PD/H2 controllers were compared with a PD controller, in order to evaluate the overshoot and settling time. / Neste trabalho é abordado o controle de estabilidade dos ângulos de atitude de um MAV (Micro Aerial Vehicle) do tipo quadrirrotor. São apresentadas as partes constituintes do veículo (estrutura mecânica e dispositivos eletrônicos), a modelagem dinâmica dos ângulos de atitude, a metodologia de projeto dos controladores de atitude e os resultados experimentais. Em relação ao projeto dos controladores de atitude, estes foram projetados usando estrutura de controle discreta combinada PD/H2, cuja sintonia do controlador H2 por realimentação de estados foi realizada a partir da resolução de um problema de otimização convexo descrito na forma de desigualdades matriciais lineares. Os experimentos mostraram que as respostas dinâmicas dos ângulos de atitude poderiam ser alteradas com o ajuste dos controladores H2, sem modificar os ganhos da parte proporcional-derivativa. Os resultados obtidos para os controladores PD/H2 foram comparados com os controladores PD, de modo a avaliar os seus desempenhos com relação a tempo de acomodação e overshoot.

The use of unmanned aerial vehicles (UAV), so-called drones, has been growingrapidly in the last decade. Today, they are used for, among other things, monitoring missions and inspections of places that are difficult for people to access. Toefficiently and robustly execute these types of missions, a swarm of drones maybe used, i.e., a collection of drones that coordinate together. However, this introduces new requirements on what solutions are used for control and navigation. Two important aspects of autonomous navigation of drone swarms are formationcontrol and collision avoidance. To manage these problems, we propose four different solution algorithms. Two of them use leader-follower control to keep formation, Artificial PotentialField (APF) for path planning and Control Barrier Function (CBF)/ExponentialControl Barrier Function (ECBF) to guarantee that the control signal is safe i.e.the drones keep the desired safety distance. The other two solutions use an optimal control problem formulation of a motion planning problem to either generate open-loop or closed-loop trajectories with a linear quadratic regulator (LQR)controller for trajectory following. The trajectories are optimized in terms of timeand formation keeping. Two different controllers are used in the solutions. Oneof which uses cascade PID control, and the other uses a combination of cascadePID control and LQR control. As a way to test our solutions, a scenario is created that can show the utilityof the presented algorithms. The scenario consists of two drone swarms that willtake on different missions executed in the same environment, where the droneswarms will be on a direct collision course with each other. The implementedsolutions should keep the desired formation while smoothly avoiding collisionsand deadlocks. The tests are conducted on real UAVs, using the open sourceflying development platform Crazyflie 2.1 from Bitcraze AB. The resulting trajectories are evaluated in terms of time, path length, formation error, smoothnessand safety.  The obtained results show that generating trajectories from an optimal control problem is superior compared to using APF+leader-follower+CBF/ECBF. However, one major advantage of the last-mentioned algorithms is that decision making is done at every time step making these solutions more robust to disturbancesand changes in the environment.