abstract: Toward the ambitious long-term goal of a fleet of cooperating Flexible Autonomous Machines operating in an uncertain Environment (FAME), this thesis addresses various perception and control problems in autonomous aerial robotics. The objective of this thesis is to motivate the use of perspective cues in single images for the planning and control of quadrotors in indoor environments. In addition to providing empirical evidence for the abundance of such cues in indoor environments, the usefulness of these perspective cues is demonstrated by designing a control algorithm for navigating a quadrotor in indoor corridors. An Extended Kalman Filter (EKF), implemented on top of the vision algorithm, serves to improve the robustness of the algorithm to changing illumination. In this thesis, vanishing points are the perspective cues used to control and navigate a quadrotor in an indoor corridor. Indoor corridors are an abundant source of parallel lines. As a consequence of perspective projection, parallel lines in the real world, that are not parallel to the plane of the camera, intersect at a point in the image. This point is called the vanishing point of the image. The vanishing point is sensitive to the lateral motion of the camera and hence the quadrotor. By tracking the position of the vanishing point in every image frame, the quadrotor can navigate along the center of the corridor. Experiments are conducted using the Augmented Reality (AR) Drone 2.0. The drone is equipped with the following componenets: (1) 720p forward facing camera for vanishing point detection, (2) 240p downward facing camera, (3) Inertial Measurement Unit (IMU) for attitude control , (4) Ultrasonic sensor for estimating altitude, (5) On-board 1 GHz Processor for processing low level commands. The reliability of the vision algorithm is presented by flying the drone in indoor corridors. / Dissertation/Thesis / Masters Thesis Electrical Engineering 2018

Nesta dissertação é apresentada a comparação do desempenho de três estimadores - o Filtro de Kalman Estendido, o Filtro de Kalman Unscented e o Filtro de Partículas - aplicados para estimar a postura de um robô diferencial. Uma câmera foi fixa no teto para cobrir todo o campo operacional do robô durante os experimentos, a fim de extrair o mapa e gerar o ground truth. Isso permitiu realizar uma análise do erro de forma precisa a cada instante de tempo. O desempenho de cada um dos estimadores foi avaliado sistematicamente e numericamente para duas trajetórias. Os resultados desse primeiro experimento demonstram que os filtros proporcionam grandes melhorias em relação à odometria e que o modelo dos sensores é crítico para obter esse desempenho. O Filtro de Partículas mostrou um desempenho melhor em relação aos demais nos dois percursos. No entanto, seu elevado custo computacional dificulta sua implementação em uma aplicação de tempo real. O Filtro de Kalman Unscented, por sua vez, mostrou um desempenho semelhante ao Filtro de Kalman Estendido durante a primeira trajetória. Porém, na segunda trajetória, a qual possui uma quantidade maior de curvas, o Filtro de Kalman Unscented mostrou uma melhora significativa em relação ao Filtro de Kalman Estendido. Foi realizado um segundo experimento, em que o robô planeja e executa duas trajetórias. Os resultados obtidos mostraram que o robô consegue chegar a um determinado local com uma precisão da mesma ordem de grandeza do que a obtida durante a estimação de estados do robô. / In this dissertation, the performance of three nonlinear-model based estimators - the Extended Kalman Filter, the Unscented Kalman Filter and the Particle Filter - applied to pose estimation of a differential drive robot is compared. A camera was placed above the operating field of the robot to record the experiments in order to extract the map and generate the ground truth so the evaluation of the error can be done at each time step with high accuracy. The performance of each estimator is assessed systematically and numerically for two robot trajectories. The first experimental results showed that all estimators provide large improvements with respect to odometry and that the sensor modeling is critical for their performance. The particle filter showed a better performance than the others on both experiments, however, its high computational cost makes it difficult to implement in a real-time application. The Unscented Kalman Filter showed a similar performance to the Extended Kalman Filter during the first trajectory. However, during the second one (a curvier path) the Unscented Kalman Filter showed a significant improvement over the Extended Kalman Filter. A second experiment was carried out where the robot plans and executes a trajectory. The results showed the robot can reach a predefined location with an accuracy of the same order of magnitude as the obtained during the robot pose estimation.

O presente trabalho trata da criação de uma nova abordagem para predição de séries temporais ruidosas, com modelo desconhecido e que apresentam grandes não-linearidades. O novo método neuro-estatístico proposto combina uma rede neural de múltiplas camadas com o método estatístico Filtro de Kalman Estendido. A justificativa para a junção dessas abordagens é o fato de possuírem características complementares para o tratamento das peculiaridades das séries descritas. Quanto ao ruído, o FKE consegue minimizar a sua influência, trabalhando com a variância do ruído extraído dos dados reais. Quanto ao modelo gerador da série, as redes neurais aproximam a sua função, aprendendo a partir de amostras dos próprios dados. Grandes não-linearidades também são tratadas pelas RNs. O método neuro-estatístico segue a estrutura do FKE, utilizando a RN como processo preditivo. Com isso, elimina-se a necessidade de conhecimento prévio da função de transição de estados. O poder de tratamento de não-linearidades da RN é mantido, utilizando-se a previsão desta como estimativa de estado e os seus valores internos para cálculo das jacobianas do FKE. As matrizes de covariâncias dos erros de estimativa e dos ruídos são utilizadas para melhora do resultado obtido pela RN. A rede é treinada com um conjunto de dados retirado do histórico da série, de maneira off-line, possibilitando o uso de poderosas estruturas de redes de múltiplas camadas. Os resultados do método neuro-estatístico são comparados com a mesma configuração de RN utilizada em sua composição, sendo ambos aplicados na série caótica de Mackey-Glass e em uma série combinada de senos. Ambas séries possuem grandes não-linearidades e são acrescidas de ruído. O novo método alcança resultados satisfatórios, melhorando o resultado da RN em todos os experimentos. Também são dadas contribuições no ajuste dos parâmetros do FKE, utilizados no novo método. O método híbrido proporciona uma melhora mútua entre a RN e o FKE, explicando os bons resultados obtidos. / This work presents a new forecast method over highly nonlinear noisy time series. The neural statistical method uses a multi-layer perceptron (NN) and the Extended Kalman Filter (EKF). The justification for the combination of these approaches is that they possess complementary characteristics for the treatment of the peculiarities of the series. The EKF minimizes the influence of noise, working with the variance of the noise obtained from the real data. The NN approximates the generating model’s function. High nonlinearities are also treated by the neural network. The neural statistical method follows the structure of the EKF, using the NN as the predictive process. Thus, it isn’t necessary previous knowledge of the state transition function. The power of treatment of nonlinearities of the NN is kept, using forecast of this as estimative of state and its internal values for calculation of the Jacobian matrix of the EKF. The error estimative covariance and the noise covariance matrixes are used to improve the NN outcome. The NN is trained offline by past observations of the series, which enable the use of powerfuls neural networks. The results of the neural statistical method are compared with the same configuration of NN used in its composition, being applied in the chaotic series of Mackey-Glass and an sine mistures series. Both series are noisy and highly nonlinear. The new method obtained satisfactory result, improving the result of the regular NN in all experiments. The method also contributes in the adjustment of the parameters of the EKF. The hybrid method has a mutual improvement between the NN and the EKF, which explains the obtained good results.

A retroalimentação de controladores preditivos que utilizam modelos em espaço de estado pode ser realizada de duas formas: (a) correção por bias, em que as saídas preditas são corrigidas adicionando-se um valor proporcional a discrepância encontrada entre o valor medido atual e sua respectiva predição e por (b) retroalimentação dos estados, onde se determinam as condições iniciais através da estimação dos estados, e a partir de uma melhor condição inicial se realizam as predições futuras usadas no cálculo das ações de controle. Nesta dissertação estas duas abordagens são comparadas utilizando a Planta Laboratorial de Seis Tanques Esféricos. As técnicas de Filtro de Kalman Estendido (EKF) e Filtro de Kalman Estendido com Restrições (CEKF) foram empregadas para estimar os estados não medidos. Inicialmente foram feitos testes off-line destes algoritmos de estimação. Para estes testes são utilizados uma série de dados da planta laboratorial do estudo de caso, na qual são estudadas as influências de diversos fatores de ajuste que determinam a qualidade final de estimação. Estes ajustes serviram de base para a aplicação destes algoritmos em tempo real, quando então, estimadores de estados estão associados ao sistema de controle do processo baseado em um algoritmo de controle preditivo. Após se ter certificado a qualidade das estimações de estado, partiu-se para sua utilização como uma alternativa de retroalimentação de controladores preditivos. Estes resultados foram comparados com os obtidos através da correção simples por bias. Os resultados experimentais apontam para uma marginal piora devido à retroalimentação por estimadores de estados frente à correção por bias, pelo menos para o caso do controlador preditivo linear utilizado na comparação. Entretanto, espera-se que resultados melhores sejam obtidos no caso de modelos preditivos não-lineares, uma vez que nestes casos o modelo é bem mais sensível à qualidade da condição inicial. / The feedback of controllers that use predictive models in state space can be accomplished in two ways: (a) bias correction, where the predicted outputs are corrected by adding a value proportional to the discrepancy found between the current measurement and its respective prediction; and by (b) state feedback, which establishes the initial conditions through the states estimation, and from a better initial condition are carried out the future predictions used in the calculation of control. In this thesis these two approaches are compared using a Laboratorial Plant of Six Spherical Tanks. The techniques of Extended Kalman Filter (EKF) and Constraint Extended Kalman Filter (CEKF) were used to estimate the unmeasured states. Initially, tests were carried out off-line for theses estimation algorithms. For such testing are used a dataset of the plant in case study, in which are studied the influences of several adjustment factors that they determine the final quality of estimation. These adjustments were used of base for the application of these algorithms in real time, when then state estimators are associated with the system of process control based on a predictive control algorithm. After having ascertained the quality of the state estimates, begins its use as an alternative for feedback of predictive controllers. These results were compared with those obtained by the simple correction of bias. The experimental results show a marginal worsening due to feedback from state estimated compared with bias correction, at least for the case of linear predictive controller used in the comparison. However, one expects that better results will be obtained in the case of non-linear predictive models, since in these cases the model is much more sensitive to the quality of the initial condition.

This thesis studies algorithms to estimate an aircraft’s position with different information from various radio stations. Because aircrafts both civilian and military are heavily dependant on GNSS signals, it can be interfered from hostile sources. The aircraft shall then be able to navigate without the GNSS signals. This thesis focuses on three radio navigation systems, DME,VOR and TACAN. With the measurements from these three radio stations and measurements from the inertial navigation system one can estimate a position with an estimation filter. In this thesis two types of filters will be used, the linear Kalman filter and the Extended Kalman filter. The linear Kalman filter will be used when converting the TACAN measurements to a pseudo position and the Extended Kalman filter will be used for the DME,VOR and TACAN measurements. The results shows that the converted TACAN measurements and TACAN measurements estimates very well in both north and east direction. When using only DME measurements the filter estimates the position fairly well in the direction towards the station and poorly in the orthogonal direction. For the VOR measurements the filter estimates the position quite poorly in the direction of the radio station and well in the orthogonal direction. In conclusion the converted TACAN measurement and TACAN measurement algorithm can be used for navigation purposes by its own measurements. However, the DME and VOR measurement algorithms need to be combined or using multiple stations at different locations to get better estimates in both directions. All of the filter could use some better tuning to get the optimal filter, but it is not necessary.

A permanent magnet synchronous motor is a type of motor that is used in several different application areas, not least in an autonomous robots where it is the motor that drives the wheels. Today, many actors choose simulation as a tool to save money and time when product tests are performed. This thesis covers both the process of modeling a permanent magnet synchronous motor and regulating it at low speeds, in a simulation environment. As previously mentioned, the motor is a permanent magnet synchronous motor and is a direct-driven outrunner, which means that the motor and the wheel are combined and that the rotor is spinning outside the stator. On current robots in production, there is a gear ratio between the motor and wheels to be able to regulate the motor at higher speeds and thus generate a torque. The gearing contributes to losses and is an extra cost, so the examination of a direct-drive motor is interesting. The direct-drive motor has a lower working speed and is therefore by some reasons more difficult to regulate when applying torque load to the motor. The motor is equipped with current sensors and a position sensor, which has a certain resolution. The position sensor is speed-dependent in the sense that at lower RPMs fewer measurements are obtained, which is a problem when regulating the motor. The thesis examines two different control strategies, one of which is a more classic PI control that is often used on the market in various systems and the other is model predictive control (MPC). The latter is an online optimization where, with the help of information about the system, an optimal input signal is calculated and applied. Two different non-linear Kalman filters are also examined, which are implemented with the two different control strategies, to estimate the speed with the help of the measurements from current and the position sensor. The conclusion is an ideal motor model that mimics the physical motor. MPC is able to regulate the motor between 0-50 RPM, both with and without applied torque and even better with speed estimation from a Kalman filter. The PI controller is not able to regulate the motor at 2 RPM but for speeds at 10 RPM and greater, however with over-/undershoot after an acceleration.

No description available.

When operating an aircraft, inaccurate measurements can have devastating consequences. For example, when measuring airspeed using a pitot tube, icing effects and other faults can result in erroneous measurements. Therefore, this master thesis aims to create an alternative method which utilizes known flight mechanical equations and sensor fusion to create an estimate of the airspeed during flight. For validation and generation of flight data, a simulation model developed by SAAB AB, called ARES, is used.  Two models are used to describe the aircraft behavior. One of which is called the dynamic model and utilizes forces acting upon the aircraft body in the equations of motion. The other model, called the kinematic model, instead describes the motion with accelerations of the aircraft body. The measurements used are the angle of attack (AoA), side-slip angle (SSA), GPS velocities, and angular rates from an inertial measurement unit (IMU). The dynamic model assumes that engine thrust and aerodynamic coefficients are already estimated to calculate resulting forces, meanwhile the kinematic model instead uses body fixed accelerations from the IMU. These models are combined with filters to create estimations of the airspeed. The filters used are the extended Kalman filter (EKF) and unscented Kalman filter (UKF). These are combined with the two models to create in total four methods to estimate the airspeed.  The results show no major difference in the performance between the filters except for computational time, for which the EKF has the fastest. Further, the result show similar airspeed estimation performance between the models, but differences can be seen. The kinematic model manages to estimate the wind with higher details and to converge faster, compared to the dynamic model. Both models suffer from an observability problem. This problem entails that the aircraft needs to be maneuvered to excite the AoA and SSA in order for the estimation methods to evaluate the wind, which is crucial for accurate airspeed estimation. The robustness of the dynamic model regarding errors in engine thrust and aerodynamic coefficients are also researched, which shows that the model is quite robust against errors in these values.

The project explores the possibilities on how to improve the accuracy of GPS velocity data by using sensor fusion with an extended Kalman filter. The proposed solution in this project is a sensor fusion between the GPS and IMU of the system, where the extended Kalman filter was used to estimate the velocity from the sensor data. The hardware used for the data acquisition to the proposed solution was a Pixhawk 4 (PX4), which has an IMU consisting of accelerometers, gyroscopes and magnetometers. The PX4:s corresponding GPS module was also used to collect accurate velocity data. The data was logged using Simulink and later processed with MATLAB. The sensor fusion using the extended Kalman filter gave good estimates upon constant acceleration but had problems with estimating over varying acceleration. This was initially planned to be solved using smoothing filters, which is an essential part of the fusion process, but was never implemented due to time constraints. The constructed filter acts as a foundation towards future improvement. Other methods such as unscented Kalman filter, particle filter and neural network could also be explored to improve the estimation of the velocity due to these filters being known to have better performance. However, most of these alternatives need more computing power and are generally harder to implement compared to the extended Kalman filter. This project would be beneficial to QTAGG, since increasing the velocity resolution and accuracy of the system can provide possibilities of better optimization. It is also a commonly implemented solution where there are many state of the art implementations available.

The fast-growing electric vehicles (EVs) market demands huge efforts from car manufacturers to develop and improve their current products’ systems. A fast charge of the battery pack is one of the challenges encountered due to the battery limitations regarding behaviour and additional degradation when exposed to such a rough situation. In addition, the outcome of a study performed on a battery does not apply to others, especially if their chemistries are different. Hence, extensive testing is required to understand the influence of design decisions on the particular energy storage device to be implemented. Due to batteries’ nonlinear behaviour that is highly dependent on external variables such as temperature, the dynamic load and aging, another defying task is the widely studied state of charge (SOC) estimation, commonly considered one of the most significant functions in a battery management system (BMS). This thesis presents an extensive battery fast charging aging test study equipped with promising current charging profiles from published literature to minimize aging. Four charging protocols are carefully designed to charge the cell from 10 to 80% SOC within fifteen minutes and have their performances discussed. A dual state estimation algorithm is modelled to estimate the SOC with the assistance of a capacity state of health (SOHcap) estimation. Finally, the dual state estimation model is validated with the fast charging aging test data. / Thesis / Master of Science in Mechanical Engineering (MSME)