Airborne wind energy (AWE) is an emerging field that aims to revolutionize the wind energy sector.This work focuses on the aspects of modeling AWE kite flight behavior. Research on such modelsis important since new or improved models can potentially reduce development time and costs andprovide insight in the role of key system parameters.This research investigates the effects of the mass and drag of the kite control unit (KCU) on the kite’sflight behavior since it is hypothesized that the mass of the KCU causes an outward swing of the KCUduring turns and that KCU drag reduces the angle of attack during the retraction phase.For this investigation, the simulation of a kite and tether system is required. Existing simulation environ-ments could not be used due to legacy code, limited validity or unsuitable model complexity. Therefore,a suitable kite and tether model need to be selected for implementation in a simulation environment(SE). Two kite models have been found suitable: Fechner’s and Ruppert’s model.In the early stages of this research the aim was to use a model that has been fitted to experimental flightdata. Due to the unavailability of experimental flight data, the parameters of both models are identifiedusing simulated flight data. This approach provides Kitepower with a method to identify the modelparameters with experimental flight data in the future. Kitepower is an AWE company that facilitatedon this research.The groundstation controller and flight controller signals of Kitepower have been integrated in the SE.Two experiments have been performed in the SE. For the first experiment, flights are performed withdifferent and increasing values for the drag coefficient of the KCU. In the second experiment the massof the KCU and kite have been varied independently. Both experiments have been performed in a lowand high wind condition.The high wind condition experiment showed that and increasing drag results in a reduction of 7% ofthe average cycle power. The KCU drag force can reach up to to 5% of the kite drag force for a KCUdrag coefficient of 1.5.Increasing the combined mass of the kite and KCU results in an increase of >150% in average cyclepower output for the low wind conditions and an increase of >40% for the high wind condition. Theincrease in power is caused by a reduction in elevation angle in combination with the assumption of aflat wind profile. Additionally an outward swing of the KCU of 4.5 ∘ has been observed.It is recommended to compare Fechner’s and Ruppert’s rigid body kite models through using experi-mental flight data in order to investigate which model reproduces actual kite flight the best. Furthermoreit is recommended to investigate the optimal altitude or elevation angle at which to operate the AWEsystem for maximal power output considering a non-flat wind profile, tether drag and kite and KCUdynamics. It is suggested that these findings should then be incorporated in the flight controller ofKitepower in order to increase the system power output.

Recently, the AlphaGo algorithm has managed to defeat the top level human player in the game of Go. Achieving professional level performance in the game of Go has long been considered as an AI milestone. The challenging properties of high state-space complexity, long reward horizon and high action branching factor in the game of Go are also shared by many other complex planning problems, such as robotics applications. This makes the algorithmic solutions of AlphaGo particularly interesting for further research. One of the key innovations in the algorithm is the combination of Monte Carlo tree search (MCTS) with deep learning. The main hypothesis of the thesis is that the success of the algorithm can be attributed to the combination of the generalization capacity of deep neural networks and the local information of tree search. This hypothesis is evaluated through the application of the AlphaZero algorithm (extension of Alphago) in single-player, deterministic and fully-observable reinforcement learning environments. The thesis presents answers to two research questions. First, what changes need to be made to transfer the AlphaZero algorithm to these environments. The changes in the reward support in these new environments can cause failure of learning, since assumption in the MCTS algorithm are violated. The thesis offers solutions that deal with this problem, including adaptive return normalization. The second research question examines what is the relative importance of the locality and generalization in the performance of the AlphaZero algorithm. This research question is answered by comparing the performance of search trees of varying sizes in several RL environments under fixed time budgets. While building small trees support generalization, through allowing more frequent training of the neural network, building larger trees provide more accurate estimates. This creates a trade-off between improved generalization capacity and more accurate local information under a fixed time budget. The experiment results show that mid-size trees achieve the best performance, which suggests that balancing local information and generalization is key to the success of the algorithm. Based on this results, possible extensions to the algorithm are proposed. At last, the thesis also highlights the relevance of the two-component system from a broader perspective and discusses the possible future impact of the algorithm.

This thesis provides an exposition of the theory of Active Inference in a control theoretic context. Active Inference is a remarkably powerful neuroscientific theory that unifies many characteristics of the biological brain. As such, Active Inference provides a valid inspiration in search of improvements in bio-inspired robot control algorithms. The literature on Active Inference however is narrow and complex. The goal of this thesis is to open the door research of Active Inference
in robotics, by applying the theory to linear state space systems and exposing the relations and differences with established engineering paradigms. We provide a detailed account of several critical details, mainly the concept of generalized motions, that are commonly not understood from the scientific literature. A start in the performance analysis of the algorithm is made, by studying the effect of changes in several tuning parameters. Additionally, with Active Inference reformulated as a state space control system, it is shown that standard behavior
such as stabilization and tracking can be achieved.

In this MSc. thesis, state estimation for a nanosatellite-class reaction wheel has been investigated. Two LPV observers, using different sensor information, have been synthesized. The performance of the designed observers has been evaluated using a detailed simulation of the reaction wheel (including disturbances that occur during the reaction wheel operation).

This study focuses on automated malaria diagnosis in low quality blood smear images, captured by a low-cost smartphone based microscope system. The aim is to localize and classify the healthy and infected erythrocytes (red blood cells) in order to evaluate the parasitaemia in an infected blood smear. Due to the lower quality of the smartphone microscope system compared to traditional high-end light microscopes, conventional algorithms fail to process these images. We propose a framework using a convolutional neural network as a pixel classifier to localize the erythrocytes. Afterwards we classify them accordingly, using a convolutional neural network as an object classifier. Such a system can offer in the-field malaria diagnosis without human intervention or can act as an aid for human experts to lower workload and increase diagnosis accuracy. The algorithm successfully localizes the erythrocytes with an average sensitivity of 97.31% and precision of 92.21%. Classification performed inadequate, in terms of low agreement with two human experts. This can be due to the low image quality or the small amount of training data available at the time.