Creating models about others is a sophisticated human ability that robotic companions need to develop in order to have successful interactions. This thesis proposes user modelling frameworks to personalise the interaction between a robot and its user and devises novel scenarios where robotic companions may apply these user modelling techniques. We tackle the creation of user models in a hierarchical manner, using a streamlined version of the Hierarchical Attentive Multiple-Models for Execution and Recognition (HAMMER) architecture to detect low-level user actions and taking advantage of Stochastic Context-Free Grammars (SCFGs) to instantiate higher-level models which recognise uncertain and recursive sequences of low-level actions. We discuss a couple of distinct scenarios for robotic companions: a humanoid sidekick for power-wheelchair users and a companion of hospital patients. Next, we address the limitations of the previous scenarios by applying our user modelling techniques and designing two further scenarios that fully take advantage of the user model. These scenarios are: a wheelchair driving tutor which models the user abilities, and the musical collaborator which learns the preferences of its users. The methodology produced interesting results in all scenarios: users preferred the actual robot over a simulator as a wheelchair sidekick. Hospital patients rated positively their interactions with the companion independently of their age. Moreover, most users agreed that the music collaborator had become a better accompanist with our framework. Finally, we observed that users' driving performance improved when the robotic tutor instructed them to repeat a task. As our workforce ages and the care requirements in our society grow, robots will need to play a role in helping us lead better lives. This thesis shows that, through the use of SCFGs, adaptive user models may be generated which then can be used by robots to assist their users.

An unprecedented amount of renewable generation is to be connected to the UK grid in the coming decades, giving rise to new power flow patterns and warranting unprecedented amounts of transmission investment. However, significant uncertainty surrounds the state of the electricity system, primarily in terms of the size, location and type of new generators to be connected. These sources of uncertainty render the system planner unable to make fully informed decisions about future transmission investment. This thesis presents a stochastic formulation for the transmission expansion planning problem under uncertainty in future generation developments. The problem has been modelled as a multi-stage stochastic optimization problem where the expected system cost is to be minimized. Uncertainty is captured in the form of a multi-stage scenario tree that portrays a range of possible future system states and transition probabilities. A set of investment options with different upgradeability levels and construction times have been included in the formulation to capture the diverse choices present in a realistic setting, where the planner can choose to invest in an anticipatory manner. A novel multi-cut Benders decomposition scheme is used to render the model tractable for large systems with multiple scenarios and operating points. The developed tool can identify the optimal long-term investment strategy based on the triptych of economic efficiency, adequate security provision and acceptable risk. Simulation results on test systems validate that the stochastic approach can lead to further expected cost minimization when compared to methods that ignore the planner’s decision flexibility. Moreover, decisions are taken with subsequent adaptability in mind. The benefit of keeping future expansion options open is properly valued; investment paths that enable future delivery at lower costs are favoured while premature project commitment is avoided.

Trading communication with redundant computation can increase the silicon efficiency of common hardware accelerators like FPGA and GPU in accelerating sparse iterative numerical algorithms. While iterative numerical algorithms are extensively used in solving large-scale sparse linear system of equations and eigenvalue problems, they are challenging to accelerate as they spend most of their time in communication-bound operations, like sparse matrix-vector multiply (SpMV) and vector-vector operations. Communication is used in a general sense to mean moving the matrix and the vectors within the custom memory hierarchy of the FPGA and between processors in the GPU; the cost of which is much higher than performing the actual computation due to technological reasons. Additionally, the dependency between the operations hinders overlapping computation with communication. As a result, although GPU and FPGA are offering large peak floating-point performance, their sustained performance is nonetheless very low due to high communication costs leading to poor silicon efficiency. In this thesis, we provide a systematic study to minimize the communication cost thereby increase the silicon efficiency. For small-to-medium datasets, we exploit large on-chip memory of the FPGA to load the matrix only once and then use explicit blocking to perform all iterations at the communication cost of a single iteration. For large sparse datasets, it is now a well-known idea to unroll k iterations using a matrix powers kernel which replaces SpMV and two additional kernels, TSQR and BGS, which replace vector-vector operations. While this approach can provide a Θ(k) reduction in the communication cost, the extent of the unrolling depends on the growth in redundant computation, the underlying architecture and the memory model. In this work, we show how to select the unroll factor k in an architecture-agnostic manner to provide communication-computation tradeoff on FPGA and GPU. To this end, we exploit inverse-memory hierarchy of the GPUs to map matrix power kernel and present a new algorithm for the FPGAs which matches with their strength to reduce redundant computation to allow large k and hence higher speedups. We provide predictive models of the matrix powers kernel to understand the communication-computation tradeoff on GPU and FPGA. We highlight extremely low efficiency of the GPU in TSQR due to off-chip sharing of data across different building blocks and show how we can use on-chip memory of the FPGA to eliminate this off-chip access and hence achieve better efficiency. Finally, we demonstrate how to compose all the kernels by using a unified architecture and exploit on-chip memory of the FPGA to share data across these kernels. Using the Lanczos Iteration as a case study to solve symmetric extremal eigenvalue problem, we show that the efficiency of FPGAs can be increased from 1.8% to 38% for small- to-medium scale dense matrices whereas up to 7.8% for large-scale structured banded matrices. We show that although GPU shows better efficiency for certain kernels like the matrix powers kernel, the overall efficiency is even lower due to increase in communication cost while sharing data across different kernels through off-chip memory. As the Lanczos Iteration is at the heart of all modern iterative numerical algorithms, our results are applicable to a broad class of iterative numerical algorithms.

Signals in communication networks exhibit significant correlation, which can stem from the physical nature of the underlying sources, or can be created within the system. Current layered network architectures, in which, based on Shannon's separation theorem, data is compressed and transmitted over independent bit-pipes, are in general not able to exploit such correlation efficiently. Moreover, this strictly layered architecture was developed for wired networks and ignore the broadcast and highly dynamic nature of the wireless medium, creating a bottleneck in the wireless network design. Technologies that exploit correlated information and go beyond the layered network architecture can become a key feature of future wireless networks, as information theory promises significant gains. In this thesis, we study from an information theoretic perspective, three distinct, yet fundamental, problems involving the availability of correlated information in wireless networks and develop novel communication techniques to exploit it efficiently. We first look at two joint source-channel coding problems involving the lossy transmission of Gaussian sources in a multi-terminal and a time-varying setting in which correlated side information is present in the network. In these two problems, the optimality of Shannon's separation breaks down and separate source and channel coding is shown to perform poorly compared to the proposed joint source-channel coding designs, which are shown to achieve the optimal performance in some setups. Then, we characterize the capacity of a class of orthogonal relay channels in the presence of channel side information at the destination, and show that joint decoding and compression of the received signal at the relay is required to optimally exploit the available side information. Our results in these three different scenarios emphasize the benefits of exploiting correlated side information at the destination when designing a communication system, even though the nature of the side information and the performance measure in the three scenarios are quite different.

Analyzing and interpreting of thermograms has been increasingly employed in the diagnosis and monitoring of diseases thanks to its non-invasive, non-harmful nature and low cost. This thesis explores methods for thermal image analysis systems based on image registration for two medical applications; skin disease diagnosis and cooling study. In the first application, a novel system is proposed to improve the diagnosis and monitoring of morphoea based on a face sketch and the published lines of Blaschko. In the latter, a novel semi-automated system is proposed to investigate a cooling mask for maxillofacial surgery patients based on thermogram data of normal subjects. In both applications, image registration based on global and local registration methods are found inevitable. A modified normalized gradient cross-correlation (NGC) method to reduce large geometrical differences between two multimodal images of different subjects that are represented by smooth gray edge maps is proposed for the global registration approach. To correct for small displacements between the global outcomes, a simple stochastic based non-rigid affine registration (NRAR) method is proposed. The NRAR method is driven by a cost function that takes into consideration the similarity between two images by a correlation coefficient. A geometric based intensity distortion to ensure only small distortions are accepted, and an overlapping pixel rate are also incorporated. Smooth deformation controlled by an exponential Euclidean based smoothing operator is employed to only edge pixels navigated by a distance function as both the images are represented by edge maps and thus reduces computation time. The NRAR method has shown good performance in correcting for small, varying displacements between images with fairly reliable flexibility and elasticity for both convex and concave objects with the help of the NGC to minimize the initial displacements. A semi-automated approach that includes the NGC and/or the NRAR method followed by determination of cooling area based on the Otsu's thresholding and a seed-based region growing method is found to achieve reliable and reproducible cooling patterns with good correlation with the clinical assessment, for potential cooling study of maxillofacial surgery patients.

In this thesis, we investigate the problem of identifying individual link metrics in a communication network from accumulated end-to-end metrics over selected measurement paths, under the assumption that link metrics are additive and constant during the measurement, and measurement paths cannot contain cycles. We know from linear algebra that all link metrics can be uniquely identified when the number of linearly independent measurement paths equals n, the number of links. It is, however, inefficient to collect measurements from all possible paths, whose number can grow exponentially in n, as the number of useful measurements (from linearly independent paths) is at most n. The aim of this thesis is thus to characterize network identifiability by easily verifiable conditions and develop efficient algorithms for achieving and maximizing network identifiability. To characterize network identifiability in terms of the network topology and the number/placement of monitors, our main results are: (i) it is generally impossible to identify all the link metrics by using two monitors; (ii) nevertheless, metrics of all the interior links not incident to any monitor are identifiable by two monitors if the topology satisfies a set of necessary and sufficient connectivity conditions; (iii) these conditions naturally extend to a necessary and sufficient condition for identifying all the link metrics using three or more monitors. We show that these conditions not only allow efficient identifiability tests, but also enable efficient algorithm design for constructing linearly independent paths and computing individual link metrics. Specifically, we show that whenever there exists a set of n linearly independent measurement paths, there must exist a set of three pairwise independent spanning trees. We exploit this property to develop an algorithm that can construct n linearly independent, cycle-free paths between monitors without examining all candidate paths, whose complexity is quadratic in n. A further benefit of the proposed algorithm is that the generated paths satisfy a nested structure that allows linear-time computation of link metrics without explicitly inverting the measurement matrix. Next, we study a complementary problem of how to characterize network partial identifiability when n linearly independent paths cannot be found in a given network, for which we establish an efficient algorithm to determine all identifiable links in an arbitrary network under a given monitor placement. Finally, we investigate a realistic problem of how to place monitors such that the network uncertainty with respect to internal link metrics is minimized. To this end, we first develop efficient algorithm to place the minimum number of monitors in order to identify all link metrics. Our evaluations on both random and real topologies show that the proposed minimum monitor placement algorithm achieves identifiability using a much smaller number of monitors than a baseline solution. However, we observe that the complete identification of all link metrics in a network can require a large number of monitors (e.g., 60% of all nodes), even if assuming optimal placement of monitors. Therefore, we then study the problem of placing a given number of monitors (k monitors) to identify the maximum number of link metrics. For this problem, we build a polynomial-time algorithm to incrementally place monitors such that each newly placed monitor maximizes the number of additional identifiable links. The significance of this k-monitor placement algorithm is that it is provable optimal if the network is 2-vertex-connected. For other types of networks, our evaluation on various ISP topologies shows that the proposed k-monitor placement algorithm allows identification for close to the maximum number of links while incurring a much lower complexity than brute-force approaches.

Energy-based modeling and control of dynamical systems is crucial since energy is a fundamental concept in Science and Engineering theory and practice. While Interconnection and Damping Assignment Passivity-based Control (IDA-PBC) is a powerful theoretical tool to control port-controlled Hamiltonian (PCH) systems that arise from energy balancing principles, sensorless operation of energy harvesters is a promising practical solution for low-power energy generation. The thesis addresses these two problems of energy-based control and efficient energy generation. The design via IDA-PBC hinges on the solution of the so-called matching equation which is the stumbling block in making this method widely applicable. In the first part of the thesis, a constructive approach for IDA-PBC for PCH systems that circumvents the solution of the matching equation is presented. A new notion of solution for the matching equation, called algebraic solution, is introduced. This notion is instrumental for the construction of an energy function defined on an extended state-space. This yields, differently from the classical solution, a dynamic state-feedback that stabilizes a desired equilibrium point. In addition, conditions that preserve the PCH structure in the extended closed-loop system have been provided. The theory is validated on four examples: a two-dimensional nonlinear system, a magnetic levitated ball, an electrostatic microactuator and a third order food-chain system. For these systems damping structures that cannot be imposed with the standard approach are assigned. In the second part of the thesis, the design of a nonlinear observer and of an energy-based controller for sensorless operation of a rotational energy harvester is presented. A mathematical model of the harvester with its power electronic interface is developed. This model is used to design an observer that estimates the mechanical quantities from the measured electrical quantities. The gains of the observer depend on the solution of a modified Riccati equation. The estimated mechanical quantities are used in a feedback control law that sustains energy generation across a range of source rotation speeds. The proposed observer-controller scheme is assessed through simulations and experiments.

Design and control of the bipedal humanoid robot locomotion are challenging areas of research. Accurate models of the kinematics and dynamics of the robot are essential to achieve bipedal locomotion. Bipedal walking can be achieved either with flat-foot or toe-foot walking. Flat-foot walking is more stable but slower, whereas toe-foot walking produces more natural and faster motion. Furthermore in toe-foot walking it is possible to perform stretch knee walking. The mechanical structure of the robot is designed with compact modular parts so that the robot kinematics can be modelled as a multi-point-mass system, and its dynamics are modelled applying the inverted pendulum model and the zero-moment-point concept. The optimality in the gait trajectory is achieved exploiting augmented model predictive control methods taking into consideration the trade-off between walking speed and stability. The robustness and stability of the walking gaits and posture in the presence of internal or external disturbances are enhanced by adopting angular compensation with joint control techniques. The thesis develops a flat-foot optimal walking gait generation method. The effectiveness of the control technique and the passive-toe design is validated by simulation tests with the robot walking on slope, stepping over an obstacle and climbing a stair. The walking gaits are implemented on a mid-size (1.6 meter, 58 kg) bipedal robot. Experiments demonstrate the effectiveness of the new proposed Augmented Model Predictive Control (AMPC) method has improved and produced a smoother gaits tracking trajectory in comparison with existing LQR and preview methods; and at the same time the proposed algorithms are able to reduce noise interference.

High Voltage Direct Current (HVDC) transmission has to date mostly been used for point-to-point projects, with only a few select projects being designed from the outset to incorporate multiple terminals. Any future HVDC network is therefore likely to evolve out of this pool of HVDC connections. As technology improves, the voltage rating, at the point of commission, of the these connections increases. Interconnection therefore requires the DC equivalent of the transformer, to bridge the voltage levels and create a multi-terminal network. This thesis investigates new potential DC/DC converter topologies, which may be used for a range of HVDC applications. Simple interconnections of new and legacy HVDC links is unlikely to require a large voltage-step, but will be required to transfer a large amount of power. As the HVDC network develops it may become feasible for wind-farms and load-centres to directly connect to the DC network, rather than requiring new and dedicated links. Such a connection is called an HVDC tap and is typically rated at only a small fraction of the link's peak capacity (around 10\%). Such taps would connect a distribution voltage level to the HVDC network. DC/DC converters suitable for large-step ratios (>5:1) may find their application here. In this work DC/DC converters for both small and large step-ratios are investigated. Two approaches are taken to design such converters: first, an approach utilising existing converter topologies is investigated. As each project comes with a huge price-tag, their reliability is paramount. Naturally, technology that has already proven itself in the field can be modified more readily and quickly for deployment. Using two modular multilevel converters in a front-to-front arrangement has been found to work efficiently for large power transfers and low step-ratios. Such a system can be operated at higher than 50 Hz frequencies to reduce the volume of a number of passive components, making the set-up suitable for compact off-shore applications. This does however incur a significant penalty in losses reducing the overall converter efficiency. In the second approach DC/DC converter designs are presented, that are more experimental and would require significantly more development work before deployment. Such designs do not look to adapt existing converter topologies but rather are designed from scratch, purely for DC/DC applications. An evolution of the front-to-front arrangement is investigated in further detail. This circuit utilises medium frequency (>50 Hz) square current and voltage waveforms. The DC/DC step-ratio is achieved through a combination of the stacks of cells and a transformer. This split approach allows for high-step ratios to be achieved at similar system efficiencies as for the front-to-front arrangement. The topology has been found to be much more suitable for higher than 50 Hz operation from a losses perspective, allowing for a compact and efficient design.

In this thesis, control of future AC/DC systems for improved system dynamic performance is studied. The objective is to determine mechanisms for providing AC network services (e.g. frequency support, damping, etc.) through coordinated control of HVDC power converters and FACTS devices while considering increased levels of wind generation. In particular, this work addresses some of the concerns associated with the stability of the future Great Britain (GB) transmission network as it evolves to support low carbon generation scenarios and the use of DC grids to integrate offshore renewable resources and form a sub-sea interconnection across Europe. The contributions of this thesis are in two main areas: emergency control for power system stabilization and exchange of frequency support across a DC grid. Fast control of FACTS devices and HVDC links can be exercised as a post-fault corrective action to maintain system stability without the need of constraining pre-fault transfer levels. This work employs a model predictive control (MPC) scheme that relies on system wide-area measurements to preserve the system stability after critical contingencies. MPC can explicitly account for system constraints and changing operating conditions and is therefore suited for on-line applications and power electronic actuators with limited short-term overload capability. The effectiveness of the proposed approach is demonstrated using time domain simulations on representative equivalent models of the future GB transmission network. A detailed analysis of the dynamic behaviour and stability issues associated with the GB transmission grid have also been presented. In the DC grid context, this thesis investigates the provision of frequency services considering frequency droop loops in the control of the converters. The interaction between onshore AC systems and a DC grid is analyzed through an extended steady-state formulation. A methodology for providing frequency response from offshore wind farms connected through a DC grid is also proposed. The performance of this scheme is illustrated both analytically and also through simulation results.