Automated planning addresses the problem of generating a sequence of actions to satisfy given goal conditions for a constructed model of the world. In recent planning approaches heuristic guidance is used to lead the search towards the goal. The focus of this work is on domains where plan quality is assessed with plan metrics. A discussion of the impact of a popular relaxed planning graph heuristic on the quality of plans in such domains is presented. The relaxed planning graph heuristic bias towards shorter plans, irrespective of quality, is described. A novel approach to constructing the relaxed planning graph based on metric cost is presented to overcome this bias and to generate good quality plans. A notion of metric sensitivity as the ability of a planner to respond to the change of the plan metric, is introduced and methods to determine metric sensitivity are presented. Current state-of-the-art planners are evaluated in terms of their metric sensitivity. This research also tackles the problem of planning in multiobjective domains, where quality of a plan is evaluated using multiple plan metrics. For multiobjective domains the solution is no longer a single plan but a set of plans. A set of non dominated solutions is called a pareto frontier. This thesis contains a discussion on the desired properties of such sets of plans and methods of generating them. Metric sensitivity is a required property for a planner to effectively reason with user defined metrics and generate desired set of plans. The main significant contributions of the work described in the thesis are: 1. A definition and exploration of metric sensitivity in planning. 2. A context-dependent, cost-based relaxed planning graph and heuristic. 3. A compilation method from cost to temporal domains. 4. Examination of the impact of planners’ properties on the quality of plans and APFs.

The rise of the model as a rst class entity in the eld of software engineering, sparked an intense period of research into the design and development of model transformations in the 1990s and 2000s. This thesis is about a particular kind of model transformation, in which the source and target models are ordered by containment. It asserts that the monolithic proof of correctness of an ordered set of transformations is equivalent to the sum of the proofs of its elemental parts. It also demonstrates by means of a type theoretical solution that the assertion is sound.

As robots play a more pervasive role in our everyday activities more and more research emphasis is being placed on having robots interact directly with humans, whether in terms of taking care of the elderly, medical interventions or performing dangerous or hazardous tasks. Another trend is for robots to use existing human tools to perform desired actions since it is also not always possible or cost effective to design special tools for every robot. This trend has led to the development of anthropomorphic dexterous manipulators that can perform equally or better than the human hand. Thus the accelerating trend is not only to design a dexterous manipulator but to focus on its ability to grasp and manipulate different and sometimes unknown objects. One of the most researched types of grasp is the precision grasp which accounts for over 80 % of the grasps performed by humans on a daily basis. Precision grasps are grasps involving the fingertips and are generally used for tasks that require fine manipulation skills. Fingertip sensors are therefore important for dexterous manipulators since humans can identify salient properties of an object and formulate effective manipulation strategies solely by grasping the object. This PhD project focuses on developing fingertip sensors, specifically force and torque fingertip sensors that can be integrated into the fingertip of an existing dexterous manipulator and gather contact force and torque information during a grasping event. Another goal is to make the sensor magnetic resonant (MR) compatible so that it can be used in high magnetic environments, as in the case of medical, magnetic resonance imagining applications. To accomplish these goals two sensors were developed based on light intensity modulation and novel sensing structures. Optical sensing schemes were chosen because they are not susceptible to magnetic interference, the sensor and its light source can be separated by long distances without significant signal attenuation and the size and weight of the actual sensing element can be reduced since the processing electronics can be positioned far from the sensing structure. The first sensor developed, as part of this PhD work, was a 2-DOF sensor which used a combination of axially-aligned fibres and linear polarizers to modulate the light to measure the applied force and torque respectively. The use of linear polarizers as the main sensing technique for force and torque sensing is a new area of research since linear polarizers have a defined response curve and can be easily cut into any desired shape and size. The experiments conducted with the 2-DOF sensor showed that the linear polarizer response was superior to traditional axially-aligned and reflective techniques and it was tolerant of small deviations and twists in the sensing structure. The second sensor improved on the first sensor by increasing the number of degrees of freedom from two to six by using a parallel-type 3-UPS (Universal Prismatic Spherical) sensing structure to allow measuring six degrees of movement. All of the joints of the sensing structure were made of nitinol flexures to reduce friction and all of the links were made of plastic and bonded together to produce a flexible but light and strong structure. Another improvement was that all of the optical modulation sensors on the sensing structure were based on linear polarizers thereby reducing the possibility of misalignment errors caused by the transmitting and receiving fibres moving out of axial alignment. The new sensor therefore satisfies the design requirements and the experiments conducted showed that Light Intensity Modulation (LIM) using linear polarizers and an appropriate sensing structure can produce an accurate and versatile force and torque sensor.

This thesis investigates the performance of cross-layer design in data communication over wireless channels. This involves a joint use of two well known techniques; namely, the link adaptation technique of adaptive modulation and coding (AMC) at the physical layer (PHY) and the automatic repeat request (ARQ) scheme at the data link layer (DLL). The motivation behind the design arises from the time-varying characteristic of the wireless channel, where fading may lead to signicant impairment in the content of the received information, thereby aecting the quality of service (QoS) (e.g. at application layer level). Rayleigh distribution is used for modelling the channel statistics. The overall objective is to design an AMC scheme that maximises the ASE, which is a measure of performance at the physical layer, while a condition on the rate (probability) of receiving erroneous packets is satised at the data link layer of the received node; hence the term cross-layer design. This is translated into an optimization problem where the objective function is the ASE, and the constraint function involves the average or instantaneous packet error rate (PER) bounded by a given target, which in turn is set by QoS requirements from higher layers of communication protocol stack. The problem can then be solved by methods such as nonlinear programming. The result is an optimised AMC scheme, where the data rate is adapted according to the estimate of the channel signal-to-noise ratio (SNR). At rst, a continuous variable rate scheme is derived for a point-to-point link between two nodes. As a side contribution, a tight closed-form approximation for PER is obtained as a function of data rate and imperfect estimate of channel SNR. This serves as a generalised case of a practical discrete-rate adaptation. Secondly, the joint DLL and PHY design is applied to communication between two nodes under cognitive radio (CR) resource restraint, where a cooperative node assists the ARQ operation. The performance is compared to that of a single non-cooperative transmission under the same CR constraint.

Through the development of the topic of Web Science there has been interest in the evolution of networks such as the WWW and online social networks almost as ecospheres in a biological sense. However much of the value of the web comes from our ability to search and index it rapidly through the development of ranking and retrieval algorithms such as that oered by Google. Our thesis examines properties of online social networks, and in particular Twitter, whose properties have not been examined to date. However one major problem is the very large size of these networks, and limited amount of resources that are usually available when accessing them for research purposes. We show that through the use of random walks, we are able to quickly discover important portions of such networks and estimate interesting properties, while keeping computational costs low. Our thesis focuses on the following: 1. To study how to crawl massive social networks representatively with limited resources. This will allow users to get a meaningful snapshot of the network. Our methodology for this is: (a) To investigate this problem experimentally and theoretically on articial networks simulated in a controlled environment. (b) To investigate on real networks by comparing our limited designed crawls, with data we have obtained giving the complete structure of e.g. the Twitter network. 2. To investigate the graph theoretic structure of the networks, to the following ends: (a) To devise representative methods of generating articial networks as given in 1(a), both experimentally, and supported by theory. (b) To relate this structure to improving the design of algorithms for information retrieval, ranking home pages, viral advertising etc. 3. To investigate the existence of patterns in user behavior in social networks in order to: (a) Detect user groups with similar interests/behaviors. (b) Recommend activities to users based on activities of similar users.

This thesis aim is to purpose the design of an automatic close-loop control system for general anaesthesia. General anaesthesia is achieved through the administration of pharmaceutical drugs, which produce an effect on patients undergoing surgery. To achieve the aim of this research clinical data obtained from a setup assembled at King's College Hospital and used for estimation and optimization of Pharmacodynamic (PD) models associating effect-site concentrations of propofol, remifentanil and cardiac output to BIS readings. These models were estimated using two techniques, Hill equation and support vector regressors (SVRs) based models. The use of SVRs as a modelling technique allows the incorporation of additional biological signals The SVR technique may produce a nonparametric model which does not guarantee total adequacy of the estimated model as a PD model, therefore a Model Adequacy Index was proposed to assess compliance of the estimated models based on the expected clinical behaviour. PD models considering both pharmaceutical drugs estimated through the SVR technique and Gaussian radial basis kernel presents a considerably higher performance when compared to the estimated multidrug Hill model. Three proportional-integral-differential (PID) controllers are employed, namely linear PID controller, type-1 (T1) fuzzy PID controller and interval type-2 (IT2) fuzzy PID controller, to regulate the bispectral index using the nominal patient's model. The PID gains and membership functions are obtained using genetic algorithm (GA) by minimizing a cost function measuring the control performance. The best trained PID controllers are tested under different scenarios and compared in terms of control performance. Simulation results show that the IT2 fuzzy PID controller offers the best control strategy regulating the BIS index while the T1 fuzzy PID controller comes second.

Increasing demand and sophistication of wireless applications require intelligent systems which, along with performing efficient and reliable adaptive operations, should be simple to implement. Cognitive radio (CR) is one such system which has the capability of adapting to its surroundings. In this thesis, the role of different layers of network in carrying out the functionalities of CR systems is investigated and cross-layer design strategies involving the physical (PHY), the media access control (MAC), and the application are proposed. This thesis makes several contributions. Firstly, we propose novel optimal radio resource allocation (RRA) algorithms under different scenarios with deterministic and probabilistic interference violation limits based on perfect and imperfect availability of cross-link channel state information (CSI). In particular, in contrast to the ‘average case’ and ‘worst case’ estimation error scenarios in the literature, we propose a probabilistic approach to mitigate the total imposed interference on the primary service under imperfect cross-link CSI. An expression for the cumulative density function (cdf) of the received signal-tointerference-plus-noise ratio (SINR) is developed to evaluate the average spectral efficiency. Through simulation results, we investigate the achievable performance and the impact of parameters uncertainty on the overall system performance. Secondly, we implement stochastic RRA algorithms in both hybrid- (i.e., mixed underlay and overlay) and opportunistic (i.e., overlay) access orthogonal frequencydivision multiple access (OFDMA)-based CR systems. The proposed solutions allocate power and subcarrier to cognitive users over wireless fading channels in order to maximize the total transmission rate based on the probabilities of channel availability obtained through spectrum sensing. In order to protect the primary service operation from harmful intervention, stochastic transmit and interference power constrains are imposed on the cognitive users. The performance of the proposed stochastic algorithms and their advantages over the conventional hard-decision-based approaches are assessed and demonstrated through simulation results. Finally a specific cross-layer design for multi scalable video application transmission in an interference-limited spectrum sharing system is proposed. The proposed design jointly considers the parameters from the PHY and the application layers in order to maximize the overall peak signal-to-noise ratio (PSNR). Results indicate that significant improvement in secondary receivers (SRxs) average video quality is achieved through our proposed algorithm over other state-of-the-art non-qualityaware (NQA) designs in the literature. The enhanced performance was obtained whilst guaranteeing SRx minimum quality and primary receiver (PRx) prescribed quality of service (QoS) constraints.

A Wireless Sensor Network (WSN) is composed of many sensor nodes that gather measurements from the environment and transmit them to a central unit for analysis. Sensor nodes are small devices with limited energy resources. Therefore, energy is a crucial concept in a WSN. Optimised design of medium access control (MAC) protocol results in efficient energy consumption while improves delay, lifetime and reliability of data delivery. Our research considers a network with a contention-based MAC protocol. The contributions of this thesis are divided into two parts: • We analyse the concept of priority in a WSN and consider setting high priority to a special node based on the application requirements. Firstly, we assign higher priority to nodes that are running out of energy in order to reduce their activity (less transmition and reception than before) and force the low priority nodes to be more active. Our energy management approach tries to balance energy consumption in a neighbourhood and presents energy saving of around 10% which leads to lifetime expansion of the prioritised network. Secondly, in some other applications we set higher priority to nodes that are more likely to generate significant data (such as kitchen node in a fire alarm system). The high priority nodes send data sooner than other nodes and reduce the contention delay. As a result, average delay for transmitting data in a prioritised network is less compared to an equal-priority nodes network (25% for traffics of 0.3 to 0.7 sample/frame ). We extend the idea of Receiver-Based MAC protocol (RB-MAC) to propose two adaptive preamble sampling techniques, named as adaptive preamble (ap-MAC) and adaptive sampling (as-MAC) protocols. They introduce cross-layer approaches in which routing decisions can be made based on MAC layer functionalities. We study the effects of multiple receiver nodes, size of preamble and variable duty-cycle on number of data retransmissions in both single-hop and multi-hop data forwarding. The analytical and numerical results demonstrate the applicability of the derived protocols in addition to energy efficiency, while maintaining comparable reliability in data delivery.

Routing is one of the most essential functions in computer and telecommunications networks. As the network grows in size, complexity and mobility, it becomes more dicult to precisely determine the routing metrics due to networks' dynamic nature. As a result the information available for decision making of Quality of Service (QoS) routing is always inaccurate. This thesis considers that network routing metrics are naturally uncertain due to the inaccurate information. A novel concept, fuzzy non-dominance multipath routing is developed for the network routing discovery and routing optimisation. The fuzzy non-dominance multipath routing de nes network routing problem in a fuzzy weight graph. The term fuzzy non-dominance routing used in this thesis is distinct from the conventional sense of fuzzy routing. Fuzzy non-dominance routing leads to the fuzzy Pareto-optimal multipath. A labelling setting algorithm is developed to nd out the limited as well as full non-dominated set of routes for network packets forwarding. This approach provides an alternative way to deal with the network routing and multipath routing optimisation problem with less computational and management costs. This thesis proposed a framework for adopting fuzzy non-dominance routing into conventional structure networks. The simulation results covered fuzzy nondominance routing discovery by considering di erent network topologies, scales, fuzzy number designs and the grade of fuzziness. The thesis also addressed fuzzy non-dominance routing for general trac-engineering. Compare to conventional Open Shortest Part First routing, fuzzy non-dominance routing allows the network to cope with at most 60 % more demand. In addition, the thesis also studied the fuzzy non-dominance routing for optimising network routing convergence, quality of service routing and its applications in Mobile ad-hoc networks.

Undoubtedly the upward growth trend in aggregate mobile Internet IP traffic is expected to continue steadily. In this emerging mobile environment with increased data traffic and always-on applications, the limitations of battery technologies lead to drastically shorten recharging cycles of mobile devices. For mobile applications that can tolerate a moderate delay which account for a high proportion of global mobile traffic, a technique that postpones the data transmission to high-rate hotspots could effectively provide significant energy gains, which can be translated to increased battery lifetime. In this thesis, potential gains are explored resulting from utilization of two important technologies for future and emerging wireless networks, namely Cognitive Radios and Delay Tolerant Networking. Both of them are in essence opportunistic in their operation and so far have been considered in isolation. Considering that an increased number of worldwide countries are permitting operation of cognitive radio systems in the spatially vacant licensed analog TV bands, this would enable new possibilities to provision further capacity increase for wireless broadband and multimedia services. Hierarchical CR networks improve spectrum efficiency by allowing the low-priority SUs to temporarily seek the wireless spectrum that is licensed to different organizations. Once mobile devices are equipped with multiple air-interfaces allowing them to connect to cellular networks, Wi-Fi and White-Fi, they could switch among these networks to seek and use any licensed spectrum bands as long as they avoid interference being caused to TV receivers. When wireless nodes are competing for secondary access to the medium, the estimation of probability of PU arrival rate and service time is important for mobile devices (SU) to effectively occupy the primary spectrum. The mobile nodes firstly contact a trusted database for historical information about PU traffic at a specific location and time duration so as to estimate the probability for the SU connections. Then, regarding the SU traffic, it is shown that it can be modeled as an M/M/K/L queuing system which allows to analyze the capability that the system can serve users simultaneously. As location of mobile users is the key to determine the capacity of accessible wireless service for themselves, stochastic characteristics of user mobility are studied in terms of user velocity, direction changes, and route selection distribution. Moreover, when mobile terminals are moving among different cells supporting different network technologies, the performance of vertical handover and cell residence time in the coverage of Wi-Fi/White- Fi hotspots would greatly affect the overall efficiency of wireless transmission. In this scenario, if the mobile applications could tolerate some delay, the proposed schemes can significantly avoid the drain of mobile device batteries by making selective use of the nearby high-speed hotspots. Nowadays, with the surge of the diverse and ubiquitous Internet applications, mobile users expect to enjoy wireless Internet connectivity anywhere and at any time. According to the inherent mobility of mobile users, optimal stopping problem is formulated for energy-delay trade-off, which the stopping decision would be made based on channel conditions, delay constraints, and energy cost. In addition, for popular video streaming applications on portable devices that could be watched several times by one user, there are trade-offs between storing video content locally at the DRAM of the device or allowing deleting the content from the local memory and relaying in wireless streaming in near-future requests of the same content. To this end, a scheme has been proposed where the mobility of the user is taken into account together with the probability of a user requesting the same content multiple times so that a decision is taken of whether or not the content should be stored locally. Finally, since the proliferation of always-on Internet applications has put significant strain on the battery capabilities, the problem of prolonging battery lifetime of mobile devices is introduced. Previous research has revealed that the data downloading via wireless radios is a dominant energy consumption factor in mobile devices. To avoid the drain of mobile device batteries, based on the delay tolerance of mobile Internet applications, the proposed strategies are designed in which mobile terminals can intelligently switch among cellular, Wi-Fi and White-Fi interfaces, by considering the energy cost, RF coverage, capabilities, and transmission algorithm. Numerical experiments on various mobility models reveal that the energy cost of wireless transmission closely relates to user locations, mobility pattern, spectrum availabilities as well as applications’ delay tolerance and available wireless access technologies.