One of the key enablers for the extraordinary dexterity of human hands is their compliance and capability to purposefully adapt with the environment and to multiply their manipulation possibilities. This observation has also produced a significant paradigm shift for the design of robotic hands, leading to the avenue of soft endeffectors that embed elastic and deformable elements directly in their mechanical architecture. This shift has also determined a perspective change for the control and planning of the grasping phases, with respect to (w.r.t.) the classical approach used with rigid grippers. Indeed, instead of targeting an accurate analysis of the contact points on the object, an approximated estimation of the relative hand-object pose is sufficient to generate successful grasps, exploiting the intrinsic adaptability of the robotic systems to overcome local uncertainties. This chapter reports on deep learning (DL) techniques used to model human manipulation and to successfully translate these modelling outcomes for enabling soft artificial hands to autonomous grasp objects with the environment.

This chapter provides an introduction on the benefits of artificial intelligence (Al) techniques for the field of affective computing, through a case study about emotion recognition via brain (electroencephalography EEG) signals. Readers are first pro-vided with a general description of the field, followed by the main models of human affect, with special emphasis to Russell's circumplex model and the pleasur-arousal-dominance (PAD) model. Finally, an AI-based method for the detection of affect elicited via multimedia stimuli is presented. The method combines both connectivity-and channel-based EEG features with a selection method that considerably reduces the dimensionality of the data and allows for efficient classification. In particular, the relative energy (RE) and its logarithm in the spatial domain, as well as the spectral power (SP) in the frequency domain are computed for the four typically used EEG frequency bands (a, 0, y and 0) and complemented with the mutual information measured over all EEG channel pairs. The resulting features are then reduced by using a hybrid method that combines supervised and unsupervised feature selection. Detection results are compared to state-of-the-art methods on the DEAP benchmark-ing data set for emotion analysis, which is composed of labelled EEG recordings from 32 individuals, acquired while watching 40 music videos. The acquired results demonstrate the potential of AI-based methods for emotion recognition, an applica-tion that can significantly benefit the fields of human-computer interaction (HCI) and of quality-of-experience (QoE).

This chapter focuses on the use of machine-learning techniques within the field of affective computing, and more specifically for the task of emotion recognition within the context of human-horse interaction. Affective computing focuses on the detection and interpretation of human emotion, an application that could significantly benefit quantitative studies in the field of animal-assisted therapy. The chapter offers a thorough description, an experimental design, and experimental results on the use of physiological signals, such as electroencephalography (EEG), electrocardiography (ECG), and electromyography (EMG) signals, for the creation and evaluation of machine-learning models for the prediction of the emotional state of an individual during interaction with horses.

In this chapter, we will look into how robots can, have and will benefit the wider human community in tasks that perhaps were taken for granted. More specifically, we review and analyse state-of-the-art work in robotic locomotion, robotic manipulation with and without human supervision. We hope to assist readers in having a thorough related work as a basis of their research with the current state-of-the-art approaches in the aforementioned fields as well as the importance of robots today.

In this chapter, we consider the visual object-tracking problem in order to find an efficient and useful method for UAV. The aim of this study is to demonstrate the feasibility and the potential of incorporating Al into autonomous target tracking for UAVs applications. A proof-of-concept prototype was developed based on the UAV platform Parrot AR.Drone 2.0 and by using the Robot Operating System (ROS) and the package ardrone_autonomy simulated in Gazebo simulator. In the first section of this work, a fuzzy logic controller (FLC) is used for visual object tracking by using the AR.Drone 2.0 and visual information feedback. Fuzzy logic is used for its ability to shape the control surfaces generated by the rules of the fuzzy inference system. In order to guarantee a correct recognition of the target by the AR.Drone 2.0, a quick response (QR) code and ar track_alvar package in ROS have been used as target marker. Visual feedback is provided by identification of visual tags, using open-source software FuzzyLite to calculate the tag identification match and orientation. The proposed algorithm is based on a small number of input parameters, which reduces the requirements for high computing power. The developed algorithm was validated by simulation studies and in-flight tests.

Future cellular networks are expected to see enhancements that will change their operation with the advent of new technologies, such as millimetre-wave (mmWave) communications, new physical layer waveforms, and network densification, to name a few. However, despite all the benefits that these concepts will bring to future networks, other challenges will arise. One issue that has gained attention in recent years is the problem of mobility management, which is expected to become even more challenging in future networks. This occurs due to the network densification process and the short-range coverage provided by mmWaves, which will lead to more frequent and an exponential number of handovers (HOs) by end users, generating a tremendous amount of signalling which cannot be handled by conventional means. To tackle these challenges, a proactive mobility-management concept, where HO events are triggered in advance with the help of intelligent tools that are able to predict the future behaviours of users and the network have been proposed recently. Results have shown that this proactive approach helps eliminating certain steps of the HO phase, resulting in less latency and signalling overhead, leading to a better network optimisation. On the other hand, if the accuracy of the predictions is not enough, this proactive approach can provide worse results than the reactive one. As such, the accuracy of the algorithm plays a vital role in ensuring the predictive management applicable. In this chapter, both traditional and proactive mobility managements in cellular networks are presented, and a Markov-chain-based proactive HO process is proposed.

5G technologies are expected to enable new verticals, services, and business models. Recently, the use of artificial intelligence (AI) and data analytics is shown to provide massive advantages in terms of reducing network complexity and enhancing its performance. In this chapter, we provide an overview of the recent studies on AI-assisted solutions in 5G wireless networks, followed by three case studies of our original work, including a Q-learning-assisted cell selection mechanism, an AI engine that enables intelligent 5G fronthaul slicing, and a beam management protocol for a multiple radio access technology (RAT) coexistence via learning. Realizing the vital role of data and data analytics in enabling AI for wireless networks in practice, we review data analytics in the current literature and discuss how data analytics and AI enable the applications in 5G networks. The recent industry and standardization activities of using AI in 5G networks are summarized. Finally, we give our insights on the research challenges and open questions.

In this chapter, we suggest an effective way of discovering a coverage hole with the help of UAV and ML. The main purpose is to take different parameters from the radio environment and detect the coverage hole efficiently and autonomously. The simulation results show that the proposed method is successful in detecting the coverage hole. Further research for this proposed method can be extended in to many directions. For example, the UAV has detected only a single objective or only one coverage hole in this simulation. If there are more than one coverage hole in a complex radio environment then we have to consider multi-objective RL and consider additional constraints such as UAV charging stations and obstacles, e.g., MBS, trees, buildings. Also, the simulation of such complex radio environment needs to urban scenarios with multi obstacles avoidance techniques considering the speed of the UAV. Apart from these, we can also consider an on-demand UAV base station (tethered or untethered UAV) to provide coverage and capacity to a coverage hole or poor network service area. Based on the traffic requirement and available wireless backhaul, UAVs can act as a base station at the same time while flying to the coverage hole area in a shortest distance.

In this chapter, we have designed an NB classifier for a UWB-based localization system. With the help of NB classifier and RMSE, the data are classified into three categories: high, medium, and low accuracy. ROCs are plotted to show the effec-tiveness of the NB classifier. As our developed technique obtains more than 90% classification accuracy, we have tested it into two different environments: LOS and partial NLOS conditions. Furthermore, to test the accuracy, small-sized and medium-sized rooms were used. From our measurements, it is observed that the accuracy of the developed NB classifier is dependent upon the environment. For LOS and NLOS envi-ronments, the accuracy are around 97% and 87.38%, respectively. Our future research will concentrate on technique that can further improve the localization classification and improve the positioning accuracy of the IPS.

This chapter developed a machine learning approach for indoor environment classification based on real-time measurements of the RF signal. Several machine learning classification methods were contemplated, including DTs, SVM, and k-NN, using different RF features. Results obtained show that a machine learning approach using k-NN method, utilizing CTF and FCF, outperforms the other methods in identifying the type of the indoor environment with a classification accuracy of 99.3%. The predication time was obtained to be less than 10 u,s, which verifies that the embraced algorithm is successful for real-time deployment scenarios. The results of this chapter facilitate an efficient deployment of IoT applications in dynamic channels

This chapter discusses the effects of template ageing in electroencephalography (EEG)-based biometrics. This chapter also serves as an introduction to general biometrics and its main tasks: identification and verification. To do so, we investigate different characterisations of EEG signals and examine the difference of performance in subject identification between single-session and cross-session identification experiments. In order to do this, EEG signals are characterised with common state-of-the-art features, i.e. mel frequency cepstral coefficients (MFCCs), autoregression coefficients, and power-spectral-density (PSD)-derived features. The samples were later classified using various classifiers, including support vector machines and k-nearest neighbours (kNN) with different parametrisations. Results show that performance tends to be worse for cross-session identification compared to single-session identification. This finding suggests that temporal permanence of EEG signals is limited and thus more sophisticated methods are needed in order to characterise EEG signals for the task of subject identification.

The advent of high-speed internet connections, advanced video coding algorithms, and consumer-grade computers with high computational capabilities has led videostreaming-over-the-internet to make up the majority of network traffic. This effect has led to a continuously expanding video streaming industry that seeks to offer enhanced quality-of-experience (QoE) to its users at the lowest cost possible. Video streaming services are now able to adapt to the hardware and network restrictions that each user faces and thus provide the best experience possible under those restrictions. The most common way to adapt to network bandwidth restrictions is to offer a video stream at the highest possible visual quality, for the maximum achievable bitrate under the network connection in use. This is achieved by storing various preencoded versions of the video content with different bitrate and visual quality settings. Visual quality is measured by means of objective quality metrics, such as the mean squared error (MSE), peak signal-to-noise ratio (PSNR), structural similarity (SSIM) index, visual information fidelity (VIF), and others, which can be easily computed analytically. Nevertheless, it is widely accepted that although these metrics provide an accurate estimate of the statistical quality degradation, they do not reflect the viewer's perception of visual quality accurately. As a result, the acquisition of user ratings in the form of mean opinion scores (MOSs) remains the most accurate depiction of human-perceived video quality, albeit very costly and time consuming, and thus cannot be practically employed by video streaming providers that have hundreds or thousands of videos in their catalogues. A recent very promising approach for addressing this limitation is the use of machine learning techniques in order to train models that represent human video quality perception more accurately. To this end, regression techniques are used in order to map objective quality metrics to human video quality ratings, acquired for a large number of diverse video sequences. Results have been very promising, with approaches like the Video Multimethod Assessment Fusion (VMAF) metric achieving higher correlations to user-acquired MOS ratings compared to traditional widely used objective quality metrics. In this chapter, we examine the performance of VMAF and its potential as a replacement for common objective video quality metrics.

Autonomous driving is inherently a multitask learning (MTL) problem. In the current work, we propose a generalized MTL framework for the estimation of various parameters needed for autonomous driving. This framework generates different networks for the estimation of a different set of tasks based on their relationship. The relationship among tasks to be learned is handled by including shared layers in the architecture. Later, the network separates into different branches to handle the difference in the behavior of each task. More specifically, we provide a solution for the estimation of driving control parameters as well as those related to scene information. We demonstrated the performance of the proposed solution on four publicly available benchmark datasets: Comma.ai, Udacity, Berkeley Deep Drive (BDD) and Sully Chen. A synthetic dataset GTA-V for autonomous driving research has also been proposed to further evaluate the proposed approach.

Hyperkalaemia is the medical terminology for a blood potassium level above normal parameters (greater than 5.5 mmol). This can have a variety of causes; however, patients with significant renal impairment/disease are particularly at a high risk as their kidneys are compromised and unable to filter out excess potassium from the bloodstream. Hyperkalaemia is a medical emergency and requires urgent medical intervention; if left untreated, there is an extremely high risk of cardiac arrest and death as high potassium levels directly affect the electrical activity of the heart. Cardiac activity can be observed by performing an electrocardiogram (ECG) test that is routinely used in all clinical areas worldwide. This research investigates the use of ECG tests as a diagnostic tool for the early detection of hyperkalaemia, and the role of machine learning in the prediction of blood potassium levels from ECG data alone. Support vector machines (SVMs), k-nearest neighbour (k-NN), decision tree and Gaussian Naïve Bayes classifiers were used comparatively to classify ECG data as either `normokalaemia' or 'hyperkalaemia'. Results showed that the decision tree model performed the best, achieving a 90.9 per cent predictive accuracy.

Connected health is continuously developing, particularly with the advent of the Internet of Things (IoT) interconnecting various sensing nodes capable of measuring a person's vital signs such as electrocardiogram (ECG). In the years to come, the current forecasts indicate a significant increase in demand of such devices, especially among a currently underserved but significant population. Most of the existing devices performing measurement and data transmission require significant effort to integrate more intelligent processing or even decision-making, at least for data reduction and more autonomy. In this chapter, we propose to combine a simple compressed sensing (CS) measurement technique with a machine learning classification, both for data reduction and low power consumption. The classification is performed on compressed data, whereas the transmission is achieved only for warnings, by sending classification information in the case of a probable pathology detection, and if neces-sary the compressed data for further analysis. For data acquisition, we utilize a simple deterministic measurement matrix that facilitates the hardware implementation. The performance of the proposed approach is demonstrated using ECG recordings from three PhysioNet databases: MIT-BIH Arrhythmia Database, MIT-BIH Normal Sinus Rhythm Database and The BIDMC Congestive Heart Failure Database.

Machine learning such as self-organizing feature maps (SOMs) is a commonly used technique for clustering and data dimensionality reduction. In fact, their inherent property of topology preservation and unsupervised learning of processed data put them in the front of candidates for data reduction. However, the high computational cost of SOMs limits their use to offline approaches and makes the on-line real-time high-performance SOM processing more challenging. This chapter focus on the large-scale distributed and scalable SOM model adapted for distributed computing nodes and present the main challenges for its adoption in the resources limited environments.

Water is one of the basic resources required for human survival. However, pollution of water has become a global problem. 2.4 billion people worldwide live without any form of water sanitation. This work focuses on case study of water pollution in Pakistan where only 20% of the population has an access to good-quality water. Drinking bad-quality water causes diseases such as hepatitis, diarrhea and typhoid. Moreover, people living close to the industrial areas are more prone to drinking polluted water and catching diseases as a result. Yet, there is no system that can monitor the quality of water or help in disease prevention. In this work, an Internet of Things (IoT)-enabled water quality monitoring system is developed that works as a stand-alone portable solution for monitoring water quality accurately and in real time. The real-time results are stored in a cloud database. The public web portal shows these results in the form of data sheets, maps and charts for analyzing data. Further, this data along with the collected data of past water quality is used to generate machine learning (ML) models for prediction of water quality. As a consequence, a model for prediction of water quality is trained and tested on a test set. The predictions on the test set resulted in a mean squared error (MSE) of 0.264.

Artificial intelligence and machine-learning field have continuously been expanding their applications into various domains, some of which have the potential to revolutionise people's daily lives. Ever-increasing importance of automation in the digital transformation of our society, economy and industry have necessitated the use of artificial intelligence, machine learning for robotics, sensing and networking.