In this paper, we study optimization of multi-tone signals for wireless power transfer (WPT) systems. We investigate different non-linear energy harvesting models. Two of them are adopted to optimize the multi-tone signal according to the channel state information available at the transmitter. We show that a second-order polynomial curve-fitting model can be utilized to optimize the multi-tone signal for any RF energy harvester design. We consider both single-antenna and multi-antenna WPT systems. In-band co-existing communication links are also considered in this work by imposing a constraint on the received power at the nearby information receiver to prevent its RF front end from saturation. We emphasize the importance of imposing such constraint by explaining how inter-modulation products, due to saturation, can cause high interference at the information receiver in the case of multi-tone signals. The multi-tone optimization problem is formulated as a non-convex linearly constrained quadratic program. Two globally optimal solution approaches using mixed-integer linear programming and finite branch-and-bound techniques are proposed to solve the problem. The achieved improvement resulting from applying both solution methods to the multi-tone optimization problem is highlighted through simulations and comparisons with other solutions existing in the literature.

In this paper, a dual-polarized multi-antenna structure is designed at 2.45 GHz with the goal of allowing simultaneous transmission of wireless information and power. Differential feeding was used to minimize the mutual coupling due to radiation leakage in addition to a mushroom-type EBG structure for suppressing the surface waves. Simulation results for the proposed structure show a mutual coupling level lower than -40 dB between the information transmitting antenna and the power transmitting antennas for both polarizations. The isolation level between the antennas is improved by at least 22 dB and 14 dB for the E-plane and H-plane coupling, respectively.

The 2016 Clinical TempEval continued the 2015 shared task on temporal information extraction with a new evaluation test set. Our team, UtahBMI, participated in all subtasks using machine learning approaches with ClearTK (LIBLINEAR), CRF++ and CRFsuite packages. Our experiments show that CRF-based classifiers yield, in general, higher recall for multi-word spans, while SVM-based classifiers are better at predicting correct attributes of TIMEX3. In addition, we show that an ensemble-based approach for TIMEX3 could yield improved results. Our team achieved competitive results in each subtask with an F1 75.4% for TIMEX3, F1 89.2% for EVENT, F1 84.4% for event relations with document time (DocTimeRel), and F1 51.1% for narrative container (CONTAINS) relations.

At Inmotion Technologies, a special method of measuring phase currents is usedin the high power inverters for automotive applications. This method requiresa considerable amount of control logic, currently implemented with discretelogic gates distributed over a number of integrated circuits. In this thesis, thefeasibility of replacing this with programmable logic hardware in one singlepackage is investigated.The theory behind the current measurement method as well as the operationof the discrete implementation is analysed and described. Requirements ona programmable logic device to implement this was identified and a suitabledevice chosen accordingly. A prototype was developed and tested, interfacingan existing product.Benefits in terms of cost and size are evaluated as well as required changesto the existing system and the possibility for improvements brought by such achange is analysed. Since the products in question have high requirements onfunctional safety, possible impacts in this regard are discussed.

Cloud services are growing in popularity and the global public cloud services are forecasted to increase by 17% in 2020[1]. The popularity of cloud services is due to the improved resource allocation for providers and simplicity of use for the customer. Due to the increasing popularity of cloud services and its increased use by companies, the security assessment of the services is strategically becoming more critical. Assessing the security of a cloud system can be problematic because of its complexity since the systems are composed of many different technologies. One way of simplifying the security assessment is attack simulations, covering cyberattacks of the investigated system. This thesis will make use of Meta Attack language (MAL) to create the Domain- Specific Language (DLS) HypervisorLang that models the virtualisation layer in an OpenStack Nova setup. The result of this thesis is a proposed DSL HypervisorLang which uses attack simulation to model hostile usage of the service and defences to evade those. The hostile usage covers attacks such as a denial of services, buffer overflows and out-of-bound-read and are sourced via known vulnerabilities. To implement the main components of the Nova module into HypervisorLang, literature studies where performed and included components in Nova together with threat modelling. Evaluating the correctness of HypervisorLang was performed by implementing test cases to display the different attack steps included in the model. However, the results also show that some limitations of the evaluations have been found and are proposed for further research. 

Software as a Service (SaaS) is a subset of cloud services where a vendor provides software as a service to customers. The SaaS application is installed on the SaaS provider’s servers, and is often accessed via the web browser. In the context of SaaS, a customer is called tenant, which often is an organization that is accessing the SaaS application, but it could also be a single individual. A SaaS application can be classified into tenancy models. A tenancy model describes how a tenant’s data is mapped to the storage on the server-side of the SaaS application.By doing a research, the authors have drawn the conclusion that there is a lack of guidance for selecting tenancy models. The purpose of this thesis is to provide guidance for selecting tenancy models. The short-term-goal is to create a tenancy selection guide. The long-term-goal is to provide researchers and students with research material. This thesis provides a guidance model for selection of tenancy models. The model is called Tenancy Model Selection Guidelines (TMSG).TMSG was evaluated by interviewing two professionals from the software industry. The criteria used for evaluating TMSG were Interviewee credibility, Syntactic correctness, Semantic correctness, Usefulness and Model flexibility. In the interviews, both of the interviewees said that TMSG was in need of further refinements. Still they were positive to the achieved result.

The space electronics sector is shifting towards the New-Space paradigm, in which traditional space-qualified and expensive components and payloads are replaced by commercial off-the-shelf (COTS) alternatives. This change in mentality is accompanied by the development of inexpensive cubesats, lowering the entry barrier in terms of cost, enabling an increase in scientific research in space. However, also well-established and resourceful spacecraft manufacturers are adopting this trend that allows them to become more competitive in the market. Following this trend, Thales Alenia Space is developing R&D activities using COTS components. One example is the SpaceFibre In-Orbit Demonstrator, a digital board integrated in a cubesat payload that aims to test two Intellectual Property blocks implementing the new ECSS standard for high-speed onboard communication. This thesis presents the necessary steps that were taken to integrate the firmware for the demonstrator's Field-Programmable Gate Array (FPGA) that constitutes the main processing and control unit for the board. The activity is centered around the development of a Leon3 System-on-Chip in VHDL used to manage the components in the board and test the SpaceFibre technology. Moreover, it also addresses the main problem of using COTS components in the space environment: their sensitivity to radiation, that, for a FPGA results in Single-Event Upsets causing the implementation to malfunction, and a potential failure of the mission if they are not addressed. To accomplish the task, a SEU-emulation methodology based in partial reconfiguration and integrating the state of the art techniques is elaborated and applied to test the reliability of the SpaceFibre technology. Finally, results show that the mean time between failures of the SpaceFibre Intellectual Property Block using a COTS FPGA is of 170 days for Low Earth Orbit (LEO) and 2278 days for Geostationary Orbit (GEO) if configuration memory scrubbing is included in the design, enabling its usage in short LEO missions for data transmission. Moreover, tailored mitigation techniques based on the information gathered from applying the proposed methodology are presented to improve the figures.

Emerging from the symbiotic combination of nanotechnology and communications, the field of nanonetworking has come a long way since its inception more than fifteen years ago. Significant progress has been achieved in several key communication technologies as enablers of the paradigm, as well as in the multiple application areas that it opens. In this paper, the focus is placed on the electromagnetic nanonetworking paradigm, providing an overview of the advances made in wireless nanocommunication technology from microwave through terahertz to optical bands. The characteristics and potential of the compared technologies are then confronted with the requirements and challenges of the broad set of nanonetworking applications in the Internet of NanoThings (IoNT) and on-chip networks paradigms, including quantum computing applications for the first time. Finally, a selection of cross-cutting issues and possible directions for future work are given, aiming to guide researchers and practitioners towards the next generation of electromagnetic nanonetworks.

Graph Neural Networks (GNNs) have demonstrated state-of-the-art performance in various graph representation learning tasks. Recently, studies revealed their vulnerability to adversarial attacks. In this work, we theoretically define the concept of expected robustness in the context of attributed graphs and relate it to the classical definition of adversarial robustness in the graph representation learning literature. Our definition allows us to derive an upper bound of the expected robustness of Graph Convolutional Networks (GCNs) and Graph Isomorphism Networks subject to node feature attacks. Building on these findings, we connect the expected robustness of GNNs to the orthonormality of their weight matrices and consequently propose an attack-independent, more robust variant of the GCN, called the Graph Convolutional Orthonormal Robust Networks (GCORNs). We further introduce a probabilistic method to estimate the expected robustness, which allows us to evaluate the effectiveness of GCORN on several real-world datasets. Experimental experiments showed that GCORN outperforms available defense methods. Our code is publicly available at: https://github.com/Sennadir/GCORN.

Commercial-Off-The-Shelf (COTS) embedded devices rely on vendor-specific firmware to perform essential tasks. These firmware have been under active analysis by researchers to check security features and identify possible vendor backdoors. However, consistently unpacking newly created filesystem formats has been exceptionally challenging. To thwart attempts at unpacking, vendors frequently use encryption and obfuscation methods. On the other hand, when handling encrypted, obfuscated, big endian cramfs, or custom filesystem formats found in firmware under test, the available literature and tools are insufficient. This study introduces DUDE, an automated framework that provides novel functionalities, outperforming cutting-edge tools in the decryption, unpacking, deobfuscation, and endian conversion of firmware. For big endian compressed romfs filesystem formats, DUDE supports endian conversion. It also supports deobfuscating obfuscated signatures for successful unpacking. Moreover, decryption support for encrypted binaries from the D-Link and MOXA series has also been added, allowing for easier analysis and access to the contents of these firmware files. Additionally, the framework offers unpacking assistance by supporting the extraction of special filesystem formats commonly found in firmware samples from various vendors. A remarkable 78% (1424 out of 1814) firmware binaries from different vendors were successfully unpacked using the suggested framework. This performance surpasses the capabilities of commercially available tools combined on a single platform.

This study focuses on analyzing and optimizing LC filter components for the triangular current mode (TCM)-based zero voltage switching (ZVS) inverter. The integration of ZVS and TCM control techniques in TCM-based ZVS two-level three-phase inverters significantly reduces switching losses, enhances overall system efficiency and power density, and mitigates electromagnetic interference (EMI) issues. An LC filter is essential for operating the inverter in TCM-based ZVS mode, ensuring soft switching throughout. Optimized LC filters play a crucial role by providing ZVS at turn-on and turn-off, enhancing motor efficiency by achieving sinusoidal waveforms and minimizing harmonics. Additionally, the design increases copper space in motor winding slots, enabling lower temperature operation, extended lifespan, and reduced reliance on cooling systems. Performance analysis, including output waveform examination, fast fourier transform (FFT), total harmonic distortion (THD) evaluation, and EMI highlights the superior effectiveness of this LC filter-optimized TCM-based ZVS inverter configuration over conventional hard-switched inverters.

The demand for highly efficient and dynamic electric vehicles (EVs) has increased dramatically. The traction inverter, a pivotal component in an EV powertrain, plays a crucial role. This study is dedicated to designing a traction inverter with focus on achieving high efficiency and elevated power density and mitigating electromagnetic interference (EMI) issues. To realize these objectives, autonomous gate drivers (AGDs) are proposed and designed using LTspice simulation software. The aim is to achieve zero voltage switching (ZVS) at both turn-on and turn-off through the utilization of triangular current mode (TCM) control on the gate driver. The AGDs implement a current modulation scheme by sensing the current and voltage and generating gate-source voltage signals with minimal delays. The implemented current modulation scheme by the AGDs results in an efficiency exceeding 99% for a 10 kW power rating. The sinusoidal output waveforms not only contribute to extending the motor lifespan by mitigating sharp-edge voltages but also bring advantages such as reduced switch stress, decreased EMI, and simplified thermal management.

This study compares shunt-based and SiC MOSFET Rds(on)-based autonomous gate drivers (AGDs) for triangular current mode (TCM)-based zero voltage switching (ZVS) two-level three-phase inverters for electric vehicle (EV) drive systems. Integrating ZVS and TCM control techniques significantly reduces switching losses, enhances overall system efficiency, and mitigates electromagnetic interference (EMI) issues. The AGDs, based on a current modulation scheme, sense switch voltage for turn-on at ZVS and switch current for turn-off, using either a shunt or SiC MOSFET Rds(on). When the switch current exceeds a predefined value, the AGD facilitates ZVS turn-off with external snubber capacitance. Performance evaluation includes gate-source voltage generation, optimum blanking time, ZVS turn-on/off, and efficiency supported by simulation results. The discussion covers the advantages, limitations, and impact of the proposed inverter design on EVs, providing valuable insights for the selection and implementation of AGDs in EV drive systems.

This paper investigates the performance and applicability of a triangular current mode (TCM)–based zero voltage switching (ZVS) three-phase inverter, incorporated with novel autonomous gate drivers (AGDs), for electric vehicle (EV) drive systems. By integrating the controller into the gate driver, delays are minimized, enabling ZVS during the turn-on and turn-off operations. Sinusoidal output waveforms aid in reducing electromagnetic interference (EMI). This study examines the potential interference of the LC filter with torque control and evaluates the inverter’s ability to respond quickly and accurately to torque commands without introducing delays, overshoots, or oscillations. The proposed 10 kW inverter achieved over 99% efficiency without including filter inductor losses. Torque steps were executed significantly faster than required for practical applications. No delays or oscillations were observed, indicating the inverter’s suitability for EV drive systems.

This paper introduces a novel approach to designing autonomous gate drivers for soft-switched buck converters. The objective is to reduce switching losses, enhance converter efficiency, and reduce electromagnetic interference (EMI). The uniqueness of this converter is that the pulse-width modulation is performed autonomously on the gate driver. The gate driver makes quick decisions on switching times, capitalizing on the minimal time delay between measurements and switching. In the proposed buck converter configuration, the gate driver senses both the current and voltage across the switches to avoid delay. When a slightly negative voltage is detected across the switch, it rapidly turns on, resulting in a zero-voltage switching (ZVS). With an external snubber capacitor placed across the switches, the turn-off switching losses are zero (ZVS). Hence, both the turn-on and turn-off of the switch are soft. To enable the switch to turn off, a reference value of the switch current needs to be sent out to the gate driver using a galvanically isolated current sensor. Through this approach, the efficiency of the 7 kW buck converter has been calculated to exceed 99% without including the filter losses. Additional benefits include reduced switch stresses, diminished electromagnetic interference (EMI), and simplified thermal management.

This paper introduces a novel triangular-current mode (TCM), zero-voltage switching (ZVS) two-level three-phase inverter, specifically designed to enhance the performance of the electric vehicle (EV) drive system. The primary objective is to enhance the inverter efficiency by minimizing turn-on and turn-off switching losses while mitigating electromagnetic interference (EMI) by generating sinusoidal output waveforms. The distinctive feature of this inverter lies in its gate driver, which executes the current modulation scheme. Achieving ZVS during turn-on involves the gate driver sensing the switch voltage and turning it on at zero voltage, utilizing TCM. For turn-off ZVS, the gate driver monitors the switch current, turning it off when it exceeds a predefined reference value. With a carefully placed snubber capacitor, turn-off ZVS is achieved. The implemented current modulation scheme yields an efficiency exceeding 99% for a 10 kW power rating. The sinusoidal output waveforms not only enhance motor lifespan by safeguarding against sharp-edge voltages but also offer benefits like reduced switch stress and simplified thermal management.

Heterostructures offer superior photocatalytic characteristics over their constituent counterparts due to their charge separation abilities. Here, we conduct a systematic study of a recently synthesized novel family of palladium-based pentagonal air-stable 2D monolayers, PdSe2, PdPSe, and PdPS, and their heterostructures using first-principles calculations for photocatalytic applications. Electronic band structure calculations reveal moderate bandgaps of 2.27 eV for PdSe2, 2.01 eV for PdPSe, and 2.25 eV for PdPS, indicating their suitability for photocatalytic water splitting. Moreover, to spatially separate and reduce the recombination possibility of photoinduced electron-hole pairs, we propose three van der Waals heterostructures: PdPSe/PdSe2, PdPS/PdSe2, PdPS/PdPS, with the corresponding bandgaps of 1.84 eV, 1.64 eV, and 1.65 eV, respectively. Based on work functions and the staggered band alignment of constituting monolayers, all three heterostructures are identified as type-II photocatalysts, which makes them notable photocatalyst. Additionally, band-edge potentials of PdPSe/PdSe2 and PdPS/PdSe2 confirm their suitability for overall water splitting via photocatalysis, whereas PdPS/PdPSe is suitable for oxygen evolution reactions only. The optical absorption spectra show the ability of each system to operate under a wide range of the spectrum, from visible light to high-energy (UV) regions. These characteristics make these systems valuable and attractive for photocatalytic applications.

Recent years have witnessed a massive increase in the amount of data generated by the Internet of Things (IoT) and social media. Processing huge amounts of this data poses non-trivial challenges in terms of the hardware and performance requirements of modern-day applications. The data we are dealing with today is of massive scale, high intensity and comes in various forms. MapReduce was a popular and clever choice of handling big data using a distributed programming model, which made the processing of huge volumes of data possible using clusters of commodity machines. However, MapReduce was not a good fit for performing complex tasks, such as graph processing, iterative programs and machine learning. Modern data processing frameworks, that are being popularly used to process complex data and perform complex analysis tasks, overcome the shortcomings of MapReduce. Some of these popular frameworks include Apache Spark for batch and stream processing, Apache Flink for stream processing and Tensor Flow for machine learning.

In this thesis, we deal with complex analytics on data modeled as time series, graphs and streams. Time series are commonly used to represent temporal data generated by IoT sensors. Analysing and forecasting time series, i.e. extracting useful characteristics and statistics of data and predicting data, is useful for many fields that include, neuro-physiology, economics, environmental studies, transportation, etc. Another useful data representation we work with, are graphs. Graphs are complex data structures used to represent relational data in the form of vertices and edges. Graphs are present in various application domains, such as recommendation systems, road traffic analytics, web analysis, social media analysis. Due to the increasing size of graph data, a single machine is often not sufficient to process the complete graph. Therefore, the computation, as well as the data, must be distributed. Graph partitioning, the process of dividing graphs into subgraphs, is an essential step in distributed graph processing of large scale graphs because it enables parallel and distributed processing.

The majority of data generated from IoT and social media originates as a continuous stream, such as series of events from a social media network, time series generated from sensors, financial transactions, etc. The stream processing paradigm refers to the processing of data streaming that is continuous and possibly unbounded. Combining both graphs and streams leads to an interesting and rather challenging domain of streaming graph analytics. Graph streams refer to data that is modelled as a stream of edges or vertices with adjacency lists representing relations between entities of continuously evolving data generated by a single or multiple data sources. Streaming graph analytics is an emerging research field with great potential due to its capabilities of processing large graph streams with limited amounts of memory and low latency. 

In this dissertation, we present graph partitioning techniques for scalable streaming graph and time series analysis. First, we present and evaluate the use of data partitioning to enable data parallelism in order to address the challenge of scale in large spatial time series forecasting. We propose a graph partitioning technique for large scale spatial time series forecasting of road traffic as a use-case. Our experimental results on traffic density prediction for real-world sensor dataset using Long Short-Term Memory Neural Networks show that the partitioning-based models take 12x lower training time when run in parallel compared to the unpartitioned model of the entire road infrastructure. Furthermore, the partitioning-based models have 2x lower prediction error (RMSE) compared to the entire road model. Second, we showcase the practical usefulness of streaming graph analytics for large spatial time series analysis with the real-world task of traffic jam detection and reduction. We propose to apply streaming graph analytics by performing useful analytics on traffic data stream at scale with high throughput and low latency. Third, we study, evaluate, and compare the existing state-of-the-art streaming graph partitioning algorithms. We propose a uniform analysis framework built using Apache Flink to evaluate and compare partitioning features and characteristics of streaming graph partitioning methods. Finally, we present GCNSplit, a novel ML-driven streaming graph partitioning solution, that uses a small and constant in-memory state (bounded state) to partition (possibly unbounded) graph streams. Our results demonstrate that \ours provides high-throughput partitioning and can leverage data parallelism to sustain input rates of 100K edges/s. GCNSplit exhibits a partitioning quality, in terms of graph cuts and load balance, that matches that of the state-of-the-art HDRF (High Degree Replicated First) algorithm while storing three orders of magnitude smaller partitioning state.

Short-term traffic prediction allows Intelligent Transport Systems to proactively respond to events before they happen. With the rapid increase in the amount, quality, and detail of traffic data, new techniques are required that can exploit the information in the data in order to provide better results while being able to scale and cope with increasing amounts of data and growing cities. We propose and compare three models for short-term road traffic density prediction based on Long Short-Term Memory (LSTM) neural networks. We have trained the models using real traffic data collected by Motorway Control System in Stockholm that monitors highways and collects flow and speed data per lane every minute from radar sensors. In order to deal with the challenge of scale and to improve prediction accuracy, we propose to partition the road network into road stretches and junctions, and to model each of the partitions with one or more LSTM neural networks. Our evaluation results show that partitioning of roads improves the prediction accuracy by reducing the root mean square error by the factor of 5. We show that we can reduce the complexity of LSTM network by limiting the number of input sensors, on average to 35% of the original number, without compromising the prediction accuracy.

Neural networks are used for different machine learning tasks, such as spatial time-series forecasting. Accurate modelling of a large and complex system requires large datasets to train a deep neural network that causes a challenge of scale as training the network and serving the model are computationally and memory intensive. One example of a complex system that produces a large number of spatial time-series is a large road sensor infrastructure deployed for traffic monitoring. The goal of this work is twofold: 1) To model large amount of spatial time-series from road sensors; 2) To address the scalability problem in a real-life task of large-scale road traffic prediction which is an important part of an Intelligent Transportation System.We propose a partitioning technique to tackle the scalability problem that enables parallelism in both training and prediction: 1) We represent the sensor system as a directed weighted graph based on the road structure, which reflects dependencies between sensor readings, and weighted by sensor readings and inter-sensor distances; 2) We propose an algorithm to automatically partition the graph taking into account dependencies between spatial time-series from sensors; 3) We use the generated sensor graph partitions to train a prediction model per partition. Our experimental results on traffic density prediction using Long Short-Term Memory (LSTM) Neural Networks show that the partitioning-based models take 2x, if run sequentially, and 12x, if run in parallel, less training time, and 20x less prediction time compared to the unpartitioned model of the entire road infrastructure. The partitioning-based models take 100x less total sequential training time compared to single sensor models, i.e., one model per sensor. Furthermore, the partitioning-based models have 2x less prediction error (RMSE) compared to both the single sensor models and the entire road model. 

Graph partitioning is an essential yet challenging task for massive graph analysis in distributed computing. Common graph partitioning methods scan the complete graph to obtain structural characteristics offline, before partitioning. However, the emerging need for low-latency, continuous graph analysis led to the development of online partitioning methods. Online methods ingest edges or vertices as a stream, making partitioning decisions on the fly based on partial knowledge of the graph. Prior studies have compared offline graph partitioning techniques across different systems. Yet, little effort has been put into investigating the characteristics of online graph partitioning strategies.

In this work, we describe and categorize online graph partitioning techniques based on their assumptions, objectives and costs. Furthermore, we employ an experimental comparison across different applications and datasets, using a unified distributed runtime based on Apache Flink. Our experimental results showcase that model-dependent online partitioning techniques such as low-cut algorithms offer better performance for communication-intensive applications such as bulk synchronous iterative algorithms, albeit higher partitioning costs. Otherwise, model-agnostic techniques trade off data locality for lower partitioning costs and balanced workloads which is beneficial when executing data-parallel single-pass graph algorithms.

Real-time road congestion detection allows improving traffic safety and route planning. In this work, we propose to use streaming graph processing algorithms for road congestion detection and evaluate their accuracy and performance. We represent road infrastructure sensors in the form of a directed weighted graph and adapt the Connected Components algorithm and some existing graph processing algorithms, originally used for community detection in social network graphs, for the task of road congestion detection. In our approach, we detect Connected Components or communities of sensors with similarly weighted edges that reflect different states in the traffic, e.g., free flow or congested state, in regions covered by detected sensor groups. We have adapted and implemented the Connected Components and community detection algorithms for detecting groups in the weighted sensor graphs in batch and streaming manner. We evaluate our approach by building and processing the road infrastructure sensor graph for Stockholm's highways using real-world data from the Motorway Control System operated by the Swedish traffic authority. Our results indicate that the Connected Components and DenGraph community detection algorithms can detect congestion with accuracy up to approximate to 94% for Connected Components and up to approximate to 88% for DenGraph. The Louvain Modularity algorithm for community detection fails to detect congestion regions for sparsely connected graphs, representing roads that we have considered in this study. The Hierarchical Clustering algorithm using speed and density readings is able to detect congestion without details, such as shockwaves.

Traffic congestion is a problem in day to day life, especially in big cities. Various traffic control infrastructure systems have been deployed to monitor and improve the flow of traffic across cities. Real-time congestion detection can serve for many useful purposes that include sending warnings to drivers approaching the congested area and daily route planning. Most of the existing congestion detection solutions combine historical data with continuous sensor readings and rely on data collected from multiple sensors deployed on the road, measuring the speed of vehicles. While in our work we present a framework that works in a pure streaming setting where historic data is not available before processing. The traffic data streams, possibly unbounded, arrive in real-time. Moreover, the data used in our case is collected only from sensors placed on the intersections of the road. Therefore, we investigate in creating a real-time congestion detection and reduction solution, that works on traffic streams without any prior knowledge. The goal of our work is 1) to detect traffic jams in real-time, and 2) to reduce the congestion in the traffic jam areas.In this work, we present a real-time traffic jam detection and congestion reduction framework: 1) We propose a directed weighted graph representation of the traffic infrastructure network for capturing dependencies between sensor data to measure traffic congestion; 2) We present online traffic jam detection and congestion reduction techniques built on a modern stream processing system, i.e., Apache Flink; 3) We develop dynamic traffic light policies for controlling traffic in congested areas to reduce the travel time of vehicles. Our experimental results indicate that we are able to detect traffic jams in real-time and deploy new traffic light policies which result in 27% less travel time at the best and 8% less travel time on average compared to the travel time with default traffic light policies. Our scalability results show that our system is able to handle high-intensity streaming data with high throughput and low latency.

Long queues and waiting times, overcrowded waiting rooms, confusion and misunderstanding in communication are common problems in the Swedish healthcare system. This problem has existed for many years, and at the same time it does not look like there is a clear solution for this problem which contributes to its unresolved nature. With popularity of mobile phones in modern society, they emerge as a viable tool for addressing the issue of long queues. Therefore, the purpose of this thesis is to provide a practical solution for the problem, in form of a mobile application, LifeLine. LifeLine is a solution that has potential to solve most of the issues in Swedish healthcare. LifeLine is an innovative application, designed to assists patients in identifying pain locations, streamline specialist referrals and direct users to less crowded healthcare facilities. The goal of this degree project is to provide a solution that can help reduce queues at Swedish healthcare facilities through development of LifeLine. The research method used in this degree project is qualitative. In order to achieve our goal, we used the mixed-methods approach, combining usability testing, user feedback and a discussion with experienced engineers to assess the potential impact and acceptance of our solution. As a result of this thesis, LifeLine enables its users to find the nearest, least crowded facility where patients can get specialist’s help as soon as possible. This also leads to less crowded waiting rooms, subsequently reducing the stress levels of the healthcare workers. At the same time patients do not have to wait for too long to receive help and spend hours in the waiting rooms, hoping to receive the adequate treatment.

Over the past few decades, the field of Enterprise Architecture has attracted researchers, and many Enterprise Architecture modeling frameworks have been proposed. However, in order to support the different needs, the different frameworks offer many different elements types that can be used to create an Enterprise Architecture. This abundance of elements can make it difficult for the end-user to differentiate between the usages of all the various elements in order to identify what elements they actually need. Therefore, this research analyzes existing Enterprise Architecture modeling frameworks and extract the common properties that exists in the different Enterprise Architecture modeling notations. In this study, we performed a Systematic Literature Review that aims at finding the most commonly used Enterprise Architecture modeling frameworks in the Enterprise Architecture literature. Additionally, the elements defined in these frameworks are used to create a taxonomy based on the similarities between the different Enterprise Architecture Frameworks. Our results showed that TOGAF, ArchiMate, DoDAF, and IAF are the most used modeling frameworks. Also, we managed to identify the common elements that are available in the different Enterprise Architecture Frameworks mentioned above and represent the common elements in a multilevel model. The findings of this study can make it easier for the end-user to pick the appropriate elements for their use cases, as it highlights the core elements of Enterprise Architecture modeling. Additionally, we showed how our model can be extended to support the needs of different domains. This thesis also forms the foundation for the development of an Enterprise Architecture modeling framework that can be customized and extended so that only the relevant elements are presented to the end-user.

Data-driven modeling of stochastic nonlinear systems is recognized as a very challenging problem, even when reduced to a parameter estimation problem. A main difficulty is the intractability of the likelihood function, which renders favored estimation methods, such as the maximum likelihood method, analytically intractable. During the last decade, several numerical methods have been developed to approximately solve the maximum likelihood problem. A class of algorithms that attracted considerable attention is based on sequential Monte Carlo algorithms (also known as particle filters/smoothers) and particle Markov chain Monte Carlo algorithms. These algorithms were able to obtain impressive results on several challenging benchmark problems; however, their application is so far limited to cases where fundamental limitations, such as the sample impoverishment and path degeneracy problems, can be avoided.

This thesis introduces relatively simple alternative parameter estimation methods that may be used for fairly general stochastic nonlinear dynamical models. They are based on one-step-ahead predictors that are linear in the observed outputs and do not require the computations of the likelihood function. Therefore, the resulting estimators are relatively easy to compute and may be highly competitive in this regard: they are in fact defined by analytically tractable objective functions in several relevant cases. In cases where the predictors are analytically intractable due to the complexity of the model, it is possible to resort to {plain} Monte Carlo approximations. Under certain assumptions on the data and some conditions on the model, the convergence and consistency of the estimators can be established. Several numerical simulation examples and a recent real-data benchmark problem demonstrate a good performance of the proposed method, in several cases that are considered challenging, with a considerable reduction in computational time in comparison with state-of-the-art sequential Monte Carlo implementations of the ML estimator.

Moreover, we provide some insight into the asymptotic properties of the proposed methods. We show that the accuracy of the estimators depends on the model parameterization and the shape of the unknown distribution of the outputs (via the third and fourth moments). In particular, it is shown that when the model is non-Gaussian, a prediction error method based on the Gaussian assumption is not necessarily more accurate than one based on an optimally weighted parameter-independent quadratic norm. Therefore, it is generally not obvious which method should be used. This result comes in contrast to a current belief in some of the literature on the subject. 

Furthermore, we introduce the estimating functions approach, which was mainly developed in the statistics literature, as a generalization of the maximum likelihood and prediction error methods. We show how it may be used to systematically define optimal estimators, within a predefined class, using only a partial specification of the probabilistic model. Unless the model is Gaussian, this leads to estimators that are asymptotically uniformly more accurate than linear prediction error methods when quadratic criteria are used. Convergence and consistency are established under standard regularity and identifiability assumptions akin to those of prediction error methods.

Finally, we consider the problem of closed-loop identification when the system is stochastic and nonlinear. A couple of scenarios given by the assumptions on the disturbances, the measurement noise and the knowledge of the feedback mechanism are considered. They include a challenging case where the feedback mechanism is completely unknown to the user. Our methods can be regarded as generalizations of some classical closed-loop identification approaches for the linear time-invariant case. We provide an asymptotic analysis of the methods, and demonstrate their properties in a simulation example.

The estimation problem of stochastic Wiener-Hammerstein models is recognized to be challenging, mainly due to the analytical intractability of the likelihood function. In this contribution, we apply a computationally attractive prediction error method estimator to a real-data stochastic Wiener-Hammerstein benchmark problem. The estimator is defined using a deterministic predictor that is nonlinear in the input. The prediction error method results in tractable expressions, and Monte Carlo approximations are not necessary. This allows us to tackle several issues considered challenging from the perspective of the current mainstream approach. Under mild conditions, the estimator can be shown to be consistent and asymptotically normal. The results of the method applied to the benchmark data are presented and discussed.

Identification of dynamic networks has attracted considerable interest recently. So far the main focus has been on linear time-invariant networks. Meanwhile, most real-life systems exhibit nonlinear behaviors; consider, for example, two stochastic linear time-invariant systems connected in series, each of which has a nonlinearity at its output. The estimation problem in this case is recognized to be challenging, due to the analytical intractability of both the likelihood function and the optimal one-step ahead predictors of the measured nodes. In this contribution, we introduce a relatively simple prediction error method that may be used for the estimation of nonlinear dynamical networks. The estimator is defined using a deterministic predictor that is nonlinear in the known signals. The estimation problem can be defined using closed-form analytical expressions in several non-trivial cases, and Monte Carlo approximations are not necessarily required. We show, that this is the case for some block-oriented networks with no feedback loops and where all the nonlinear modules are polynomials. Consequently, the proposed method can be applied in situations considered challenging by current approaches. The performance of the estimation method is illustrated on a numerical simulation example.

Differential-algebraic equations (DAEs) arise naturally as a result of equation-based object-oriented modeling. In many cases, these models contain unknown parameters that have to be estimated using experimental data. However, often the system is subject to unknown disturbances which, if not taken into account in the estimation, can severely affect the model's accuracy. For non-linear state-space models, particle filter methods have been developed to tackle this issue. Unfortunately, applying such methods to non-linear DAEs requires a transformation into a state-space form, which is particularly difficult to obtain for models with process disturbances. In this paper, we propose a simulation-based prediction error method that can be used for non-linear DAEs where disturbances are modeled as continuous-time stochastic processes. To the authors' best knowledge, there are no general methods successfully dealing with parameter estimation for this type of model. One of the challenges in particle filtering  methods are random variations in the minimized cost function due to the nature of the algorithm. In our approach, a similar phenomenon occurs and we explicitly consider how to sample the underlying continuous process to mitigate this problem. The method is illustrated numerically on a pendulum example. The results suggest that the method is able to deliver consistent estimates.

The first part of the paper examines the asymptotic properties of linear prediction error method estimators, which were recently suggested for the identification of nonlinear stochastic dynamical models. It is shown that their accuracy depends not only on the shape of the unknown distribution of the data, but also on how the model is parameterized. Therefore, it is not obvious in general which linear prediction error method should be preferred. In the second part, the estimating functions approach is introduced and used to construct estimators that are asymptotically optimal with respect to a specific class of estimators. These estimators rely on a partial probabilistic parametric models, and therefore neither require the computations of the likelihood function nor any marginalization integrals. The convergence and consistency of the proposed estimators are established under standard regularity and identifiability assumptions akin to those of prediction error methods. The paper is concluded by several numerical simulation examples.

The estimation problem for stochastic parametric nonlinear dynamical models is recognized to be challenging. The main difficulty is the intractability of the likelihood function and the optimal one-step ahead predictor. In this paper, we present relatively simple prediction error methods based on non-stationary predictors that are linear in the outputs. They can be seen as extensions of the linear identification methods for the case where the hypothesized model is stochastic and nonlinear. The resulting estimators are defined by analytically tractable objective functions in several common cases. It is shown that, under certain identifiability and standard regularity conditions, the estimators are consistent and asymptotically normal. We discuss the relationship between the suggested estimators and those based on second-order equivalent models as well as the maximum likelihood method. The paper is concluded with a numerical simulation example as well as a real-data benchmark problem.

In this note, we derive and compare the asymptotic covariance matrices of two parametric estimators: the Gaussian Maximum Likelihood Estimator (MLE), and the optimally weighted Least-Squares Estimator (LSE). We assume a general model parameterization where the model's mean and variance are jointly parameterized, and consider Gaussian and non-Gaussian data distributions.

Recent findings in the field of learning analytics have brought to our attention that conclusions drawn from cross-sectional group-level data may not capture the dynamic processes that unfold within each individual learner. In this light, idiographic methods have started to gain grounds in many fields as a possible solution to examine students’ behavior at the individual level by using several data points from each learner to create person-specific insights. In this study, we introduce such novel methods to the learning analytics field by exploring the possible potentials that one can gain from zooming in on the fine-grained dynamics of a single student. Specifically, we make use of Gaussian Graphical Models —an emerging trend in network science— to analyze a single student's dispositions and devise insights specific to him/her. The results of our study revealed that the student under examination may be in need to learn better self-regulation techniques regarding reflection and planning.

Student engagement has a trajectory (a timeline) that unfolds over time and can be shaped by different factors including learners' motivation, school conditions, and the nature of learning tasks. Such factors may result in either a stable, declining or fluctuating engagement trajectory. While research on online engagement is abundant, most authors have examined student engagement in a single course or two. Little research has been devoted to studying online longitudinal engagement, i.e., the evolution of student engagement over a full educational program. This learning analytics study examines the engagement states (sequences, successions, stability, and transitions) of 106 students in 1396 course enrollments over a full program. All data of students enrolled in the academic year 2014-2015, and their subsequent data in 2015-2016, 2016-2017, and 2017-2018 (15 courses) were collected. The engagement states were clustered using Hidden Markov Models (HMM) to uncover the hidden engagement trajectories which resulted in a mostly-engaged (33% of students), an intermediate (39.6%), and a troubled (27.4%) trajectory. The mostly-engaged trajectory was stable with infrequent changes, scored the highest, and was less likely to drop out. The troubled trajectory showed early disengagement, frequent dropouts and scored the lowest grades. The results of our study show how to identify early program disengagement (activities within the third decile) and when students may drop out (first year and early second year).

Recent findings in the field of learning analytics have brought to our attention that conclusions drawn from cross-sectional group-level data may not capture the dynamic processes that unfold within each individual learner. In this light, idiographic methods have started to gain grounds in many fields as a possible solution to examine students' behavior at the individual level by using several data points from each learner to create person-specific insights. In this study, we introduce such novel methods to the learning analytics field by exploring the possible potentials that one can gain from zooming in on the fine-grained dynamics of a single student. Specifically, we make use of Gaussian Graphical Models -an emerging trend in network science- to analyze a single student's dispositions and devise insights specific to him/her. The results of our study revealed that the student under examination may be in need to learn better self-regulation techniques regarding reflection and planning. 

Single-stage DC-AC power converters are gaining higher attention due to their simpler structure compared to the two-stage equivalent solution. In this paper, a single-stage DC-AC converter solution is proposed for interfacing a low voltage (LV) DC source with a higher voltage AC load or grid, where this converter has a modular structure with multilevel operation. The proposed converter, which is called boost modular multilevel converter (BMMC), comprises the boosting capability within the inversion operation, and it is mainly dedicated for interfacing LV energy storage systems, such as fuel cells and batteries, and it allows the use of LV MOSFETs (« 300 V), in order to utilize their low ON-state resistance, along with LV electrolytic capacitors. This converter is introduced and analysed in this paper, where simulation results using PLECS, considering a 10 kW three-phase BMMC, are presented in order to verify its functionality.

Autonomising industrial robots is the main goal in this paper; imagine humanoid robots that have several degrees of freedom (DOF) mechanisms as their arms. What if the humanoid's arms could be programmed to be responsive to their surrounding environment, without any hard-coding assigned? This paper presents the idea of an autonomous system, where the system observes the surrounding environment and takes action on its observation. The application here is that of rebuffing an object that is thrown towards a robotic arm's workspace. This application mimics the idea of high dynamic responsiveness of a robot's arm. This paper will present a trajectory generation framework for rebuffing incoming flying objects. The framework bases its assumptions on inputs acquired through image processing and object detection. After extensive testing, it can be said that the proposed framework managed to fulfil the real-time system requirements for this application, with an 80% successful rebuffing rate. 

The adoption of containers and container orchestration in cloud computing is motivated by many aspects, from technical and organizational to economic gains. In this climate, even security demanding organizations are interested in such technologies but need reassurance that their requirements can be satisfied. The purpose of this thesis was to investigate how separation of applications could be achieved with Docker and Kubernetes such that it may satisfy the demands of the Swedish Police Authority.

The investigation consisted of a literature study of research papers and official documentation as well as a technical study of iterative creation of Kubernetes clusters with various changes. A model was defined to represent the requirements for the ideal separation. In addition, a system was introduced to classify the separation requirements of the applications.

The result of this thesis consists of three architectural proposals for achieving segmentation of Kubernetes cluster networking, two proposed systems to realize the segmentation, and one strategy for providing host-based separation between containers. Each proposal was evaluated and discussed with regard to suitability and risks for the Authority and parties with similar demands. The thesis concludes that a versatile application isolation can be achieved in Docker and Kubernetes. Therefore, the technologies can provide a sufficient degree of separation to be used in security demanding environments.

A significant transformation has been gradually taking place within the energy sector, mainly as a result of energy policies targeting environmental objectives. Consequently, the penetration of Distributed Energy Resources (DERs) has been escalating, including self-generation, demand side management, storage, and electrical vehicles. Although the integration of DERs may create technical challenges in the operation of distribution networks, it may also provide opportunities to more efficiently manage the network and defer network reinforcements. These opportunities and challenges impose the necessity of redesigning distribution network charges to incentivize efficient customer response.

This PhD thesis focuses on the design of distribution network charges that send correct economic signals and trigger optimal responses within the context of active customers. First, a cost-reflective network charge is proposed that consists of a forward-looking locational component based on the network’s utilization level, which transmits the long-term incremental cost of network upgrades. Then, a residual cost component that recovers the remaining part of the regulated network revenues is proposed. The objective of the proposed network charge is to increase the system’s efficiency by incentivizing efficient short- and long-term customers’ reaction while ensuring network cost recovery. The Thesis presents an optimization model that simulates customers’ response to the proposed network charge in comparison to other traditional network charge designs. The model considers the operational and DER investment decisions that customers take rationally to minimize their total costs.

Secondly, an evaluation methodology based on the Analytical Hierarchy Process technique is proposed in order to assess and compare different designs of network charges with respect to four attributes: network cost recovery, deferral of network costs, efficient customer response and recognition of side-effects on customers.

Finally, a framework for Local Flexibility Mechanisms (LFM) is presented, complementing the proposed cost-reflective network charge. It aims to provide distribution-level coordination to mitigate unintended customer responses to network charges, by allowing customers to reveal their preferences and offer their flexibility services. It consists of a short-term LFM that utilizes customers’ flexibility in day-to-day network operation, and a long-term LFM that procures customers’ long-term flexibility to replace partially or fully network investments in network planning.

One of the leading global health issues is cancer. For women, by far the most pervasive variant is breast cancer. It was also one of the cancers with the highest number of new cases in 2020. To give individuals suffering from breast cancer the best chance possible to survive it is vital to detect the disease early. To do so, computer-aided diagnostics such as a machine learning model can be used to classify mammography images of the breast abnormality as either benign or malignant. In the patient’s journal, more data is available regarding the abnormality and the context of how the mammography scans were taken. This paper aims to examine how that additional data might affect the performance of a model classifying mammography images. Three different image model architectures were used, one quite simple and two utilizing transfer learning. They were then combined with a rather simple architecture used for the additional categorical data. The categorical data was divided into three classes: features describing how the images were taken (class 1), features describing the abnormality (class 2), and a combination of the two aforementioned classes (class 3). The results show that a significant increase was achieved when comparing the hybrid models to the image-only models. However, this increase seems to occur as a result of the image model’s poor performance, and the categorical data being favored by the models combining the two modalities. A slight synergy effect was observed for the hybrid models using the class 3 data. This was, however, too small for us to be able to make any general claims.

YouTube is an online video sharing platform where users can distribute and consume video and other types of content. The rapid technological advancement along with the proliferation och technological gadgets has led to the phenomenon of viral videos where videos and content garner hundreds of thousands if not million of views in a short span of time. This thesis looked at the reason for these viral content, more specifically as it pertains to videos on YouTube. This was done by building a predictor model using two different approaches and extracting important features that causes video popularity. The thesis further observed how the subsequent features impact video popularity via partial dependency plots. The knn model outperformed logistic regression model. The thesis showed, among other things that YouTube channel and title were the most important features followed by comment count, age and video category. Much research have been done pertaining to popularity prediction, but less on deriving important features and evaluating their impact on popularity. Further research has to be conduced on feature influence, which is paramount to comprehend the causes for content going viral. 

Many learners find the transition from block-based programming to text-based programming difficult. Consequently, research has investigated how block-based languages support learners when making the transition to text-based programming. It categorized the way in which block-based languages support the transition into one-way transition, dual-modality and hybrid environments. This research investigates how one-way transition environments compare to dual-modality environments with regards to learning a text-based language, and how the two modalities differ with regards to the motivational factors satisfaction, enjoyment and easiness. The results show that dual-modality environments could be a better alternative than one-way transition environment when learners make the transition from block-based to text-based programming. The results also show that solving a problem in dual-modality environments could be easier than solving them in one-way transition environments, which could potentially mean that learners experience more motivation when making the transition in a dual-modality environment. This study also investigated if there is an alternative to one-way transition, dual-modality and hybrid environments when helping learners transition from block-based to text-based programming, and what a learning activity in this alternative solution could look like. It found that Blockly Games is an alternative, and describes a learning activity built in Blockly Games. Future research should aim at gaining a deeper understanding of the differences between one-way transition, dual-modality and hybrid environments, and investigate if the approach taken by Blockly Games is a better alternative.

Modern cities must overcome complex challenges to achieve socio-economic development and to improve the quality of life as the urban population is rapidly increasing. The concept of smart citie s is a response to these challenges. Thus, emerging technologies that are key enablers for the development of a smart city are said to be IoT and 5G. To deploy such technologies , however, may be expensive and requires the involvement of multiple actors. Hence, lack of cooperation and coordination for planning, financing, deploying and managing the city’s operational networks makes it even more difficult to overcome such challenges. Further, waste management companies and parking services operators in a city have expensive operation costs and services inefficiency due to little utilization of IoT-based solutions. This paper identifies and analyzes smart city ecosyst e ms, value networks, actors, actor’s roles, and business models in order to illustrate business relationships and provide business opportunities in the development of smart and sustainable cities through cooperation and collaboration among involved actors . Target actors that this study focuse s are on Mobile Network Operators, Parking Services Operators, and Waste Management Companies, and uses smart parking and smart waste collection as use-cases. Results show several cooperative business scenarios that can lead to successful business relationships and opportunities.

Nowadays, optical network-on-chip is accepted as a promising alternative solution for traditional electrical interconnects due to lower transmission delay and power consumption as well as considerable high data bandwidth. However, silicon photonics struggles with some particular challenges that threaten the reliability of the data transmission process.The most important challenges can be considered as temperature fluctuation, process variation, aging, crosstalk noise, and insertion loss. Although several attempts have been made to investigate the effect of these issues on the reliability of optical network-on-chip, none of them modeled the reliability of photonic network-on-chip in a system-level approach based on basic element failure rate. In this paper, an analytical model-based simulator, called Reliability Assessment of Photonic Network-on-Chips (RAP-NoC), is proposed to evaluate the reliability of different 2D optical network-on-chip architectures and data traffic. The experimental results show that, in general, Mesh topology is more reliable than Torus considering the same size. Increasing the reliability of Microring Resonator (MR) has a more significant impact on the reliability of an optical router rather than a network.

Autonomous robots are expected to make a greater presence in the homes and workplaces of human beings. Unlike their industrial counterparts, autonomous robots have to deal with a great deal of uncertainty and lack of structure in their environment. A remarkable aspect of performing manipulation in such a scenario is the possibility of physical contact between the robot and the environment. Therefore, not unlike human manipulation, robotic manipulation has to manage contacts, both expected and unexpected, that are often characterized by complex interaction dynamics.

Skill learning has emerged as a promising approach for robots to acquire rich motion generation capabilities. In skill learning, data driven methods are used to learn reactive control policies that map states to actions. Such an approach is appealing because a sufficiently expressive policy can almost instantaneously generate appropriate control actions without the need for computationally expensive search operations. Although reinforcement learning (RL) is a natural framework for skill learning, its practical application is limited for a number of reasons. Arguably, the two main reasons are the lack of guaranteed control stability and poor data-efficiency. While control stability is necessary for ensuring safety and predictability, data-efficiency is required for achieving realistic training times. In this thesis, solutions are sought for these two issues in the context of contact-rich manipulation.

First, this thesis addresses the problem of control stability. Despite unknown interaction dynamics during contact, skill learning with stability guarantee is formulated as a model-free RL problem. The thesis proposes multiple solutions for parameterizing stability-aware policies. Some policy parameterizations are partly or almost wholly deep neural networks. This is followed by policy search solutions that preserve stability during random exploration, if required. In one case, a novel evolution strategies-based policy search method is introduced. It is shown, with the help of real robot experiments, that Lyapunov stability is both possible and beneficial for RL-based skill learning.

Second, this thesis addresses the issue of data-efficiency. Although data-efficiency is targeted by formulating skill learning as a model-based RL problem, only the model learning part is addressed. In addition to benefiting from the data-efficiency and uncertainty representation of the Gaussian process, this thesis further investigates the benefits of adopting the structure of hybrid automata for learning forward dynamics models. The method also includes an algorithm for predicting long-term trajectory distributions that can represent discontinuities and multiple modes. The proposed method is shown to be more data-efficient than some state-of-the-art methods. 

In this letter, we investigate learning forward dynamics models and multi-step prediction of state variables (long-term prediction) for contact-rich manipulation. The problems are formulated in the context of model-based reinforcement learning (MBRL). We focus on two aspects-discontinuous dynamics and data-efficiency-both of which are important in the identified scope and pose significant challenges to State-of-the-Art methods. We contribute to closing this gap by proposing a method that explicitly adopts a specific hybrid structure for the model while leveraging the uncertainty representation and data-efficiency of Gaussian process. Our experiments on an illustrative moving block task and a 7-DOF robot demonstrate a clear advantage when compared to popular baselines in low data regimes.

Deep reinforcement learning (DRL) has been successfully used to solve various robotic manipulation tasks. However, most of the existing works do not address the issue of control stability. This is in sharp contrast to the control theory community where the well-established norm is to prove stability whenever a control law is synthesized. What makes traditional stability analysis difficult for DRL are the uninterpretable nature of the neural network policies and unknown system dynamics. In this work, stability is obtained by deriving an interpretable deep policy structure based on the energy shaping control of Lagrangian systems. Then, stability during physical interaction with an unknown environment is established based on passivity. The result is a stability guaranteeing DRL in a model-free framework that is general enough for contact-rich manipulation tasks. With an experiment on a peg-in-hole task, we demonstrate, to the best of our knowledge, the first DRL with stability guarantee on a real robotic manipulator.