Change search
Refine search result
123 1 - 50 of 135
CiteExportLink to result list
Permanent link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Rows per page
  • 5
  • 10
  • 20
  • 50
  • 100
  • 250
Sort
  • Standard (Relevance)
  • Author A-Ö
  • Author Ö-A
  • Title A-Ö
  • Title Ö-A
  • Publication type A-Ö
  • Publication type Ö-A
  • Issued (Oldest first)
  • Issued (Newest first)
  • Created (Oldest first)
  • Created (Newest first)
  • Last updated (Oldest first)
  • Last updated (Newest first)
  • Disputation date (earliest first)
  • Disputation date (latest first)
  • Standard (Relevance)
  • Author A-Ö
  • Author Ö-A
  • Title A-Ö
  • Title Ö-A
  • Publication type A-Ö
  • Publication type Ö-A
  • Issued (Oldest first)
  • Issued (Newest first)
  • Created (Oldest first)
  • Created (Newest first)
  • Last updated (Oldest first)
  • Last updated (Newest first)
  • Disputation date (earliest first)
  • Disputation date (latest first)
Select
The maximal number of hits you can export is 250. When you want to export more records please use the Create feeds function.
  • 1.
    Tao, Jiangpeng
    KTH, School of Electrical Engineering and Computer Science (EECS).
    3D LiDAR based Drivable Road Region Detection for Autonomous Vehicles2020Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Accurate and robust perception of surrounding objects of interest, such as onroad obstacles, ground surface, curb and ditch, is an essential capability for path planning and localization in autonomous driving. Stereo cameras are often used for this purpose. Comparably, 3D LiDARs directly provide accurate depth measurements of the environment without the need for association of pixels in image pairs. In this project, disparity is used to bridge the gap between LiDAR and stereo cameras, therefore efficiently extracting the ground surface and obstacles from 3D point cloud in the way of 2D image processing. Given the extracted ground points, three kinds of features are designed to detect road structures with large geometrical variation, such as curbs, ditches and grasses. Based on the feature result, a robust regression method named least trimmed squares is used to fit the final road boundary. The proposed approach is verified with the real dataset from a 64-channel LiDAR mounted on Scania bus Klara, as well as the KITTI road benchmark, both achieving satisfying performances in some particular situations.

    Download full text (pdf)
    fulltext
  • 2.
    Khoche, Ajinkya
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Depth Estimation from Images using Dense Camera-Lidar Correspondences and Deep Learning2020Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Depth estimation from 2D images is a fundamental problem in Computer Vision, and is increasingly becoming an important topic for Autonomous Driving. A lot of research is driven by innovations in Convolutional Neural Networks, which efficiently encode low as well as high level image features and are able to fuse them to find accurate pixel correspondences and learn the scale of the objects. Current state-of-the-art deep learning models employ a semi-supervised learning approach, which is a combination of unsupervised and supervised learning. Most of the research community relies on the KITTI datasets for benchmarking of results. But the training performance is known to be limited by the sparseness of the lidar ground truth as well as lack of training data.

    In this thesis, multiple stereo datasets with increasingly denser depth maps are generated on the corpus of driving data collected at the Audi Electronics Venture GmbH. In this regard, a methodology is presented to obtain an accurate and dense registration between the camera and lidar sensors. Approaches are also outlined to rectify the stereo image datasets and filter the depth maps. Keeping the architecture fixed, a monocular and a stereo depth estimation network each are trained on these datasets and their performances are compared to other networks reported in literature. The results are competitive, with the stereo network exceeding the state-of-the-art accuracy. More work is needed though to establish the influence of increasing depth density on depth estimation performance. The proposed method forms a solid platform for pushing the envelope of depth estimation research as well as other application areas critical to autonomous driving.

    Download full text (pdf)
    fulltext
  • 3.
    Vijjappu, Srihaarika
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Distributed Decentralised Visual SLAM for Multi-Agent Systems2020Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    A key challenge in multi robot systems is performing distributed SLAM (Simultaneous Localisation and Mapping). The aim of this thesis is to be able to perform visual SLAM in a decentralised manner across multiple autonomous agents while minimising the inter-agent communication bandwidth requirement. For this purpose, the distributed multi-agent system communication protocol, Contract NET protocol has been suitably adapted to define the interaction between the autonomous agents [38]. The agents in this context could be mobile devices such as robots or autonomous cars. The agents communicate with one another by means of proposals and bids wherein each agent attempts to minimise the resources it has to spend while attempting to maximise the benefit derived from interacting with another agent in the system. Keeping this in mind, a set of rules have been defined for the agents to logic and reason independently based on the state information available to them so that they can take appropriate decisions by themselves without the need for an external intervention. This thesis work is directed towards the design and execution of a visual SLAM system in a multi-agent architecture that aims to improve the accuracy of the pose estimates of the agents, maximise the map area exploration and the accuracy of the map copies possessed by the agents while attempting to minimise the bandwidth required for the data exchange between them. The proposed multi agent system has been thoroughly studied and evaluated with multiple datasets to analyse its performance with respect to bandwidth requirements, stability, consistency, scalability, map accuracy and usage of temporal information.

    Download full text (pdf)
    fulltext
  • 4.
    Villani, Gianluca
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Analysis of an Attractor Neural Network Model for Working Memory: A Control Theory Approach2019Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Working Memory (WM) is a general-purpose cognitive system responsible for temporaryholding information in service of higher order cognition systems, e.g. decision making. Inthis thesis we focus on a non-spiking model belonging to a special family of biologicallyinspired recurrent Artificial Neural Network aiming to account for human experimentaldata on free recall. Considering its modular structure, this thesis gives a networked systemrepresentation of WM in order to analyze its stability and synchronization properties.Furthermore, with the tools provided by bifurcation analysis we investigate the role of thedifferent parameters on the generated synchronized patterns. To the best of our knowledge,the proposed dynamical recurrent neural network has not been studied before froma control theory perspective.

    Download full text (pdf)
    fulltext
  • 5.
    Huang, Mengdi
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Consistent Outdoor Environment Mapping with a 3D Laser Scanner2019Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Simultaneous Localization and Mapping(SLAM) is an essential prerequisite for various automated systems, such as self-driving cars, and planetary rovers. These autonomous machines acquire the knowledge of the environment through building a map while exploring an unknown area. Without this knowledge, they are not able to make right decisions.

    We used a 3D laser scanner with 16 channels, and encoders to collect the internal and external information. Then we estimate the trajectory that the robot has been to and build a consistent map upon sensor data. In this project, we studied and proposed several ways to improve the existing registration and optimization methods. By adding odometry information to predict the initial estimation as the input for Normal Distribution Transform(NDT), the performance is proved to be boosted. And the result witnessed a satisfying change after we distribute the error before inputing the registration result into g2o optimization framework. Besides, we proposed a method to increase the weight of certain voxels to improve the performance of NDT. We also experimented with different configurations of NDT and g2o to see how they impact the results.

    We conducted and analyzed the experiment on two datasets, urban dataset and forest dataset. The mapping is considered successful in the urban dataset.

    Download full text (pdf)
    fulltext
  • 6.
    Wei, Xiao
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Deep Active Learning for 3D Object Detection for Autonomous Driving2019Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    3D object detection is vital for autonomous driving. However, to train a 3D detector often requires a huge amount of labeled data which are extremely expensive and tedious to obtain. In order to alleviate the annotation effort while maintaining detection performance, we aim to adopt active learning framework for training a 3D object detector with the least amount of labeled data. In contrast with the conventional passive learning that a machine learning model is trained on a pre-determined training dataset, active learning allows the model to actively select the most informative samples for labeling and add them to the training set. Therefore, only a fraction of data need to be labeled. To the best of our knowledge, this thesis is the first that studies active learning for 3D detection.

    We take progressive steps towards the goal. There are three stages with increasingly complex models and learning tasks. First, we start with active learning for image classification which can be viewed as a sub-problem of object detection. Second, we investigate and build a multi-task active learning framework with a deep refinement network for multi-modal 3D object detection. Lastly, we further analyze multi-task active learning with a more complicated two-stage 3D LiDAR vehicle detector. In our experiments, we study the fundamental and important aspects of an active learning framework with an emphasis on evaluating several popular data selection strategies based on prediction uncertainty. Without bells and whistles, we successfully propose an active learning framework for 3D object detection using 3D LiDAR point clouds and accurate 2D image proposals that saves up to 60% of labeled data on a public dataset. In the end, we also discuss some underlying challenges on this topic from both academic and industrial perspectives.

    Download full text (pdf)
    fulltext
  • 7.
    Bereza-Jarocinski, Robert
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Distributed Model Predictive Control for Rendezvous Problem2019Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    This thesis investigates the potential advantages and disadvantages of using adistributed control approach to land an autonomous drone on an autonomousboat. The expected advantages include better utilisation of computational resources,as well as increased robustness towards communication delays. Inthis context, distributed control means that separate computers on the droneand boat are both involved in computing the control inputs to the system. Thisstands in contrast to an existing centralised algorithm where all computationsfor finding the control input are performed on the drone. Two new algorithmsare proposed, one using distributed model predictive control (DMPC) and oneusing a combination of DMPC with linear state-space feedback. The followingproperties of all the algorithms are tested: what the longest possible predictionhorizon with sufficiently short solution time is, how long it takes to solve optimisationproblems for the algorithms, and how quickly and safely each algorithmcan land the drone. Finally, the DMPC algorithm is shown to in certainscenarios possess improved robustness towards communication delays.

    Download full text (pdf)
    fulltext
  • 8.
    Dahlberg, John
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Explainable AI - Visualization of Neuron Functionality in Recurrent Neural Networks for Text Prediction2019Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Artificial Neural Networks are successfully solving a wide range of problems with impressive performance. Nevertheless, often very little or nothing is understood in the workings behind these black-box solutions as they are hard to interpret, let alone to explain. This thesis proposes a set of complementary interpretable visualization models of neural activity, developed through prototyping, to answer the research question ”How may neural activity of Recurrent Neural Networks for text sequence prediction be represented, transformed and visualized during the inference process to explain interpretable functionality with respect to the text domain of some individual hidden neurons, as well as automatically detect these?”. Specifically, a Vanilla and a Long Short-Term Memory architecture are utilized for character respectively word prediction as testbeds. The research method is experimental; causalities between text features triggering neurons and detected patterns of corresponding nerve impulses are investigated. The result reveals not only that there exist neurons with clear and consistent feature-specific patterns of activity, but also that the proposed models of visualization successfully may automatically detect and interpretably present some of these.

    Download full text (pdf)
    fulltext
  • 9.
    Belo, Pedro
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Heterogeneous 3D Exploration with UAVs2019Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Multi-robot exploration algorithms usually focus on exploration time minimization while ignoring map accuracy. In this thesis, it is presented a new heterogeneous multi-robot exploration strategy that finds a balance between time consumption and map accuracy. By ranking UAVs based on their sensor accuracy, it is possible to coordinate them and pick strategic points to explore rather than the most rewarding ones. In particular, with a function (in this case a Gaussian) that maps a voxel’s uncertainty to a score, it is possible to tailor a UAV’s preference (by tuning expected value and variance) towards certain features and not only unexplored space. This algorithm was compared with AEPlanner for several environments, achieving better accuracy towards complete map exploration.

    Download full text (pdf)
    fulltext
  • 10.
    Aarflot, Ludvig
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Implementation of High Current Measurement Technology for Automotive Applications in Programmable Logic2019Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    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.

    Download full text (pdf)
    fulltext
  • 11.
    Wu, Ching-An
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Investigation of Different Observation and Action Spaces for Reinforcement Learning on Reaching Tasks2019Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Deep reinforcement learning has been shown to be a potential alternative to a traditional controller for robotic manipulation tasks. Most of modern deep reinforcement learning methods that are used on robotic control mostly fall in the so-called model-free paradigm. While model-free methods require less space and have better generalization capability compared to model-based methods, they suffer from higher sample complexity which leads to the problem of sample ineffi ciency. In this thesis, we analyze three modern deep reinforcement learning, model-free methods: deep Q-network, deep deterministic policy gradient, and proximal policy optimization under different representations of the state-action space to gain a better insight of the relation between sample complexity and sample effi ciency. The experiments are conducted on two robotic reaching tasks. The experimental results show that the complexity of observation and action space are highly related to the sample effi ciency during training. This conclusion is in line with corresponding theoretical work in the field.

    Download full text (pdf)
    fulltext
  • 12.
    Du, Yipai
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Learning Dynamical Features for Vision-based Tactile Sensors2019Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Traditional tactile sensors, from resistive, capacitive to piezoelectric sensors, have not been significantly improved for years. They are not suitable to build sensing arrays as wiring can become quite complicated in large scale. Moreover the physical principles adopted are often sensitive to environmental changes like temperature, humidity, air pressure and electromagnetic field. A new trend is to leverage the recent advances in computer vision and machine learning to solve the tactile sensing problem. The approach of using cameras and machine learning models is often robust and generalizes quite well under different circumstances.In this thesis, we extend the work by Carmelo Sferrazza, "Design, Motivation and Evaluation of a Full-Resolution Optical Tactile Sensor" with a focus on dynamical data. This is the first time dynamic data and models are used and analyzed. We used the same Dense Inverse Search optical flow feature extraction as before but adapted the data processing framework to generate sequences on ETH Euler cluster to handle large amount of data. We establish three static model baselines, two of which are multilayer perceptrons and one convolutional neural network. We then successfully trained an LSTM (Long Short Term Memory) network on the given dataset. This approach shows a better capability at learning and predicting dynamical forces, but the performance is quite limited within the dataset. To make the LSTM generalize to a wide range of dynamical situations, more variations in the dataset are needed to train the network. Two newly found features of the training dataset that are essential to the dynamical model are the loading/unloading speed and force duration.

  • 13.
    Ferro, Federica
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Leveraging a service oriented architecture for automatic retrieval and processing of fault recordings to obtain information for maintenance of circuit breakers2019Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Maintenance of power system components is fundamental to ensure high qualityoperations and avoid malfunctioning. Given the crucial role of the circuitbreakers (CBs) in ensuring quality of the power systems operations, this thesisworks on the implementation of an automatic retrieval and processing offault recordings with the aim compute quantities relevant for maintenance andpreventive maintenance of the CBs. For the scope, a service oriented architecture(SOA) is developed on top of the power system and connected with twoapplications able to automatically retrieve, decode and use fault recordingsto obtain indicators on the health of the CBs. Even if the lack of a commonmetadata for fault recordings does not permit generalizations on the topic, theproject shows that the resulting layered architecture composed of power system,SOA and applications, allows to automatically obtain indicators on thestate of the CBs and consequently to improve maintenance of the analyzedarea of the substation.

    Download full text (pdf)
    fulltext
  • 14.
    Zhang, Zihan
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Modeling of VSC HVDC System for Meshed DC Grid2019Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    This thesis work deals with the modeling of a Voltage Source Converter based on High Voltage DirectCurrent (VSC-HVDC) for a meshed DC network topology. The general concept of the modeling is tointerface a tool independent control structure called ‘Common Component’ to user defined models fordifferent power system simulation tools. The control principles are developed in the language C++ andlinked into the simulation process as external functions via user model interface mechanisms. The thesisdemonstrates the modeling both in PSS/E and Power Factory.Different from the linear characteristics of a two terminal DC system, the meshed multi-terminal DC gridcontains non-linear dynamics that will play an importance role in the integration of DC grids in large ACsystems. In the first part of this report, the mathematical representation of the meshed DC grid model bothfor stability and dynamic behavior is derived. This enables the interfacing of a sequential AC/DC systempower flow algorithm, which is adapted to the multi terminal VSC HVDC system especially developedfor the meshed DC grid topology. The Gauss-Seidel algorithm is used to solve the DC power balanceequations. The differential algebraic equations (DAE) for a meshed DC terminal dynamic system arepresented and prepared for the programing interface code of the DC grid system model.The VSC-HVDC model for meshed DC network in PSS/E simulation is presented in details. In the PSS/Eload flow analysis, the VSC HVDC transmission is modeled as generic generators to represent converterswith active and reactive power levels and voltage set points. The meshed DC grid system is not explicitlyrepresented in the PSS/E built-in load flow model. The model interface is responsible for thecommunications with the control principles in ‘Common Component’ as well as the transmissionprocedure between the AC system and DC grid for steady-state and dynamic simulation.Furthermore, the network setups of VSC-HVDC transmission system as well as the meshed DC systemtopology are graphically represented through the components library in Power Factory. The convertercontrol systems are implemented as DSL models and communicate between the converters and themeasurements signals.A number of case studies, including step changes in active power reference, ac voltage reference, dcvoltage reference, and three-phase ground fault, have been used to study the system performance of theimplemented models in the simulation tools. Two meshed DC grid topologies are represented, one is threeconverters with three DC cables and the other one is five converters with seven DC cables. In theverification, the response of the proposed VSC-HVDC meshed DC network model in PSS/E is almostsame as the behavior shown in PowerFactory model. Both models performance reach satisfactory levelfor electro-mechanical simulation studies.

  • 15.
    Khays, Samir
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Motion Prediction of Surrounding Vehicles in Highway Scenarios With Deep Learning2019Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Anticipating the future positions of the surrounding vehicles is a crucial task foran autonomous vehicle in order to drive safely. To foresee complex manoeuvresfor longer time horizons, a framework that relies on high-level properties ofmotion and is able to incorporate, e.g. contextual features, is needed. In thisthesis, the problem of predicting the trajectories of the surrounding vehicles ona highway is tackled by using machine learning. The objective is to evaluate theperformance of recurrent neural networks for trajectory prediction, specificallylong-short term memory neural networks. Moreover, the goal is to investigateif contextual features can improve the predictions.The problem of predicting future trajectories is solved by using two differentapproaches, which are compared by using the same framework. The firstapproach is based on the vehicle states of the surrounding vehicles relative tothe ego-vehicle, where the reference system is in the ego-vehicle. The secondapproach is based on the velocities of the vehicles relative to the ground, wherethe reference system is in the ground. The results show that, with the proposedarchitecture, the latter approach results in a lower RMSE in the longitudinaldirection compared with the former approach. The results also show that theproposed models, overall, outperform a simple model, which is based on polynomialfitting, particularly in the lateral direction where the proposed modelsare significantly better than the polynomial models. Furthermore, contextualfeatures do not improve the predictions significantly. However, the results indicatethat contextual information has a positive impact on the predictions inspecific scenarios.

    Download full text (pdf)
    fulltext
  • 16.
    Shady Ahmed, Samy
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Navigational system for visually impaired people in a swimming pool2019Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    In this thesis conducted at IBM in Amsterdam we explore the ability Computer Vision has to assist visually impaired people in navigating a swimming pool. We examine different Computer Vision techniques and develop an algorithm to navigate a swimmer in a pool. In cooperation with a center for visually impaired people we collect a video-dataset that reflects the use-case at hand for testing, and to be able to utilize data-driven algorithms. The Computer Vision algorithm designed was implemented using Deep Learning (CNN) and statistical methods like Kalman filtering. Evaluation of the algorithm was done using both the dataset and by comparing the algorithm to the state of the art in pedestrian tracking using the MOT benchmark. The MOT benchmark was used in lack of standardized tests for tracking in pools, it provided an outlook of the algorithm’s performance in comparison to other methods. The results showed that the tracker could compete with the state of the art in pedestrian tracking as well as navigate swimmers in a pool. While the dataset needs to be expanded to perfect the algorithm, the thesis concludes that data-driven Computer Vision techniques can in a robust way navigate a swimmer in a pool with the help of statistical filtering. This is an important step to make visually impaired people more autonomous and in consequence healthier.

    Download full text (pdf)
    fulltext
  • 17.
    Tuul, Viktor
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Online Collaborative Radio-enhanced Visual-inertial SLAM2019Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Simultaneous localization and mapping (SLAM) allows robots and other devices to localize and navigate in environments by using a map which itself generates. SLAM for single agent applications has matured and is showing promising results, thus the interest for collaborative SLAM has increased.This thesis proposes a framework for online collaborative radio-enhanced visual-inertial (VI) SLAM where multiple agents can collaborate by having their individually built maps merged and shared amongst each other. The framework is centralized with the aim to allow multiple agents to be managed by a single machine, also rendering it feasible to use the framework with agents that have limited computational resources, e.g. nano drones. Furthermore, radio technology is implemented in the framework which augments the SLAM solution by fusing ultra-wideband (UWB) anchor information into the built maps. This enables agents to query relevant parts of potentially large maps based on their contemporary radio activity.Four individual experiments are conducted to thoroughly evaluate the proposed solution. The results show that the collaborative SLAM system successfully allow agents to localize on parts of a map that other agents have built, running simultaneously. Moreover, the results also show that fusing UWB information into a visual-inertial map allow agents to perform partial-map queries, restricting the search area for visual matches between camera images and the map, reducing the risk of false re-localizations.

    Download full text (pdf)
    fulltext
  • 18.
    Masso, Gabriel
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Optimising energy consumption on straight roads using regression analysis2019Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Cloud computation together with robotics has opened up possibilitiesto process large amount of data (big data) generated by the greatnumber of robotic systems. Todays vehicles are equipped withhundreds of sensors generating a lot of data that needs to beprocessed. The data can further be analysed and used to obtainmodels predicting the dynamics of the vehicles. It is thereforepossible to optimise the vehicle performance by studying thepredictive behaviour and finding the best combination of the vehicleparameters. In this thesis, the energy efficiency of an electric racingvehicle is studied on straight road whereafter an optimal velocityprofile is to be found. By using a multiple linear regression togetherwith regularization methods on previously recorded data, apredictive model managed to be obtained with an accuracy of 79.1 %.Having used this model in optimisation process, a velocity profilewas obtained which is shown that can enhance the efficiency of thesystem by 4.08%.

    Download full text (pdf)
    fulltext
  • 19.
    Hongisto, Jonas
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Performance Analysis of Elastic Band Based Time Optimal Control Formulations for Industrial Robots2019Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Industrial robots are becoming an integral part of the production industry. Efficientoperation with respect to fast movements is critical to increase the economicbenefits of automating the production line. Facilitating near optimalitywith regards to time has high computational demands however and multipleframeworks have been suggested to remedy this. In this thesis we consider oneof these frameworks, namely the elastic band framework. We investigate howthe elastic band time optimal control framework performs regarding computationaltime for point-to-point movements on a SCARA type robot with threerevolute and one prismatic joint. We compare an unconstrained elastic bandformulation with a constrained formulation in the open loop, along with simulatingperformance in the closed-loop. We show that a constrained formulationwhich considers the sparseness of underlying matrices in the optimizationproblem has the lowest computational time. Additionally, we show that the unconstrainedformulation benefits from early stopping. Finally, we show that acontroller implementing this formulation can be used in a model predictivecontroller, although the computational time is still too high for commercialuse on the hardware used in testing.

    Download full text (pdf)
    fulltext
  • 20.
    Lindståhl, Simon
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Reinforcement Learning with Imitation for Cavity Filter Tuning: Solving problems by throwing DIRT at them2019Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Cavity filters are vital components of radio base stations and networks.After production, they need tuning, which has proven to be a difficultprocess to do manually and even more so to automate. Previously, attemptsto automate this process with Reinforcement Learning have beenmade but have failed to reach consistent performance on anything butthe simplest filter models. This Master thesis builds upon these resultsand aims to improve them. Multiple methods are tested and evaluated,including introducing a pre-processing step, tuning hyperparameters anddividing the problem into multiple sub-tasks. In particular, by using Imitationlearning as an initial phase, a semi-realistic filter model with 13tuning screws is tuned, fulfilling both insertion loss and return loss requirements.On this problem, this algorithm has a greater efficiency thanany previously published results on Reinforcement Learning for Cavityfilter tuning.

    Download full text (pdf)
    fulltext
  • 21.
    Jiang, Frank
    KTH, School of Electrical Engineering and Computer Science (EECS).
    The Automated Vehicle Traffic Control Tower2019Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    This thesis presents and motivates the integration of automated vehicle traffic control towers (AVTCT) into transportation networks to support the operation of automated driving systems. Loosely analogous to air traffic control towers, the AVTCT provides a human-driven exception-handling layer to automated driving systems for safe operation in unexpected or uncertain, safety-critical scenarios.  In the thesis, a case-study of plausible scenarios is first presented that either pose difficulty for automated systems or require the supervision of human operators. An outline of the AVTCT concept and its different potential roles in the support of automated driving systems is then discussed.  Using reachability analysis, a concrete example from the case-study is formulated to show how the AVTCT would support an automated driving system that has reached an infeasibility in its automation. Finally, a demonstration of the example using an experimental implementation of the AVTCT and a scaledvehicle is presented.

    Download full text (pdf)
    fulltext
  • 22.
    Yokobori Sävö, Andreas
    KTH, School of Electrical Engineering and Computer Science (EECS).
    User Plane Selection for Core Networks using Deep Reinforcement Learning2019Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Allocating service functions to a core network upon users’ various demands isof importance in 5G networks. In this thesis work, we have studied reinforcementlearning models to solve this allocation problem. More precisely, 1) webuild a simple version of an MDP model for allocation in 5G core networks,2) we train an agent using a family of deep-Q learning (DQN) algorithms.When the number of nodes in the core network is large, one critical challengeis overcoming the sampling inefficiency due to a high dimensional actionspace, i.e., most of the exploratory allocations made by the agent gives zeroreward. To deal with such reward sparsity, we applied two techniques: prioritizedexperience replay (PER) and hindsight experience replay (HER).Our study shows that a DQN agent trained with both HER and PER providesa reasonable allocation in a larger sized networks, whereas a vanillaDQN agent works only for a very limited case where the number of nodesis small.

    Download full text (pdf)
    fulltext
  • 23.
    Stefansson, Thor
    KTH, School of Electrical Engineering and Computer Science (EECS), Robotics, perception and learning, RPL.
    3D obstacle avoidance for drones using a realistic sensor setup2018Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Obstacle avoidance is a well researched area, however most of the works only consider a 2D environment. Drones can move in three dimensions. It is therefore of interest to develop a system that ensures safe flight in these three dimensions. Obstacle avoidance is of highest importance for drones if they are intended to work autonomously and around humans, since drones are often fragile and have fast moving propellers that can hurt humans. This project is based on the obstacle restriction algorithm in 3D, and uses OctoMap to conveniently use the sensor data from multiple sensors simultaneously and to deal with their limited field of view. The results show that the system is able to avoid obstacles in 3D.

    Download full text (pdf)
    3D obstacle avoidance for drones using a realistic sensor setup
  • 24.
    Peng, Ming
    KTH, School of Electrical Engineering and Computer Science (EECS).
    A Game Theoretic Approach to Power Control in Vehicula rCommunication Systems2018Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Vehicular-to-network (V2N) communications are a key enabler in various intelligenttransportation system services currently investigated by global standards bodiesand European projects. In this master thesis, we study the impact of pilot and datapower setting on the uplink performance of V2N communications. Specifically, weconsider the “urban” and “rural” scenarios of the 5GCAR project, using the recommendedbase station and vehicular user equipment parameters in the 5.9 GHz band.We study a distributed non-cooperative game theoretic algorithm to determine thepilot-data power ratio for each vehicle. Numerical results indicate that the proposeddistributed algorithm converges to the Nash equilibrium, and that the mean squarederror (MSE) of the received data symbols decreases by 0.6 − 1.2 dB as comparedwith the MSE obtained by a centralized benchmark algorithm. The proposed algorithmachieves this improvement over a centralized benchmarking algorithm at theexpense of only a few iterations.

    Download full text (pdf)
    fulltext
  • 25.
    Lin, Tengfan
    KTH, School of Electrical Engineering and Computer Science (EECS).
    A GUI Design of Robot Motion and Task Planning Based on Linear Temporal Logic2018Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Recent approaches solve the problem of robot motion and task planningby using formal methods-based model checking algorithms. Inthis work, we consider the software package P-MAS-TG, an automatictool to generate correct-by-design controllers for robot motion and taskplanning. The robot motion can be modelled as a finite-state transitionsystem and the task can be represented by a linear temporal logic formula.Then, using a model checking algorithm, an accepting path forthe robot motion can be found so that the robot satisfies the specificlinear temporal logic task. In this thesis, the process of searching forthe accepting path of the robot motion by P-MAS-TG is visualized bymeans of a graphical user interface (GUI) in form of an rqt plug-inwhere all requested inputs for the P-MAS-TG package can be definedin the GUI so that the process is simplified and user-friendly. Moreover,the GUI provides a simple hybrid control mechanism betweenthe initial linear temporal logic task and a temporary task that is insertedby a human.All functions of the GUI are implemented and demonstrated onthe TurtleBot platform. The experiment shows that the GUI and theP-MAS-TG are successfully integrated. Features of the P-MAS-TG areperformed with the GUI to conduct motion planning and task planningon the TurtleBot.

    Download full text (pdf)
    fulltext
  • 26.
    Nilsson, Mao-Wei
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Autonomous Docking and Navigation of ships using Model Predictive Control2018Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Autonomous shipping is a coming field, where it will be important tooperate a ship without manual intervention. Although there are manyissues yet to be solved, not the least the legal ones, it would be interestingto investigate functions that already now would be possibleto use in today’s ship operation. One such field is autonomous navigationin narrow areas. The purpose of this study is to implement amotion control system to navigate marine vessel autonomously, and aGuidance, Navigation and Control system (GNC) is implemented fordocking and navigating vessels. Voronoi diagram is used for generatinga waypoint list for waypoint tracking. MPC with integral actionis applied to control the vessel for reducing model mismatches andconstant disturbance from current and wind. We performed the GNCsystem for South Harbour of Helsinki, and shown that the vessel isnavigated and docked at port. Moreover, we studied the effects of disturbanceto keep the controller stabilized and suggested an upper limitfor the disturbance.

    Download full text (pdf)
    fulltext
  • 27.
    Choudhary, Abhishek
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Autonomous Exploration and Data Gathering with a Drone2018Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Unmanned Aerial Vehicles (UAV) are agile and are able to fly in and out of areas that are either dangerous for humans or have complex terrains making ground robots unsuitable. For their autonomous operation, the ability to explore unmapped areas is imperative. This has applications in data gathering tasks, search and rescue etc. 

    The objective of this thesis is to ascertain that it is, in fact, possible and feasible to use UAVs equipped with 2D laser scanners to perform autonomous exploration tasks in indoor environments. The system is evaluated by testing it in different simulated and real environments. The results presented show that the system is capable of completely and safely exploring unmapped and/or unexplored regions.

    Download full text (pdf)
    fulltext
  • 28.
    Mattsson, Filip
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Centralized Model Predictive Control of a Vehicle Platoon2018Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    A centralized model predictive controller for longitudinal control of a vehicleplatoon is designed based on previous work on a distributed platooningcontroller. The vehicles in the platoon track a varying speed referenceand a constant timegap to the preceding vehicle. The designed controller isimplemented and compared to the distributed controller in simulation experimentsand in a simple practical setup. The experiments show that thecontroller works but does in general not outperform the distributed controller.

    Download full text (pdf)
    fulltext
  • 29.
    Roussel, Nicolas
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Denoising of Dual Energy X-ray Absorptiometry Images and Vertebra Segmentation2018Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Dual Energy X-ray Absorptiometry (DXA) is amedical imaging modality used to quantify bone mineral density and to detect fractures. It is widely used due to its cheap cost and low radiation dose, however it produces noisy images that can be difficult to interpret for a human expert or a machine. In this study, we investigate denoising on DXA lateral spine images and automatic vertebra segmentation in the resulting images. For denoising, we design adaptive filters to avoid the frequent apparition of edge artifacts (cross contamination), and validate our results with an observer experiment. Segmentation is performed using deep convolutional neural networks trained on manually segmented DXA images. Using few training images, we focus on depth of the network and on the amount of training data. At the best depth, we report a 94 % mean Dice on test images, with no post-processing. We also investigate the application of a network trained on one of our databases to the other (different resolution). We show that in some cases, cross contamination can degrade the segmentation results and that the use of our adaptive filters helps solving this problem. Our results reveal that even with little data and a short training, neural networks produce accurate segmentations. This suggests they could be used for fracture classification. However, the results should be validated on bigger databases with more fracture cases and other pathologies.

    Download full text (pdf)
    fulltext
  • 30.
    Lotz, Max
    KTH, School of Electrical Engineering and Computer Science (EECS), Robotics, perception and learning, RPL.
    Depth Inclusion for Classification and Semantic Segmentation2018Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    The  majority  of  computer  vision  algorithms  only  use  RGB  images  to  make  inferencesabout  the  state  of  the  world.  With  the  increasing  availability  of  RGB-D  cameras  it  is  im-portant  to  examine  ways  to  effectively  fuse  this  extra  modality  for  increased  effective-ness.  This  paper  examines  how  depth  can  be  fused  into  CNNs  to  increase  accuracy  in  thetasks  of  classification  and  semantic  segmentation,  as  well  as  examining  how  this  depthshould  best  be  effectively  encoded  prior  to  inclusion  in  the  network.  Concatenating  depthas  a  fourth  image  channel  and  modifying  the  dimension  of  the  initial  layer  of  a  pretrainedCNN  is  initially  examined.  Creating  a  separate  duplicate  network  to  train  depth  on,  andfusing  both  networks  in  later  stages  is  shown  to  be  an  effective  technique  for  both  tasks.The  results  show  that  depth  concatenation  is  an  ineffective  strategy  as  it  clamps  the  ac-curacy  to  the  lower  accuracy  of  the  two  modalities,  whilst  late  fusion  can  improve  thetask  accuracy  beyond  that  of  just  the  RGB  trained  network  for  both  tasks.  It  is  also  foundthat  methods  such  as  HHA  encoding  which  revolve  around  calculating  geometric  prop-erties  of  the  depth,  such  as  surface  normals,  are  a  superior  encoding  method  than  sim-pler  colour  space  transformations  such  as  HSV.  This  only  holds  true  when  these  depthimages  are  normalised  over  the  maximum  depth  of  the  dataset  as  opposed  to  the  maxi-mum  depth  of  each  individual  image,  thus  retaining  geometric  consistency  between  im-ages.  The  reverse  holds  true  for  simpler  colour  space  transformations.

    Download full text (pdf)
    fulltext
  • 31.
    Pontusson, Magnus
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Design and Implementation of the SAX, a Robotic Measurement System for On-Chip Antennas at 140-325 GHz2018Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    There is currently a demand of mm‑wave on‑chip antennas to enable all kinds of new applications in several different areas. But the development requires, among other things, special equipment used during the measurement phase due to the small dimensions and the high frequencies.

    In this project a robotic measurement system, SAX (Single Arm eXtra), was designed and constructed at Micro and Nanosystems (MST) department at KTH Royal Institute of Technology (Sweden). The purpose of the SAX is to enable radiation pattern measurements of on‑chip antennas ( 140 GHz to 325 GHz ), whether the boresight is vertical or horizontal along with other requirements, by moving a converter with the measurement antenna around the antenna in question.

    Several alternative designs for the basic construction, both from other works and invented by the author, were analyzed based on the requirements for this project and other limitations. The chosen unique design, the SAX, is very compact and uses only one stepper motor. Several parts have been developed in this project to ensure the proper functionality of the SAX. That includes a main operator program, a motor input signal generating program, a motor input signal executing system, a security system, and a system for controlled rotation of the SAX. For the input signal to the motor two different algorithms to generate the time delays were developed and tested. They were adapted to make the motor manage the sweeps of an ever‑changing load with high inertia during acceleration and deceleration. One of them was developed to make the time delay array generation much more efficient albeit with larger approximation error.

    The SAX worked well and should be rather easy‑to‑use regarding the operation of the system, from the physical maneuvering to utilizing the sub‑systems to the running of the main operator program. It fulfilled the specific requirements by enable a cross pattern measurement from  -60° to +60°  both from above and from the side, adjustment of the radius between 15cm to 45cm , adjustment 10cm in height, to be rotated along the floor in steps of 1°, measurement steps of 1° with an accuracy of less than 0,5° (the largest error was measured to be ≤ 0,461°). However, some calibration work needs to be done before the optimal performance of the system is reached.

    As a verification of the operation of the system data from measurements of open‑ended waveguides was presented.

    Download full text (pdf)
    fulltext
  • 32.
    Shilo, Albina
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Detection and tracking of unknown objects on the road based on sparse LiDAR data for heavy duty vehicles2018Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Environment perception within autonomous driving aims to provide a comprehensive and accurate model of the surrounding environment based on information from sensors. For the model to be comprehensive it must provide the kinematic state of surrounding objects. The existing approaches of object detection and tracking (estimation of kinematic state) are developed for dense 3D LiDAR data from a sensor mounted on a car. However, it is a challenge to design a robust detection and tracking algorithm for sparse 3D LiDAR data. Therefore, in this thesis we propose a framework for detection and tracking of unknown objects using sparse VLP-16 LiDAR data which is mounted on a heavy duty vehicle. Experiments reveal that the proposed framework performs well detecting trucks, buses, cars, pedestrians and even smaller objects of a size bigger than 61x41x40 cm. The detection distance range depends on the size of an object such that large objects (trucks and buses) are detected within 25 m while cars and pedestrians within 18 m and 15 m correspondingly. The overall multiple objecttracking accuracy of the framework is 79%.

    Download full text (pdf)
    fulltext
  • 33.
    Drollinger, Nadine
    KTH, School of Electrical Engineering and Computer Science (EECS), Automatic Control.
    Developing a System for Robust Planning using Linear Temporal Logic2018Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Human robot-collaborative search missions have gotten more and more attention in recent years.Especially in scenarios where the robot first scouts the scene before sending in human agents. Thissaves time and avoids unnecessary risks for the human agents. One possible configuration of such arescue team is, a human operator instructing a unmanned aerial vehicle (UAV) via speech-commandshow to traverse through an environment to investigate areas of interest. A first step to address thisproblem is presented in this master thesis by developing a framework for mapping temporal logicinstructions to physical motion of a UAV.The fact that natural language has a strong resemblance to the logic formalism of Linear-TemporalLogic (LTL) is exploited. Constraints expressed as an LTL-formula are imposed on a provided labeledmap of the environment. An LTL-to-cost-map converter including a standard input-skeleton is developed.Respective cost maps are obtained and a satisfaction-measure of fulfilling these constraints ispresented. The input-skeleton and the map-converter are then combined with a cost-map-based pathplanning algorithm in order to obtain solution sets. A clarification request is created such that theoperator can choose which solution set should be executed. The proposed framework is successivelyvalidated, first by MATLAB-experiments to ensure the validity of the cost-map-creation followed bysimulation experiments in ROS incorporating the entire framework. Finally, a real-world experimentis performed at the SML to validate the proposed framework.

    Download full text (pdf)
    fulltext
  • 34.
    Maloo, Shreyans
    KTH, School of Electrical Engineering (EES), Automatic Control.
    Developing a voice-controlled home-assisting system for KTH Live-in Labs2018Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    The following master thesis is conducted on behalf of KTH Royal Instituteof Technology and KTH Live-in Lab with the purpose of developing avoice-controlled home-assisting system for the KTH Live-in Lab. The labis being designed to serve as a testbed for products and services that canbe tested and veried within an optimal space which can simulate a reallifeusage. Being designed as a bridge between industry and academia, itaims to create a greater ease to which new products are tested and are researchedwhile involving KTH students in the process. Having innovationat its core the KTH Live-in Lab needs a mode of communication betweenthe user/occupant and the appliances in the space. That is why this thesisproposes to design a voice-controlled system that can control the appliancesand execute the commands provided by the user. The system will be createdaround a Speech to text service and improving its performance through variousmodications/integrations. The solution will be installed in the KTHLive-in Lab and integrated with the central controller once the sensor placementand controller installation is done.To make the system more robust and accurate, a new variable called,\Failure Factors" were dened for a voice-controlled system. The prototypeswere designed and improved with these factors as a basis. The main aimof the project is to make a system capable of handling a set of pre-denedsimple commands. For testing purpose, only 3 appliances were considered {light, heater and music. Also, the output is observed on LEDs rather thanon real appliances for the testing. These limitations were adapted to keepour focus on the prime motive of this project and that was to make the voicerecognitionas consistent and accurate as possible. Future work will consist ofmaking the system capable of handling complex user commands and havingan active feedback mechanism such that the user can have conversation withthe system.

    Download full text (pdf)
    fulltext
  • 35.
    Elanjimattathil Vijayan, Aravind
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Dynamic Locomotion of Quadrupedal Robots over Rough Terrain2018Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Previous works have enabled locomotion of quadrupedal robots usingthe ZMP-based motion optimization framework on flat terrain withvarious gait patterns. Locomotion over rough terrain brings in newchallenges such as planning safe footholds for the robot, ensuring kinematicstability during locomotion and avoiding foot slippage over roughterrain etc. In this work, terrain perception is integrated into the ZMPbasedmotion optimization framework to enable robots to perform dynamiclocomotion over rough terrain.In a first step, we extend the foothold optimization framework touse processed terrain information to avoid planning unsafe footholdpositions while traversing over rugged terrain. Further, to avoid kinematicviolations during locomotion over rugged terrain, we presentadditional constraints to the ZMP-based motion optimization frameworkto solve for kinematically feasible motion plans in real-time. Weadd nonlinear kinematic constraints to existing nonlinear ZMP motionoptimization framework and solve a Sequential Quadratic Programming(SQP) problem to obtain feasible motion plans. Lastly, to avoidfoot contact slippage, we drop the approximated terrain normal anduse measured terrain normal at foot contact position to compute thefriction polygon constraints.The proposed algorithms are tested in simulation and on hardwarewith dynamic gaits to validate the effectiveness of this approach toenable quadrupedal robots to traverse rugged terrain safely. The computationaltime and performance of the proposed algorithms were analyzedunder various scenarios and presented as part of this thesis.

    Download full text (pdf)
    fulltext
  • 36.
    Eriksson, Urban
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Dynamic Path Planning for Autonomous Unmanned Aerial Vehicles2018Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    This thesis project investigates a method for performing dynamic path planning in three dimensions, targeting the application of autonomous unmanned aerial vehicles (UAVs).  Three different path planning algorithms are evaluated, based on the framework of rapidly-exploring random trees (RRTs): the original RRT, RRT*, and a proposed variant called RRT-u, which differs from the two other algorithms by considering dynamic constraints and using piecewise constant accelerations for edges in the planning tree. The path planning is furthermore applied for unexplored environments. In order to select a path when there are unexplored parts between the vehicle and the goal, it is proposed to test paths to the goal location from every vertex in the planning graph to get a preliminary estimate of the total cost for each partial path in the planning tree. The path with the lowest cost given the available information can thus be selected, even though it partly goes through unknown space. For cases when no preliminary paths can be obtained due to obstacles, dynamic resizing of the sampling region is implemented increasing the region from which new nodes are sampled. This method using each of the three different algorithms variants, RRT, RRT*, and RRT-u, is tested for performance and the three variants are compared against each other using several test cases in a realistic simulation environment.  Keywords

    Download full text (pdf)
    fulltext
  • 37.
    Thai Do, Hoang
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Energy Management of Parallel Hydraulic Hybrid Wheel Loader2018Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Hybridization of driveline system is one possible solution to increase fuel eciency.In this thesis a parallel hybrid hydraulic wheel loader concept was studied. A highpressure accumulator was added to the system and acted as a second source of energy.By adding the high pressure accumulator, regenerative braking energy canbe stored for later utilization. A backward facing simulation model was developedwhere the high pressure accumulator's State Of Charge (SOC) as state variableand hydraulic pump/motor's displacement as control input. Furthermore, dierentenergy management strategies: Dynamic Programming (DP), rule-based andEquivalent Consumption Minimization Strategy (ECMS) were developed. Thesestrategies were evaluated and compared to each other all with respect to the fuelconsumption. The result from conventional machine acted as the benchmark forother strategies to compare with. From simulation results, rule-based strategiesshowed to be the most robust, resulted in lower fuel consumption in every testeddriving cycle. For ECMS, the performance varied from cycle to cycle. A reductionin fuel consumption was observed for short-loading cycles. Especially in one cycle,ECMS result outclassed rule-based and was almost the same as DP. However, asmall increment was observed for long-carry cycle. Here the introduction of lock-upfeature in the torque converter yielded instead the most fuel saving. These valuableconclusions acted perfectly as a good starting point for future product development.

    Download full text (pdf)
    fulltext
  • 38.
    Chintha, Cheerudeep
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Facilitating Automatic Setup in a Robotised Test Framework for Autonomous Vehicles by Path Planning and Real-Time Trajectory Generation2018Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    The research in the field of autonomous vehicles and self-driving carsis growing at a rapid pace and strong initiatives are being taken to verifythe capability and functionality of such autonomous vehicles.Withcontinuous development being carried out in the field of AdvancedDriver Assist Systems (ADAS) and Autonomous Drive (AD) functions,ensuring safety, robustness and reliability of these functions is challengingand it requires advanced ways of verification and testing beforethese functions are deployed on the vehicle and delivered to thecustomer. Testing of these modern features can be done either on testtrack, real driving roads or in simulations by Computer Aided Engineering(CAE) . But testing a high-risk scenario in the real-worldwould be challenging due to safety concerns. Also, high regressionand continuous testing requires a test framework where the developmentand testing can be done in an efficient way.At Volvo Cars, it is envisioned that the best approach to test theAD vehicles is by subjecting the vehicle under test to several high riskscenarios by simulation based engineering and replicate the subset ofthese tests on a closed-loop test framework developed on the test track.This thesis is a part of FFI Funded Research Project called CHRONOS2where Volvo Car Corporation and other project partners aim to developthe closed-loop test framework for verification of AD Vehicles.This thesis work focuses on ensuring efficient and reproducible testingin the said test framework by accurate path planning and trajectorygeneration to drive the multiple test objects to their starting positionsin an unstructured test environment. The algorithm developedfor path planning should also ensure the generation of a safe path inreal-time for the test objects in case of failure or error in the test framework.The path-planning algorithm has been successfully implementedtaking the unstructured environment and vehicle dimensions into considerationresulting in a safe path avoiding obstacles and satisfyingnonholonomic constraints of the vehicle. The implemented architectureutilizes the parallel-process framework of Robot Operation System(ROS) and results in a algorithm which can run in real-time.

    Download full text (pdf)
    fulltext
  • 39.
    Borsub, Jatesada
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Hardened Registration Process for Participatory Sensing2018Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Participatory sensing systems need to gather information from a largenumber of participants. However, the openness of the system is a doubleedgedsword: by allowing practically any user to join, the system can beabused by an attacker who introduces a large number of virtual devices.This work proposes a hardened registration process for participatory sensingto raise the bar: registrations are screened through a number of defensivemeasures, towards rejecting spurious registrations that do not correspondto actual devices. This deprives an adversary from a relatively easytake-over and, at the same time, allows a flexible and open registrationprocess. The defensive measures are incorporated in the participatorysensing application.

    Download full text (pdf)
    fulltext
  • 40.
    Elfving, Maria
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Hydraulic closed loop control2018Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    The purpose of the thesis work is to investigate methods for closedloop control of hydraulic pressure in transmissions to make them bemore precise. This is desirable since it decreases the fuel consumptionas well as emissions, and improves the driving performance.To be able to study the behaviour of the transmission, a Simulink modelis designed with the parts relevant to the problem, and from this a linearmodel is obtained. Three different controllers are designed andimplemented in the Simulink model, to compare and analyze differentsolutions. The controllers implemented are a PI controller, a PIDcontroller and a LQR controller.The results from the simulation with the different controllers showstep responses to be able to evaluate their individual performance. Theresults show that all of the controllers meet the requirements for a stepreponse under better conditions, but under worse ones the LQR controllerperforms best of the three. The LQR controller is therefore themost suitable of the three controllers for this particular problem.

    Download full text (pdf)
    fulltext
  • 41.
    Liu, Feiyang
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Implementation and verification of the Information Bottleneck interpretation of deep neural networks2018Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    Although deep neural networks (DNNs) have made remarkable achievementsin various elds, there is still not a matching practical theory that is able toexplain DNNs' performances. Tishby (2015) proposed a new insight to analyzeDNN via the Information bottleneck (IB) method. By visualizing how muchrelevant information each layer contains in input and output, he claimed thatthe DNNs training is composed of tting phase and compression phase. Thetting phase is when DNNs learn information both in input and output, andthe prediction accuracy goes high during this process. Afterwards, it is thecompression phase when information in output is preserved while unrelatedinformation in input is thrown away in hidden layers. This is a tradeo betweenthe network complexity (complicated DNNs lose less information in input) andprediction accuracy, which is the same goal with the IB method.In this thesis, we verify this IB interpretation rst by reimplementing Tishby'swork, where the hidden layer distribution is approximated by the histogram(binning). Additionally, we introduce various mutual information estimationmethods like kernel density estimators. Based upon simulation results, we concludethat there exists an optimal bound on the mutual information betweenhidden layers with input and output. But the compression mainly occurs whenthe activation function is \double saturated", like hyperbolic tangent function.Furthermore, we extend the work to the simulated wireless model where thedata set is generated by a wireless system simulator. The results reveal that theIB interpretation is true, but the binning is not a correct tool to approximatehidden layer distributions. The ndings of this thesis reect the informationvariations in each layer during the training, which might contribute to selectingtransmission parameter congurations in each frame in wireless communicationsystems.

    Download full text (pdf)
    fulltext
  • 42.
    Lewenhaupt, Johan
    KTH, School of Electrical Engineering and Computer Science (EECS).
    Learning Operational Goals for Propulsion System Using Reinforcement Learning2018Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
    Abstract [en]

    This degree project, conducted at ABB, aims to analyze and solve differentsituations that a crew on board a vessel might face by controllingits propulsion system. The propulsion system is viewed as static,transition-deterministic, as well as stochastic when measuring data.This system is then used to formulate a decision problem using a finiteMarkov Decision Process, which is attempted to be tackled usingQ-learning, Speedy Q-learning and Double Q-learning for three differentobjectives that are relevant to the system’s behaviour and performance.The objective policie