1,721,035 research outputs found
Resolution Enhancement of Polar Ice Core Micro CT Scans via Deep Learning: A Comparative Study
No full textAccurately acquiring microstructure parameters of porous materials is key to microstructure analysis. Polar ice cores are frozen time capsules holding invaluable information about past climates of our planet in their microstructure. Via micro CT scanning it is possible to capture ice core microstructure parameters, however, scanning the whole firn column and shallow bubbly ice ( ) with high-resolution (HR) is practically impossible and low-resolution (LR) scans lead to unacceptable errors in microstructure analysis. Therefore, enhancing the resolution before analysis is essential. First, we scanned the selected specimens two times (LR m, and HR m). A data set of LR and HR scans was generated and aligned with image registration. Next, several SOTA deep learning models (SRCNN, SRUnet, DCSRN, SRResnet) were selected and modified for 3D microstructure super-resolution (SR) tasks. Finally, these models were tested on samples with various porosity against the HR data. Models were compared on both pixel-wise metrics (e.g. PNSR, SSI) and microstructure parameters (e.g. SSA, Permeability, Tortuosity). It was shown that SR models can detect tiny channels in the firn samples. Also, it was found that bubbly ice has the most difficult microstructure for resolution enhancement and SR models could improve bubble identification by 15%. Among models, SRResnet had the best performance on all metrics. These SR models improved the image quality (12% PSNR), leading to more accurate porosity (57% in ice), Euler number (10% in firn), and Tortuosity (36% in snow). Also, the estimated weight had less than 1% error compared to the measured weights of samples
Design und Aufbau eines hominiden Roboters, welcher zur Verbesserung seiner Lokomotionsfähigkeiten mit lokalen Kontrollreglern in seinen Fußstrukturen ausgestattet ist
No full textWith increasing mechanization of our daily lives, the expectations and demands in robotic systems increase in the general public and in scientists alike. In recent events such as the Deepwater Horizon''-accident or the nuclear disaster at Fukushima, mobile robotic systems were used, e.g., to support local task forces by gaining visual material to allow an analysis of the situation. Especially the Fukushima example shows that the robotic systems not only have to face a variety of different tasks during operation but also have to deal with different demands regarding the robot's mobility characteristics. To be able to cope with future requirements, it seems necessary to develop kinematically complex systems that feature several different operating modes. That is where this thesis comes in: A robotic system is developed, whose morphology is oriented on chimpanzees and which has the possibility due to its electro-mechanical structure and the degrees of freedom in its arms and legs to walk with different gaits in different postures. For the proposed robot, the chimpanzee was chosen as a model, since these animals show a multitude of different gaits in nature. A quadrupedal gait like crawl allows the robot to traverse safely and stable over rough terrain. A change into the humanoid, bipedal posture enables the robot to move in man-made environments. The structures, which are necessary to ensure an effective and stable locomotion in these two poses, e.g., the feet, are presented in more detail within the thesis. This includes the biological model and an abstraction to allow a technical implementation. In addition, biological spines are analyzed and the development of an active, artificial spine for the robotic system is described. These additional degrees of freedom can increase the robot's locomotion and manipulation capabilities and even allow to show movements, which are not possible without a spine. Unfortunately, the benefits of using an artificial spine in robotic systems are nowadays still neglected, due to the increased complexity of system design and control. To be able to control such a kinematically complex system, a multitude of sensors is installed within the robot's structures. By placing evaluation electronics close by, a local and decentralized preprocessing is realized. Due to this preprocessing is it possible to realize behaviors on the lowest level of robot control: in this thesis it is exemplarily demonstrated by a local controller in the robot's lower leg. In addition to the development and evaluation of robot's structures, the functionality of the overall system is analyzed in different environments. This includes the presentation of detailed data to show the advantages and disadvantages of the local controller. The robot can change its posture independently from a quadrupedal into a bipedal stance and the other way around without external assistance. Once the robot stands upright, it is to investigate to what extent the quadrupedal walking pattern and control structures (like the local controller) have to be modified to contribute to the bipedal walking as well
Maschinelles Lernen für Gang-Klassifikation
No full textMachine learning is a powerful tool for making predictions and has been widely used for solving various classification problems in last decades. As one of important applications of machine learning, gait classification focuses on distinguishing different gait patterns by investigating the quality of gait of individuals and categorizing them as belonging to particular classes. The most studied gait pattern classes are the normal gait patterns of healthy people, i.e., gait of people who do not have any gait disability caused by an illness or an injury, and the pathological gait of patients suffering from illnesses which cause gait disorders such as neurodegenerative diseases (NDDs). There has been significant research work trying to solve the gait classification problems using advanced machine learning techniques, as the results may be beneficial for the early detection of underlined NDDs and for the monitoring of the gait rehabilitation progress. Despite the huge development in the field of gait analysis and classification, there are still a number of challenges open to further research. One challenge is the optimization of applied machine learning strategies to achieve better classification results. Another challenge is to solve gait classification problems even in the case when only limited amount of data are available. Further, a challenge is the development of machine learning-based methods that could provide more precise results to evaluate the level of gait quality or gait disorder, in contrast of just classifying gait pattern as belonging to healthy or pathological gait. The focus of this thesis is on the development, implementation and evaluation of a novel and reliable solution for the complex gait classification problems by addressing the current challenges. This solution is presented as a classification framework that can be applied to different types of gait signals, such as lower-limbs joint angle signals, trunk acceleration signals, and stride interval signals. Developed framework incorporates a hybrid solution which combines two models to enhance the classification performance. In order to provide a large number of samples for training the models, a sample generation method is developed which could segments the gait signals into smaller fragments. Classification is firstly performed on the data sample level, and then the results are utilized to generate the subject-level results using a majority voting scheme. Besides the class labels, a confidence score is computed to interpret the level of gait quality. In order to significantly improve the gait classification performances, in this thesis a novel feature extraction methods are also proposed using statistical methods, as well as machine learning approaches. Gaussian mixture model (GMM), least square regression, and k-nearest neighbors (kNN) are employed to provide additional significant features. Promising classification results are achieved using the proposed framework and the extracted features. The framework is ultimately applied to the management of patients and their rehabilitation, and is proved to be feasible in many clinical scenarios, such as the evaluation of medication effect on Parkinsona s disease (PD) patientsa gait, the long-term gait monitoring of the hereditary spastic paraplegia (HSP) patient under physical therapy
Image Registration in Centrifuge-Based Adhesive Fracture Testing
No full textThis thesis investigates the application of image registration techniques in centrifuge-based adhesive fracture testing, aiming to develop a framework for fully automated analysis of adhesive joints. Adhesive bonding is widely used in various industries, yet the evaluation of adhesive performance remains challenging, particularly in aligning fractured surfaces for analysis. The proposed framework utilizes image registration algorithms to automate the alignment process, enhancing the accuracy and efficiency of fracture pattern classification. A comprehensive evaluation of different algorithms shows that gradient descent-based optimization methods, particularly those utilizing Mutual Information (MI) as a distance measure, yield the most reliable results. This work not only contributes to the automation of fracture analysis but also provides insights into the integration of image registration techniques into standard adhesive testing procedures, ultimately promoting the reliability and consistency of adhesive evaluations
Underwater Visual Multi-Modal 3D Sensing
No full textIn the underwater domain, optical sensors are extremely limited with respect to range, resolution, and accuracy in comparison to most terrestrial remote sensors. The reason for this is the medium water, which heavily interacts with electromagnetic signals and therefore reduces their corresponding signal-to-noise ratio. Also, many underwater areas can only be visited by remotely operated vehicles due to water pressure, turbidity, and or strong currents, posing a high risk for humans. This combination considerably increases the complexity of underwater metrology, and many applications currently require highly skilled personnel and large support vessels. Here, a simplification of these applications is presently effectively prevented by the performance gap of underwater optical systems in comparison to their terrestrial counterparts.
Motivated by the above limitations, this research work evaluates different optical sensing modalities when applied to the underwater domain and identifies their possible sweet spots. Based on these considerations, several novel fusion strategies for passive-active optical systems are presented able to reconstruct whole underwater scenes with high accuracy without relying on additional navigation systems. For their evaluation, a novel self-referenced optical 3D underwater scanner is implemented and used for several test setups as well as real-world scenarios. The implementation also includes a novel method for in-air calibration of flat-port cameras and integration into bundle adjustment frameworks for visual pose estimation. Here, the evaluation demonstrates that passive-active optical systems outperform standard methods when underwater sensor motion is a critical design parameter. The most significant advantage of self-referenced optical 3D scanners is that they compensate sensor motion in the same sensor domain as 3D measurements take place. This reduces the complexity of sensor co-calibration, ensures a similar accuracy for sensor pose and scene depth estimation, and broadens their possible application to smaller sensor platforms
Adaptive user interaction methods for semi-autonomous task execution within rehabilitation scenarios
No full textThe ever increasing performance of modern computing systems enables the realization of more challenging functionalities in software and mechatronic systems. This tendency results in an increase in system complexity and also makes the operation by users more difficult. Therefore, recent developments are focusing more strongly on the usability of technical systems, especially in case of systems that do not only communicate with users via a user interface but also interact with them physically. Systems that support social reintegration of persons with disabilities, so-called rehabilitation or support robots, fall into this area. This thesis focuses on the development of methods for adaptive user interactions within a software architecture for rehabilitation robots. The objective is the development of a software framework that acts as a basis for the adaptability of the graphical user interface. The methods presented to realize adaptivity are based on a user interface modularization by encapsulating all functionalities into modules. These modules can be activated or deactivated during run-time depending on the availability of resources. Furthermore, a bi-directional communication channel between the user interface and active modules as well as among modules is established. Thus it becomes possible to source common functionality out into modules and to have it reused by other modules. The communication is based on a specification language that has been developed to enable validation and to reach robust run-time behavior. An extensive review of the software architecture used for the target system identified open problems that previously prevented the realization of adaptivity within the user interface. By using another specification language, finding solutions for those open problems becomes possible as well as achieving the set objective. The development is based on an abstraction layer between the user interface and the remaining layers of the software architecture. This realizes full decoupling of the user interface from system specificfunctionality. To proof the concept for adaptivity within the user interface, the implementation of a module integrating an algorithm for pattern recognition is exemplarily shown with the aim to predict future actions of the user by evaluating previous actions
Adaptive Freiheitsgrade zur gemeinsamen Steuerung assistiver Roboter
No full textRobots have become a prominent part of our daily lives, be it as a versatile tool at work or as an obedient household helper. This is especially exciting for people with physical limitations, since designated assistive robots harbour a large potential to enhance their users' autonomy and quality of life.
Following this line of thought, the field of assistive robotics introduces mechanical assistants to people who would otherwise struggle with activities of daily living.
However, this necessitates adequate and potentially personalised control methods.
Focusing on wheelchair-mounted robotic arms, this thesis discusses currently available options, evaluates directly applicable manual alternatives, and proposes a novel shared control based on adaptive degrees of freedom (DoFs).
With a participatory design, each element is developed and evaluated in close collaboration with the target group, thus allowing for appropriate integrations and realistic assessments.
For the contemporary manual analysis, users evaluated the manufacturer-provided input device in comparison to a gamepad, 3D mouse, and a command-based voice control.
Here, the participants expressed an eagerness for personalised adaptability, as well as an explicit willingness to train in the use of more complex but capable systems, such as a 3D mouse.
Heeding this, this thesis introduces the novel shared control approach of Adaptive DoFs:
A camera-based sensor system probabilistically analyses the current situation to generate the most likely directions of robot motion and maps these to the user's input device, effectively replacing the classically available cardinal DoF.
For the user, this feels like the system anticipating their next move and providing support without taking over control.
An implementation of this proposed adaptive control was thoroughly evaluated with the target group, using e.g. smart glasses, showcasing high user acceptance with a steep learning curve and high success rates
Dynamische Schlagregelung eines redundanten Ballspielroboters
No full textThis thesis shows a control algorithm for coping with a ball batting task for an entertainment robot. The robot is a three jointed robot with a redundant degree of freedom and its name is "Doggy". Doggy because of its dog-like costume. Design, mechanics and electronics were developed by us. DC-motors control the tooth belt driven joints, resulting in elasticities between the motor and link. Redundancy and elasticity have to be taken into account by our developed controller and are demanding control tasks. In this thesis we show the structure of the ball playing robot and how this structure can be described as a model. We distinguish two models: One model that includes a flexible bearing, the other does not. Both models are calibrated using the toolkit Sparse Least Squares on Manifolds (SLOM) - i.e. the parameters for the model are determined. Both calibrated models are compared to measurements of the real system. The model with the flexible bearing is used to implement a state estimator - based on a Kalman filter - on a microcontroller. This ensures real time estimation of the robot states. The estimated states are also compared with the measurements and are assessed. The estimated states represent the measurements well. In the core of this work we develop a Task Level Optimal Controller (TLOC), a model-predictive optimal controller based on the principles of a Linear Quadratic Regulator (LQR). We aim to play a ball back to an opponent precisely. We show how this task of playing a ball at a desired time with a desired velocity at a desired position can be embedded into the LQR principle. We use cost functions for the task description. In simulations, we show the functionality of the control concept, which consists of a linear part (on a microcontroller) and a nonlinear part (PC software). The linear part uses feedback gains which are calculated by the nonlinear part. The concept of the ball batting controller with precalculated feedback gains is evaluated on the robot. This shows successful batting motions. The entertainment aspect has been tested on the Open Campus Day at the University of Bremen and is summarized here shortly. Likewise, a jointly developed audience interaction by recognition of distinctive sounds is summarized herein. In this thesis we answer the question, if it is possible to define a rebound task for our robot within a controller and show the necessary steps for this
Learning task constraints for whole-body control of robotic systems
No full textToday, Whole-Body Control (WBC) has become the standard method for controlling robots with redundant degrees of freedom, such as humanoids or mobile manipulators. WBC enables the simultaneous execution of multiple tasks by formulating them as constraints or within the cost function of an instantaneous optimization problem.
However, it requires a lot of expertise to model the optimization problem in such a way that the desired robot behavior is achieved. A human expert must analyze the task, derive appropriate task models, define constraints, and assign suitable priorities to the tasks. This process is commonly performed by hand, which is time-consuming and prone to errors. Moreover, the solutions developed are usually limited to certain situations. If the given task or the environment of the robot changes, these manually designed solutions may fail and the task specification must be adapted.
In this thesis, we address these very problems to improve usability, adaptability, and generality of existing WBC approaches. First, we introduce a programming by demonstration (PbD) approach for whole-body controllers. The approach derives a part of the optimization problem, namely the task constraints and their associated priorities, from user demonstrations. The demonstrations are performed in varying conditions, which we refer to as contexts. Using the acquired data, we can derive probabilistic models that allow generalization of task constraints, and their associated priorities with respect to novel, previously unseen contexts. That is, the whole-body controller learns to adapt to unknown situations. The proposed method not only significantly reduces the effort required to design the optimization problem, but it also improves the performance of the robot in dynamic environments. As a second contribution we present different methods for black-box optimization of task priorities, which may increase the performance of the derived whole-body controller when deployed on the target robot. Third, we integrate these contributions in a modular Whole-Body Control framework named ARC-OPT, which allows us to automatically derive, adapt, and optimize whole-body behaviors, while preserving the positive features of classical WBC approaches
- …
