ASU Electronic Theses and Dissertations
This collection includes most of the ASU Theses and Dissertations from 2011 to present. ASU Theses and Dissertations are available in downloadable PDF format; however, a small percentage of items are under embargo. Information about the dissertations/theses includes degree information, committee members, an abstract, supporting data or media.
In addition to the electronic theses found in the ASU Digital Repository, ASU Theses and Dissertations can be found in the ASU Library Catalog.
Dissertations and Theses granted by Arizona State University are archived and made available through a joint effort of the ASU Graduate College and the ASU Libraries. For more information or questions about this collection contact or visit the Digital Repository ETD Library Guide or contact the ASU Graduate College at gradformat@asu.edu.
Displaying 1 - 10 of 126
Description
The development of advanced, anthropomorphic artificial hands aims to provide upper extremity amputees with improved functionality for activities of daily living. However, many state-of-the-art hands have a large number of degrees of freedom that can be challenging to control in an intuitive manner. Automated grip responses could be built into artificial hands in order to enhance grasp stability and reduce the cognitive burden on the user. To this end, three studies were conducted to understand how human hands respond, passively and actively, to unexpected perturbations of a grasped object along and about different axes relative to the hand. The first study investigated the effect of magnitude, direction, and axis of rotation on precision grip responses to unexpected rotational perturbations of a grasped object. A robust "catch-up response" (a rapid, pulse-like increase in grip force rate previously reported only for translational perturbations) was observed whose strength scaled with the axis of rotation. Using two haptic robots, we then investigated the effects of grip surface friction, axis, and direction of perturbation on precision grip responses for unexpected translational and rotational perturbations for three different hand-centric axes. A robust catch-up response was observed for all axes and directions for both translational and rotational perturbations. Grip surface friction had no effect on the stereotypical catch-up response. Finally, we characterized the passive properties of the precision grip-object system via robot-imposed impulse perturbations. The hand-centric axis associated with the greatest translational stiffness was different than that for rotational stiffness. This work expands our understanding of the passive and active features of precision grip, a hallmark of human dexterous manipulation. Biological insights such as these could be used to enhance the functionality of artificial hands and the quality of life for upper extremity amputees.
ContributorsDe Gregorio, Michael (Author) / Santos, Veronica J. (Thesis advisor) / Artemiadis, Panagiotis K. (Committee member) / Santello, Marco (Committee member) / Sugar, Thomas (Committee member) / Helms Tillery, Stephen I. (Committee member) / Arizona State University (Publisher)
Created2013
Description
This research is focused on two separate but related topics. The first uses an electroencephalographic (EEG) brain-computer interface (BCI) to explore the phenomenon of motor learning transfer. The second takes a closer look at the EEG-BCI itself and tests an alternate way of mapping EEG signals into machine commands. We test whether motor learning transfer is more related to use of shared neural structures between imagery and motor execution or to more generalized cognitive factors. Using an EEG-BCI, we train one group of participants to control the movements of a cursor using embodied motor imagery. A second group is trained to control the cursor using abstract motor imagery. A third control group practices moving the cursor using an arm and finger on a touch screen. We hypothesized that if motor learning transfer is related to the use of shared neural structures then the embodied motor imagery group would show more learning transfer than the abstract imaging group. If, on the other hand, motor learning transfer results from more general cognitive processes, then the abstract motor imagery group should also demonstrate motor learning transfer to the manual performance of the same task. Our findings support that motor learning transfer is due to the use of shared neural structures between imaging and motor execution of a task. The abstract group showed no motor learning transfer despite being better at EEG-BCI control than the embodied group. The fact that more participants were able to learn EEG-BCI control using abstract imagery suggests that abstract imagery may be more suitable for EEG-BCIs for some disabilities, while embodied imagery may be more suitable for others. In Part 2, EEG data collected in the above experiment was used to train an artificial neural network (ANN) to map EEG signals to machine commands. We found that our open-source ANN using spectrograms generated from SFFTs is fundamentally different and in some ways superior to Emotiv's proprietary method. Our use of novel combinations of existing technologies along with abstract and embodied imagery facilitates adaptive customization of EEG-BCI control to meet needs of individual users.
Contributorsda Silva, Flavio J. K (Author) / Mcbeath, Michael K (Thesis advisor) / Helms Tillery, Stephen (Committee member) / Presson, Clark (Committee member) / Sugar, Thomas (Committee member) / Arizona State University (Publisher)
Created2013
Description
Myoelectric control is lled with potential to signicantly change human-robot interaction.
Humans desire compliant robots to safely interact in dynamic environments
associated with daily activities. As surface electromyography non-invasively measures
limb motion intent and correlates with joint stiness during co-contractions,
it has been identied as a candidate for naturally controlling such robots. However,
state-of-the-art myoelectric interfaces have struggled to achieve both enhanced
functionality and long-term reliability. As demands in myoelectric interfaces trend
toward simultaneous and proportional control of compliant robots, robust processing
of multi-muscle coordinations, or synergies, plays a larger role in the success of the
control scheme. This dissertation presents a framework enhancing the utility of myoelectric
interfaces by exploiting motor skill learning and
exible muscle synergies for
reliable long-term simultaneous and proportional control of multifunctional compliant
robots. The interface is learned as a new motor skill specic to the controller,
providing long-term performance enhancements without requiring any retraining or
recalibration of the system. Moreover, the framework oers control of both motion
and stiness simultaneously for intuitive and compliant human-robot interaction. The
framework is validated through a series of experiments characterizing motor learning
properties and demonstrating control capabilities not seen previously in the literature.
The results validate the approach as a viable option to remove the trade-o
between functionality and reliability that have hindered state-of-the-art myoelectric
interfaces. Thus, this research contributes to the expansion and enhancement of myoelectric
controlled applications beyond commonly perceived anthropomorphic and
\intuitive control" constraints and into more advanced robotic systems designed for
everyday tasks.
Humans desire compliant robots to safely interact in dynamic environments
associated with daily activities. As surface electromyography non-invasively measures
limb motion intent and correlates with joint stiness during co-contractions,
it has been identied as a candidate for naturally controlling such robots. However,
state-of-the-art myoelectric interfaces have struggled to achieve both enhanced
functionality and long-term reliability. As demands in myoelectric interfaces trend
toward simultaneous and proportional control of compliant robots, robust processing
of multi-muscle coordinations, or synergies, plays a larger role in the success of the
control scheme. This dissertation presents a framework enhancing the utility of myoelectric
interfaces by exploiting motor skill learning and
exible muscle synergies for
reliable long-term simultaneous and proportional control of multifunctional compliant
robots. The interface is learned as a new motor skill specic to the controller,
providing long-term performance enhancements without requiring any retraining or
recalibration of the system. Moreover, the framework oers control of both motion
and stiness simultaneously for intuitive and compliant human-robot interaction. The
framework is validated through a series of experiments characterizing motor learning
properties and demonstrating control capabilities not seen previously in the literature.
The results validate the approach as a viable option to remove the trade-o
between functionality and reliability that have hindered state-of-the-art myoelectric
interfaces. Thus, this research contributes to the expansion and enhancement of myoelectric
controlled applications beyond commonly perceived anthropomorphic and
\intuitive control" constraints and into more advanced robotic systems designed for
everyday tasks.
ContributorsIson, Mark (Author) / Artemiadis, Panagiotis (Thesis advisor) / Santello, Marco (Committee member) / Greger, Bradley (Committee member) / Berman, Spring (Committee member) / Sugar, Thomas (Committee member) / Fainekos, Georgios (Committee member) / Arizona State University (Publisher)
Created2015
Description
Brain Computer Interfaces are becoming the next generation controllers not only in the medical devices for disabled individuals but also in the gaming and entertainment industries. In order to build an effective Brain Computer Interface, which accurately translates the user thoughts into machine commands, it is important to have robust and fail proof signal processing and machine learning modules which operate on the raw EEG signals and estimate the current thought of the user.
In this thesis, several techniques used to perform EEG signal pre-processing, feature extraction and signal classification have been discussed, implemented, validated and verified; efficient supervised machine learning models, for the EEG motor imagery signal classification are identified. To further improve the performance of system unsupervised feature learning techniques have been investigated by pre-training the Deep Learning models. Use of pre-training stacked autoencoders have been proposed to solve the problems caused by random initialization of weights in neural networks.
Motor Imagery (imaginary hand and leg movements) signals are acquire using the Emotiv EEG headset. Different kinds of features like mean signal, band powers, RMS of the signal have been extracted and supplied to the machine learning (ML) stage, wherein, several ML techniques like LDA, KNN, SVM, Logistic regression and Neural Networks are applied and validated. During the validation phase the performances of various techniques are compared and some important observations are reported. Further, deep Learning techniques like autoencoding have been used to perform unsupervised feature learning. The reliability of the features is analyzed by performing classification by using the ML techniques mentioned earlier. The performance of the neural networks has been further improved by pre-training the network in an unsupervised fashion using stacked autoencoders and supplying the stacked autoencoders’ network parameters as initial parameters to the neural network. All the findings in this research, during each phase (pre-processing, feature extraction, classification) are directly relevant and can be used by the BCI research community for building motor imagery based BCI applications.
Additionally, this thesis attempts to develop, test, and compare the performance of an alternative method for classifying human driving behavior. This thesis proposes the use of driver affective states to know the driving behavior. The purpose of this part of the thesis was to classify the EEG data collected from several subjects while driving simulated vehicle and compare the classification results with those obtained by classifying the driving behavior using vehicle parameters collected simultaneously from all the subjects. The objective here is to see if the drivers’ mental state is reflected in his driving behavior.
In this thesis, several techniques used to perform EEG signal pre-processing, feature extraction and signal classification have been discussed, implemented, validated and verified; efficient supervised machine learning models, for the EEG motor imagery signal classification are identified. To further improve the performance of system unsupervised feature learning techniques have been investigated by pre-training the Deep Learning models. Use of pre-training stacked autoencoders have been proposed to solve the problems caused by random initialization of weights in neural networks.
Motor Imagery (imaginary hand and leg movements) signals are acquire using the Emotiv EEG headset. Different kinds of features like mean signal, band powers, RMS of the signal have been extracted and supplied to the machine learning (ML) stage, wherein, several ML techniques like LDA, KNN, SVM, Logistic regression and Neural Networks are applied and validated. During the validation phase the performances of various techniques are compared and some important observations are reported. Further, deep Learning techniques like autoencoding have been used to perform unsupervised feature learning. The reliability of the features is analyzed by performing classification by using the ML techniques mentioned earlier. The performance of the neural networks has been further improved by pre-training the network in an unsupervised fashion using stacked autoencoders and supplying the stacked autoencoders’ network parameters as initial parameters to the neural network. All the findings in this research, during each phase (pre-processing, feature extraction, classification) are directly relevant and can be used by the BCI research community for building motor imagery based BCI applications.
Additionally, this thesis attempts to develop, test, and compare the performance of an alternative method for classifying human driving behavior. This thesis proposes the use of driver affective states to know the driving behavior. The purpose of this part of the thesis was to classify the EEG data collected from several subjects while driving simulated vehicle and compare the classification results with those obtained by classifying the driving behavior using vehicle parameters collected simultaneously from all the subjects. The objective here is to see if the drivers’ mental state is reflected in his driving behavior.
ContributorsManchala, Vamsi Krishna (Author) / Redkar, Sangram (Thesis advisor) / Rogers, Bradley (Committee member) / Sugar, Thomas (Committee member) / Arizona State University (Publisher)
Created2015
Description
In a collaborative environment where multiple robots and human beings are expected
to collaborate to perform a task, it becomes essential for a robot to be aware of multiple
agents working in its work environment. A robot must also learn to adapt to
different agents in the workspace and conduct its interaction based on the presence
of these agents. A theoretical framework was introduced which performs interaction
learning from demonstrations in a two-agent work environment, and it is called
Interaction Primitives.
This document is an in-depth description of the new state of the art Python
Framework for Interaction Primitives between two agents in a single as well as multiple
task work environment and extension of the original framework in a work environment
with multiple agents doing a single task. The original theory of Interaction
Primitives has been extended to create a framework which will capture correlation
between more than two agents while performing a single task. The new state of the
art Python framework is an intuitive, generic, easy to install and easy to use python
library which can be applied to use the Interaction Primitives framework in a work
environment. This library was tested in simulated environments and controlled laboratory
environment. The results and benchmarks of this library are available in the
related sections of this document.
to collaborate to perform a task, it becomes essential for a robot to be aware of multiple
agents working in its work environment. A robot must also learn to adapt to
different agents in the workspace and conduct its interaction based on the presence
of these agents. A theoretical framework was introduced which performs interaction
learning from demonstrations in a two-agent work environment, and it is called
Interaction Primitives.
This document is an in-depth description of the new state of the art Python
Framework for Interaction Primitives between two agents in a single as well as multiple
task work environment and extension of the original framework in a work environment
with multiple agents doing a single task. The original theory of Interaction
Primitives has been extended to create a framework which will capture correlation
between more than two agents while performing a single task. The new state of the
art Python framework is an intuitive, generic, easy to install and easy to use python
library which can be applied to use the Interaction Primitives framework in a work
environment. This library was tested in simulated environments and controlled laboratory
environment. The results and benchmarks of this library are available in the
related sections of this document.
ContributorsKumar, Ashish, M.S (Author) / Amor, Hani Ben (Thesis advisor) / Zhang, Yu (Committee member) / Yang, Yezhou (Committee member) / Arizona State University (Publisher)
Created2017
Description
Computer Vision as a eld has gone through signicant changes in the last decade.
The eld has seen tremendous success in designing learning systems with hand-crafted
features and in using representation learning to extract better features. In this dissertation
some novel approaches to representation learning and task learning are studied.
Multiple-instance learning which is generalization of supervised learning, is one
example of task learning that is discussed. In particular, a novel non-parametric k-
NN-based multiple-instance learning is proposed, which is shown to outperform other
existing approaches. This solution is applied to a diabetic retinopathy pathology
detection problem eectively.
In cases of representation learning, generality of neural features are investigated
rst. This investigation leads to some critical understanding and results in feature
generality among datasets. The possibility of learning from a mentor network instead
of from labels is then investigated. Distillation of dark knowledge is used to eciently
mentor a small network from a pre-trained large mentor network. These studies help
in understanding representation learning with smaller and compressed networks.
The eld has seen tremendous success in designing learning systems with hand-crafted
features and in using representation learning to extract better features. In this dissertation
some novel approaches to representation learning and task learning are studied.
Multiple-instance learning which is generalization of supervised learning, is one
example of task learning that is discussed. In particular, a novel non-parametric k-
NN-based multiple-instance learning is proposed, which is shown to outperform other
existing approaches. This solution is applied to a diabetic retinopathy pathology
detection problem eectively.
In cases of representation learning, generality of neural features are investigated
rst. This investigation leads to some critical understanding and results in feature
generality among datasets. The possibility of learning from a mentor network instead
of from labels is then investigated. Distillation of dark knowledge is used to eciently
mentor a small network from a pre-trained large mentor network. These studies help
in understanding representation learning with smaller and compressed networks.
ContributorsVenkatesan, Ragav (Author) / Li, Baoxin (Thesis advisor) / Turaga, Pavan (Committee member) / Yang, Yezhou (Committee member) / Davulcu, Hasan (Committee member) / Arizona State University (Publisher)
Created2017
Description
With the rise of the Big Data Era, an exponential amount of network data is being generated at an unprecedented rate across a wide-range of high impact micro and macro areas of research---from protein interaction to social networks. The critical challenge is translating this large scale network data into actionable information.
A key task in the data translation is the analysis of network connectivity via marked nodes---the primary focus of our research. We have developed a framework for analyzing network connectivity via marked nodes in large scale graphs, utilizing novel algorithms in three interrelated areas: (1) analysis of a single seed node via it’s ego-centric network (AttriPart algorithm); (2) pathway identification between two seed nodes (K-Simple Shortest Paths Multithreaded and Search Reduced (KSSPR) algorithm); and (3) tree detection, defining the interaction between three or more seed nodes (Shortest Path MST algorithm).
In an effort to address both fundamental and applied research issues, we have developed the LocalForcasting algorithm to explore how network connectivity analysis can be applied to local community evolution and recommender systems. The goal is to apply the LocalForecasting algorithm to various domains---e.g., friend suggestions in social networks or future collaboration in co-authorship networks. This algorithm utilizes link prediction in combination with the AttriPart algorithm to predict future connections in local graph partitions.
Results show that our proposed AttriPart algorithm finds up to 1.6x denser local partitions, while running approximately 43x faster than traditional local partitioning techniques (PageRank-Nibble). In addition, our LocalForecasting algorithm demonstrates a significant improvement in the number of nodes and edges correctly predicted over baseline methods. Furthermore, results for the KSSPR algorithm demonstrate a speed-up of up to 2.5x the standard k-simple shortest paths algorithm.
A key task in the data translation is the analysis of network connectivity via marked nodes---the primary focus of our research. We have developed a framework for analyzing network connectivity via marked nodes in large scale graphs, utilizing novel algorithms in three interrelated areas: (1) analysis of a single seed node via it’s ego-centric network (AttriPart algorithm); (2) pathway identification between two seed nodes (K-Simple Shortest Paths Multithreaded and Search Reduced (KSSPR) algorithm); and (3) tree detection, defining the interaction between three or more seed nodes (Shortest Path MST algorithm).
In an effort to address both fundamental and applied research issues, we have developed the LocalForcasting algorithm to explore how network connectivity analysis can be applied to local community evolution and recommender systems. The goal is to apply the LocalForecasting algorithm to various domains---e.g., friend suggestions in social networks or future collaboration in co-authorship networks. This algorithm utilizes link prediction in combination with the AttriPart algorithm to predict future connections in local graph partitions.
Results show that our proposed AttriPart algorithm finds up to 1.6x denser local partitions, while running approximately 43x faster than traditional local partitioning techniques (PageRank-Nibble). In addition, our LocalForecasting algorithm demonstrates a significant improvement in the number of nodes and edges correctly predicted over baseline methods. Furthermore, results for the KSSPR algorithm demonstrate a speed-up of up to 2.5x the standard k-simple shortest paths algorithm.
ContributorsFreitas, Scott (Author) / Tong, Hanghang (Thesis advisor) / Maciejewski, Ross (Committee member) / Yang, Yezhou (Committee member) / Arizona State University (Publisher)
Created2018
Description
The performance of most of the visual computing tasks depends on the quality of the features extracted from the raw data. Insightful feature representation increases the performance of many learning algorithms by exposing the underlying explanatory factors of the output for the unobserved input. A good representation should also handle anomalies in the data such as missing samples and noisy input caused by the undesired, external factors of variation. It should also reduce the data redundancy. Over the years, many feature extraction processes have been invented to produce good representations of raw images and videos.
The feature extraction processes can be categorized into three groups. The first group contains processes that are hand-crafted for a specific task. Hand-engineering features requires the knowledge of domain experts and manual labor. However, the feature extraction process is interpretable and explainable. Next group contains the latent-feature extraction processes. While the original feature lies in a high-dimensional space, the relevant factors for a task often lie on a lower dimensional manifold. The latent-feature extraction employs hidden variables to expose the underlying data properties that cannot be directly measured from the input. Latent features seek a specific structure such as sparsity or low-rank into the derived representation through sophisticated optimization techniques. The last category is that of deep features. These are obtained by passing raw input data with minimal pre-processing through a deep network. Its parameters are computed by iteratively minimizing a task-based loss.
In this dissertation, I present four pieces of work where I create and learn suitable data representations. The first task employs hand-crafted features to perform clinically-relevant retrieval of diabetic retinopathy images. The second task uses latent features to perform content-adaptive image enhancement. The third task ranks a pair of images based on their aestheticism. The goal of the last task is to capture localized image artifacts in small datasets with patch-level labels. For both these tasks, I propose novel deep architectures and show significant improvement over the previous state-of-art approaches. A suitable combination of feature representations augmented with an appropriate learning approach can increase performance for most visual computing tasks.
The feature extraction processes can be categorized into three groups. The first group contains processes that are hand-crafted for a specific task. Hand-engineering features requires the knowledge of domain experts and manual labor. However, the feature extraction process is interpretable and explainable. Next group contains the latent-feature extraction processes. While the original feature lies in a high-dimensional space, the relevant factors for a task often lie on a lower dimensional manifold. The latent-feature extraction employs hidden variables to expose the underlying data properties that cannot be directly measured from the input. Latent features seek a specific structure such as sparsity or low-rank into the derived representation through sophisticated optimization techniques. The last category is that of deep features. These are obtained by passing raw input data with minimal pre-processing through a deep network. Its parameters are computed by iteratively minimizing a task-based loss.
In this dissertation, I present four pieces of work where I create and learn suitable data representations. The first task employs hand-crafted features to perform clinically-relevant retrieval of diabetic retinopathy images. The second task uses latent features to perform content-adaptive image enhancement. The third task ranks a pair of images based on their aestheticism. The goal of the last task is to capture localized image artifacts in small datasets with patch-level labels. For both these tasks, I propose novel deep architectures and show significant improvement over the previous state-of-art approaches. A suitable combination of feature representations augmented with an appropriate learning approach can increase performance for most visual computing tasks.
ContributorsChandakkar, Parag Shridhar (Author) / Li, Baoxin (Thesis advisor) / Yang, Yezhou (Committee member) / Turaga, Pavan (Committee member) / Davulcu, Hasan (Committee member) / Arizona State University (Publisher)
Created2017
Description
The sensor industry is a growing industry that has been predicted by Allied Market Research to be a multi-billion industry by 2022. One of the many key drives behind this rapid growth in the sensor industry is the increase incorporation of sensors into portable electrical devices. The value for sensor technologies are increased when the sensors are developed into innovative measuring system for application uses in the Aerospace, Defense, and Healthcare industries. While sensors are not new, their increased performance, size reduction, and decrease in cost has opened the door for innovative sensor combination for portable devices that could be worn or easily moved around. With this opportunity for further development of sensor use through concept engineering development, three concept projects for possible innovative portable devices was undertaken in this research. One project was the development of a pulse oximeter devise with fingerprint recognition. The second project was prototyping a portable Bluetooth strain gage monitoring system. The third project involved sensors being incorporated onto flexible printed circuit board (PCB) for improved comfort of wearable devices. All these systems were successfully tested in lab.
ContributorsNichols, Kevin William (Author) / Redkar, Sangram (Thesis advisor) / Rogers, Brad (Committee member) / Sugar, Thomas (Committee member) / Arizona State University (Publisher)
Created2018
Description
In this work, different methods for fabrication of flexible sensors and sensor characterization are studied. Using materials and equipment that is unconventional, it is shown that different processes can be used to create sensors that behave like commercially available sensors. The reason unconventional methods are used is to cut down on cost to produce the sensors as well as enabling the manufacture of custom sensors in different sizes and different configurations. Currently commercially available sensors are expensive and are usually designed for very specific applications. By creating these same types of sensors using new methods and materials, these new sensors will show that flexible sensor creation for many uses at a fraction of the cost is achievable.
ContributorsCasanova, Lucas Montgomery (Author) / Redkar, Sangram (Thesis advisor) / Rogers, Bradley (Committee member) / Sugar, Thomas (Committee member) / Arizona State University (Publisher)
Created2018