71446 research outputs found
Sort by
Capacitive Strain Sensor System for Soft-Rigid Hybrid Robotic Grippers
No full textSoft-rigid hybrid robotic grippers offer several advantages over traditional rigid grippers, including the ability to conform to diverse object geometries, exhibit human-safe operation, and operate at potentially lower cost. However, despite recent progress in robotic gripper technologies, soft-rigid hybrid systems still face significant challenges in miniaturization, robustness, speed, sensing integration, and control.
This work addresses these challenges by integrating capacitive strain sensors into a soft-rigid robotic finger and developing a learning-based method for mapping frequency-dependent electrical signals to localized strain states. These sensors enable reconstruction of the gripper’s joint configuration through a single electrical interface, supporting control policies for executing more precise and dexterous manipulations across a range of object geometries and stiffness profiles.
Measurements of parallel capacitance and dissipation factor versus frequency were collected at various local strain states using an LCR meter and later replicated using a capacitance-to-digital converter for real-time implementation. Using this frequency-sweep data, an artificial neural network was trained in TensorFlow to accurately infer the sensor’s strain state. After sensor integration into multi-segment robotic fingers, the system achieved 99.05% classification accuracy and a 2.02° mean absolute error (MAE) in regression.
This proprioceptive sensing system enables robotic grippers to adjust their grip dynamically and autonomously. For each finger, only a single sensor and electrical interface are required to localize and quantify deformation across multiple joints along the finger unit. By eliminating the need for multiple sensing modules per joint, this architecture significantly simplifies both design and manufacturability. The proposed sensing framework has potential applications across robotics, prosthetics, assistive technologies, and wearable systems.Mechanical Engineering S
Molecular Insights into Noncanonical Histone Chaperone Complexes
No full textTwo fundamental macromolecular complexes required for the survival of eukaryotic cells are nucleosomes, which organize our genetic material (DNA), and ribosomes, which translate the information encoded by genes into functional proteins. The protein subunits of each of these complexes must be faithfully assembled together with DNA (nucleosomes) or RNA (ribosomes) for correct function. My dissertation research focused on two factors, UBE2O and CDAN1, that interact with specific chaperones of histones, the building blocks of nucleosomes, to perform functions linked to protein complex assembly that are particularly important for red blood cell (RBC) production.
In Chapter 2, I focus on how UBE2O mediates the quality control of unassembled ribosomal subunit proteins (RPs). RPs that fail to assemble into the ribosome, also called orphan RPs, must be recognized and degraded by the ubiquitin-proteasome system (UPS), a cellular protein quality control pathway. UBE2O is a unique ubiquitylation enzyme implicated in RP quality control and ribosome clearance during RBC differentiation. However, the mechanisms of specific RP client selection by UBE2O remain unclear. I describe how the histone chaperone NAP1L1 is repurposed as a substrate adaptor for UBE2O to promote ubiquitylation of orphan RP clients. Our discovery of NAP1L1 as an E3 ligase adaptor for orphan RPs represents a new noncanonical function of this histone chaperone.
In Chapter 3, I focus on CDAN1, which interacts with the histone chaperone ASF1. CDAN1 is an essential protein of unknown function that has been reported to bind CDIN1, a predicted endonuclease, and ASF1. Mutations in the CDAN1 or CDIN1 genes cause a specific form of congenital dyserythropoietic anemia (CDA). Although ASF1 is required for efficient nucleosome assembly, the role of the CDAN1:ASF1 interaction and its relationship to CDIN1 and histones remains unclear. I combined structural and biochemical approaches to investigate the CDAN1:CDIN1:ASF1 (C:C:A) complex. We demonstrate that the CDAN1 complex blocks multiple functional interfaces of ASF1, including regions that recruit histones, thus forming a noncanonical complex that prevents the binding of typical ASF1 clients. We also find that CDAN1 forms a dimer capable of recruiting multiple copies of ASF1 through previously unappreciated conserved binding motifs that discriminate between the ASF1 paralogs, ASF1A and ASF1B. Altogether, my thesis research reveals new molecular insights into noncanonical histone chaperone interactors.Medical Science
Boundless Weaving: Buddhist Visual Culture in a Tibetophone World, 9th to 13th Century
No full textThis dissertation examines the pivotal role of Tibetan Buddhists in shaping the Buddhist visual culture of medieval Inner Asia from the mid-ninth to the early thirteenth century, a period traditionally perceived as politically fragmented and unstable. Challenging the conventional historiographical models that portray Tibet as experiencing religious and cultural rupture after the collapse of the Tibetan Empire, this study argues that it was during the period that Tibetans emerged as key intermediaries in the transmission and transformation of Indian esoteric texts and images, weaving them into an expansive religious and visual network throughout Inner Asia.
Integrating visual, textual, and archaeological sources, this dissertation is organized into four case studies. Chapter One reassesses the Tibetan “Dark Age” by analyzing the recently excavated Bumpache Paintings from Central Tibet, revealing artistic and cultural continuities in the post-imperial period Tibet that question the traditional narratives of the region’s sudden decline. Chapters Two through Four trace how esoteric Buddhist imagery was transformed through innovative and localized reinterpretations among Tibetophone Buddhist communities across Inner Asia. Chapter Two examines the evolving iconography of the Crowned Buddha, demonstrating its fluid adaptability across diverse textual traditions. Chapter Three explores the evolution of Sitātapatrā from a protective dhāraṇī to a fully personified icon, a transformative shift inspired by a significant collection of newly translated sādhana texts, which provide unique and precise depictions of various tantric deities. Chapter Four revisits the so-called “Dance of the Sixteen Heavenly Devils” at the Mongol court, not only by tracing its origins to Indo-Tibetan esoteric ritual manuals, but also by studying how the visual representations of the assembly were formalized and standardized in the mid-twelfth century Tangut-Xixia State.
At the core of this study is the concept of the “Tibetophone world,” a transregional and transcultural network in which the Tibetan language functioned as a lingua franca for the study and practice of Indo-Tibetan Esoteric Buddhism in medieval Inner Asia. Through case studies in Dunhuang, the Tangut-Xixia State, and Mongol-ruled China, this dissertation not only foregrounds the agency of Tibetan Buddhists in shaping a cohesive religious and visual tradition. At the same time, it also demonstrates that the Tibetophone world was not a monolithic Tibetan construct, but a collaborative space where diverse ethnic groups—including Tibetans, Chinese, Tanguts, Mongols, and others—actively participated in transforming Buddhism and its visual culture. By doing so, this study redefines the concept of “Tibetan,” not as a fixed ethnic or geographic category, but as a fluid and evolving cultural identity shaped by sustained transregional exchange.Inner Asian and Altaic Studie
The Point That Makes a Difference: Interpreting Influence in Linear Off-Policy Evaluation
No full textEvaluating how a decision-making policy will perform without actually running it is a core challenge in reinforcement learning, especially when working with pre-collected data. This task, known as offline policy evaluation (OPE), becomes particularly difficult when the data doesn’t fully cover the situations the new policy might encounter, making certain datapoints disproportionately important. This thesis explores how individual datapoints influence value estimates under two value-based OPE algorithms—Least-Squares Temporal Difference (LSTD) and Fitted Q-Evaluation (FQE)—using linear function approximation. We derive exact leave-one-out (LOO) estimates for both methods and, through experiments on structured environments, show how these influence measures can be used to understand convergence issues and offer insight to the reliability of OPE estimates. In particular, we show that FQE convergence provides a strong signal for when LSTD estimates can be trusted and offer an interpretation through influences for cases of FQE divergence. We further introduce a clipped version of FQE that bounds predicted values, reducing variance while maintaining stability, and provide exact LOO estimates for it as well.Computer Scienc
VoxelCompress: Learned Implicit Neural Compression of 3D Connectomics Data
No full textConnectomics, the study of comprehensive maps of neural connections in the brain, faces a significant computational bottleneck in data acquisition and storage. Current methods for imaging brain tissue at the required nanoscale resolution via serial section electron microscopy demand an estimated 10,000 years and require 1000 exabytes to image a complete human brain. This thesis introduces VoxelCompress, a deep learning-based compression framework that drastically reduces storage requirements while preserving high-fidelity reconstructions of volumetric electron microscopy (EM) data. Through extensive experiments, we demonstrate that VoxelCompress achieves a 2048× compression ratio and 99.95% overall compression, with efficiency far surpassing traditional lossless methods (only 1.68×) and accuracy exceeding state-of-the-art learned approaches (PSNR 38+ dB). Notably, VoxelCompress is the first generalizable model that does not require retraining between volumes, making its incorporation into existing imaging pipelines highly practical. Additionally, we show that denoising the input data significantly enhances compression performance—improving the peak signal-to-noise ratio (PSNR) by 6.64 dB and enabling a reduction in model depth from six layers to five without compromising reconstruction quality.Computer Scienc
DESIGNING THE FUNGAL CITY: A framework for connecting mycorrhizal networks in the built environment
No full textUrban infrastructure has replaced what once was biologically rich soil with compacted, sterile substrates severing the underground networks that facilitate mutualistic relationships between plants and mycorrhizal fungi. This dissertation argues that one way of reintegrating mycorrhizal networks – and thereby restoring the health and resilience of the urban biota—is by redesigning underground environments. An extensive literature review was first undertaken to develop a comprehensive understanding of current research across the fields of landscape architecture, microbiology and ecology. The review revealed a lack of cross disciplinary collaboration and highlighted the need for a more integrative approach to studying and designing urban environments. Central to this dissertation is the design, development and testing of the soil conduit—a new form of green infrastructure designed to connect trees physically in order to help foster the growth of fungal networks in urban conditions. By creating a connected rhizosphere, the soil conduit enables trees to exchange nutrients, which encourages health and resilience through cooperative dynamics. To demonstrate the feasibility and impact of the soil conduit, this research employed two proof of concepts at different scales. At the micro scale, a controlled experiment was conducted in a greenhouse with 53 oak saplings to evaluate the effects of the soil conduit on tree health and fungal communities. Over the course of the study, samples were collected from the soil and the root zones, to be analysed using molecular and microscopic techniques.
The results confirmed the presence of ectomycorrhizal fungi in the soil conduit, in addition to improved tree health-scores and growth measurements in connected treatments. These findings suggest that the conduit is effective in promoting underground connectivity and ecological resilience in urban landscapes. At the macro scale, a mapping exercise was undertaken to analyse the spatial and infrastructural conditions that support soil connectivity and tree health. A study explored trees growing in parks and sidewalks to determine the effects of fragmentation in an urban context on the structure of the soil as well as the health score and growth measurements of trees. Here mapping is also suggested as a tool to begin speculating about potential underground fungal networks in the built environment and how a soil conduit might enhance this connectivity, thereby enhancing the resilience of the urban forest. This research demonstrates the soil conduit’s potential to transform how we design for urban ecosystems, reimagining the underground landscapes as a site of connection rather than isolation by offering a scalable and design-oriented framework for reintegrating mycorrhizal fungi—and the vital ecological functions they support—into the urban fabric.Advanced Studies Progra
Structure of the Southern Canadian Rocky Mountain Foreland and Western San Joaquin Basin Fold-and-Thrust Belts
No full textThis thesis investigates the geometry and kinematic evolution of fold-and-thrust belts from tectonic to human timescales in the Front Ranges of the Southern Canadian Rocky Mountains and the South-Central San Joaquin Valley of California. Findings advance the field’s capabilities for developing internally consistent 2D and 3D models of both ancient and active fold and thrust belts and have important implications for earthquake hazards in California.
In chapter one, we investigate the geometry and kinematic evolution of the Southern Front Ranges of the Canadian Rockies, and orogen that formed over tens of millions of years associated with subduction of the Pacific Plate and accretion of tectonic terrains to western North America. Due to its spectacular field exposures, this mountain belt has long served as a prototypical example of a thin-skinned fold-and-thrust belt. Nevertheless, many questions remain unanswered regarding its detailed structure and kinematic evolution. In conjunction with 10 m and 1 m/pixel digital elevation models, high-resolution satellite imagery enables unparalleled quantitative definition of the fold geometry and bedding-fault relationships in the Front Ranges. Using these fundamental quantitative relationships, we develop a series of new, kinematically viable cross sections and 3D models that both reinforce, and in some cases overturn, previous paradigms about this fold and thrust belt. Traditionally, the region is thought to contain a single basal detachment and develop through a break forward sequence of thrusting. Our analysis shows that this fold and thrust belt is underlain by multiple basal detachments and developed through a complex sequence of thrusting that fundamentally controls the map patterns expressed within the Front Ranges. The methods we develop for remote data collection and analysis can be applied to develop and refine interpretations of fold and thrust belts worldwide.
Chapters two and three investigate the geometry and seismic hazards associated with Central California's active Southern San Joaquin Fold and Thrust Belt. In Chapter 2, we develop an internally consistent structural model for the system of blind-thrust faults and their overlying, en-echelon anticlines. Portions of these faults generated a series of Mw 5.5-6.5 earthquakes in the 1980s, yet the activity and earthquake potential of along-strike and down dip segments of the system remain uncertain. Our model reveals that the fold changes along strike from a structural wedge to a composite growth fault-bend fold. The wedge locally consumes much of the slip on the underlying blind thrust fault, whereas the fault-bend folds send slip eastward into the basin on an upper detachment surface. Consistent with this, we observe distinct variations in the eastern extent of deformation in the basin. Specifically, slip into the basin on the upper detachment is manifest in both fault-related folds and additional thrust ramps. To the west and at depth, these fault segments sole to two primary basal detachment levels that extend beneath the Temblor Ranges.
In chapter three, we couple the structural model from chapter two with groundwater wells to determine the activity of various segments of the Southern San Joaquin Fold and Thrust Belt. We find that sediments of late Pleistocene age are folded through active axial surfaces above each of the major blind-thrust segments. Our observation indicates that the entire system of ramps and detachments in this fold and thrust belt are active. Based on the size of the fault system, it is capable of generating Mw ≈7.7 earthquakes. We additionally identify new active thrust faults in the basin that have the potential to generate damaging earthquakes.Earth and Planetary Science
Learning to Adapt: Representation-Based Reinforcement Learning for Multi-Task Skill Transfer
No full textReinforcement learning (RL) has demonstrated remarkable success in learning complex control policies, yet its applicability to real-world robotics remains limited due to sample inefficiency and poor generalization across tasks. This thesis explores representation-based reinforcement learning as a means to address these challenges, with a particular focus on multi-task learning and policy adaptation. We begin by conducting a comprehensive review of state-of-the-art offline RL algorithms, highlighting the evolution from traditional Soft Actor-Critic (SAC) to advanced representation-based methods such as Contrastive Learned Representation Soft Actor-Critic (CTRL-SAC). Motivated by the limitations of existing approaches, we propose a novel multi-skill representation learning framework that builds upon CTRL-SAC, more specifically known as RepMT-SAC, enabling efficient skill transfer by leveraging shared task-independent dynamics. To evaluate the effectiveness of our framework, we conduct extensive empirical studies in order to demonstrate that our approach improves policy adaptation and generalization, reducing the need for task-specific retraining. By combining theoretical insights with rigorous empirical validation, this work contributes to the advancement of reinforcement learning for autonomous control, paving the way for more efficient and versatile robotic systems.Computer Scienc
Harnessing UniAnalytics: Exploring the Role of Engagement and Collaboration in Student Success in Jupyter Notebook Environments
No full textWhether and how students engage, and the ways in which they collaborate during learning activities, are critical determinants of academic success. These factors become even more significant in computational learning environments where digital tools mediate interactions. In this thesis, we examine the relationship between student engagement, collaboration, and academic success in a Jupyter Notebook-based learning environment. We utilize regression and multilevel models to identify key engagement and collaboration behaviors observing their impact on three key measures of student success: performance on assignments, on the final exam, and overall course grade. From our analyses, we find that certain engagement metrics such as execution frequency are strong predictors of success. We also found that other metrics such as higher error rates and prolonged idle time are negatively associated with all 3 of our measures of success. Our chosen collaboration metrics do not show significant correlations with student success, raising questions about how collaborative learning is assessed in digital environments. This thesis hopes to contribute to the growing field of learning analytics by providing empirical evidence on how digital engagement behaviors impact student success. The ultimate goal of this work is to offer insights that can enhance teaching strategies and educational outcomes in computational classrooms that rely on digital tool such as Jupyter Notebook.Applied Mathematic
From the lab to the field: studying fish locomotor dynamics using animal-borne dataloggers
No full textUnderstanding the mechanics and efficiencies associated with how animals move is essential to understand links between form and function and ultimately understand where and why animals move. However, most current studies of animal biomechanics are focused on stereotyped behaviors and dominated by small-bodied animals due to logistical constraints of laboratory experiments. Instead of bringing an animal into the lab and recording its movements with a camera, we can directly measure the body movement of animals using small acceleration, angular velocity, and magnetic field data loggers. This method allows us to tag large animals and collect spatio-temporally unbiased estimates of their kinematics in both lab conditions and in the wild. In my dissertation, I identify the problems, solutions, and benefits to using this method and I show how we can expand our knowledge of in situ kinematics that would be impossible to document using traditional methods.
In Chapter 1, I demonstrate that individual dataloggers temporally drift apart from each other at a rate that meaningfully disrupts their use for kinematics. I show that these errors are quantifiable in both the lab and field and how to correct them.
In Chapter 2, I deploy multiple dataloggers on a soft robotic system and three diverse species of fish (an agnathan, an elasmobranch, and a teleost), show the diverse metrics that can be generated using multiple dataloggers during both routine and high performance swimming in near-field conditions, and evaluate these metrics against traditional video data.
In Chapters 3 and 4, I develop a novel multitag package and deploy it on smooth dogfish in the wild. In Chapter 3, I use this dataset to test a previous hypothesis that suggested that the anterior and posterior body regions of sharks oscillate at different frequencies. I observed no evidence of different oscillation frequencies between the head and tail and suggest these previous observations are due to short monitoring periods and small errors in location and frequency estimation.
In Chapter 4, I use this multi datalogger data set to recreate the kinematics of sharks in the wild using five smooth dogfish released into Massachusetts Bay. I use data from three data loggers to measure how different body parts move and demonstrate how movement along the body varies with tailbeat frequency during routine swimming. Additionally, I document maximal performance and unsteady behaviors such as turns and show that unsteady behaviors are a significant part of a wild animal’s behavior that are poorly captured in laboratory studies.
Overall, this dissertation demonstrates that important biomechanical metrics can be calculated in situ from multiple data loggers attached to animals in the field, and opens the door for documentation of new behaviors, in new species. This work advances and directly links the biomechanics, behavior and ecology of animals.Biology, Organismic and Evolutionar