12023 research outputs found
Sort by
Layered Control Architectures: Constructive Theory and Application to Legged Robots
No full textFueled in part by the imagination of science fiction, every decade since the 1950s has expected robots to enter our everyday lives in the subsequent decade. Despite this anticipation, the widespread adoption of robots has consistently fallen short of societal expectations. This delay is attributable to the sheer variety of complexities in robotics --- perception, contact-rich dynamics, human-robot interactions. Each sub-discipline of robotics poses unique challenges that must be addressed to achieve general autonomy. As progress is made in these sub-fields, it is increasingly important to adopt a layered architecture perspective that combines isolated controller blocks into a unified framework.
This thesis argues that on the road to general autonomy, adopting layered architectures enables three key benefits: efficiency, feasibility, and generalizability. We root our discussion in a general problem in robotics: the design of a controller that navigates a robot to a goal state while satisfying all state and input constraints that are present. Throughout the thesis, we focus on solutions that are both general --- applicable across a wide variety of robotic platforms --- and concrete --- deployed and tested on specific hardware platforms. As such, we aim to not only propose a framework for reasoning about this problem, but also methods to synthesize controllers that solve it in practice for legged robots.
We begin by motivating and formalizing the notion of layered architectures and use this to build our control stack from the bottom up. We start with low-level planning and tracking layers that stabilize the system within a tracking tube for both the actuated and underactuated states of legged robots. We then introduce high-level planning and tracking layers that generate and follow sparse, dynamically feasible graphs for coarse global navigation through cluttered environments. By decomposing the global control problem into interacting levels and layers, each operating with disparate timescales and system abstractions, we enable tractable, reliable, and extensible robot autonomy.
Throughout this thesis, an emphasis will be placed on mathematical structure, constructive synthesis, and experimental validation. We demonstrate that adopting a layered architecture perspective is not merely an implementation convenience, but a fundamental organizing principle that can enable true robot autonomy.</p
Neural Operator for Scientific Computing
No full textScientific computing, which aims to accurately simulate complex physical phenomena, often requires substantial computational resources. By viewing data as continuous functions, we leverage the smoothness structures of function spaces to enable efficient large-scale simulations. We introduce the neural operator, a universal machine learning framework designed to approximate solution operators in infinite-dimensional spaces, achieving scalable physical simulations. The thesis begins with the introduction and definition of neural operators. Chapters 2-4 discuss architecture designs of neural operators including graph neural operator, multipole neural operator, and Fourier neural operator. Chapters 5-7 discuss physics-based learning techniques such as dissipative loss, physics-informed loss, and scale consistency loss. Chapters 8-10 discuss geometric neural operators with various boundary shapes, including latent space embedding, learned deformation, and optimal transport. Chapters 11-12 discuss further applications of neural operator in weather forecast and carbon capture storage
Probing the Origins of Directly Imaged Planets and Brown Dwarfs: From Atmospheric Compositions to Binarity
Get PDFHigh-contrast imaging has revealed a population of substellar companions, generally classified as giant planets (~2-13 MJup) or brown dwarfs (~13-75 MJup), orbiting at large separations (~3-1000 au) from their host stars. Past studies have mostly relied on low-resolution spectroscopy (R~20-100) to study their atmospheres, but encountered hurdles in measuring reliable atmospheric abundances. In my thesis, I work to overcome these challenges by studying these objects using high-resolution spectroscopy from Keck/KPIC, a unique single-mode fiber feed into NIRSPEC that provides R~35,000 spectra in the near-infrared. Besides studying substellar atmospheres with KPIC, I contributed significantly to its data reduction pipeline and calibration procedure.
With KPIC, I used atmospheric retrievals to characterize a large sample of planetary-mass companions and brown dwarfs to shed light on their formation history. First, I measured the carbon and oxygen abundances of high-mass brown dwarfs and low-mass M dwarfs (m≈60-90 MJup) and showed they are chemically homogeneous to their host stars (Chapters 2 and 3). I also made one of the first estimates of the vertical mixing rate in a L/T transition brown dwarf companion from its relative H₂O, CO, and CH₄ abundances (Chapter 2). Next, I carried out a survey of eight planetary-mass companions with estimated masses between 10-30 MJup (Chapter 4). I found that these companions also have C and O abundances clustered around the solar value, similar to abundances of stars in the same star-forming regions. In these studies, I made several isotopologue ratio measurements including ¹²CO/¹³CO and showed that a late-M dwarf companion has the same ¹²C/1¹³C and ¹⁶O/¹⁸O as its K6V host star. Overall, my KPIC studies show that companions with m≳10 MJup likely form as the tail-end of star formation, consistent with the conclusions from demographic and orbital architecture studies of substellar companions.
Next, I worked on addressing the over-massive brown dwarf problem, an emerging phenomenon where several brown dwarf companions have dynamical masses higher than predictions from evolutionary models given their luminosities. This problem can be solved if these objects are not single entities. Using VLTI/GRAVITY and VLT/CRIRES+, I resolved the first brown dwarf companion, Gliese 229B, into two nearly-equal mass brown dwarfs, Gliese 229 Ba and Bb, on a 12 day orbit (Chapter 5). Gliese 229Bab is the tightest substellar binary orbiting a star, and indicates that other over-massive brown dwarfs might also be unresolved, tight binaries. As a follow-up study, I analyzed JWST/MIRI spectrum (5-14 µm) of Gliese 229 Bab to show that both brown dwarfs have similar C/O and metallicities as their host star, as expected for a star-like formation scenario (Chapter 6).</p
Computational Design of Wearable Chemical Sensors for Personalized Healthcare
Get PDFWearable sweat sensors have the potential to revolutionize precision medicine as they can non-invasively collect molecular information closely associated with an individual’s health status. However, the majority of clinically relevant biomarkers cannot be continuously detected in situ using existing wearable approaches. Molecularly imprinted polymers (MIPs) are a promising candidate to address this challenge but haven’t yet gained widespread use due to their complex design and optimization process yielding variable selectivity. Despite their promise, MIPs have historically been known to be exceedingly difficult to optimize. Changes in the monomer/monomers used, polymerization solvent, and crosslinking agent have been shown to change the performance of MIP sensors significantly. This is particularly a concern in sweat-based sensors where the concentration of analytes is very low and chemical diversity is very high as a drop of sweat can contain vitamins, hormones, and amino acids. Consequentially, any sweat based sensor must exhibit high sensitivity (ability to detect low analyte concentrations) and selectivity (ability to distinguish one analyte from another). Computational methods have been introduced to design MIP sensitivity alone, however these prior methods do not cover all aspects essential for using a sensor in a wearable device such as selectivity optimization, detection of non-electroactive analytes, and scalable manufacturing. Here, we introduce a full computational method that allows for high throughput materials discovery for wearable devices. We will describe how to design novel sensing materials with QuantumDock, an automated computational framework for universal MIP development toward wearable applications. Then we delve into further technical details on signal transduction and scalable manufacturing approaches for these wearable devices. We present a number of novel devices designed with these computational methods including a wearable non-invasive phenylalanine monitoring system (the first of its kind), a wearable nutritional tracker ‘Nutritrek’ capable of monitoring a range of metabolic disorders, and an implantable pharmaceutical drug monitoring system for cancer patients
Thermal Kinetic Inductance Detectors (TKIDs) for Cosmic Microwave Background (CMB) Polarimetry
Get PDFThe modern era of precision cosmology has been driven by advances in detector technology and observing techniques. Observational cosmology is experiencing a rapid growth in detector numbers. New architectures are emerging
for low-loading applications such as far-infrared spectroscopy, ultra-sensitive low-threshold sensors for particle astrophysics, and dark matter investigations. Current millimeter-wave observatories use kilo-pixel arrays of detectors to measure the polarization of the Cosmic Microwave Background (CMB). There is a strong push within the CMB community to deploy new experiments with hundreds of thousands of detectors to achieve novel scientific outcomes.
However, for over a decade, CMB detectors have been limited by background noise, where fluctuations in the photon flux incident on the camera overshadow internal detector noise. As a result, improving instrument sensitivity now requires increasing the number of pixels. This focal plane size and
detector density increase significantly complicates integration and readout. Thermal Kinetic
Inductance Detectors (TKIDs) are an innovative solution for scaling up detector counts, offering high sensitivity and ease of multiplexing. TKIDs are narrow-bandwidth superconducting resonators that can be multiplexed and read out using a single transmission line via microwave frequency division multiplexing.
In this thesis, I present the design, development, and laboratory characterization of a TKID polarimeter for CMB studies at 150 GHz with a 25% bandwidth. I provide
a detailed physical model of TKID operation and readout, accurately predicting detector noise and responsivity. Three generations of prototype detectors were developed and tested, leading to the final tile design. The first generation demonstrated the feasibility of fabricating TKIDs with internal noise low enough for background-limited performance given the expected optical loading on our telescope. The second generation validated the scalability of the initial design to larger arrays and was crucial for refining fabrication processes, cosmic ray susceptibility testing, and readout development. The third generation
integrated the tested detector design with a polarization-sensitive planar phased-array antenna. This required precise fabrication of sub-micron microstrip lines and an in-depth understanding of both the antenna and detector fabrication processes. We show that antenna-coupled TKIDs achieve end-to-end optical efficiency comparable to existing Transition Edge Sensor (TES) detectors and exhibit smooth Gaussian
antenna beams matching the design spectral response.
Our efforts culminate in the design of a 64-pixel dual-polarization TKID array, intended for CMB observations in a telescope observing from the South Pole. This camera will be the first demonstration of TKIDs in the millimeter-wave regime, advancing the technology for future cosmological and astrophysical applications. I present results from in-lab dark and optical testing of the TKID focal plane, along
with design methodologies, electromagnetic simulations, and fabrication procedures for achieving high-yield, uniform TKID arrays
Advancing Structural Analysis with Computational Methods Development
Get PDFIn this thesis, a set of computational methods is developed to extend structural techniques beyond their conventional practice. First, we build in silico simulations and image processing protocols to design a new data acquisition workflow in cryo-electron tomography. This enables in situ visualization of macromolecular complexes at sub-nanometer resolution in a micron-scale field of view. Then, we demonstrate the applicability of a novel machine-learning algorithm in processing small molecule electron diffraction data for the first time. For most molecules tested, the correct ab initio structures can be obtained without the common practice of manual dataset curation. Finally, molecular dynamics simulations using crystallographic structures of protein and drug molecule complexes are performed to investigate the fundamental principles of a ternary binding property. A minimal forcefield with multi-scale coarse-graining enables alchemical free energy calculations at an unconventional size of perturbation while providing physical insight into the role of the drug linker and protein shapes
Shear-Normal Coupled Deformations in Anisotropic Structured Materials
Get PDFThe advent of additive manufacturing has allowed the design and engineering of a new class of materials known as metamaterials, or structured/architected materials. These metamaterials exhibit unique functionalities, such as ultrahigh strength-to-density ratios, which their base materials cannot achieve. Often designed to exhibit near-isotropic behavior, metamaterials derive their special properties from the distinctive deformation, dynamic motion, and elastic energy distribution of their micro- and meso-architectures. However, designing metamaterials for anisotropy, despite their ability to attain unique properties, is challenging. Fully characterizing anisotropic stiffness in planar loading requires six independent elastic tensor moduli. This high number of independent elastic stiffness parameters also expands the design space of structured materials and leads to unusual phenomena, such as materials that can shear under uniaxial compression. This direction-dependent shear-axial coupling is crucial for many applications such as shape-morphing, elastic wave manipulation devices and impact redirection.
This thesis aims to understand the fundamental limits of shear-normal coupled deformations in anisotropic structured materials. Currently, there are no established upper and lower bounds on anisotropic moduli achieving extreme elastic anisotropy, similar to the Hashin-Shtrikman bounds in isotropic composites. This range is known as G-closure and provides limits for the achievable tensors. To date, there are no experimental methods that can measure the stiffness parameters of fully anisotropic structured materials from a single experiment. To address these challenges, we first introduce a method to generate two-phase periodic anisotropic unit cell geometries and construct a database of unit cells with a diverse range of effective elasticity tensors. The constructed database is compared with the properties achieved by hierarchical laminates and identify the regions where hierarchical designs are necessary to reach a specific extreme elasticity tensor.
We then propose an experimental methodology to evaluate the anisotropic material properties. Our technique, which utilizes the virtual fields method, allows for the determination of six separate stiffness tensor parameters of two-dimensional structured materials using just one tension test. This method thus eliminates the need for multiple experiments as is typical in traditional methods. We show the accuracy of our method using synthetic data generated from finite element simulations as well as by conducting experiments on four additively manufactured specimens. The approach requires no stress data and uses the full-field displacement data measured using digital image correlation and global force data.
We present a method for creating functionally graded anisotropic structures that smoothly transition between unit cells with distinct patterns. Isotropic materials with spatially varying density gradients have been shown to exhibit unique characteristics such as superior energy absorption. However, achieving smooth spatial gradients in the anisotropic mechanical properties while ensuring the connectivity of adjacent meso-architectures is non-trivial. This method allows for independent control of several functional gradients, such as porosity, anisotropic moduli, and symmetry. We show that certain nonlinearly graded structures when designed with unit cells positioned at distinct corners of the property space boundary exhibit novel mechanical behaviors. We conclude by designing specific functionally graded structures that demonstrate peculiar behaviors such as selective strain energy localization, localized rotations, compressive strains under tension, and longitudinal-shear wave mode conversion.</p
Electron Dynamics in Molecular Qubits and Catalytic Films
Get PDFTwo significant areas within molecular and materials chemistry are explored: spin-phonon coupling in molecular qubits and electrocatalysis in cobalt oxyhydroxide thin films, detailed over four chapters and an appendix. The first chapter reviews advancements in molecular quantum information science, focusing on decoherence mechanisms in transition metal complex-based qubits, and introduces a dynamic ligand field model that categorizes decoherence regimes and designs qubits for various environments. The second chapter develops and enhances a ligand field theory model to quantify spin-phonon interactions in transition metal complexes, correlating theoretical insights with experimental data to improve quantum coherence in molecular qubits. The third chapter investigates the electron transport and dynamic defect states in cobalt-phosphate and cobalt-borate oxyhydroxide films, crucial for understanding their photoexcited states in oxygen evolution catalysis. The fourth chapter presents a novel magneto-electrochemical setup that quantifies magnetoenhancement in electrocatalytic current for water splitting, highlighting the potential of magnetic fields in enhancing electrocatalytic processes. This work provides both a physical inorganic framework and experimental insights for ongoing and future developments in molecular quantum information science and energy conversion
Development and Application of Proteomic and Genomic Methods in RNA Biology
Get PDFThis thesis contains three interrelated projects. Chapter 1 describes the development of a novel RNA-proteomics method: RNA-antisense purification followed by mass spectrometry (RAP-MS 2.0). It contains results of a RAP-MS 2.0 study profiling the protein partners of eight RNAs (7SL, 7SK, RMRP, U1, U2, U6, U7, and Xist) as well as a detailed, step-by-step protocol for the new method. Chapter 2 describes a quality control method for Split and Pool Identification of RBP targets (SPIDR). It identifies an underappreciated failure point in SPIDR experiments (the equal loading of antibody-IDs onto beads), and describes a method for monitoring and resolving this issue. Chapter 3 describes the application of SPIDR to ribosome-associated proteins in human cells. The study both validates existing structures and identifies novel interactions between nucleolar proteins and immature ribosomal RNA, and between protein trafficking factors and the large ribosomal subunit
Spectroscopic Investigation, Kinetic Analysis, and Ligand Field Theory Rationalization of Catalytic Reactivity for Data-Driven Methodology Development
Get PDFFirst-row transition metal catalysis can leverage one or two-electron redox chemistry to catalyze selective C–C bond formation between two stereoelectronically differentiated substrates. Owing to this redox flexibility, many competing reaction pathways could occur, leading to the formation of both desired and undesired products. The electronic structure of the catalytic intermediates and reaction conditions are empirically recognized to modulate product distributions, but identifying the underlying design principle is often challenging. Mechanistic elucidation of the catalytic cycle and spectroscopic elucidation of important factors that influence catalytic reactivity could be beneficial to this endeavor. With the aid of ligand field theory and molecular orbital theory, a direct relationship may be established between the electronic structures of the metal catalysts and the thermodynamic or kinetic parameters of the elementary transformation they catalyze. To this end, this thesis describes the effort of combining spectroscopy, reactivity interpretation, and reaction kinetics to understand Ni-catalyzed reductive alkenylation and acylation of benzylic electrophiles and Cu-catalyzed allylic alkylation of γ-butyric lactone. The research approach and the results described herein are anticipated to aid the emergent effort of data-driven reaction development