Hundreds of millions of people worldwide are affected by visual disabilities, due to various factors, including genetics, injury, infection, or aging. Majority of visual impairment are avoidable with timely intervention and early detection. The high prevalence of avoidable visual impairment is primarily due to underutilization of eye care services, stemming from challenges related to availability, accessibility, and affordability. Inadequate numbers of trained professionals pose a significant obstacle in expanding access to eye care services, particularly in rural areas with higher demand. Additionally, individuals with disabilities, and those facing financial and socioeconomic barriers encounter greater challenges in accessing eye care services, further exacerbating the issue.

Ocular imaging modalities, such as Optical Coherence Tomography (OCT) and fundus imaging, are vital for diagnosing retinal diseases. OCT generates high-resolution images of the retina, enabling early detection of abnormalities, while fundus imaging captures photographs of the eye's back, useful for screening retinal diseases. However, clinical OCT systems are large and require dedicated space, limiting their use in compact or field settings. Patients may also face challenges due to the need for head stabilization, especially those with movement difficulties or individuals with mental, physical or developmental disabilities.

A virtual reality (VR)-based headset that can perform multi-modal retinal imaging is designed, specifically OCT and scanning laser ophthalmoscopy. The headset is designed to be worn by patients while watching videos, enabling the simultaneous capture of OCT and fundus images. The project aims to overcome limitations by eliminating the requirement for mechanical head stabilization and skilled operators. Moreover, the use of portable VR headsets can enhance the affordability of eye care by reducing the need for dedicated space and trained professionals, and potentially introducing automation for image capture and diagnosis. Optical simulations are performed to test the performance of the proposed system in terms of retinal coverage and resolution. The feasibility of pupil tracking and the ability to control the user's gaze to improve optical performance is tested.

An interesting feature about birds is their ability to fly with open wings, but their footprintis almost the same as the bird’s body when perched on a branch. This feature is more prominent in birds such as the albatrosses. Albatrosses can fly 10,000 miles or 16,000 km in a stretch [1]. These long distances are possible due to their long, thin wings - up to 11 feet from wingtip to wingtip - capable of using oceanic air currents. However, near the ground, their footprint is as small as the bird’s. This reduced footprint is because of the efficient tuck-in motion of their wings. This ability to manage footprint as required is the prime inspiration of this project. These birds are also known to flap their wings 10% of the time, showing that oceanic currents are harvested efficiently to minimize the energy usage required to stay airborne.

Multirotors, a class of uncrewed aerial vehicles(UAVs), can take off and land in a smallfootprint. However, their nature of flight expends quite some energy, which affects their flight time. Fixed-wing aircraft cannot take off land in a small footprint, but they can glide when powered down due to the nature of their flight. Combining these two features resulted in hybrid vehicles called VTOLs or Vertical Takeoff Landing Aircraft. This project proposes bringing UAVs closer to operating in small footprints using wing morphing techniques. The ability to morph wings allows UAVs to be more agile in a low-footprint configuration(folded wings) and to land in denser spaces, like urban environments, rainforests, etc. Opening the wings to its entire span provides an efficient, gliding flight when required. This paper proposes a novel morphing technique, with its design driving criteria being to keep the mechanism simple to allow scalability and ease of manufacturing and prototypes the proposed morphing wing technique for aircraft with long wingspans. The mechanism involves a system of wing sections supported by torsional springs and actuated by nylon string and pulley. During takeoff and landing, the wings can stay in a tucked-in position, allowing for small footprint operations, and when required, these long, thin wings can harvest thermal updrafts to stay airborne and un-powered for an extended amount of time.

The concern about air pollution has led government agencies to design and implement mandates to replace some commercial fleets’ gasoline vehicles with Alternative Fuel Vehicles (AFVs). In Part One of this dissertation, I investigate the diffusion of AFV’s in the commercial sector. Commercial fleets are frequently the first target of government regulation because policy agencies can target a large number of vehicles while regulating fewer establishments relative to the household sector. Using stated preference survey data from over 2000 commercial and local government fleets in California, I estimate multinomial logit and nested logit models of fuel choice that predict the probability of choosing each type of AFV. Given certain assumptions about vehicle technology, these models predict that starting in year 2010, almost 17% of new vehicle purchases by the commercial and local government fleets will be electric, about 20% will be compressed natural gas, and almost 21% will be methanol vehicles.

I find that fuel choice probabilities differ depending on the market structure. Public agencies seem to be more AFV friendly than private firms. Important factors in fleet vehicle choice are the degree of familiarity of the firm’s staff with the AFV operation, the size of the establishment, government regulations, and the availability of the refueling infrastructure.

In Part Two, I review hypotheses about the determinants of local government agencies’ efficiency and use the stated preference survey data to test these hypotheses. Public choice models predict systematic differences among government agencies regarding their cost considerations and sensitivity to environmental issues. The empirical evidence identifies two factors that affect government agencies’ performance. The first factor is jurisdiction: an agency that has a more rigid boundary, such as a city or a county, seems to operate more efficiently than an agency that has more flexible geographic boundaries, as is the case with the special districts. The second factor is direct citizen voting: an agency director who is subject to re-election seems to coordinate a more efficient agency operation than one that is appointed to the job as a career position.

This work deals with optical devices in medicine and industrial automation. The first chapter describes a design to integrate retinal imaging into a virtual reality headset. Conventional methods of retinal imaging require patients to sit still and follow instructions. The headset is meant to facilitate testing of toddlers and autistic children. The second chapter covers optics in industrial automation, specifically the localization of objects. The feasibility of a rapid, non-contact, high accuracy stereo vision system is explored. Active stereo vision using zoom and rotation of cameras is modelled. Inherent errors in sensors reduce the potential accuracy. This chapter calculates the highest possible accuracy achievable using off the shelf components, when the cameras are placed one - five meters away.

Autonomous driving is a challenging task given the variety and complexity of drive sce- narios. Meta-learning, particularly few-shot learning, is a common approach to tackle a classification task with limited number of training data for each classes. In this project, we focus on a specific scenario of visually navigating vehicles along an unseen route through recognizing the waypoints with only a few available waypoint images and no need to re-train for a new course. To attack the targeted task, we proposed a hierarchical framework with two deep models in which the highlight is the StampNet, a few-shot learning architecture with a covariance estimator for recognizing waypoints. To train the models, we collected an associated indoor dataset with images along the routes in various buildings on UC Davis campus, and quantified the performance of the tests to explore the best hyperparameters. Additionally, to demonstrate the effectiveness of the proposed approach, the framework was implemented on a customized mini vehicle for an online test that run the models in real time to navigate the vehicle in unseen courses.

Due to a high demand for robotic manipulation, many research groups and companies have designed robotic hands to showcase these cutting-edge marvels of mechatronics. However, most of these developments have an inclination towards the prosthetics or biomedical usage, such as rehabilitation. The development of high-dexterity humanoid robotic hands is also crucial for the advancement of automation industries, ranging from electronics to aerospace. In this research, a novel design of a robotic humanoid hand is presented, with fully actuated fingers and high dexterity design to meet the requirements of an industrial assembly environment. One of the key features of the design is its ability to actuate each finger phalange independently, which allows it to perform various grasps poses that are currently difficult or underactuated in the traditional tendon-driven robotic hands. To carry this out, a state-of-the-art lead screw actuation system is developed for each phalange. The robotic hand also includes fingers’ abduction/adduction for power grasps of larger objects, wrist radial-ulnar deviation and wrist flexion/extension, which totally constitutes 18 Degrees of Freedom (DoF) for the whole hand. The developed humanoid robotic hand will be useful in industrial automation and autonomous teleoperation due to its unique functions and economical design.

This thesis focuses on the effect of the environment of galaxies on their star formation activity and the metal content of their interstellar medium. I develop a technique to reconstruct the underlying number density field of galaxies in the Cosmic Assembly Near-infrared Deep Extragalactic Legacy Survey (CANDELS). I then use these measurements to estimate stellar mass and environmental quenching efficiencies out to z~ 3.5. I find that the environmental quenching efficiency increases with increasing stellar mass and decreasing redshift. I show that a dense environment can quench very massive galaxies as early as z~3. These observations provide a supporting argument for a scenario where the termination of cool gas accretion happens in a dense environment, and the galaxy starts to consume its remaining gas reservoir in depletion time. The depletion time is shorter for massive galaxies, so they are expected to become quenched faster. I also study the gas-phase metallicity of galaxies in different environments using near-IR spectroscopy of the MOSFIRE Deep Evolution Field (MOSDEF) survey. Based on gas-phase oxygen abundance computed from rest-frame optical emission lines, I find that at a fixed stellar mass, galaxies in overdense regions have lower metallicity than their field counterparts at z~ 2.3, but they become more metal-rich as they evolve to z~1.5. My results suggest that the efficient gas cooling mechanisms at high redshifts result in the prominent accretion of primordial metal-poor gas into the galaxies in overdensities. However, as galaxies evolve to the lower redshifts (z<2), the shock-heated gas in overdensities with massive halos cannot cool down efficiently, ramping up the gas-phase metallicity of galaxies. Future surveys will provide statistically significant samples of high redshift galaxies in diverse environments that can further test results from the present study. I build a machine learning model to transfer the knowledge gained in fields with a wealth of observations to those which lack such extensive observations. The technique provides valuable information to optimize the observing strategy for future surveys and offers complimentary data in the wavebands not accessible by these surveys.

With the increasing complexity and dimensionality of data across many fields and applications, traditional data analysis methods fail to properly reveal the insight concealed within the data, made further inaccessible through the ``curse of dimensionality,'' complications inherent to high-dimensional spaces. Motivated by the recognition that high-dimensional data often lie near lower-dimensional structures or manifolds within the high-dimensional space, manifold learning algorithms aim to provide more meaningful and efficient representations by capturing the intrinsic geometry of data, facilitating downstream analysis and visualization of data.

Building on manifold learning and recognizing the fact that there does not exist a universally good representation of data, deep representation learning considers learning parameterized mappings in the biologically inspired form of neural networks and through the mechanism of gradient descent, often referred to as backpropagation. This extends the manifold learning representation capability beyond strictly attempting to recover the low-dimensional geometry of the data and allows for both tailoring the objective to the desired outcome and better introducing inductive biases, which are structures and mechanisms that are designed to prompt the learning system to favor solutions that meet certain priors, assumptions, and presuppositions about the task or data.

In the first part of this thesis, we demonstrate how using manifold learning techniques enables building solutions to problems in complex systems. Specifically, we first show how a class of localization problems in wireless sensor networks can be efficiently tackled by learning low-dimensional diffusion map representations through implicit pairwise relationships, vastly simplifying the original combinatorial optimization problem. Next, we analyze the US census data and show that the features learned through manifold learning can serve as potential multivariate socioeconomic indicators that are robust against partisan manipulation given that the design choices are few and abstract, and in particular we demonstrate interesting consequences of considering one such feature as a deprivation index.

In the second part, we propose algorithms for deep representation learning and demonstrate their competitive performance. The first chapter in this part considers the problem of anomaly detection, where we demonstrate through theoretical analysis, numerical simulations, and experimental benchmarking how a simple modification of the objective and anomaly score function of an autoencoder driven by a basic principle dramatically improves its anomaly detection performance, even beyond considerably more sophisticated algorithms with additional components. Further, we develop an improved scheme for metric few-shot classification within a statistical framework and inspired by human cognition and perception and show its competitive performance. Finally, we incorporate a modified variant of the developed few-shot classifier in a novel description-based navigation system design, which aims at significantly reducing the need for sensors and positioning devices, as well as the need for training in the target environment or exposure to substantial amounts of data from it.