Satellite missions generally require real-time knowledge of the satellite’s orientation in space. A star tracker, which operates by identifying constellations of stars in photographs, is generally the most accurate attitude determination system and therefore the preferred method. However, it is not uncommon for small satellite missions to have limited computing resources, such as radiation-hardened CPUs, as well as a limited budget. Commercial star trackers can handle these limitations, but most are prohibitively expensive and do not have public documentation of their software. Existing open-source star trackers exhibit good performance only in specific scenarios or require more powerful computing hardware. In order to reproducibly evaluate star tracking algorithms that are capable of running on low-compute satellite missions, and to provide reusable open-source implementations of these algorithms, we developed LOST: Open-source Star Tracker. We compare a suite of star tracking algorithms on performance metrics such as speed, accuracy, and memory usage, with a testing framework capable of generating realistic star images with various noise sources. Our evaluation determines which algorithms have the strongest performance under varying conditions such as motion blur, centroiding error, and number of false stars. In a scenario representative of a low-cost star tracker, the evaluation finds a set of algorithms that are able to identify over 95% of photos in less than 1 millisecond per image, with a peak memory usage of less than 1 MiB, backed by a database of less than 500 KiB. These results indicate that the algorithms implemented in LOST are suitable for running on embedded or radiation-hardened systems with very limited memory and compute power, while still achieving the high accuracy that is characteristic of star trackers.

The ZaP-HD Experiment combines the research fields of electric space propulsion and nuclear fusion by stabilizing the confinement of high energy density plasmas by accelerating ions to high axial velocities. This is known as the Sheared Flow Stabilized (SFS) Z-pinch, and has potential applications in both clean energy production and interstellar space propulsion. ZaP-HD employs numerous diagnostics to record plasma behavior with the goal of understanding how to improve the SFS Z-pinch. My research on ZaP-HD aims to create spatiotemporally resolved "contour maps" of the radial electron density distributions at fixed axial locations in the plasma flow. To collect data for these maps, I use a method known as Helium-Neon (HeNe) Interferometry, which records the interference experienced by a HeNe laser beam shining through a plasma to create a time-resolved curve of the chord integrated electron density of the plasma along the path of the beam. For each "pulse" of experimental plasma, I record four of these curves using four parallel laser beams which pass through the plasma perpendicular to, and at different distances from, the center axis of the Z-pinch. By recording electron density measurements under different experimental parameters, I can gain insight on how to increase the plasma density in future experiments. Increasing plasma density directly contributes to increasing the fusion reaction rate, and thus the energy output of a fusion plasma. I have recorded chord-integrated electron number densities on the order of 10^21 electrons/m^2, and I have observed a trend of the electron density increasing radially inward towards a peak at the center axis. My research aims to contribute to the global push for the advancement of clean and renewable energy production, as well as the development of spacecraft propulsion systems which could enable humanity to finally achieve interstellar space exploration.

Teaching robots to efficiently search for a target object (such as medication) in cluttered environments (such as a house) with limited prior information is a challenging yet important task, with applications ranging from home assistance to search and rescue. An ideal search policy, i.e. strategy, maximizes the accumulation of the target object while travelling along an efficient search route. We formulate object-search as a combinatorial-optimization problem known as the Stochastic Orienteering Problem (SOP), which is a graph traversal problem where an agent must identify and traverse a subset of nodes (such as furniture items, in our case) in a graph data-structure, with each edge associated with some cost and each node associated with some expected reward. The agent must choose a path that maximizes expected reward and keeps total travel cost under some prespecified bound, where the bound is informed by an unavoidable real-world constraint such as battery-usage. In our formulation, we call each node a "container", a catch-all term for any distinct area that can hold an object (such as a cabinet), and the edge costs represent the distances between each container. In this work, we mathematically show how this SOP can describe object-search tasks at a high-level, and present a simulated agent trained in a basic grid-world environment. We leverage the powerful reward-maximization capabilities of deep reinforcement-learning (a subfield of machine learning) to achieve near-optimal performance for solving this object-search SOP. The broader implication of this work is that real-world robotic object-search tasks are well-described by SOPs, since the multi-objective nature of SOPs forces the agent to choose search policies that both have a high-likelihood of finding the object(s) and do not exceed a hard constraint, such as energy expenditure.

The application of engineered origami structures has become increasingly popular for the past decades. Among the variety of origami patterns, the Resch pattern recently began to reveal its potential in the field. It is a planar tessellation composed of pre-defined polygons. Some of its interesting properties feature controllable morphability and self-supporting reformability. However, its static and dynamic response to an external load remains to be unveiled. Therefore, this project aims at studying the Resch-patterned origami structure’s folding behavior, associated stored potential energy, and impact mitigation capability. We first constructed its kinematic model in order to accurately predict the folding motion of a tessellation. The model was then augmented with a torsional spring embedded in each crease line to predict the force-displacement relationship under external loads. We fabricated a series of prototypes with polymers and conducted static compression tests to compare and calibrate the kinematic model. The force-displacement curve generated from the kinematic model was fitted to the experimental result, that the two curves shared a very similar profile. As for the physical model, it demonstrated consistent force-displacement and energy dissipation properties over cyclic compression-expansion tests. After studying the fundamental behavior of the Resch pattern, we performed dynamic impact tests on our physical model to explore its potential for impact mitigation. A cylindrical weight was dropped on the center of our Resch pattern at its natural posture, and by tracking the motion of the impactor, we determined the energy and momentum dispersed in the impact. In summary, the Resch-patterned origami structure’s unique properties exhibit great potential for impact-mitigating structures for deployable panels with repeated loads. We envision that the energy handling mechanism of the Resch pattern investigated herein can be employed in numerous engineering structures, including lightweight deployable architecture.