The control of multiple unmanned aerial vehicles (UAVs) in a swarm or cooperative team scenario has been a topic of great interest for well over a decade, growing steadily with the advancements in UAV technologies. In the academic community, a majority of the studies conducted rely on simulation to test developed control strategies, with only a few institutions known to have nurtured the infrastructure required to propel multiple UAV control studies beyond simulation and into experimental testing. With the Cal Poly UAV FLOC Project, such an infrastructure was created, paving the way for future experimentation with multiple UAV control systems. The control system architecture presented was built on concepts developed in previous work by Cal Poly faculty and graduate students. An outer-loop formation flight controller based on a virtual waypoint implementation of potential function guidance was developed for use on an embedded microcontroller. A commercially-available autopilot system, designed for fully autonomous waypoint navigation utilizing low cost hardware and open source software, was modified to include the formation flight controller and an inter-UAV communication network. A hardware-in-the-loop (HIL) simulation was set up for multiple UAV testing and was utilized to verify the functionality of the modified autopilot system. HIL simulation results demonstrated leader-follower formation convergence to 15 meters as well as formation flight with three UAVs. Several sets of flight tests were conducted, demonstrating a successful leader-follower formation, but with relative distance convergence only reaching a steady state value of approximately 35 +/- 5 meters away from the leader.

Benefits of developing missions with multiple formation flying spacecraft as an alternative to a traditional monolithic vehicle are becoming apparent. In some cases, these missions can lower cost and increase flexibility among other situational advantages. However, there are various limitations that are imposed by these missions that are centered on the concept of maintaining the necessary formation. One such limitation is that of the propulsion system required for each spacecraft. To mitigate the complexity and mass of the onboard propulsion, the pairing of electromagnetic actuators and differential drag to replace the functionality of a propulsive system is investigated. By using COTS magnetorquer boards to command satellite orientation, a scenario in which two 3U CubeSats are initially deployed from the ISS NanoRacks at an altitude of 400 km. They are then commanded to achieve a relative separation of 1 km and hold the spacing to demonstrate the capability of formation flight. The scenario was simulated through the MATLAB/Simulink platform and the magnitude of the necessary command torques were determined. By comparison to the ISIS magnetorquer board, the necessary command torques seem relatively high than compared to what the actuator is capable of. The ISIS board may supply ~5e-6 Nm of torque while the mission requires as much as 3e-3 Nm at times. However, by extending the settling time of the control law at the expense of absolute orientation control, the control torques necessary to carry out the simulated mission are well within the bounds of the ISIS magnetorquer boards as well as other COTS boards. With this alteration, mission feasibility is determined. It should be noted that further analysis should be conducted regarding concerns with CubeSat detumble to further confirm feasibility.

In addition to serving as a brief overview of aspects relevant to reduced-order modeling (in particular balanced-state and modal techniques) as applied to structural finite element models, this work produced tools for visualizing the relationship between the modes of a model and the states of its balanced representation. Specifically, error contour and mean error plots were developed that provide a designer with frequency response information absent from a typical analysis of a balanced model via its Hankel singular values. The plots were then used to analyze the controllability and observability aspects of finite element models of an illustrative system from a modal perspective -- this aided in the identification of computational artifacts in the models and helped predict points at which to halt the truncation of balanced states. Balanced reduced-order reference models of the illustrative system were implemented as part of a direct adaptive control algorithm to observe the effectiveness of the models. It was learned that the truncation point selected by observing the mean error plot produced the most satisfactory results overall -- the model closely approximated the dominant modes of the system and eliminated the computational artifacts. The problem of improving the performance of the system was also considered. The truncated balanced model was recast in modal form so that its damping could be increased, and the settling time decreased by about eighty percent.

Space debris has become an increasingly hazardous obstacle to continued spaceflight operations. In an effort to mitigate this problem an investigation of the feasibility of a microsatellite active debris removal system was conducted. Through proposing a novel concept of operation, utilizing a grapple-and-tug system architecture, and by analyzing each resultant mission phase in the frame of a representative example, it was found that microsatellite scale systems are capable of fulfilling the active debris removal mission. Analysis of rendezvous, docking, control and deorbit mission requirements determined that the design of a grapple-and-tug system will be driven by sizing of the propellant required to deorbit the target vehicle. Further sensitivity analysis determined that target altitude and mass are critical factors in determining the capabilities of a microsatellite mission. Preliminary sizing demonstrated that hardware considerations for both satellite core and mission related activities do not impede microsatellite feasibility. Further investigation of microsatellite debris removal missions including detailed design analysis and engineering is suggested.

The Pyramidal Reaction Wheel Platform, or PRWP, is used to simulate three-axis controls in a torque free space-like environment. The primary purpose of the system will be to evaluate the effects of conjoining sensors to maximize pointing accuracy. Furthermore, the system will incorporate a star tracker in conjunction with a Simulated Star Field (SSF) to better estimate the PRWP orientation. For the sake of this document, however, the goal is to implement a gyroscope, wheel rate sensors, and a make-shift accelerometer—to the PRWP—and integrate a controls algorithm such that three-axis controls are achieved for the PRWP. Three sensors were either better integrated into the system or added altogether. Tachometers were created as a form of hardware circuitry to measure each wheel rate with an accuracy of approximately 2.5 Hz (nearly 15 radians per second). The TAC board circuitry converted each motors encoder output into a speed by use of a frequency to voltage converter. Additionally, although three gyroscopes had been implemented previously, the system was better incorporated into the model such that it was directly transformed via a ROBOSTIX ADC converter before being relayed to SIMULINK via a Bluetooth link. The MEMS gyroscopes allowed for very accurate rate measurements—with a minimum resolution of approximately 0.25 radians per second. Finally, a makeshift accelerometer was incorporated into the system for the purpose of system identification. The accelerometer was incorporated into the system by utilizing a discrete time derivative of the gyroscope readings. However, thankfully a system of two accelerometers can be later utilized to achieve an accuracy of approximately 6 degrees per second-second in the x-axis and 2-3 degrees per second-second in the y- and z-axes. A controls test was performed where the starting location was qo=[0, 0, sqrt(2)/2, sqrt(2)/2] and the target location was qc=[0, 0, 0, 1]. At 80 seconds, the pointing accuracy was 70 degrees around the target and the system was unable to settle during the 80 second trial. The inaccuracy was because of the low frequency of operation of the system—1 Hz. Additionally, the platform reacts slowly to sensor readings and commands. The coupling of these issues causes the pointing accuracy to high. Furthermore, through experimental testing, the maximum wheel rate was found to be approximately 6400 RPM at a duty cycle of 50% at an 8000Hz PWM application due to the Pololu MD01B design limitations: low voltage range (up to 16V), low limit current limiter (5A), and high susceptibility to overheating for large currents.

The autonomous driving (AD) industry is advancing at a rapid pace. New sensing technology for tracking vehicles, controlling vehicle behavior, and communicating with infrastructure are being added to commercial vehicles. These new automotive technologies reduce on road fatalities, improve ride quality, and improve vehicle fuel economy. This research explores two types of automotive sensor fusion systems: a novel radar/camera sensor fusion system using a long shortterm memory (LSTM) neural network (NN) to perform data fusion improving tracking capabilities in a simulated environment and a traditional radar/camera sensor fusion system that is deployed in Mississippi State’s entry in the EcoCAR Mobility Challenge (2019 Chevrolet Blazer) for an adaptive cruise control system (ACC) which functions in on-road applications. Along with vehicles, pedestrians, and cyclists, the sensor fusion system deployed in the 2019 Chevrolet Blazer uses vehicle-to-everything (V2X) communication to communicate with infrastructure such as traffic lights to optimize and autonomously control vehicle acceleration through a connected corridor

This thesis outlines an orbit determination and 3-axis attitude determination system for use on orbit as applicable to 1U CubeSats and other picosatellites. The constraints imposed by the CubeSat form factor led to the need for a simple configuration and relaxed accuracy requirements. To design a system within the tight mass, volume, and power constraints inherent to CubeSats, a balance between hardware complexity, software complexity and accuracy is sought. The proposed solution consists of a simple orbit propagator, magnetometers with a magnetic field look-up table, Sun sensors with an analytic Sun direction model, and the TRIAD method to combine vector observations into attitude information. The orbit propagator is a simple model of a circular trajectory with several frequently updated parameters and can provide orbital position data with average and maximum errors—when compared to SGP4—of less than 3.7km and 10.7km for 14 days. The magnetic field look up table provides useful information from a small memory footprint; only 480 data points provide a mean error of approximately 0.2° and a maximum error of approximately 2°—when compared to the IGRF model. The Sun’s direction is modeled, and as expected, can be modeled simply and accurately. Combining the magnetic field and Sun direction models with inaccurate sensors and the TRIAD method results in useful attitude information from a very simple system. A system with Sun sensor error standard deviation of 1° and magnetometer error standard deviation of 5° yields results with average error of only 2.74°, and 99% of the errors in this case are less than approximately 13°. The system outlined provides crude attitude determination with software and hardware requirements that are well within the capabilities of current 1U CubeSats—something that many other systems, such as Kalman filters or star trackers, cannot do. It also provides an excellent starting point for future ADCS systems, which will significantly increase the ability of CubeSats.

Recent interest surrounding large scale satellite constellations has increased analysis efforts to create the most efficient designs. Multiple studies have successfully optimized constellation patterns using equations of motion propagation methods and genetic algorithms to arrive at optimal solutions. However, these approaches are computationally expensive for large scale constellations, making them impractical for quick iterative design analysis. Therefore, a minimalist algorithm and efficient computational method could be used to improve solution times. This thesis will provide a tool for single target constellation optimization using spherical trigonometry propagation, and an evolutionary genetic algorithm based on a multi-objective optimization function. Each constellation will be evaluated on a normalized fitness scale to determine optimization. The performance objective functions are based on average coverage time, average revisits, and a minimized number of satellites. To adhere to a wider audience, this design tool was written using traditional Matlab, and does not require any additional toolboxes. To create an efficient design tool, spherical trigonometry propagation will be utilized to evaluate constellations for both coverage time and revisits over a single target. This approach was chosen to avoid solving complex ordinary differential equations for each satellite over a long period of time. By converting the satellite and planetary target into vectors of latitude and longitude in a common celestial sphere (i.e. ECI), the angle can be calculated between each set of vectors in three-dimensional space. A comparison of angle against a maximum view angle, , controlled by the elevation angle of the target and the satellite’s altitude, will determine coverage time and number of revisits during a single orbital period. Traditional constellations are defined by an altitude (a), inclination (I), and Walker Delta Pattern notation: T/P/F. Where T represents the number of satellites, P is the number of orbital planes, and F indirectly defines the number of adjacent planes with satellite offsets. Assuming circular orbits, these five parameters outline any possible constellation design. The optimization algorithm will use these parameters as evolutionary traits to iterate through the solutions space. This process will pass down the best traits from one generation to the next, slowly evolving and converging the population towards an optimal solution. Utilizing tournament style selection, multi-parent recombination, and mutation techniques, each generation of children will improve on the last by evaluating the three performance objectives listed. The evolutionary algorithm will iterate through 100 generations (G) with a population (n) of 100. The results of this study explore optimal constellation designs for seven targets evenly spaced from 0° to 90° latitude on Earth, Mars and Jupiter. Each test case reports the top ten constellations found based on optimal fitness. Scatterplots of the constellation design solution space and the multi-objective fitness function breakdown are provided to showcase convergence of the evolutionary genetic algorithm. The results highlight the ratio between constellation altitude and planetary radius as the most influential aspects for achieving optimal constellations due to the increased field of view ratio achievable on smaller planetary bodies. The multi-objective fitness function however, influences constellation design the most because it is the main optimization driver. All future constellation optimization problems should critically determine the best multi-objective fitness function needed for a specific study or mission.

Recently, various self-driving and driving assistance systems such as Advanced Driver Assistance System (ADAS) have been developed with the intent to reduce the number of motor vehicle accidents. While self-driving systems have been proven to reduce traffic accidents, the systems sometimes make other drivers confused because of their mechanical behavior. To avoid confusion and possible error, it is necessary to construct self-driving systems that exhibit human-like behaviors. Risk Potential theory has been used to construct models that successfully represent driver behavior, especially expert behavior. This project uses Risk Potential theory to construct and evaluate a collision avoidance driver model which uses braking to avoid potential collisions with pedestrians. As a first step, a basic driver model which uses Risk Potential theory is constructed and evaluated using metrics such as collision avoidance, comfortability, and false alarm avoidance. Second, human driving data is collected to observe driver’s risk perception during interactions with a pedestrian. Finally, our proposed driver models improve on standard RP model’s performance but comparisons of the models with observed human performance reveal opportunities for further improvement.

Augmented reality is a field of technology in which the real world is overlaid with additional information from a computer generated display. Enhancements to augmented reality technology presently support limited mobility which is expected to increase in the future to provide much greater real world functionality. This work reports on a set of experiments that investigate performance in search and rescue navigating tasks using augmented reality. Augmentation consisted of a spatially and temporally registered map of a maze that was overlaid onto a real world maze. Participants were required to traverse the maze, answer spatially oriented questions in the maze, acquire a target object, and exit. Pre and post hoc questionnaires were administered. Time and accuracy data from one hundred twenty participants were collected across six treatments. The between subject treatments, which had an equal number of male and female participants, were a control condition with only a compass, a control condition with a paper map available prior to maze traversal and four experimental conditions consisting of combinations of egocentric and exocentric maps, and a continuously on and on demand map display. Data collected from each participant consisted of time to traverse the maze, percent of the maze covered, estimations of euclidian distance and direction, estimations of cardinal direction, and spatial recall. Data was also collected via pre and post hoc questionnaires. Results indicate that best performance with respect to time was in the control condition with a map. The small size of the maze could have facilitated this result through route memorization. Augmented reality can offer enhancement to performance as navigational tasks become more complex and saturate working memory. Augmented reality showed best performance in accuracy by facilitating participants' coverage of the maze. Exocentric maps generally exhibited better performance than egocentric maps. On demand displays also generally resulted in better performance than continuously on displays. Gender differences also were evident with males exhibiting better performance than females. Participants reporting an initial tendency to not rotate maps exhibited better performance than those reporting a tendency to rotate maps. Enhancements being made to augmented reality and related technologies will result in more features, improved form factor for users, and improved performance in the future. Guidelines provided in this work seek to ensure augmented reality systems continue to progress in enhancing performance