Browsing by Author "Atkins, Ella"

Now showing 1 - 5 of 5
  • Thumbnail Image
    The Application of Reinforcement Learning for Interceptor Guidance
    Porter, Daniel Michael (Virginia Tech, 2024-10-04)
    The progression of hypersonic vehicle research and development has presented a challenge to modern missile defenses. These attack vehicles travel at speeds of Mach 5+, have low trajectories that result in late radar detections, and can be highly maneuverable. To counter this, new interceptors must be developed. This work explores using machine learning for the guidance of these interceptors through applied steering commands, with the intent to improve upon traditional guidance methods. Specifically, proximal policy optimization (PPO) was selected as the reinforcement learning algorithm due to its advanced and efficient nature, as well as its successful use in related work. A framework was developed and tuned for the interceptor guidance problem, combining the PPO algorithm with a specialized reward shaping method and tuned parameters for the engagements of interest. Low-fidelity vehicle models were used to reduce training time and narrow the scope of work towards improving the guidance algorithms. Models were trained and tested on several case studies to understand the benefits and limitations of an intelligently guided interceptor. Performance comparisons between the trained guidance models and traditional methods of guidance were made for cases with supersonic, hypersonic, weaving, and dynamically evasive attack vehicles. The models were able to perform well with initial conditions outside of their training sets, but more significant differences in the engagements needed to be included in training. The models were therefore found to be more rigid than desired, limiting their effectiveness in new engagements. Compared to the traditional methods, the PPO-guided interceptor was able to intercept the attacker faster in most cases, and had a smaller miss distance against several evasive attackers. However, the PPO-guided interceptor had a lower percent kill against nonmaneuvering attackers, and typically required larger lateral acceleration commands than traditional methods. This work acts as a strong foundation for using machine learning for guiding missile interceptors, and presents both benefits and limitations of a current implementation. Proposals for future efforts involve increasing the fidelity and complexity of the vehicles, engagements, and guidance methods.
  • Thumbnail Image
    Efficient Vertical Structure Correlation and Power Line Inference
    Flanigen, Paul; Atkins, Ella; Sarter, Nadine (MDPI, 2024-03-05)
    High-resolution three-dimensional data from sensors such as LiDAR are sufficient to find power line towers and poles but do not reliably map relatively thin power lines. In addition, repeated detections of the same object can lead to confusion while data gaps ignore known obstacles. The slow or failed detection of low-salience vertical obstacles and associated wires is one of today’s leading causes of fatal helicopter accidents. This article presents a method to efficiently correlate vertical structure observations with existing databases and infer the presence of power lines. The method uses a spatial hash key which compares an observed tower location to potential existing tower locations using nested hash tables. When an observed tower is in the vicinity of an existing entry, the method correlates or distinguishes objects based on height and position. When applied to Delaware’s Digital Obstacle File, the average horizontal uncertainty decreased from 206 to 56 ft. The power line presence is inferred by automatically comparing the proportional spacing, height, and angle of tower sets based on the more accurate database. Over 87% of electrical transmission towers were correctly identified with no false negatives.
  • Thumbnail Image
    Enhanced Navigation Using Aerial Magnetic Field Mapping
    Owens, Dillon Joseph (Virginia Tech, 2024-01-23)
    This thesis applies the methods of previous work in aerial magnetic field mapping and use in state estimation to the Virginia Tech Swing Space motion capture indoor facility. State estimation with magnetic field data acquired from a quadrotor is comparatively performed with Gaussian process regression, a multiplicative extended Kalman filter, and a particle filter to estimate the position and attitude of an uncrewed aircraft system (UAS) at any point in the motion capture testing environment. Motion capture truth data is used in the analysis. The first experimental method utilized in this thesis is Gaussian process regression. This machine learning tool allows us to create three-dimensional magnetic field maps of the indoor test space by collecting magnetic field vector data with a small UAS. Here, the maps illustrate the 3D magnetic field strengths and directions in the Virginia Tech Swing Space motion capture lab. Also, the magnetic field spatial variation of the test space is analyzed, yielding higher magnetic field gradient at lower heights above the ground. Next, the multiplicative extended Kalman filter is used with our Gaussian process regression magnetic field maps to estimate the attitude of the quadrotor. The results indicate an increase in attitude estimation accuracy when magnetic field mapping is utilized compared to when it is not. Here, results show that the addition of aerial magnetic field mapping leads to enhanced attitude estimation. Finally, the particle filter is utilized with support from our magnetic field maps to estimate the position of a small quadrotor UAS. The magnetic field maps allow us to obtain UAS position vectors by tracking UAS movement through magnetic field data. The particle filter gives three-dimensional position estimates to within 0.2 meters for five out of our eight test flights. The root mean square error is within 0.1 meters for each test flight. The effects of magnetic field spatial variation are also analyzed. The accuracy of position estimation is higher for two out the four flights in the maximum magnetic gradient area, while the accuracy is similar in both minimum and maximum gradient regions for the remaining two flights. There is evidence to support an increase in accuracy for high magnetic variation areas, but further work is needed to confirm utility for practical applications.
  • Thumbnail Image
    A Study of the Wind Sensing Performance of Small Pusher and Puller Hexacopters
    González-Rocha, Javier; Sharma, Prashin; Atkins, Ella; Woolsey, Craig A. (American Institute of Aeronautics and Astronautics, 2023-09)
  • Thumbnail Image
    Time Delay Mitigation in Aerial Telerobotic Operations Using Predictors and Predictive Displays
    Sakib, Nazmus (Virginia Tech, 2024-05-23)
    Semi-autonomous uncrewed aerial vehicles (UAVs) are telerobotic operations by definition where the UAV assumes the role of a telerobot and the human assumes the role of a supervisor. All telerobotic operations are susceptible to time delays due to communication, mechanical, and other constraints. Typically, these delays are small and do not affect the telerobotic operation for most of the tasks. However, for long-distance telerobotic operations like interplanetary rovers, deep underwater vehicles, etc. the delays can be so significant that they can render the entire operation void. This dissertation investigates the use of a novel heterogeneous stereo-vision system to mitigate the effects of time delays in a UAV-based visual interface presented to a human operator. The heterogeneous stereo-vision system consists of an omnidirectional camera and a pan-tilt-zoom camera. Two predictive display setups were developed that modify the delayed video imagery that would otherwise be presented to the operator in a way that provides an almost immediate visual response to the operator's control actions. The usability of the system is determined through human performance testing with and without the predictive algorithms. The results indicate that the predictive algorithm allows more efficient, accurate, and user-friendly operation. The second half of the dissertation deals with improving the performance of the predictive display and expanding the concept of the prediction from a stationary gimbal-camera system to a moving 6 DoF aircraft. Specifically, it talks about a novel extended Kalman filter (EKF)-based nonlinear predictor – the extended Kalman predictor (EKP) – and compares its performance with two linear predictors, the Smith predictor (SP) and the Kalman predictor (KP). This dissertation provides the mathematical formulation of the EKP, as well as the two linear predictors, and describes their use with simulated flight data obtained using a nonlinear motion model for a small, fixed-wing UAV. The EKP performs comparably to the KP when the aircraft motion experiences small perturbations from a nominal trajectory, but the EKP outperforms the KP for larger excursions. The SP performs poorly in every case.