Dissertation

Spacecraft State Estimation and Stealth through Orbit-Perturbing Maneuvers: A Game Theory Approach

Reiter, Jason A

2020

Overview

In space mission architectures, redundancy of subsystems mitigates the risk of isolated malfunctions, but not the risk of the entire platform being targeted by hostile forces. This dissertation explores the use of deception as a tactical defense mechanism. The scenario of a satellite accomplishing an unknown mission while evading a dedicated ground sensor is modeled as a two-player zero-sum game, which supports a robust performance assessment based on both the satellite and the sensor optimizing against each other. Moreover, the methodology determines the optimal strategies and associated performance based on assumed constraints that can be varied parametrically, so the method can be adapted to a range of specific scenarios as well as to advances in underlying technologies. The two player game, which featured infinite strategy choices for both players, was solved using a reinforcement learning/neural network approach, specifically proximal policy optimization, capable of high-fidelity strategy tuning. Parametric sensitivity analysis on the results, computationally impractical with proximal policy optimization, was instead accomplished with coevolution, a genetic algorithm approach. The primary result is that, under reasonable technological assumptions for both players, an evading spacecraft can expect to avoid detection by an optimally tracking optical sensor just under 60% of the time, and can accomplish this evasion expending, on average, just under 16 m/s of ΔV every 5 days. Analysis also indicated that this result is more sensitive to the sensor’s parameters than the satellite’s. Additional results compare the computational methodologies employed. The impact of the research as a methodological innovation is discussed and future direction is offered for applying the methodology to other problems, as well as possible refinements in the context of this application.

• Publisher: ProQuest Dissertations & Theses
• Subject: Aerospace engineering / Algorithms / Applied Mathematics / Artificial intelligence / Computational physics / Game theory / Neural networks / Optimization / Sensors / Simulation / Space surveillance / Standard deviation / Surveillance / Velocity
• ISBN: 9798535586564