Date of Award

2026-07-01

Degree Name

Doctor of Philosophy

Department

Mechanical Engineering

Advisor(s)

John J. Bird

Second Advisor

Angel Flores-Abad

Abstract

State estimation is fundamental to autonomous systems operating in uncertain environments, where navigation, coordination, and decision-making depend on accurately inferring internal system states from incomplete and noisy observations. While classical estimation theory primarily focuses on reconstructing hidden states from available measurements, practical autonomous systems must also contend with sensing interruptions, communication constraints, limited computational resources, and nonlinear system dynamics that reduce the availability and usability of estimation-relevant information. This dissertation adopts an information-centric perspective on state estimation by introducing the concept of an Information Lifecycle, in which information is generated through system behavior, utilized by estimation algorithms under operational constraints, and analyzed through computational observability methods. Within this framework, observability serves as the mathematical foundation for quantifying the information available for state inference.

The first contribution develops an observability-aware autonomous control framework that enables agents to intentionally generate informative behaviors while accomplishing mission objectives. By embedding a state estimator within a reinforcement learning training architecture, the proposed method encourages control policies that improve the observability of hidden system states, enhancing estimation performance without requiring additional sensing resources.

The second contribution addresses information utilization in resource-constrained environments by developing a covariance-constrained observation decimation framework for state estimation under scheduled measurement updates. The proposed formulation predicts steady-state estimation performance through a single discrete algebraic Riccati equation evaluation, enabling a tradeoff between measurement assimilation frequency and estimation performance.

The third contribution advances information analysis by establishing a mathematical relationship between the Unscented Transformation and empirical observability analysis. The resulting sigma-point framework recovers empirical observability information directly from Unscented Transformation covariance propagation and extends the formulation using the Spherical Simplex Unscented Transformation to reduce the required number of system simulations while preserving trajectory-based observability analysis.

Collectively, these contributions establish a unified theoretical and computational framework for improving the generation, utilization, and analysis of information for state estimation in nonlinear autonomous systems. By treating information as an engineering resource throughout the estimation process, this dissertation provides new methods that improve estimation performance, reduce computational complexity, and support the development of more capable autonomous systems operating under sensing, communication, and computational constraints.

Language

en

Provenance

Received from ProQuest

File Size

129 p.

File Format

application/pdf

Rights Holder

Andres Enriquez Fernandez

Download

Included in

Engineering Commons

Share

COinS