Skip navigation
Please use this identifier to cite or link to this item:
Title: Methods Toward the Design of Estimation and Control for Networked Multiagent Systems
Authors: Savas, Anthony John
Advisors: LeonardPoor, NaomiHarold EV
Contributors: Mechanical and Aerospace Engineering Department
Keywords: Actuator and sensor allocation
Decentralized control
Distributed estimation
Graph theory
Multiagent systems
Networked systems
Subjects: Mechanical engineering
Issue Date: 2022
Publisher: Princeton, NJ : Princeton University
Abstract: This dissertation explores distributed estimation and control of networked multiagent systems. We consider collaborative systems in which agents access information from their neighbors and use this information to better understand and manipulate the environment. We consider distributed filtering of a scalar linear stochastic process over a networked multiagent system and focus on the setting where communication between agents is corrupted by Gaussian noise. We show that communication noise is not easily handled by a two-stage consensus filter in the literature and propose a novel algorithm which does not suffer performance degradation under such communication noise. We discuss how to optimally tune two fixed gains to minimize the asymptotic error covariance of the filter. We consider designing a network of agents to perform distributed estimation and control of linear time-invariant systems. We develop a framework in which agents have the flexibility to change their feedback control parameters without needing to redesign their estimation strategies. We use the small-gain theorem and the bounded real lemma to characterize conditions under which this is possible using linear matrix inequalities. We show how linear consensus dynamics can be used to further extend the operating regime of this framework. Leveraging our distributed estimation and control framework, we develop a methodology to distinguish agents for the purpose of actuator and sensor selection. We use hypothesis testing to rigorously compare a set of centrality measures as selection tools, focusing on the actuator selection problem. We show that under our framework there is strong statistical evidence that, given sufficient edge density, betweenness centrality is a good metric for actuator selection over Erdos-Renyi random graphs in terms of minimizing a key matrix norm. We further show that these results broadly extend to other graph generation methods and are robust to actuator failure as well as network scale. When considering scenarios in which individual agents do not have full controllabilty of the system, we show that there is strong statistical evidence that degree centrality, rather than betweenness centrality, is a good selection heuristic for actuator placement.
Type of Material: Academic dissertations (Ph.D.)
Language: en
Appears in Collections:Mechanical and Aerospace Engineering

Files in This Item:
File Description SizeFormat 
Savas_princeton_0181D_14331.pdf8.55 MBAdobe PDFView/Download

Items in Dataspace are protected by copyright, with all rights reserved, unless otherwise indicated.