Distributed Detection and Control of Unexpected/Emergent Behaviors in Multiagent Systems
Abstract
In this project, we propose to develop an integrated model-based approach for distributed detection and control of unexpected/emergent behaviors in multiagent systems. The overall goal of the project is to enhance the understanding of emergent behaviors in distributed multiagent systems through rigorous analysis for modeling, detection, learning, and estimation, and control of interaction dynamics and interaction topologies. We consider multiagents that are modeled as a network of dynamical systems. One of the salient characteristics of multiagent systems is that local behaviors of individual agents may lead to certain emergent global behaviors through intermittent interactions among agents. While significant progresses have been made in the design and analysis of multiagent systems under limited sensing/communication topologies, there are very limited results available in the literature to address issues on how collective behaviors emerge. The multiagent system has some unique characteristics which make it remarkably difficult to analyze by simply using standard tools. The question then is how to detect and analyze the emergent behaviors of multiagent systems induced by system dynamics and by agent interactions. That is, two dominating factors involved in the study of emergent behaviors of multiagent systems are agents interaction dynamics and interaction topologies. Hence, it is of paramount importance to develop reliable strategies for detecting emergent behaviors of autonomous multiagent systems by identifying, designing, and controlling interaction dynamics and interaction topologies. We propose to explore the following two major elements for determining emergent behaviors: interaction rules and interaction topologies. The contributions of our proposed research are summarized below: - First, model-based validation and analysis will be conducted for finding out how system model parameters may affect emergent behaviors. A distributed estimation strategy and extremum seeking control will be designed to estimate the behavior objective functions for multiagent systems, which are instrumental in determining the emergent behaviors. - Second, nonlinear agent dynamics with nonlinear interaction rules will be considered. Model-based analysis will be performed through estimation of unknown interaction dynamics and analysis of control parameters in nonlinear interaction rules. - Third, we will address the problem of detecting emergent behaviors by distributed estimation of spectral properties of interaction topology matrices, which are a promising indicator for group behaviors and cluster behaviors. A predictive control strategy will be developed for the distributed estimation of behavior variables in the presence of interaction delays. - Fourth, we propose a group size monitoring algorithm. Accordingly, the model-based analysis will be enhanced for detecting emergent behaviors. The proposed work will be validated through rigorous theoretical analysis, extensive simulations using the AgentFly platform, and experimental tests. We anticipate that the results of this research will help us better understanding of the resulting structure, evolution, design and control of emergent group behaviors in multiagent systems.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 11, 2016
- Source ID
- FA87501510143
Entities
People
- Jing Wang
Organizations
- Bradley University
- Rome Laboratory
- United States Air Force