Consensusability of linear interconnected multi-agent systems

Mustafa Sahin Turan, Liang Xu, Giancarlo Ferrari-TrecatearXiv 2020 · OpenReview

TL;DR

This paper investigates how physical connections between agents in a multi-agent system affect their ability to reach consensus. The authors derive new necessary and sufficient conditions for consensusability in these "interconnected" systems, showing that weaker physical coupling and faster communication generally promote consensus.

What Problem Does This Paper Address?

Many real-world multi-agent systems (MASS), like power grids or robotic swarms, don't just rely on communication links for coordination; their agents are also physically interconnected. This physical coupling means the agents' dynamics influence each other directly, not just through information exchange. Existing research on "consensusability" (whether a system can reach agreement) has largely focused on systems where agents are only coupled via communication. This paper tackles the more complex problem of understanding and characterizing consensusability in linear MASS where agents have both physical and communication interconnections. The key challenge is to disentangle the effects of these two types of coupling on the system's ability to achieve a common state.

Methods

The authors approach this problem by first formulating the dynamics of linear interconnected multi-agent systems (LIMASs) that include both physical coupling (represented by a physical graph Laplacian, $L_p$) and communication coupling (represented by a communication graph Laplacian, $L_c$). They then analyze the conditions under which a linear, static, decentralized control gain $K$ can drive the system to consensus.

A crucial step involves simplifying the problem by assuming that the physical and communication Laplacians commute ($L_p$ and $L_c$ commute). This assumption allows for simultaneous diagonalization of these matrices, transforming the problem into a set of independent sub-problems. For higher-order systems (where agent states are vectors), they derive a sufficient condition for consensusability based on the eigenvalues of the transformed system matrices and the control gain. They also establish a necessary condition by analyzing the determinant of the system's state matrix.

For the simpler case of first-order subsystems (scalar states), they relax the commutativity assumption and derive both necessary and sufficient conditions for consensusability. These conditions are expressed in terms of the agent's intrinsic dynamics, the eigenvalues of the physical and communication Laplacians, and the control gain. The analysis leverages concepts from algebraic graph theory and linear systems theory, particularly focusing on matrix stability and controllability.

Datasets

Not discussed in the paper. The paper uses simulated systems for validation.

Key Insights and Ideas

  • Physical coupling complicates consensus: The presence of physical interconnections ($L_p \otimes A_p$) significantly alters the consensusability conditions compared to systems with only communication coupling.
  • Weak physical coupling is beneficial: The derived conditions suggest that weaker physical coupling (smaller $a$ or smaller eigenvalues of $A_p$) generally favors consensusability. This is intuitive, as strong physical links can introduce conflicting dynamics.
  • Fast information diffusion is key: Similar to existing results, faster information diffusion rates, indicated by eigenvalues of the communication graph Laplacian ($L_c$) being closer to each other (approaching a complete graph), also promote consensusability.
  • Interplay matters: The consensusability is not solely determined by the physical or communication graph in isolation but by their interplay with the agent dynamics and the control strategy.
  • Scalar systems offer insights: Analyzing first-order subsystems, while less general, provides a clearer understanding of how the separation of eigenvalues in the physical and communication Laplacians affects consensus.

Key Results

  • Sufficient Condition (Higher-Order, Commuting Laplacians): Under assumptions of commuting Laplacians and controllable subsystems, consensusability is achieved if a specific inequality involving eigenvalues of $L_p$, $L_c$, and the agent dynamics is met. This condition is related to the simultaneous stabilizability of a set of linear systems.
  • Necessary Condition (Higher-Order): A necessary condition for consensusability is derived, which involves the determinant of the transformed system matrices and a ratio of eigenvalues from the communication graph Laplacian. This condition can be used to rule out consensusability.
  • Necessary and Sufficient Conditions (First-Order): For scalar systems, explicit conditions (C1 and C2) are provided for consensusability, along with the range of possible control gains. A necessary condition is also given, which is a generalization of the higher-order case.
  • Simulation Validation: A simulation example demonstrates that a LIMAS satisfying the sufficient conditions can indeed reach consensus, with the deviation from the consensus state converging to zero.

Limitations and Future Work

  • Commutativity Assumption: The analysis for higher-order systems relies on the assumption that the physical and communication Laplacians commute. The authors acknowledge that this is a strong assumption and plan to remove it in future work.
  • Physical Coupling Matrix Structure: The sufficient condition for higher-order systems also relies on a specific structure for the physical coupling matrix ($A_p = aA$). This assumption is also a target for future relaxation.
  • Eliminating Assumptions: A primary future research direction is to eliminate these technical assumptions to obtain more general results.
  • Beyond Eigenvalue Information: The authors aim to relate consensusability conditions to other graph properties beyond just eigenvalue information.
  • Multi-Layer Consensus: Investigating multi-layer consensus schemes, where subsystems are coupled through multiple graphs, is another promising future perspective.

Summary updated

Abstract

Consensusability is an important property for many multi-agent systems (MASs) as it implies the existence of networked controllers driving the states of MAS subsystems to the same value. Consensusability is of interest even when the MAS subsystems are physically coupled, which is the case for real-world systems such as power networks. In this paper, we study necessary and sufficient conditions for the consensusability of linear interconnected MASs. These conditions are given in terms of the parameters of the subsystem matrices, as well as the eigenvalues of the physical and communication graph Laplacians. Our results show that weak coupling between subsystems and fast information diffusion rates in the physical and communication graphs favor consensusability. Technical results are verified through computer simulations.
BibTeX
@misc{turan2020consensusability,
  author = {Mustafa Sahin Turan and Liang Xu and Giancarlo Ferrari-Trecate},
  title = {{Consensusability of linear interconnected multi-agent systems}},
  year = {2020},
  eprint = {2003.07121},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2003.07121},
}
Figure