How improving performance may imply losing consistency in event-triggered consensus

David Meister, Duarte Antunes, Frank Allgöwer

TL;DR

This paper investigates event-triggered control (ETC) for multi-agent systems aiming for consensus. It reveals a surprising trade-off: while using more information locally can improve performance, it can paradoxically make a previously consistent ETC scheme inconsistent. This finding helps reconcile seemingly contradictory results in existing literature.

What Problem Does This Paper Address?

Event-triggered control (ETC) promises efficiency by reducing communication and computation compared to traditional time-triggered control (TTC). A key claim is that ETC can achieve similar or better performance with fewer "events" (transmissions). This desirable property is termed "consistency." However, formally proving consistency, especially in complex multi-agent systems, is challenging. Existing research has shown conflicting results for ETC in consensus problems: some demonstrate clear performance advantages, while others show these advantages vanish or even reverse under certain conditions. This paper aims to bridge this gap by analyzing a fundamental multi-agent consensus problem and explaining why these seemingly contradictory outcomes arise.

Methods

The authors tackle this problem by analyzing a system of single-integrator agents with noise, aiming for consensus. They use a quadratic performance measure (average deviation from consensus) and a level-triggering rule for ETC. The core of their method involves:

  • Optimal Control Input Derivation: They derive optimal control inputs for both ETC and TTC schemes under two distinct information scenarios for local controllers. This ensures they are comparing against the best possible controllers for each setup.
  • Information Scenarios:
    1. Broadcast Information Only (Ip): Controllers only use information broadcast by other agents up to the last triggering instant.
  • Broadcast + Local State at Triggering (Ib): Controllers can use broadcast information and their own local state at the moment of triggering.
  • Consistency Analysis: They formally analyze the consistency of the ETC schemes by comparing their performance to the optimal TTC under equal average triggering rates. This involves deriving closed-form expressions for performance measures.
  • Reconciliation of Literature: A key methodological contribution is explaining how the differing information assumptions in previous works ([4] vs. [18]) lead to their contrasting consistency results.

Datasets

Not discussed in the paper. The paper focuses on theoretical analysis and simulation-based validation of its findings.

Key Insights and Ideas

  • The Information Trade-off: The most striking insight is that while using more information locally (scenario Ib) generally improves performance for both ETC and TTC, it can break the consistency of an ETC scheme. This means the ETC might perform worse than TTC under equal triggering rates, a counter-intuitive result.
  • Reconciling Contradictory Results: The paper provides a unified framework to explain why studies like [4] (showing ETC consistency) and [18] (showing inconsistency in a similar setup) arrived at different conclusions. The difference lies solely in the information available to the local controllers.
  • Optimal Control is Key: By deriving optimal control inputs for each information scenario, the authors ensure their consistency claims are robust. They are not comparing a suboptimal ETC to an optimal TTC, but rather the best possible controllers under each information assumption.
  • Consistency is Not Guaranteed: The work highlights that the benefits of ETC (lower triggering rates, better performance) are not automatic and depend critically on the system setup, particularly the information available to controllers.

Key Results

  • Scenario Ip (Broadcast Only): The ETC scheme with this information set is shown to be consistent. It outperforms the optimal TTC under equal average triggering rates by a factor of 3.
  • Scenario Ib (Broadcast + Local State): The ETC scheme with this information set can become inconsistent for a sufficiently large number of agents. While it still improves performance over the TTC scheme, the ETC's performance advantage over TTC can vanish or even reverse.
  • Analytical Proofs: The paper provides analytical proofs for its consistency claims, moving beyond numerical simulations for some aspects.
  • Simulation Validation: Numerical simulations confirm the theoretical findings, illustrating the performance differences and the emergence of inconsistency in scenario Ib as the number of agents increases. Table 1 shows concrete cost ratios. For instance, the cost ratio $J_{ETC}/J_{TTC}$ is approximately $1/3$ for scenario Ip, but can be greater than 1 for scenario Ib when $n=50$.

Limitations and Future Work

  • Specific Setup: The analysis is limited to a continuous-time single-integrator consensus problem with a complete communication graph and a level-triggering rule.
  • Generalization: The authors explicitly state that generalizing these findings to more complex system dynamics, communication topologies, performance measures, or triggering rules is a crucial direction for future research.
  • Practical Implications: While the paper provides theoretical insights, further work is needed to explore how these phenomena manifest in real-world distributed systems where information availability might be constrained. The authors suggest investigating how the uncovered phenomenon can provide new insights into the performance properties of decentralized ETC and inspire new research.

Summary updated

Abstract

Event-triggered control is often argued to lower the average triggering rate compared to time-triggered control while still achieving a desired control goal, e.g., the same performance level. However, this property, often called consistency, cannot be taken for granted and can be hard to analyze in many settings. In particular, the performance properties of decentralized event-triggered control schemes with respect to time-triggered control remain mostly unexplored. Therefore, in this paper, we examine these performance properties for a consensus problem considering single-integrator agent dynamics, a level-triggering rule, and a complete communication graph. We consider the long-term average quadratic deviation from consensus as a performance measure. For this setting, we show that enriching the information the local controllers use improves the performance of the consensus algorithm but renders a previously consistent event-triggered control scheme inconsistent. In addition, we do so while deploying optimal control inputs which we derive for both information cases and triggering schemes. With this insight, we can furthermore explain the relationship between two seemingly contrasting consistency results from the literature.
BibTeX
@article{meister2026how,
  author = {David Meister and Duarte Antunes and Frank Allgöwer},
  title = {{How improving performance may imply losing consistency in event-triggered consensus}},
  journal = {Automatica},
  year = {2026},
  doi = {10.1016/j.automatica.2025.112768},
}
Figure