Time- versus event-triggered consensus of a single-integrator multi-agent system

David Meister, Frank Aurzada, Mikhail Lifshits, Frank Allgöwer

TL;DR

In a surprising twist on control theory "common wisdom," this paper proves that Event-Triggered Control (ETC)—usually the gold standard for communication efficiency—actually performs worse than simple periodic Time-Triggered Control (TTC) in large-scale multi-agent consensus networks. The authors demonstrate that as the number of agents grows, the decentralized nature of triggering decisions leads to a performance "tipping point" where the presumed advantages of event-based updates vanish.

What Problem Does This Paper Address?

In Networked Control Systems (NCS), engineers face a constant trade-off: how do you maintain high control performance while minimizing communication over a shared, limited-bandwidth network?

The standard answer has long been Event-Triggered Control (ETC). Instead of sending updates at fixed intervals (Time-Triggered), agents only communicate when a specific "event" occurs—for example, when an agent's state deviates too far from its last broadcast value. For a single agent (the "single-loop" case), it is mathematically proven that ETC provides better performance for the same average communication rate.

However, the authors identify a critical gap: does this superiority hold when you move from one agent to a large Multi-Agent System (MAS) trying to reach consensus? In a decentralized network, each agent makes its own triggering decision based on local info. The authors tackle whether the "distributedness" of the system fundamentally changes the performance relationship between ETC and TTC.

Methods

The researchers analyzed a system of $N$ "single-integrator" agents (simple systems where the rate of change is directly controlled) perturbed by Brownian motion (random noise). The goal is consensus: keeping all agents at the same state despite the noise.

Key Approaches:

  • Control Law: They use an "impulsive" control strategy. Whenever a trigger occurs, the system instantaneously resets all agents to a common consensus point.
  • Triggering Rules:
    • TTC: Updates happen at a constant, periodic frequency.
  • ETC: Updates happen via a "level-triggering" rule—if any single agent's local state deviates from its last broadcast by more than a threshold $\Delta$, a network-wide update is triggered.
  • Performance Measure: The long-term average of the quadratic deviation from consensus (essentially, how "spread out" the agents get between updates).
  • Mathematical Analysis: The authors used stochastic calculus and Extreme Value Theory. Specifically, they modeled the inter-event time in ETC as the "first exit time" of $N$ independent Brownian motions.

Datasets

This is a theoretical and computational study, so no external real-world datasets were used. Instead, the authors validated their proofs using extensive Monte Carlo simulations:

  • Scope: Simulations covered networks ranging from $N=2$ to $N=80$ agents.
  • Volume: They performed 10,000 runs to estimate average triggering times and 250,000 runs to estimate the specific performance costs for each network size.
  • Method: The Euler–Maruyama method was used to simulate the stochastic differential equations governing the agents.

Key Insights and Ideas

  • The "First-to-Exit" Problem: In a single-loop system, the trigger depends on one agent. In a decentralized MAS, the entire network triggers if any of the $N$ agents hits the threshold. As $N$ increases, the probability that at least one agent hits the threshold very quickly becomes extremely high.
  • Logarithmic Scaling: The authors found that the average time between events in ETC shrinks as the number of agents increases, scaling at a rate of $1/\ln N$.
  • The Tipping Point ($N_0$): The core "aha moment" is the proof of Theorem 2: there exists a critical number of agents ($N_0$) beyond which TTC provably outperforms ETC. While ETC is better for a small group of robots, it becomes a liability for a massive swarm.
  • Decentralization is the Culprit: The performance loss isn't due to the consensus goal itself, but rather the lack of global information in the triggering decision. Because agents trigger updates based only on their local "unhappiness," the system over-communicates or communicates at suboptimal times compared to a simple periodic clock.

Key Results

  • Theoretical Superiority of TTC: The paper provides a complete proof that for any undirected connected graph, TTC eventually outperforms ETC as $N$ grows.
  • Asymptotic Order: Both TTC and ETC costs scale as $O(1/\ln N)$ for large $N$, but the constant factor for ETC is higher.
  • Simulation Findings: In their specific setup (with a threshold $\Delta=1$), the simulation showed that ETC was superior for small groups, but the "tipping point" occurred between 30 and 55 agents. Beyond this range, the cost ratio of ETC/TTC rose above 1.0, signaling a performance disadvantage for the event-triggered approach.

Limitations and Future Work

  • Idealized Conditions: The study assumes an "ideal" network with zero communication delay and no packet loss. In reality, delays might further penalize ETC or change the $N_0$ threshold.
  • Impulsive Control: The assumption that agents can be reset to consensus instantaneously is a mathematical abstraction. Real actuators have limits.
  • Triggering Rule Variety: The authors only tested "level-triggering." They suggest that different triggering norms (like $p$-norms) or rules that incorporate more neighbor information might recover some of ETC's lost performance.
  • Future Directions: The authors aim to investigate the "root cause" more deeply—specifically, whether providing agents with a small amount of global information could restore ETC's superiority in large networks.

Summary updated

Abstract

Event-triggered control has shown the potential for providing improved control performance at the same average sampling rate when compared to time-triggered control. While this observation motivates numerous event-triggered control schemes, proving it from a theoretical perspective has only been achieved for a limited number of settings. Inspired by existing performance analyses for the single-loop case, we provide a first fundamental performance comparison of time- and event-triggered control in a multi-agent consensus setting. For this purpose, we consider undirected connected network topologies without communication delays, a level-triggering rule for event-triggered control, and the long-term average of the quadratic deviation from consensus as a performance measure. The main finding of our analysis is that time-triggered control provably outperforms event-triggered control beyond a certain number of agents in our particular setting. We thereby provide an illustrative distributed problem setup in which event-triggered control results in a performance disadvantage when compared to time-triggered control in the case of large networks. Moreover, we derive the asymptotic order of the performance measure under both triggering schemes which gives more insights into the cost relationship for large numbers of agents. Thus, by presenting an analysis for a particular setup, this work points out that transferring an event-triggering scheme from the single-loop to the multi-agent setting can lead to a loss of the often presumed superiority of event-triggered control over time-triggered control. In particular, the design of performant decentralized event-triggering schemes can therefore pose additional challenges when compared to the analogue single-loop case.
BibTeX
@inproceedings{meister2024time,
  author = {David Meister and Frank Aurzada and Mikhail Lifshits and Frank Allgöwer},
  title = {{Time- versus event-triggered consensus of a single-integrator multi-agent system}},
  booktitle = {Nonlinear},
  year = {2024},
  doi = {10.1016/j.nahs.2024.101494},
}

Loading related content...

Figure