pith. sign in

arxiv: 2604.06299 · v1 · submitted 2026-04-07 · 📡 eess.SY · cs.SY

An Evolutionary Algorithm for Actuator-Sensor-Communication Co-Design in Distributed Control

Pengyang Wu , Jing Shuang Li This is my paper

Pith reviewed 2026-05-10 19:11 UTC · model grok-4.3

classification 📡 eess.SY cs.SY
keywords evolutionary algorithmdistributed controlactuator selectionsensor selectioncommunication co-designLQR pruningsparse controlswing equation
0
0 comments X

The pith

An evolutionary algorithm prunes a dense LQR controller to jointly minimize control cost and the number of actuators, sensors, and communication links in distributed systems.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops an evolutionary algorithm that begins with a complete LQR controller for a partitioned networked plant and removes unnecessary actuators, sensors, and communication links to lower material costs while preserving low linear-quadratic performance cost. It supplies convergence analysis for the search process and a modification that prevents instability when pruning is applied to unstable plants. Simulations confirm the method works on both stable and unstable examples, with the standout demonstration that a 98-state swing-equation model finishes its full co-design in seconds on ordinary hardware and cuts material use by more than half relative to simple pruning. A reader would care because many real distributed systems, from power grids to vehicle platoons, pay heavily for every extra actuator or link; an automated way to find the smallest set that still delivers good closed-loop behavior could reduce both expense and complexity.

Core claim

The paper claims that an evolutionary algorithm can perform actuator-sensor-communication co-design by iteratively pruning a baseline dense LQR controller, minimizing the sum of the LQ cost and a term counting the retained components. Convergence of the evolutionary process is analyzed, and a safeguard modification is introduced to maintain closed-loop stability when the plant is unstable. The resulting sparse configurations are validated through simulation on multiple plants, including a 98-state swing equation model where the entire procedure runs in seconds on a standard laptop and yields more than 50 percent better material efficiency than naive pruning.

What carries the argument

The evolutionary pruning algorithm that starts from a full LQR solution and selectively removes actuators, sensors, and communication links while tracking the combined LQ-plus-material cost

Load-bearing premise

Selective pruning of the baseline dense LQR controller through the evolutionary process will reliably keep the closed-loop system stable for both stable and unstable plants and will not create hidden performance losses outside the LQ cost metric.

What would settle it

A simulation of an unstable plant where the modified pruning algorithm still produces an unstable closed-loop trajectory, or a pruned controller whose LQ cost is low yet exhibits large tracking errors when tested under disturbances absent from the original cost function.

Figures

Figures reproduced from arXiv: 2604.06299 by Jing Shuang Li, Pengyang Wu.

Figure 1
Figure 1. Figure 1: Results of running Algorithm 1 on three different plants. [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Results of running Algorithm 1 (“EA without repairs”) and [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
read the original abstract

This paper studies the co-design of actuators, sensors, and communication in the distributed setting, where a networked plant is partitioned into subsystems each equipped with a sub-controller interacting with other sub-controllers. The objective is to jointly minimize control cost (measured by LQ cost) and material cost (measured by the number of actuators, sensors, and communication links used). We approach this using an evolutionary algorithm to selectively prune a baseline dense LQR controller. We provide convergence and stability analyses for this algorithm. For unstable plants, controller pruning is more likely to induce instability; we provide an algorithm modification to address this. The proposed methods is validated in simulations. One key result is that co-design of a 98-state swing equation model can be done on a standard laptop in seconds; the co-design outperforms naive controller pruning by over 50%.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes an evolutionary algorithm to co-design actuators, sensors, and communication links for distributed control by selectively pruning a baseline dense LQR controller. The objective is to jointly minimize the LQ control cost and a material cost counting the number of actuators, sensors, and links. Convergence and stability analyses are provided for the algorithm; a modification to the pruning step is introduced for unstable plants to reduce instability risk. The approach is validated in simulations, with a key result being that co-design for a 98-state swing-equation model completes in seconds on a standard laptop and outperforms naive pruning by over 50%.

Significance. If the stability modification and analyses hold, the work provides a practical, scalable heuristic for a combinatorially difficult co-design problem in networked control, with demonstrated applicability to moderately large systems such as power-grid swing models. The reported laptop-scale runtime and empirical outperformance constitute a concrete strength, as does the explicit handling of unstable plants. The inclusion of convergence analysis adds rigor beyond pure heuristics.

major comments (2)
  1. [Abstract and algorithm modification for unstable plants] Abstract and the section describing the algorithm modification for unstable plants: the LQ cost penalizes only nominal state and input deviations under the closed-loop dynamics; the modification is claimed to address instability risk, yet no description is given of an explicit post-pruning check (e.g., eigenvalue verification of the closed-loop matrix) inside the acceptance step. This leaves open the possibility that accepted designs satisfy the reported cost but violate closed-loop stability or robustness margins not captured by the LQ metric.
  2. [Simulation results (98-state example)] Simulation results for the 98-state swing-equation example: the claim of >50% outperformance over naive controller pruning is central to the empirical validation, but the exact definition of the combined cost (weighting between LQ term and material count) and the precise baseline naive method are not stated with sufficient quantitative detail to allow independent reproduction or assessment of whether the gain is robust across random seeds or plant parameters.
minor comments (2)
  1. [Problem formulation] The combined objective function (LQ cost plus material cost) should be written explicitly with its weighting parameter in the problem formulation to clarify how the evolutionary fitness is computed.
  2. [Simulation results] Figure captions for the simulation plots would benefit from inclusion of the specific plant parameters (e.g., swing-equation constants) and the number of independent runs used to generate the reported averages.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments, which help improve the clarity and rigor of our manuscript. We address each major comment below and will incorporate the suggested clarifications in the revised version.

read point-by-point responses

  1. Referee: [Abstract and algorithm modification for unstable plants] Abstract and the section describing the algorithm modification for unstable plants: the LQ cost penalizes only nominal state and input deviations under the closed-loop dynamics; the modification is claimed to address instability risk, yet no description is given of an explicit post-pruning check (e.g., eigenvalue verification of the closed-loop matrix) inside the acceptance step. This leaves open the possibility that accepted designs satisfy the reported cost but violate closed-loop stability or robustness margins not captured by the LQ metric.

    Authors: We appreciate the referee pointing out this potential gap in the description of the acceptance step. Our stability analysis provides theoretical conditions under which pruned controllers remain stable, and the modification for unstable plants adjusts the pruning probability to reduce the likelihood of instability. However, we agree that explicitly incorporating a post-pruning stability verification (such as checking the eigenvalues of the closed-loop matrix A-BK) in the acceptance criterion would provide an additional practical safeguard and address robustness aspects not fully captured by the LQ cost. We will revise the manuscript to include this explicit check in the algorithm description and update the abstract accordingly. revision: yes

  2. Referee: [Simulation results (98-state example)] Simulation results for the 98-state swing-equation example: the claim of >50% outperformance over naive controller pruning is central to the empirical validation, but the exact definition of the combined cost (weighting between LQ term and material count) and the precise baseline naive method are not stated with sufficient quantitative detail to allow independent reproduction or assessment of whether the gain is robust across random seeds or plant parameters.

    Authors: We agree that additional quantitative details are needed for reproducibility. The combined cost is defined as J = J_LQ + λ * C_material, where J_LQ is the infinite-horizon LQR cost, C_material counts the total number of actuators, sensors, and communication links, and λ is a tunable weighting parameter (set to 0.1 in the 98-state example). The naive baseline prunes the dense LQR controller by randomly removing the same total number of components without evolutionary optimization. We will add these exact definitions, the value of λ, and results from multiple random seeds (showing consistent >50% improvement with standard deviation) to the simulation section in the revision. revision: yes

Circularity Check

0 steps flagged

No significant circularity; evolutionary pruning and analyses are independent of fitted inputs

full rationale

The paper applies an evolutionary search to prune a precomputed dense LQR controller for joint actuator-sensor-communication minimization, reports empirical runtime and performance gains on the 98-state example, and supplies separate convergence and stability analyses. No derivation step equates a reported outcome (e.g., the >50% outperformance or laptop-scale runtime) to a quantity defined by the algorithm's own fitted parameters or self-referential normalizations. The LQ cost and material cost are external metrics; the pruning modification for unstable plants is an algorithmic adjustment whose correctness is asserted via analysis rather than by construction. Self-citations, if present, are not load-bearing for the central empirical claim.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Based solely on the abstract, the approach rests on standard LQR optimality and evolutionary search operators; no explicit free parameters, domain axioms, or new invented entities are stated.

pith-pipeline@v0.9.0 · 5441 in / 1179 out tokens · 68528 ms · 2026-05-10T19:11:00.354915+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

10 extracted references · 10 canonical work pages

  1. [1]

    An explicit parametrization of controllers with sparsity constraints,

    E. Jensen and B. Bamieh, “An explicit parametrization of controllers with sparsity constraints,”IEEE Transactions on Automatic Control, vol. 67, no. 8, pp. 3790–3805, 2022

    work page 2022

  2. [2]

    System level synthesis,

    J. Anderson, J. C. Doyle, S. H. Low, and N. Matni, “System level synthesis,”Annual Reviews in Control, vol. 47, pp. 364–393, 2019

    work page 2019

  3. [3]

    Near-optimal distributed linear-quadratic regulator for networked systems,

    S. Shin, Y . Lin, G. Qu, A. Wierman, and M. Anitescu, “Near-optimal distributed linear-quadratic regulator for networked systems,”SIAM Journal on Control and Optimization, vol. 61, no. 3, pp. 1113–1135, 2023

    work page 2023

  4. [4]

    Minimum cost constrained input-output and control configuration co-design problem: A structural systems approach,

    S. Pequito, S. Kar, and G. J. Pappas, “Minimum cost constrained input-output and control configuration co-design problem: A structural systems approach,” in2015 American control conference (ACC). IEEE, 2015, pp. 4099–4105

    work page 2015

  5. [5]

    An iterative approach to the optimal co-design of linear control systems,

    Y . Jiang, Y . Wang, S. A. Bortoff, and Z.-P. Jiang, “An iterative approach to the optimal co-design of linear control systems,”Interna- tional Journal of Control, vol. 89, no. 4, pp. 680–690, 2016

    work page 2016

  6. [6]

    Approximating constrained minimum cost input–output selection for generic arbitrary pole placement in structured systems,

    S. Moothedath, P. Chaporkar, and M. N. Belur, “Approximating constrained minimum cost input–output selection for generic arbitrary pole placement in structured systems,”Automatica, vol. 107, pp. 200– 210, 2019

    work page 2019

  7. [7]

    Feedback control of linear systems with optimal sensor and actuator selection,

    T. Singh, M. De Mauri, W. Decr ´e, J. Swevers, and G. Pipeleers, “Feedback control of linear systems with optimal sensor and actuator selection,”Journal of Vibration and Control, vol. 27, no. 11-12, pp. 1250–1264, 2021

    work page 2021

  8. [8]

    An overview of evolutionary algorithms: practical issues and common pitfalls,

    D. Whitley, “An overview of evolutionary algorithms: practical issues and common pitfalls,”Information and software technology, vol. 43, no. 14, pp. 817–831, 2001

    work page 2001

  9. [9]

    Genetic algorithms: A 10 year perspective,

    K. De Jong, “Genetic algorithms: A 10 year perspective,” inProceed- ings of the first International Conference on Genetic Algorithms and their Applications. Psychology Press, 2014, pp. 169–177

    work page 2014

  10. [10]

    Global convergence of policy gradient methods for the linear quadratic regulator,

    M. Fazel, R. Ge, S. Kakade, and M. Mesbahi, “Global convergence of policy gradient methods for the linear quadratic regulator,” in International conference on machine learning. PMLR, 2018, pp. 1467–1476

    work page 2018