Pith. sign in

REVIEW 5 major objections 6 minor 55 references

Reinforcement learning of a three-bead biflagellate model shows that speed-optimal strokes are synchronized, breaststroke-like beats that outpace circular flagella orbits and produce a weakly extensile (pusher) time-averaged flow field.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Reinforcement learning on a three-bead biflagellate model yields quasi-synchronized, symmetric beating strokes with a pusher-type averaged flow field, outperforming predefined circular flagellar motions.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection A clean RL study of a model biflagellate that finds symmetric beating and a weak pusher far-field; the best-of-20 selection needs more scrutiny, but the core result holds up. the 5 major comments →

arxiv 2508.15561 v1 pith:YXGGNS3E submitted 2025-08-21 cond-mat.soft physics.bio-phphysics.comp-ph

Reinforcement learning of a biflagellate model microswimmer

Sridhar Bulusu , Andreas Z\"ottl This is my paper
classification cond-mat.soft physics.bio-phphysics.comp-ph
keywords microswimmersreinforcement learninglow-Reynolds-number locomotionbiflagellate algaethree-bead swimmerforce dipolepusher-pullerNEAT
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

Using a three-bead hydrodynamic model (one spherical body, two flagella beads) at zero Reynolds number, the paper asks what force pattern between beads maximizes swimming speed. The answer, found by evolving small neural networks rather than prescribing beat shapes, is a synchronized, quasi-symmetric breaststroke: both flagella beads beat in phase, and the gait is faster than the circular flagella orbits used in earlier three-bead studies. The study's sharpest claim is that the stroke-averaged far-field flows of all optimized swimmers are weakly extensile (pusher, p̄ > 0), even though the instantaneous flow oscillates between puller-like power strokes and pusher-like recovery strokes as in Chlamydomonas. This matters because it shows that breaststroke-like shape changes by themselves do not determine the averaged far-field type; stroke details do. The optimized policies are simple enough to be written as two sign functions of the two arm lengths.

Core claim

The paper claims that, for a model biflagellate swimmer in which two small flagella beads and a larger body bead interact through Rotne-Prager hydrodynamic mobilities and pairwise arm forces, the fastest learned strokes are synchronized beats: the two flagella beads move quasi-symmetrically, with essentially equal body-flagella distances at all times. This holds even when the learning is free to break symmetry (non-symmetric runs), and the resulting speeds almost match the symmetric-constrained optimum. The learned gaits are not circular: they keep the flagella beads near the body longer during recovery and then sweep them outward during the power stroke, giving a larger net forward displace

What carries the argument

Central object: a three-bead force-based microswimmer with pairwise active arm forces f_ij(t) generated by an artificial neural network; total forces are force- and torque-free by construction. Hydrodynamic coupling between beads enters through Rotne–Prager mobilities. The learning machinery is NEAT (neuroevolution of augmenting topologies), which evolves both weights and network topology of a population of 100 ANNs over 2000 generations, with fitness equal to the mean x-displacement over a finite run; the best of 20 runs is taken. The key output is a low-dimensional policy mapping the three instantaneous arm lengths to three arm forces, which the paper collapses to approximate sign-function

Load-bearing premise

The whole result rests on assuming that the evolved network policies are close to the true speed-optimal strokes; the paper explicitly says no strict optimum can be guaranteed, and several independent non-symmetric training runs never converge.

What would settle it

Do an exhaustive or analytically optimal search over the same three-arm force space for the same body sizes; a non-synchronized gait with a higher stroke-averaged speed, or a negative time-averaged force dipole, would overturn the paper's claims.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If the optimal speed stroke is indeed synchronized, then restricting attention to symmetric beats sacrifices little speed, and symmetric training is a cheaper, more reliable route for optimizing other quantities.
  • The stroke-averaged pusher result implies that breaststroke morphology alone is not enough to classify a microswimmer; laboratory classifications of puller/pusher should be tied to measured stroke specifics.
  • The simple two-rule policy can be ported to other bead-spring swimmers as a closed-form controller, avoiding the need to run neuroevolution for every parameter set.
  • The observed one-step-back-two-step-forward kinematics with larger backward recovery motion but even larger forward power stroke explains why the optimized gait beats circular orbits: it trades stronger recovery for an even stronger power stroke.
  • Physical-unit estimates place the model's swimming speed roughly an order of magnitude below real Chlamydomonas, suggesting that slender, anisotropically dragged flagella and near-field effects contribute significantly to biological speed.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper: I would read the emergence of synchronized beating as indirect evidence that, for speed maximization in this bead geometry, synchrony is itself hydrodynamically optimal; a natural extension is to test whether adding thermal noise destabilizes this synchrony, which the paper lists as future work.
  • Beyond the paper: Because p̄ is only slightly positive, the pusher/puller classification sits close to zero; changing the mobility approximation (e.g., including near-field lubrication or flagella anisotropy) might flip the sign, so a robustness check with higher-order multipoles is a concrete next test.
  • Beyond the paper: Rewarding efficiency instead of speed would be a discriminating test: with power close to fmax for all gaits, an efficiency-optimal stroke may favor asymmetric or different-timed beating, potentially reversing the averaged dipole.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

5 major / 6 minor

Summary. The paper applies neuroevolution (NEAT) to optimize the swimming speed of a three-bead biflagellate microswimmer model at low Reynolds number. The swimmer consists of a large body bead and two smaller flagella beads; the policy maps instantaneous arm lengths to active arm forces, subject to maximum-force and passive-spring constraints. Two learning setups are considered: symmetric (enforcing l12=l13 and f12=f13) and non-symmetric. For several body-to-flagella radius ratios, the best of 20 NEAT runs is selected. The authors report that the non-symmetric policies converge to nearly symmetric, quasi-synchronized flagella beating; stroke-averaged speed, power, and efficiency are analyzed; the instantaneous flow fields alternate between puller-type and pusher-type, while the stroke-averaged force-dipole strength is slightly positive (pusher); and the learned strokes outperform prescribed circular flagella-bead trajectories.

Significance. If robust, the work is a valuable contribution: it shows that a generic RL method can discover simple, interpretable near-optimal strokes for a model biflagellate, and it provides a concrete counterexample to the expectation that breast-stroke-like motion necessarily produces a time-averaged puller far field. The extracted analytic policy, Eqs. (10)-(11), is a useful falsifiable prediction for bead-spring swimmers. Strengths include transparent Rotne-Prager hydrodynamics, multiple independent NEAT runs, and a direct comparison with circular flagella-bead motion. The main caveat is that the central claims currently rest on a single selected run per morphology without uncertainty quantification, and one benchmark equation contains a dimensional inconsistency.

major comments (5)
  1. [Sec. 3, Fig. 2; Sec. 4.1, 4.3] All reported results are for the single best of 20 NEAT runs per morphology. Fig. 2 shows that non-symmetric runs frequently do not converge and that even converged runs have a fitness spread. Since the central claims are that the quasi-optimum is generically symmetric/quasi-synchronized and that p-bar>0 is 'slightly positive', please report distributions or error bars over converged runs for f, v, P, efficiency, p-bar, and a quantitative symmetry defect (e.g. max |l12-l13|/mean l12). Without this, the conclusions are not separated from the properties of one selected ANN. A convergence criterion and selection rule for 'converged' runs should also be stated.
  2. [Sec. 4.3, Fig. 6] The claim 'for all our considered microswimmers slightly positive, p-bar>0' is not supported numerically in the paper: Fig. 6 shows p(t) only for Rb=6Rf, and no p-bar values or error bars are given for other Rb or for the S/N cases. Please provide a table of p-bar with uncertainties, and state the sign convention and the approximation p=F1(t).d(t). This is important because the sign is the entire basis of the pusher classification.
  3. [Sec. 4.5, Eq. (12)] Equation (12) is dimensionally inconsistent for a prescribed circular motion: the x-components v2x and v3x contain Rc cos(omega t), which has units of length, while v1x and Rc omega sin(omega t) have units of velocity. The intended circular trajectory presumably uses Rc omega cos(omega t). Please correct the equation and confirm that the simulations used the dimensionally correct form; otherwise the comparison with circular flagella-bead strokes is not reproducible.
  4. [Sec. 4.5, Abstract] The comparison to 'previously used biflagellate microswimmer models relying on predefined circular flagella bead motion' is made against one specific non-optimized circular benchmark whose radius, center, and phase are derived from the learned trajectory. This does not establish superiority over the full family of circular strokes or over the specific models of Refs. [22-25]. Please either optimize the circular-stroke parameters (or scan a representative range) and/or soften the claim to 'outperforms the circular benchmark considered here'.
  5. [Sec. 4.1] The training reward is x1(TT)/TT with TT=260 tau, not the stroke-averaged velocity v reported in Fig. 3(b). As the paper notes, TT is not an integer multiple of the stroke period and is only about 11.8T for the longest period. The observed inversion for Rb=4,5 (N has lower fitness but higher v than S) shows that the finite horizon matters. Please quantify the residual finite-time effect, e.g. by re-evaluating top policies over multiple periods or by reporting how v varies with phase/TT, so that 'quasi-optimized' is referenced to the correct objective.
minor comments (6)
  1. [Throughout] Typographical errors: 'minmum', 'chracterized', 'Data A vailability'; 'henceforth' in Sec. 4.1 should be 'furthermore' or similar.
  2. [Fig. 3] The black dots in panel (b) are not identified in the caption; please add a legend entry for the circular-stroke velocities.
  3. [Sec. 4.4, Fig. 8] The sentence 'while in the respective red regions they are contracted' should presumably read 'blue regions'; the color description is confusing.
  4. [Sec. 2] The notation fij is used both as a scalar force strength and as a vector fij rhat_ij; please distinguish clearly (e.g. f_ij and bold f_ij).
  5. [Sec. 4.4] The constants la, lb, ca in Eqs. (10)-(11) are said to depend on Rb but are not tabulated; reporting them would improve reproducibility.
  6. [Data Availability] The results and data are available only on reasonable request. Depositing the trained policies and analysis scripts would strengthen the reproducibility of the central claims.

Circularity Check

0 steps flagged

No significant circularity: the RL objective is swimming speed only, and the symmetry/pusher results are post-hoc observables, not fitted targets.

full rationale

Training maximizes the finite-time mean body displacement (x1(TT)-x1(0))/TT (Sec. 3), while the reported stroke-averaged velocity v, power P, efficiency epsilon, and force-dipole strength p-bar are computed afterward from the selected policies and do not enter the reward (Secs. 4.1-4.3). In particular, the pusher result p-bar>0 is not a fitted target: p=F1*d is evaluated on the learned strokes and is not constrained by the speed objective. The symmetric (S) runs do impose f12=f13 by construction, but the paper labels them as constrained and the novel synchronization claim rests on the unconstrained non-symmetric (N) runs, which also converge to quasi-symmetric strokes (Secs. 3, 4.2, 5). The circular-bead comparison uses matched area, inter-bead distances, and stroke frequency as a controlled baseline, so the RL advantage is not an artifact of the metric (Sec. 4.5). Self-citations [33,43,44] are methodological (NEAT population size, restoring-force forms, prior RL swimmer studies) and are not load-bearing evidence for the central claims; there is no imported uniqueness theorem. The paper's own caveat that 2000-generation NEAT runs give only quasi-optima and that some N runs fail to converge (Sec. 3, Fig. 2) is a robustness/correctness limitation, not a circularity.

Axiom & Free-Parameter Ledger

3 free parameters · 6 axioms · 0 invented entities

The model does not introduce new physical entities; it uses standard low-Reynolds-number hydrodynamics with hand-chosen constraint parameters. The central claims rest mainly on the modeling assumptions listed above, particularly the arm-length constraints and the finite-horizon RL convergence.

free parameters (3)
  • Maximum arm force fmax = 10
    Sets force scale; paper states it does not influence optimum stroke kinematics but only scaling.
  • Spring constant k = 10
    Chosen sufficiently stiff; affects effective arm length constraints by allowing overshoot of about 1 length unit.
  • Arm length constraints (lmin_bf, lmin_ff, lmax_bf, lmax_ff) = Based on Rb, Rf, Lf=3Rb
    Define the allowed shape space; the optimized gaits depend on these boundaries.
axioms (6)
  • domain assumption Stokes flow linearity and force/torque-free conditions
    Sec. 2: bead velocities are linear in forces via mobility tensor; forces from arms ensure zero net force and torque.
  • domain assumption Rotne-Prager approximation for hydrodynamic interactions
    Sec. 2, Eq. (4): used for cross-mobilities; a standard far-field approximation that neglects near-field lubrication.
  • domain assumption Neglect of thermal fluctuations, bead rotation, and 3D flagellum shape
    Sec. 2: bead rotation and noise are ignored; flagella are modeled as small spheres, losing anisotropic drag of slender flagella.
  • domain assumption Motion restricted to x-y plane
    Sec. 2: initial positions and forces are in plane, yielding planar motion.
  • ad hoc to paper Arm forces limited to [-fmax, fmax] with passive harmonic restoring forces beyond allowed lengths
    Sec. 2, Eqs. (1-2): introduces a specific constraint mechanism that shapes the set of possible strokes.
  • ad hoc to paper NEAT algorithm with population 100 and 2000 generations converges to quasi-optimum
    Sec. 3: training converges in selected runs; the paper acknowledges non-convergence in some non-symmetric runs.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Reinforcement learning of a biflagellate model microswimmer." pith.science (2026). https://pith.science/paper/YXGGNS3E

@misc{pith2026250815561,
  author       = {Pith},
  title        = {Pith review of: Reinforcement learning of a biflagellate model microswimmer},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YXGGNS3E}},
  note         = {Machine review of arXiv:2508.15561}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Many microswimmers are able to swim through viscous fluids by employing periodic non-reciprocal deformations of their appendages. Here we use a simple microswimmer model inspired by swimming biflagellates which consists of a spherical cell body and two small spherical beads representing the motion of the two flagella. Using reinforcement learning we identify for different microswimmer morphologies quasi-optimized swimming strokes. For all studied cases the identified strokes result in symmetric and quasi-synchronized beating of the two flagella beads. Interestingly, the stroke-averaged flow fields are of pusher type, and the observed swimming gaits outperform previously used biflagellate microswimmer models relying on predefined circular flagella bead motion.

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

55 extracted references · 46 canonical work pages · 1 internal anchor

  1. [1]

    Lauga, E., Powers, T.R.: The hydrodynamics of swimming microorganisms. Rep. Prog. Phys. 72, 096601 (2009) https://doi.org/10.1088/0034-4885/72/9/096601

    doi:10.1088/0034-4885/72/9/096601 2009

  2. [2]

    Elgeti, J., Winkler, R.G., Gompper, G.: Physics of microswimmers—single par- ticle motion and collective behavior: a review. Rep. Prog. Phys. 78(5), 56601 (2015) https://doi.org/10.1088/0034-4885/78/5/056601

    doi:10.1088/0034-4885/78/5/056601 2015

  3. [3]

    Z¨ ottl, A., Stark, H.: Emergent behavior in active colloids. J. Phys. Condens. Matter 28(25), 253001 (2016) https://doi.org/10.1088/0953-8984/28/25/253001

    doi:10.1088/0953-8984/28/25/253001 2016

  4. [4]

    Purcell, E.M.: Life at low Reynolds Number. Am. J. Phys. 45, 3–11 (1977) https: //doi.org/10.1119/1.10903

    doi:10.1119/1.10903 1977

  5. [5]

    Brennen, C., Winet, H.: Fluid Mechanics of Propulsion by Cilia and Flagella. Ann. Rev. Fluid Mech. 9(Lehninger 1971), 339–398 (1977) https://doi.org/10. 1146/annurev.fl.09.010177.002011

    work page arXiv 1971

  6. [6]

    Lighthill, J.: Flagellar hydrodynamics* (2) (1976) 19

    work page 1976

  7. [7]

    Annual Review of Fluid Mechanics 47, 343–375 (2015) https://doi.org/10.1146/ annurev-fluid-010313-141426

    Goldstein, R.E.: Green algae as model organisms for biological fluid dynamics. Annual Review of Fluid Mechanics 47, 343–375 (2015) https://doi.org/10.1146/ annurev-fluid-010313-141426

    work page 2015

  8. [8]

    Nature Reviews Physics 2(2), 74–88 (2020) https://doi.org/10.1038/ s42254-019-0129-0

    Gilpin, W., Bull, M.S., Prakash, M.: The multiscale physics of cilia and flagella. Nature Reviews Physics 2(2), 74–88 (2020) https://doi.org/10.1038/ s42254-019-0129-0

    work page 2020

  9. [9]

    Z¨ ottl, A., Stark, H.: Modeling Active Colloids: From Active Brownian Particles to Hydrodynamic and Chemical Fields. Annu. Rev. Condens. Matter Phys. 14, 109–127 (2023)

    work page 2023

  10. [10]

    Guasto, J.S., Johnson, K.a., Gollub, J.P.: Oscillatory flows induced by microor- ganisms swimming in two dimensions. Phys. Rev. Lett. 105(16), 18–21 (2010) https://doi.org/10.1103/PhysRevLett.105.168102

    work page doi:10.1103/physrevlett.105.168102 2010

  11. [11]

    Drescher, K., Goldstein, R.E., Michel, N., Polin, M., Tuval, I.: Direct mea- surement of the flow field around swimming microorganisms. Phys. Rev. Lett. 105(October), 1–4 (2010) https://doi.org/10.1103/PhysRevLett.105.168101

    doi:10.1103/physrevlett.105.168101 2010

  12. [12]

    Physical Review E - Statistical, Nonlinear, and Soft Matter Physics 92(6), 1–6 (2015) https://doi.org/10.1103/PhysRevE.92.063019

    Klindt, G.S., Friedrich, B.M.: Flagellar swimmers oscillate between pusher- and puller-type swimming. Physical Review E - Statistical, Nonlinear, and Soft Matter Physics 92(6), 1–6 (2015) https://doi.org/10.1103/PhysRevE.92.063019

    work page doi:10.1103/physreve.92.063019 2015

  13. [13]

    Spagnolie, S.E., Lauga, E.: Hydrodynamics of self-propulsion near a boundary: predictions and accuracy of far-field approximations. J. Fluid Mech. 700, 105 (2012) https://doi.org/10.1017/jfm.2012.101

    work page doi:10.1017/jfm.2012.101 2012

  14. [14]

    Schaar, K., Z¨ ottl, A., Stark, H.: Detention Times of Microswimmers Close to Surfaces: Influence of Hydrodynamic Interactions and Noise. Phys. Rev. Lett. 115(3), 038101 (2015) https://doi.org/10.1103/PhysRevLett.115.038101

    work page doi:10.1103/physrevlett.115.038101 2015

  15. [15]

    Hernandez-Ortiz, J.P., Stoltz, C.G., Graham, M.D.: Transport and collective dynamics in suspensions of confined swimming particles. Phys. Rev. Lett. 95(November), 204501 (2005) https://doi.org/10.1103/PhysRevLett.95.204501

    work page doi:10.1103/physrevlett.95.204501 2005

  16. [16]

    Saintillan, D., Shelley, M.J.: Orientational order and instabilities in suspensions of self-locomoting rods. Phys. Rev. Lett. 99(August), 058102 (2007) https://doi. org/10.1103/PhysRevLett.99.058102

    doi:10.1103/physrevlett.99.058102 2007

  17. [17]

    Z¨ ottl, A., Stark, H.: Hydrodynamics determines collective motion and phase behavior of active colloids in quasi-two-dimensional confinement. Phys. Rev. Lett. 112(11), 118101 (2014) https://doi.org/10.1103/PhysRevLett.112.118101

    doi:10.1103/physrevlett.112.118101 2014

  18. [18]

    Cates, M.E., Tailleur, J.: Motility-Induced Phase Separation. Annu. Rev. Condens. Matter Phys. 6, 219–244 (2015) https://doi.org/10.1146/ annurev-conmatphys-031214-014710 20

    work page 2015

  19. [19]

    Aranson, I.S.: Bacterial active matter. Rep. Prog. Phys. 85(7), 076601 (2022) https://doi.org/10.1088/1361-6633/ac723d

    doi:10.1088/1361-6633/ac723d 2022

  20. [20]

    Proceedings of the National Academy of Sciences of the United States of America 108(3), 1001–1006 (2011) https://doi.org/10.1073/pnas.1011185108

    Tam, D., Hosoi, A.E.: Optimal feeding and swimming gaits of biflagellated organ- isms. Proceedings of the National Academy of Sciences of the United States of America 108(3), 1001–1006 (2011) https://doi.org/10.1073/pnas.1011185108

    work page doi:10.1073/pnas.1011185108 2011

  21. [21]

    Biophysical Journal 100(11), 2716–2725 (2011) https://doi.org/10.1016/ j.bpj.2011.05.001

    Bayly, P.V., Lewis, B.L., Ranz, E.C., Okamoto, R.J., Pless, R.B., Dutcher, S.K.: Propulsive forces on the flagellum during locomotion of chlamydomonas rein- hardtii. Biophysical Journal 100(11), 2716–2725 (2011) https://doi.org/10.1016/ j.bpj.2011.05.001

    work page 2011

  22. [22]

    Friedrich, B.M., J¨ ulicher, F.: Flagellar synchronization independent of hydrody- namic interactions. Phys. Rev. Lett. 109(September), 1–5 (2012) https://doi. org/10.1103/PhysRevLett.109.138102

    work page doi:10.1103/physrevlett.109.138102 2012

  23. [23]

    New Journal of Physics 15 (2013) https://doi.org/10.1088/1367-2630/15/ 4/045005

    Polotzek, K., Friedrich, B.M.: A three-sphere swimmer for flagellar synchroniza- tion. New Journal of Physics 15 (2013) https://doi.org/10.1088/1367-2630/15/ 4/045005

    doi:10.1088/1367-2630/15/ 2013

  24. [24]

    Bennett, R.R., Golestanian, R.: Emergent run-and-tumble behavior in a simple model of Chlamydomonas with intrinsic noise. Phys. Rev. Lett. 110(14), 1–5 (2013) https://doi.org/10.1103/PhysRevLett.110.148102

    work page doi:10.1103/physrevlett.110.148102 2013

  25. [25]

    New Journal of Physics 15(7) (2013) https://doi.org/10.1088/1367-2630/15/7/075028

    Bennett, R.R., Golestanian, R.: Phase-dependent forcing and synchronization in the three-sphere model of Chlamydomonas. New Journal of Physics 15(7) (2013) https://doi.org/10.1088/1367-2630/15/7/075028

    work page doi:10.1088/1367-2630/15/7/075028 2013

  26. [26]

    Physical Review E 96(5), 1–8 (2017) https://doi.org/10.1103/PhysRevE.96.052610

    Jibuti, L., Zimmermann, W., Rafa¨ ı, S., Peyla, P.: Effective viscosity of a suspen- sion of flagellar-beating microswimmers: Three-dimensional modeling. Physical Review E 96(5), 1–8 (2017) https://doi.org/10.1103/PhysRevE.96.052610

    work page doi:10.1103/physreve.96.052610 2017

  27. [27]

    Proceedings of the National Academy of Sciences of the United States of America 107(17), 7669–7673 (2010) https: //doi.org/10.1073/pnas.0912455107

    Kotar, J., Leoni, M., Bassetti, B., Lagomarsino, M.C., Cicuta, P.: Hydrodynamic synchronization of colloidal oscillators. Proceedings of the National Academy of Sciences of the United States of America 107(17), 7669–7673 (2010) https: //doi.org/10.1073/pnas.0912455107

    work page doi:10.1073/pnas.0912455107 2010

  28. [28]

    Proceedings of the National Academy of Sciences of the United States of America 108(38), 15727–15732 (2011) https://doi.org/10.1073/pnas.1107889108

    Osterman, N., Vilfan, A.: Finding the ciliary beating pattern with optimal effi- ciency. Proceedings of the National Academy of Sciences of the United States of America 108(38), 15727–15732 (2011) https://doi.org/10.1073/pnas.1107889108

    work page doi:10.1073/pnas.1107889108 2011

  29. [29]

    European Physical Journal E 34(4) (2011) https://doi.org/10.1140/ epje/i2011-11042-7

    Wollin, C., Stark, H.: Metachronal waves in a chain of rowers with hydrodynamic interactions. European Physical Journal E 34(4) (2011) https://doi.org/10.1140/ epje/i2011-11042-7

    work page 2011

  30. [30]

    Phys- ical Review E 97(2), 1–9 (2018) https://doi.org/10.1103/PhysRevE.97.023102 21

    Rizvi, M.S., Farutin, A., Misbah, C.: Three-bead steering microswimmers. Phys- ical Review E 97(2), 1–9 (2018) https://doi.org/10.1103/PhysRevE.97.023102 21

    work page doi:10.1103/physreve.97.023102 2018

  31. [31]

    Physical Review E 98(4), 1–9 (2018) https://doi.org/10.1103/PhysRevE.98.043104

    Rizvi, M.S., Farutin, A., Misbah, C.: Size and shape affect swimming of a triangular bead-spring microswimmer. Physical Review E 98(4), 1–9 (2018) https://doi.org/10.1103/PhysRevE.98.043104

    work page doi:10.1103/physreve.98.043104 2018

  32. [32]

    New Journal of Physics 21(11) (2019) https://doi.org/10.1088/ 1367-2630/ab4fc2

    Ziegler, S., Hubert, M., Vandewalle, N., Harting, J., Smith, A.S.: A general per- turbative approach for bead-based microswimmers reveals rich self-propulsion phenomena. New Journal of Physics 21(11) (2019) https://doi.org/10.1088/ 1367-2630/ab4fc2

    work page 2019

  33. [33]

    H¨ ubl, M.: A Neural Network Approach to Microswimmer Locomotion, Master Thesis, TU Wien (2021)

    work page 2021

  34. [34]

    Najafi, A., Golestanian, R.: Propulsion at low Reynolds number. J. Phys. Con- dens. Matter 17, 1203–1208 (2005) https://doi.org/10.1088/0953-8984/17/14/ 009

    work page doi:10.1088/0953-8984/17/14/ 2005

  35. [35]

    SIAM Review 18(2), 161–230 (1976) https://doi.org/10.1137/1018040

    Lighthill, J.: Flagellar Hydrodynamics. SIAM Review 18(2), 161–230 (1976) https://doi.org/10.1137/1018040

    work page doi:10.1137/1018040 1976

  36. [36]

    Journal of Fluid Mechanics 66(2), 391–415 (1974) https://doi.org/10.1017/ S0022112074000279

    Pironneau, O., Katz, D.F.: Optimal swimming of flagellated micro-organisms. Journal of Fluid Mechanics 66(2), 391–415 (1974) https://doi.org/10.1017/ S0022112074000279

    work page 1974

  37. [37]

    Journal of Nonlinear Science18(3), 277–302 (2008) https: //doi.org/10.1007/s00332-007-9013-7

    Alouges, F., Desimone, A., Lefebvre, A.: Optimal strokes for low reynolds number swimmers: An example. Journal of Nonlinear Science18(3), 277–302 (2008) https: //doi.org/10.1007/s00332-007-9013-7

    work page doi:10.1007/s00332-007-9013-7 2008

  38. [38]

    IEEE Transactions on Robotics35(4), 1014–1033 (2019)

    Ramasamy, S., Hatton, R.L.: The Geometry of Optimal Gaits for Drag- Dominated Kinematic Systems. IEEE Transactions on Robotics35(4), 1014–1033 (2019)

    work page 2019

  39. [39]

    Journal of the Physical Society of Japan 92(12), 1–13 (2023) https: //doi.org/10.7566/JPSJ.92.121005

    Moreau, C.: Controllability and Optimal Control of Microswimmers: Theory and Applications. Journal of the Physical Society of Japan 92(12), 1–13 (2023) https: //doi.org/10.7566/JPSJ.92.121005

    work page doi:10.7566/jpsj.92.121005 2023

  40. [40]

    European Physical Journal E 28(3), 279–284 (2009) https: //doi.org/10.1140/epje/i2008-10406-4

    Alouges, F., DeSimone, A., Lefebvre, A.: Optimal strokes for axisymmetric microswimmers. European Physical Journal E 28(3), 279–284 (2009) https: //doi.org/10.1140/epje/i2008-10406-4

    work page doi:10.1140/epje/i2008-10406-4 2009

  41. [41]

    Applied Sciences (Switzerland) 9(19) (2019) https://doi.org/10.3390/ app9194023

    Wang, Q.: Optimal strokes of low reynolds number linked-sphere swim- mers. Applied Sciences (Switzerland) 9(19) (2019) https://doi.org/10.3390/ app9194023

    work page 2019

  42. [42]

    Tsang, A.C.H., Tong, P.W., Nallan, S., Pak, O.S.: Self-learning how to swim at low Reynolds number. Phys. Rev. Fluids 5, 074101 (2020) https://doi.org/10. 1103/PhysRevFluids.5.074101 22

    work page 2020

  43. [43]

    Hartl, B., H¨ ubl, M., Kahl, G., Z¨ ottl, A.: Microswimmers learning chemotaxis with genetic algorithms. Proc. Natl. Acad. Sci. U. S. A. 118, 2019683118 (2021) https://doi.org/10.1073/pnas.2019683118

    work page doi:10.1073/pnas.2019683118 2021

  44. [44]

    Hartl, B., Levin, M., Z¨ ottl, A.: Neuroevolution of Decentralized Decision-Making in N-Bead Swimmers Leads to Scalable and Robust Collective Locomotion. Commun. Phys. 8, 194 (2025) https://doi.org/10.1038/s42005-025-02101-5

    work page doi:10.1038/s42005-025-02101-5 2025

  45. [45]

    Robotics and Autonomous Systems 175(February), 104659 (2024) https://doi

    Jebellat, I., Jebellat, E., Amiri-Margavi, A., Vahidi-Moghaddam, A., Nejat Pishkenari, H.: A Reinforcement Learning Approach to Find Optimal Propulsion Strategy for Microrobots Swimming at Low Reynolds Number. Robotics and Autonomous Systems 175(February), 104659 (2024) https://doi. org/10.1016/j.robot.2024.104659

    work page arXiv 2024

  46. [46]

    Physics of Fluids 33(6) (2021) https://doi.org/10.1063/5.0053563

    Liu, Y., Zou, Z., Tsang, A.C.H., Pak, O.S., Young, Y.N.: Mechanical rotation at low Reynolds number via reinforcement learning. Physics of Fluids 33(6) (2021) https://doi.org/10.1063/5.0053563

    work page doi:10.1063/5.0053563 2021

  47. [47]

    Commu- nications Physics 5(1), 1–9 (2022) https://doi.org/10.1038/s42005-022-00935-x

    Zou, Z., Liu, Y., Young, Y.N., Pak, O.S., Tsang, A.C.H.: Gait switching and targeted navigation of microswimmers via deep reinforcement learning. Commu- nications Physics 5(1), 1–9 (2022) https://doi.org/10.1038/s42005-022-00935-x

    work page doi:10.1038/s42005-022-00935-x 2022

  48. [48]

    Physics of Fluids 35(3) (2023) https://doi.org/ 10.1063/5.0140662

    Qin, K., Zou, Z., Zhu, L., Pak, O.S.: Reinforcement learning of a multi-link swim- mer at low Reynolds numbers. Physics of Fluids 35(3) (2023) https://doi.org/ 10.1063/5.0140662

    work page doi:10.1063/5.0140662 2023

  49. [49]

    Enabling microrobotic chemotaxis via reset-free hierarchical reinforcement learning

    Xiong, T., Liu, Z., Ong, C.J., Zhu, L.: Enabling microrobotic chemotaxis via reset-free hierarchical reinforcement learning. arXiv preprint arXiv:2408.07346, 2408–07346 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  50. [50]

    Com- munications in Nonlinear Science and Numerical Simulation 128(October 2023), 107666 (2024) https://doi.org/10.1016/j.cnsns.2023.107666

    Zou, Z., Liu, Y., Tsang, A.C.H., Young, Y.N., Pak, O.S.: Adaptive micro- locomotion in a dynamically changing environment via context detection. Com- munications in Nonlinear Science and Numerical Simulation 128(October 2023), 107666 (2024) https://doi.org/10.1016/j.cnsns.2023.107666

    work page arXiv 2023

  51. [51]

    Scientific Reports 13(1), 1–10 (2023) https://doi.org/10.1038/s41598-023-36305-y

    Liu, Y., Zou, Z., Pak, O.S., Tsang, A.C.H.: Learning to cooperate for low- Reynolds-number swimming: a model problem for gait coordination. Scientific Reports 13(1), 1–10 (2023) https://doi.org/10.1038/s41598-023-36305-y

    work page doi:10.1038/s41598-023-36305-y 2023

  52. [52]

    Dover Publications Inc., Mineaola, New York (2005)

    Kim, S., Karila, S.J.: Microhydrodynamics: Principles and Selected Applications. Dover Publications Inc., Mineaola, New York (2005)

    work page 2005

  53. [53]

    Expert Systems with Applications 231(April), 120495 (2023) https://doi.org/10.1016/j.eswa.2023.120495

    Shakya, A.K., Pillai, G., Chakrabarty, S.: Reinforcement learning algorithms: A brief survey. Expert Systems with Applications 231(April), 120495 (2023) https://doi.org/10.1016/j.eswa.2023.120495

    work page arXiv 2023

  54. [54]

    Evolutionary Computation 10(2), 99–127 (2002) https://doi.org/10

    Stanley, K.O., Miikkulainen, R.: Evolving neural networks through augmenting 23 topologies. Evolutionary Computation 10(2), 99–127 (2002) https://doi.org/10. 1162/106365602320169811

    work page 2002

  55. [55]

    379–425 (2022) 24

    Montesinos L´ opez, O.A., Montesinos L´ opez, A., Crossa, J.: Fundamentals of Artificial Neural Networks and Deep Learning, pp. 379–425 (2022) 24

    work page 2022

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.