pith. sign in

arxiv: 2606.05984 · v1 · pith:HWMRIBXJnew · submitted 2026-06-04 · ⚛️ physics.bio-ph

Emergent swimming strategies of a smart three-bead swimmer

Ruma Maity , Maximilian Huebl , Julian Lemmel , Benedikt Hartl , Gerhard Kahl This is my paper

Pith reviewed 2026-06-27 22:39 UTC · model grok-4.3

classification ⚛️ physics.bio-ph
keywords microswimmersreinforcement learninglow Reynolds numberswimming gaitsneural networksautonomous locomotionhydrodynamic modelneuroevolution
0
0 comments X

The pith

A three-bead microswimmer learns five swimming gaits with neural networks of under ten nodes and weights.

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

The paper trains a two-dimensional three-bead hydrodynamic model to find autonomous swimming strategies via reinforcement learning and neuroevolution techniques that favor minimal controller complexity. Five distinct gaits emerge from the process: three produce directed locomotion at different efficiencies and two produce rotation. Every gait is realized by a neural network with fewer than ten nodes and weights, demonstrating that low-Reynolds-number swimming can be both efficient and robust under tight computational limits. This result matters for artificial microswimmers that must operate without external commands in variable environments. It also supplies a minimal model that may help interpret locomotion in simple biological organisms.

Core claim

In a two-dimensional hydrodynamic model of a three-bead swimmer, reinforcement learning via neuroevolution produces five characteristic gaits. Three gaits yield directed locomotion of varying efficiency and two gaits produce rotational, inefficient motion. All five gaits are achieved by very simple neural networks containing fewer than ten nodes and weights, indicating that low-Reynolds-number swimming can be performed efficiently and robustly with only minimal computational power.

What carries the argument

Neuroevolution of minimal neural-network controllers for a three-bead low-Reynolds-number swimmer model.

If this is right

  • Artificial microswimmers can be made autonomous with very low onboard computation.
  • Experimental designs can use simple controllers instead of complex ones for efficient swimming.
  • The same minimal models may be applied to understand locomotion in microorganisms such as Chlamydomonas reinhardtii.

Where Pith is reading between the lines

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

  • If the 2D gaits survive in 3D viscous fluids, the approach could guide fabrication of real microswimmers.
  • The rotational gaits might serve as built-in reorientation mechanisms in navigation tasks.
  • Minimal network size suggests energy savings when the controllers are realized in hardware.

Load-bearing premise

The two-dimensional three-bead hydrodynamic model plus the chosen reward function in the reinforcement learning setup are sufficient to produce strategies that remain effective when transferred to three-dimensional physical experiments or real biological fluids.

What would settle it

Transferring the learned neural controllers to a physical three-bead device in a viscous fluid and checking whether the predicted directed or rotational motions appear.

Figures

Figures reproduced from arXiv: 2606.05984 by Benedikt Hartl, Gerhard Kahl, Julian Lemmel, Maximilian Huebl, Ruma Maity.

Figure 1
Figure 1. Figure 1: FIG. 1. Microswimmer model and training setup. (a) Sketch of the microswimmer model, consist [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Visualization of the characteristic features of the flapping (panels (a) and (b)), the chiral [PITH_FULL_IMAGE:figures/full_fig_p017_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Forces [PITH_FULL_IMAGE:figures/full_fig_p018_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. Forces [PITH_FULL_IMAGE:figures/full_fig_p019_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5. Visualization of the characteristic features of the rotational – panels (a) and (b) – and of [PITH_FULL_IMAGE:figures/full_fig_p020_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6. Forces [PITH_FULL_IMAGE:figures/full_fig_p021_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7. Forces [PITH_FULL_IMAGE:figures/full_fig_p022_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8. Trajectories of the COM of the [PITH_FULL_IMAGE:figures/full_fig_p023_8.png] view at source ↗
read the original abstract

Low-Reynolds-number microswimmers have recently attracted much interest for their ubiquity in biology and their applications in biotechnology and medicine. However, a key obstacle for the design and deployment of artificial microswimmers lies in their autonomy: to successfully perform tasks in any real-world scenario, these swimmers need to be able to interact with and adapt to their environment without external control. Here, we train a simple two-dimensional model microswimmer (consisting of three-bead) to learn autonomous swimming strategies via Reinforcement Learning, focusing on neuroevolution techniques to derive controller architectures with minimal complexity. We identify five different, characteristic swimming gaits: three of these gaits lead to directed locomotion with varying grades of efficiency and two gaits result in a rotational, inefficient movement. Remarkably, all of these gaits can be achieved by very simple neural networks (with less than ten nodes and weights), showing that low-Reynolds-number swimming can be achieved efficiently and robustly while requiring only minimal computational power. These results are of particular interest to the experimental design of artificial microswimmers and may have implications for modeling biological microorganisms such as Chlamydomonas reinhardtii.

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 / 1 minor

Summary. The paper trains a two-dimensional three-bead microswimmer via reinforcement learning and neuroevolution to discover autonomous swimming strategies. It reports the emergence of five characteristic gaits (three directed with varying efficiency, two rotational and inefficient), all realizable by neural networks with fewer than ten nodes and weights, and claims this demonstrates efficient, robust low-Re swimming with minimal computational power, with implications for artificial microswimmer design and biological modeling such as Chlamydomonas reinhardtii.

Significance. If the results hold under scrutiny, the demonstration that minimal-complexity controllers suffice for directed locomotion in the model would be useful for experimental microswimmer engineering and could inform reduced-order models of biological swimmers.

major comments (2)
  1. [Abstract and §3 (Results)] Abstract and §3 (Results): the central claim that the gaits are achieved 'efficiently and robustly' with '<10 nodes and weights' is presented without any quantitative performance metrics (e.g., mean speed, efficiency, or success rate across runs), baseline comparisons to hand-designed controllers, sensitivity tests on reward parameters, or error bars; this absence makes it impossible to judge whether the reported simplicity is load-bearing or an artifact of the specific training setup.
  2. [§2 (Model)] §2 (Model): the hydrodynamic interactions are computed in two dimensions, where the mobility tensor exhibits logarithmic far-field decay; this differs qualitatively from the 1/r Stokeslet decay in three-dimensional Stokes flow, so the phase relations, stroke timing, and minimal controller complexity found here may not transfer, yet the manuscript provides no 3D control simulations, no comparison to existing three-bead 3D results, and no experimental transfer test to support the robustness claim for real low-Re swimming.
minor comments (1)
  1. [Abstract] Abstract: the phrase 'varying grades of efficiency' is used without defining the efficiency metric or providing numerical values for the three directed gaits.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major point below and indicate planned revisions where appropriate.

read point-by-point responses

  1. Referee: [Abstract and §3 (Results)] Abstract and §3 (Results): the central claim that the gaits are achieved 'efficiently and robustly' with '<10 nodes and weights' is presented without any quantitative performance metrics (e.g., mean speed, efficiency, or success rate across runs), baseline comparisons to hand-designed controllers, sensitivity tests on reward parameters, or error bars; this absence makes it impossible to judge whether the reported simplicity is load-bearing or an artifact of the specific training setup.

    Authors: We agree that quantitative metrics are needed to support the claims of efficiency and robustness. In the revised manuscript we will add mean swimming speeds and hydrodynamic efficiencies for each of the five gaits, success rates and standard deviations across at least ten independent neuroevolution runs, and a direct comparison of net displacement per cycle against a simple hand-designed three-bead controller with fixed phase lag. A short sensitivity study on the two main reward weights will also be included in the supplementary material. These additions will demonstrate that the minimal-controller result is reproducible and not an artifact of the chosen training protocol. revision: yes

  2. Referee: [§2 (Model)] §2 (Model): the hydrodynamic interactions are computed in two dimensions, where the mobility tensor exhibits logarithmic far-field decay; this differs qualitatively from the 1/r Stokeslet decay in three-dimensional Stokes flow, so the phase relations, stroke timing, and minimal controller complexity found here may not transfer, yet the manuscript provides no 3D control simulations, no comparison to existing three-bead 3D results, and no experimental transfer test to support the robustness claim for real low-Re swimming.

    Authors: The work is explicitly framed as a two-dimensional model study chosen to enable exhaustive neuroevolution searches at modest computational cost. While we acknowledge that 2D logarithmic hydrodynamics differ from 3D Stokes flow, the central finding—that directed low-Re locomotion can be realized by neural networks with fewer than ten parameters—remains a general proof-of-principle that does not rely on the precise far-field decay. We will revise the discussion section to state the 2D limitation more explicitly and to moderate claims about immediate transfer to three-dimensional or biological systems. Full 3D neuroevolution and experimental validation lie outside the scope of the present manuscript. revision: partial

Circularity Check

0 steps flagged

No circularity; results emerge from independent RL training process

full rationale

The paper trains neural network controllers via reinforcement learning on a 2D three-bead Stokes model to produce swimming gaits. The reported gaits and the finding that networks with fewer than ten nodes/weights suffice are direct outputs of the training runs, not quantities defined by or fitted to the same data in a self-referential loop. No equations, self-citations, or ansatzes are quoted that reduce the central claims to inputs by construction. The derivation chain is therefore self-contained against external benchmarks such as the RL simulator itself.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Only the abstract is available, so the ledger records the minimal domain assumptions implied by the described approach; no explicit free parameters or invented entities are stated.

axioms (1)
  • domain assumption A two-dimensional three-bead hydrodynamic model captures the essential physics needed to discover functional swimming strategies.
    The abstract relies on this model without further justification of its fidelity to three-dimensional or biological conditions.

pith-pipeline@v0.9.1-grok · 5745 in / 1283 out tokens · 25135 ms · 2026-06-27T22:39:31.706815+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

93 extracted references · 8 canonical work pages

  1. [1]

    Purcell, E. M. Life at low Reynolds number.Am. J. Phys.45, 3-11 (1977)

    1977

  2. [2]

    Desai, N., Shaik, V. A. and Ardekani, A. M. Hydrodynamics-mediated trapping of micro- swimmers near drops.Soft matter14(2), 264-278 (2018)

    2018

  3. [3]

    and Pedley, T

    Magar, V., Goto, T. and Pedley, T. J. Nutrient uptake by a self-propelled steady squirmer. Q. J. Mech. Appl. Math.,56(1), 65-91 (2003)

    2003

  4. [4]

    The mechanics and statistics of active matter.Annu

    Ramaswamy, S. The mechanics and statistics of active matter.Annu. Rev. Condens. Matter Phys.1(1), 323-345 (2010)

    2010

  5. [5]

    Waggett, R. J. and Buskey, E. J. Calanoid copepod escape behavior in response to a visual predator.Mar. Biol.150, 599-607 (2007)

    2007

  6. [6]

    The mechanics of amphibian locomotion.J

    Barclay, O.R. The mechanics of amphibian locomotion.J. Exp. Biol.23(2), 177-203 (1946)

    1946

  7. [7]

    and Seymour, J

    Stocker, R. and Seymour, J. R. Ecology and physics of bacterial chemotaxis in the ocean. Microbiol. Mol. Biol.,76(4), 792-812 (2012)

    2012

  8. [8]

    Elgeti, J., Winkler, R. G. and Gompper, G. Physics of microswimmers—single particle motion and collective behavior: a review.Rep. Prog. Phys.78(5), 056601 (2015)

    2015

  9. [9]

    L., Fermigier, M., Stone, H

    Dreyfus, R., Baudry, J., Roper, M. L., Fermigier, M., Stone, H. A. and Bibette, J. Microscopic artificial swimmers.Nature437(7060), 862-865 (2005)

    2005

  10. [10]

    S., Gao, W., Wang, J

    Pak, O. S., Gao, W., Wang, J. and Lauga, E. High-speed propulsion of flexible nanowire motors: Theory and experiments.Soft Matter7(18), 8169-8181 (2011)

    2011

  11. [11]

    M., Gompper, G

    Alvarez, L., Friedrich, B. M., Gompper, G. and Kaupp, U. B. The computational sperm cell. Trends Cell Biol.24(3), 198-207 (2014)

    2014

  12. [12]

    Elgeti, J., Kaupp, U. B. and Gompper, G. Hydrodynamics of sperm cells near surfaces.Bio- phys. J.99(4), 1018-1026 (2010)

    2010

  13. [13]

    Berg, H. C. E. coli in Motion (Springer Science and Business Media, New York, 2008)

    2008

  14. [14]

    and Baroud, C

    Hamel, A., Fisch, C., Combettes, L., Dupuis-Williams, P. and Baroud, C. N. Transitions between three swimming gaits in Paramecium escape.Proc. Natl. Acad. Sci. U.S. A.108(18), 25 7290-7295 (2011)

    2011

  15. [15]

    Blake, J. R. and Sleigh, M. A. Mechanics of ciliary locomotion.Biol. Rev.49(1), 85-125 (1974)

    1974

  16. [16]

    Taylor, G. I. Analysis of the swimming of microscopic organisms.Proc. R. Soc. A: Math. Phys. Eng. Sci.209(1099), 447-461 (1951)

    1951

  17. [17]

    Lighthill, M. J. On the squirming motion of nearly spherical deformable bodies through liquids at very small Reynolds numbers.Commun. Pure Appl. Math.5(2), 109-118 (1952)

    1952

  18. [18]

    Blake, J. R. A spherical envelope approach to ciliary propulsion.J. Fluid Mech.46(1), 199-208 (1971)

    1971

  19. [19]

    and Golestanian, R

    Najafi, A. and Golestanian, R. Simple swimmer at low Reynolds number: Three linked spheres. Phys. Rev. E69(6), 062901 (2004)

    2004

  20. [20]

    Johnson, R. E. and Brokaw, C. J. Flagellar hydrodynamics. A comparison between resistive- force theory and slender-body theory.Biophys. J.25(1), 113-127 (1979)

    1979

  21. [21]

    and He, X

    Zhao, Y., Xuan, C., Qian, X., Alsaid, Y., Hua, M., Jin, L. and He, X. Soft phototactic swimmer based on self-sustained hydrogel oscillator.Sci. Robot.4(33), eaax7112 (2019)

    2019

  22. [22]

    and Pan, G

    Luo, Y., Xiao, Q., Zhu, Q. and Pan, G. Jet propulsion of a squid-inspired swimmer in the presence of background flow.Phys. Fluids.33(3), 031909 (2021)

    2021

  23. [23]

    C., Bae, J., Kim, H., Choi, H., Song, S

    Choi, Y., Park, C., Lee, A. C., Bae, J., Kim, H., Choi, H., Song, S. W., Jeong, Y., Choi, J., Lee, H. and Kwon, S. Photopatterned microswimmers with programmable motion without external stimuli.Nat. Commun.12(1), 4724 (2021)

    2021

  24. [24]

    and Thiffeault, J

    Chen, H. and Thiffeault, J. L. Shape matters: a Brownian microswimmer in a channel.J. Fluid Mech.916, A15 (2021)

    2021

  25. [25]

    and Komura, S

    Era, K., Koyano, Y., Hosaka, Y., Yasuda, K., Kitahata, H. and Komura, S. Autonomous elastic microswimmer.Europhys. Lett.133(3), 34001 (2021)

    2021

  26. [26]

    and Newhall, K

    Daftari, K. and Newhall, K. A. Self-avoidant memory effects on enhanced diffusion in a stochastic model of environmentally responsive swimming droplets.Phys. Rev. E105(2), 024609 (2022)

    2022

  27. [27]

    and Celani, A

    Borra, F., Biferale, L., Cencini, M. and Celani, A. Reinforcement learning for pursuit and evasion of microswimmers at low Reynolds number.Phys. Rev. Fluids7(2), 023103 (2022)

    2022

  28. [28]

    Kuramoto model with run-and-tumble dynamics.Phys

    Frydel, D. Kuramoto model with run-and-tumble dynamics.Phys. Rev. E104(2), 024203 (2021). 26

    2021

  29. [29]

    S., Farutin, A

    Rizvi, M. S., Farutin, A. and Misbah, C. Three-bead steering microswimmers.Phys. Rev. E 97(2), 023102 (2018)

    2018

  30. [30]

    and Sen, A

    Patra, D., Sengupta, S., Duan, W., Zhang, H., Pavlick, R. and Sen, A. Intelligent, self-powered, drug delivery systems.Nanoscale,5(4), 1273-1283 (2013)

    2013

  31. [31]

    and Fischer, P

    Walker, D., K¨ asdorf, B.T., Jeong, H.H., Lieleg, O. and Fischer, P. Enzymatically active biomimetic micropropellers for the penetration of mucin gels.Sci. Adv.,1(11), e1500501- e1500501 (2015)

    2015

  32. [32]

    Small8, 460–467 (2012)

    Gao, W.et al.Cargo-towing fuel-free magnetic nanoswimmers for targeted drug delivery. Small8, 460–467 (2012)

    2012

  33. [33]

    and Gracias, D.H

    Ghosh, A., Yoon, C., Ongaro, F., Scheggi, S., Selaru, F.M., Misra, S. and Gracias, D.H. Stimuli-responsive soft untethered grippers for drug delivery and robotic surgery.Front. Mech. Eng.,3, 7 (2017)

    2017

  34. [34]

    and Abbott, J.J

    Nelson, B.J., Kaliakatsos, I.K. and Abbott, J.J. Microrobots for minimally invasive medicine. Annu. Rev. Biomed. Eng.,12(1), 55-85 (2010)

    2010

  35. [35]

    and Couraud, P.-O

    Nassif, X., Bourdoulous, S., Eug` ene, E. and Couraud, P.-O. How do extracellular pathogens cross the blood–brain barrier?Trends Microbiol.,10, 227–232 (2002)

    2002

  36. [36]

    Mirbagheri, S. A. and Fu, H. C. Helicobacter pylori couples motility and diffusion to actively create a heterogeneous complex medium in gastric mucus.Phys. Rev. Lett.116, 198101 (2016)

    2016

  37. [37]

    and Volpe, G

    Cichos, F., Gustavsson, K., Mehlig, B. and Volpe, G. Machine learning for active matter.Nat. Mach. Intell.2, 94–103 (2020)

    2020

  38. [38]

    Tsang, A. C. H., Demir, E., Ding, Y. and Pak, O. S. Roads to smart artificial microswimmers. Adv. Intell. Syst.2, 1900137 (2020)

    2020

  39. [39]

    and Tsang, A.C., Learning to cooperate for low-Reynolds-number swimming: a model problem for gait coordination.Sci

    Liu, Y., Zou, Z., Pak, O.S. and Tsang, A.C., Learning to cooperate for low-Reynolds-number swimming: a model problem for gait coordination.Sci. Rep.,13(1), 9397 (2023)

    2023

  40. [40]

    and Tsang, A.C., Learning to Navigate in Chemical Fields Without A Map at Low Reynolds Numbers.Adv

    Liu, Y., Pak, O.S. and Tsang, A.C., Learning to Navigate in Chemical Fields Without A Map at Low Reynolds Numbers.Adv. Sci.,12(41), e10092 (2025)

    2025

  41. [41]

    and Zhu, L., Chemotactic navigation in robotic swim- mers via reset-free hierarchical reinforcement learning.Nat

    Xiong, T., Liu, Z., Wang, Y., Ong, C.J. and Zhu, L., Chemotactic navigation in robotic swim- mers via reset-free hierarchical reinforcement learning.Nat. Commun.,16(1), 5441 (2025)

    2025

  42. [42]

    and Tsang, A.C., Reinforcement learning selects multimodal locomotion strategies for bioinspired microswimmers.Soft Matter,21(12), 2363 (2025)

    Liu, Y., Wang, Z. and Tsang, A.C., Reinforcement learning selects multimodal locomotion strategies for bioinspired microswimmers.Soft Matter,21(12), 2363 (2025). 27

    2025

  43. [43]

    and Young, Y.N., Mechanical rotation at low Reynolds number via reinforcement learning.Phys

    Liu, Y., Zou, Z., Tsang, A.C.H., Pak, O.S. and Young, Y.N., Mechanical rotation at low Reynolds number via reinforcement learning.Phys. Fluids,33(6) (2021)

    2021

  44. [44]

    Mohamed, O., Tsang, A. C. H, Reinforcement learning of biomimetic navigation: a model problem for sperm chemotaxis.Eur. Phys. J. E,47, 59 (2024)

    2024

  45. [45]

    and Komura, S., Emergence of odd elasticity in a mi- croswimmer using deep reinforcement learning.Phys

    Lin, L.S., Yasuda, K., Ishimoto, K. and Komura, S., Emergence of odd elasticity in a mi- croswimmer using deep reinforcement learning.Phys. Rev. Res.,6(3), 033016 (2024)

    2024

  46. [46]

    and Man, Y., Navigation of a three-link microswimmer via deep reinforcement learning.Phys

    Lai, Y., Heydari, S., Pak, O.S. and Man, Y., Navigation of a three-link microswimmer via deep reinforcement learning.Phys. Rev. Fluids.,10(6), 064103 (2025)

    2025

  47. [47]

    and Choi, H., Autonomous 3D positional control of a magnetic microrobot using reinforcement learning.Nat

    Abbasi, S.A., Ahmed, A., Noh, S., Gharamaleki, N.L., Kim, S., Chowdhury, A.M.B., Kim, J.Y., Pan´ e, S., Nelson, B.J. and Choi, H., Autonomous 3D positional control of a magnetic microrobot using reinforcement learning.Nat. Mach. Intell.,6(1), 92 (2024)

    2024

  48. [48]

    and Liebchen, B., Optimal active particle navigation meets machine learning.Europhys

    Nasiri, M., L¨ owen, H. and Liebchen, B., Optimal active particle navigation meets machine learning.Europhys. Lett.,142(1), 17001 (2023)

    2023

  49. [49]

    and Pak, O.S., Reinforcement learning of a multi-link swimmer at low Reynolds numbers.Phys

    Qin, K., Zou, Z., Zhu, L. and Pak, O.S., Reinforcement learning of a multi-link swimmer at low Reynolds numbers.Phys. Fluids,35(3) (2023)

    2023

  50. [50]

    and Liu, W., An efficient deep reinforcement method for smart particle navi- gation in complex flows.New J

    Lu, X., Liu, J. and Liu, W., An efficient deep reinforcement method for smart particle navi- gation in complex flows.New J. Phys., (2026)

    2026

  51. [51]

    and Tsang, A.C

    Zou, Z., Liu, Y., Young, Y.N., Pak, O.S. and Tsang, A.C. Gait switching and targeted navi- gation of microswimmers via deep reinforcement learning.Commun. Phys.,5(1), 158 (2022)

    2022

  52. [52]

    and Kirkegaard, J.B

    Alonso, A. and Kirkegaard, J.B. Learning optimal integration of spatial and temporal infor- mation in noisy chemotaxis.PNAS nexus,3(7), pgae235 (2024)

    2024

  53. [53]

    Intelligent Navigation of a Magnetic Microrobot with Model-Free Deep Reinforcement Learning in a Real-World Environment.Micromachines,15, 112 (2024)

    Salehi, A.; Hosseinpour, S.; Tabatabaei, N.; Soltani Firouz, M.; Yu, T. Intelligent Navigation of a Magnetic Microrobot with Model-Free Deep Reinforcement Learning in a Real-World Environment.Micromachines,15, 112 (2024)

    2024

  54. [54]

    & Schulman, J

    Cobbe, K., Klimov, O., Hesse, C., Kim, T. & Schulman, J. Quantifying generalization in reinforcement learning.Proc. 36th Int. Conf. Mach. Learn.97, 1282–1289 (2019)

    2019

  55. [55]

    & Meger, D

    Henderson, P., Islam, R., Bachman, P., Pineau, J., Precup, D. & Meger, D. Deep reinforcement learning that matters.Proc. AAAI Conf. Artif. Intell.32, 3207–3214 (2018)

    2018

  56. [56]

    and Z¨ ottl, A

    Hartl, B., H¨ ubl, M., Kahl, G. and Z¨ ottl, A. Microswimmers learning chemotaxis with genetic algorithms.Proc. Natl. Acad. Sci. U.S. A.118(19), e2019683118 (2021). 28

    2021

  57. [57]

    Three-sphere low-Reynolds-number swimmer with a cargo container.Eur

    Golestanian, R. Three-sphere low-Reynolds-number swimmer with a cargo container.Eur. Phys. J. E.25, 1-4 (2008)

    2008

  58. [58]

    Stanley, K. O. and Miikkulainen, R. Evolving neural networks through augmenting topologies. Evol. Comput.10(2), 99-127 (2002)

    2002

  59. [59]

    S., Farutin, A

    Rizvi, M. S., Farutin, A. and Misbah, C. Size and shape affect swimming of a triangular bead-spring microswimmer.Phys. Rev. E98(4), 043104 (2018)

    2018

  60. [60]

    S., Nait-Ouhra, A., Farutin, A., Peyla, P., Rafai, S

    Rizvi, M. S., Nait-Ouhra, A., Farutin, A., Peyla, P., Rafai, S. and Misbah, C. Rheological signature of microswimmer phase-locking under flow.Phys. Rev. Fluids4(10), 103302 (2019)

    2019

  61. [61]

    Evolutionary algorithms in theory and practice: evolution strategies, evolutionary programming, genetic algorithms (Oxford university press, 1996)

    Back, T. Evolutionary algorithms in theory and practice: evolution strategies, evolutionary programming, genetic algorithms (Oxford university press, 1996)

    1996

  62. [62]

    Eiben, A. E. and Smith, J. E. Introduction to evolutionary computing (springer, 2015)

    2015

  63. [63]

    Propulsion at low Reynolds number.J

    Najafi, A., and Golestanian, R. Propulsion at low Reynolds number.J. Condens. Matter Phys. 17(14), S1203 (2005)

    2005

  64. [64]

    and Mikhailov, A

    Iima, M. and Mikhailov, A. S. Propulsion hydrodynamics of a butterfly micro-swimmer.Eu- rophys. Lett.85(4), 44001 (2009)

    2009

  65. [65]

    and Bartolo, D

    Lauga, E. and Bartolo, D. No many-scallop theorem: Collective locomotion of reciprocal swimmers.Phys. Rev. E Stat. Nonlin. Soft Matter.78(3), 030901 (2008)

    2008

  66. [66]

    M., Kenneth, O., Gat, O., and Avron, J

    Leshansky, A. M., Kenneth, O., Gat, O., and Avron, J. E., A frictionless microswimmer.New J. Phys.9(5), 145 (2007)

    2007

  67. [67]

    Active Matter and Artificial Intelligence

    Volpe, G., Cichos, F., & Volpe, G. Active Matter and Artificial Intelligence. In bookActive Colloids: From Fundamentals to Frontiers(Royal Society of Chemistry, 2024)

    2024

  68. [68]

    & and Bian, X

    Mo, C., Li, G. & and Bian, X. Challenges and attempts to make intelligent microswimmers Front. Phys.11, (2023)

    2023

  69. [69]

    C., Quah, T

    Takatori, S. C., Quah, T. & Rawlings, J. B. Feedback Control of Active MatterAnnu. Rev. Condens. Matter Phys.16, 319-341 (2025)

    2025

  70. [70]

    Collective intelligence: A unifying concept for integrating biology across scales and substrates.Commun

    McMillen, P., and Levin, M. Collective intelligence: A unifying concept for integrating biology across scales and substrates.Commun. Biol.7, 378 (2024)

    2024

  71. [71]

    & Risi, S

    Pontes-Filho, S., Walker, K., Najarro, E., Nichele, S. & Risi, S. A unified substrate for body–brain co-evolution. In *ICLR workshop on Cells2Societies* (2022)

    2022

  72. [72]

    & Efros, A.A

    Pathak, D., Lu, C., Darrell, T., Isola, P. & Efros, A.A. Learning to control self-assembling morphologies: a study of generalization via modularity. InProc. 33rd Int. Conf. Neural 29 Inf. Process. Syst., 206 (Curran Associates, Red Hook, NY, 2019). Available at:https: //pathak22.github.io/modular-assemblies/

    2019

  73. [73]

    & Risi, S

    Sudhakaran, S., Grbic, D., Li, S., Katona, A., Najarro, E., Glanois, C. & Risi, S. Growing 3D Artefacts and Functional Machines with Neural Cellular Automata.Artificial Life Conference Proceedings: ALIFE 2021, 108 (2021). DOI: 10.1162/isal a 00451

    work page doi:10.1162/isal 2021

  74. [74]

    Morphological change in machines accelerates the evolution of robust behavior

    Bongard, J. Morphological change in machines accelerates the evolution of robust behavior. Proc. Natl Acad. Sci. USA108, 1234-1239 (2011). DOI: 10.1073/pnas.1015390108

    work page doi:10.1073/pnas.1015390108 2011

  75. [75]

    2024 Bayesian polynomial neural networks and polynomial neural ordinary differential equations.PLOS Computational Biology20, e1012414

    Brodbeck, L., Hauser, S. & Iida, F. Morphological evolution of physical robots through model-free phenotype development.PLoS ONE10, e0128444 (2015). DOI: 10.1371/jour- nal.pone.0128444

    work page doi:10.1371/jour- 2015

  76. [76]

    & Sandini, G

    Lungarella, M., Metta, G., Pfeifer, R. & Sandini, G. Developmental robotics: a survey.Con- nection Science15, 151–190 (2003). DOI: 10.1080/09540090310001655110

    work page doi:10.1080/09540090310001655110 2003

  77. [77]

    Neuroevolution of decentralized decision-making in N -bead swimmers leads to scalable and robust collective locomotion.Commun

    Hartl, B., Levin, M., and Z¨ ottl, A. Neuroevolution of decentralized decision-making in N -bead swimmers leads to scalable and robust collective locomotion.Commun. Phys.8, 194 (2025)

    2025

  78. [78]

    & Bongard, J

    Kriegman, S., Blackiston, D., Levin, M. & Bongard, J. A scalable pipeline for design- ing reconfigurable organisms.Proc. Natl Acad. Sci. USA117, 1853–1859 (2020). DOI: 10.1073/pnas.1910837117

    work page doi:10.1073/pnas.1910837117 2020

  79. [79]

    & Bongard, J

    Kriegman, S., Blackiston, D.J., Levin, M. & Bongard, J. Kinematic self-replication in reconfigurable organisms.Proc. Natl Acad. Sci. USA118, e2112672118 (2021). DOI: 10.1073/pnas.2112672118

    work page doi:10.1073/pnas.2112672118 2021

  80. [80]

    Sci.11, 2303575 (2024)

    Gumuskaya, G.et al.Motile living biobots self-construct from adult human somatic progenitor seed cells.Adv. Sci.11, 2303575 (2024). DOI: 10.1002/advs.202303575

    work page doi:10.1002/advs.202303575 2024

Showing first 80 references.