Pith. sign in

REVIEW 2 major objections 5 minor 110 references

Microcosmos models artificial life as elastic filaments in a viscous fluid that is both GPU-scalable and fully differentiable.

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 →

T0 review · grok-4.5

2026-07-12 05:51 UTC pith:MFN5I4ZT

load-bearing objection Solid systems paper that actually fills the three-way gap it claims (differentiable Cosserat filaments + resolved LBM fluid + evolutionary search) and ships open code; the hydrodynamics are approximate but not load-bearing for the platform claim. the 2 major comments →

arxiv 2607.02954 v1 pith:MFN5I4ZT submitted 2026-07-03 cs.NE

Microcosmos: Reimagining Artificial Life for the GPU Era

classification cs.NE
keywords artificial lifeelastic filamentslattice Boltzmannimmersed boundary methoddifferentiable simulationquality diversityopen-ended evolutionGPU scaling
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.

Artificial life has long been stuck between abstract rule systems that evolve easily but feel unphysical, and detailed physics engines that cannot support the huge populations open-ended evolution needs. Microcosmos tries to close that gap by representing every creature as a chain of elastic filaments living in a two-dimensional viscous fluid. The fluid is solved with a lattice-Boltzmann method, the filaments are Cosserat rods updated by position-based dynamics, and the whole pipeline is written in JAX so it runs on GPUs and can be differentiated end-to-end. Hand-designed gaits recover known low-Reynolds constraints, gradient descent folds filaments into MNIST shapes, and quality-diversity search invents a menagerie of swimming and chemotaxis strategies. Runtime grows only linearly with particle count. The authors release the engine as an open platform whose long-term aim is large-scale, physically grounded open-ended evolution.

Core claim

A single simulation engine can simultaneously be physically grounded in viscous fluid dynamics, fully end-to-end differentiable, and linear-scaling on modern GPUs, thereby supporting both gradient-based morphology optimization and evolutionary discovery of diverse locomotion and chemotaxis behaviors.

What carries the argument

Elastic Cosserat filaments coupled to a lattice-Boltzmann fluid via the immersed-boundary method, with all constraints and forces expressed so that JAX can differentiate through the entire time-stepping loop.

Load-bearing premise

The combination of grid-based soft repulsion and diffuse immersed-boundary forcing is accurate enough that the evolutionary dynamics remain physically credible even though fluid can still leak through the zero-thickness filaments and collisions are never exact.

What would settle it

Run the same hand-designed reciprocal and non-reciprocal gaits at progressively higher viscosity and check whether net displacement of reciprocal strategies collapses to zero while non-reciprocal strategies continue to propel, exactly as Purcell's scallop theorem requires; any persistent reciprocal locomotion would falsify the fluid coupling.

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

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

2 major / 5 minor

Summary. Microcosmos is a JAX-based, GPU-native, end-to-end differentiable simulator for artificial life in which organisms are elastic Cosserat filament chains immersed in a 2D viscous fluid solved by lattice Boltzmann (D2Q9) with immersed-boundary coupling. The authors validate the engine with four experiments: hand-designed gaits that recover the qualitative signature of Purcell’s scallop theorem under a viscosity sweep; gradient-based folding of 1000-node filaments into MNIST digit shapes; NEAT/CPPN neuroevolution and MAP-Elites quality-diversity search that discover diverse swimming and chemotaxis behaviors; and wall-clock measurements showing linear scaling with particle count up to 500k particles. The platform is released open-source with the long-term aim of supporting large-scale open-ended evolution in a physically grounded substrate.

Significance. If the engineering claims hold, the paper supplies a missing middle ground between abstract ALife substrates and high-fidelity but non-scalable biophysical simulators. The combination of differentiable Cosserat rods, a resolved LBM fluid, linear scaling, and open code is a concrete platform contribution that the community can immediately build upon for morphology and controller co-evolution. Explicit strengths include the open GitHub release, the end-to-end differentiability demonstrated by MNIST folding through PBD and fluid, the qualitative recovery of scallop-theorem behavior, and the linear scaling data. These are falsifiable, reproducible engineering results rather than purely aspirational claims.

major comments (2)
  1. Methods (Fluid-filament Interaction) and Discussion acknowledge fluid leakage through zero-thickness filaments and non-exact collisions from grid-based steric repulsion. The hand-designed locomotion experiment (Figure 2) recovers only the qualitative scallop-theorem distinction. For a platform paper this is acceptable, but the manuscript should state more explicitly that the hydrodynamics are approximate and that quantitative drag coefficients or force-free swimming speeds have not been validated against analytic Stokes solutions or established Cosserat-fluid benchmarks (e.g., SophT, PyElastica). Without that caveat the claim of “physically plausible” locomotion risks overstatement for readers who expect quantitative fidelity.
  2. The strongest novelty claim (“first system to combine differentiable flexible filament simulation with a resolved fluid solver for both morphology and controller optimization”) is asserted in the Introduction and Related Work. The comparisons to SophT, PyElastica, Diff-FlowFSI and DiffAqua are useful, yet a short table or paragraph that enumerates which of the three ingredients (differentiable Cosserat, resolved fluid, evolutionary morphology/controller search) each prior system lacks would make the gap claim fully checkable rather than narrative.
minor comments (5)
  1. Figure 5 caption and text claim linear scaling; the plot itself is not shown in the supplied manuscript text. Ensure the figure and any fit statistics appear in the final version.
  2. Equations (1)–(4) for the PBD bending and position passes are clear, but the shear-stiffness blending in Eq. (3) would benefit from a one-sentence geometric interpretation of the Kirchhoff–Love residual.
  3. The MAP-Elites behavioral descriptors E_bend and E_pos (Eqs. 5–6) are sums of squared rest-parameter changes; a brief note on why these particular descriptors were chosen over, e.g., center-of-mass trajectory features would help reproducibility.
  4. Typographical: “M üller et al.” appears with a space artifact; “Guti érrez” likewise. Standardize author names and accents.
  5. The Discussion mentions restriction to cactus graphs; a short forward-looking sentence on how general graphs or self-assembly would be added would clarify the roadmap without over-claiming current capability.

Circularity Check

0 steps flagged

No significant circularity; engine is validated against external physical constraints and empirical benchmarks rather than tautological self-definition.

full rationale

Microcosmos is an engineering systems paper that introduces a GPU-native, differentiable filament-fluid simulator and exercises it on four independent validation tasks. Hand-designed gaits recover the qualitative signature of Purcell’s scallop theorem (reciprocal motion fails at high viscosity; non-reciprocal succeeds), an external physical constraint not encoded in the free parameters. Filament folding optimizes rest angles/lengths via Adam through the full PBD+LBM pipeline to match external MNIST targets; success demonstrates differentiability and expressivity rather than predicting a quantity already fitted. Neuroevolution/QD maximize external fitness (net displacement or energy collected) and discover diverse gaits; the archive is an empirical outcome, not a construction. Linear wall-clock scaling is a measured runtime property. No equation reduces a claimed prediction to a fitted constant by construction, no uniqueness theorem is imported from overlapping authors to forbid alternatives, and self-citations (NEAT, CPPN, MAP-Elites) are ordinary method references, not load-bearing premises. The approximate hydrodynamics (grid steric repulsion, diffuse IBM) are acknowledged limitations of fidelity, not circularities in the derivation chain. The paper is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 1 invented entities

The paper is primarily an engineering systems contribution. Its load-bearing content rests on standard continuum-mechanics and lattice-Boltzmann approximations plus a collection of practical numerical choices (diffuse interfaces, artificial thickening, grid steric fields) that trade exactness for GPU scalability. No new physical constants or particles are postulated; the free parameters are ordinary simulation and optimization hyper-parameters.

free parameters (6)
  • Adam learning rate and update budget for folding = 0.04 / 300 steps
    Set to 0.04 and 300 steps; directly controls whether MNIST folding succeeds and therefore the differentiability claim.
  • Simulation horizon for folding loss = 250 steps
    250 steps; the loss is evaluated only at the final frame, so the horizon is a free choice that affects gradient signal.
  • Viscosity / Reynolds sweep range = 10^{-3}–2
    10^{-3} to 2; chosen to straddle the inertial-to-viscous transition and to illustrate scallop-theorem behavior.
  • Behavioral descriptors E_bend and E_pos for MAP-Elites
    Sum of squared changes in rest angle and rest length; the archive geometry and claimed diversity depend on this hand-chosen feature space.
  • Grid resolution and particle counts for scaling = 256×256 / ≤500k
    Fixed 256×256 grid, up to 500 k particles; the linear-scaling claim is measured under these concrete settings.
  • Per-edge stiffnesses and rest lengths/angles of hand-designed gaits
    Manually tuned to produce worm, tadpole, cilia, jellyfish and ray behaviors used as physical-correctness evidence.
axioms (5)
  • domain assumption D2Q9 lattice Boltzmann with TRT collision approximates the incompressible Navier-Stokes equations at the viscosities used
    Invoked throughout Methods (Fields) and the viscosity-sweep experiment; standard but not re-derived.
  • domain assumption Position-based dynamics iterations sufficiently enforce Cosserat stretch/shear/bend constraints for the chosen time-step
    Core of the filament solver (Methods: Filaments); unconditional stability is cited from Müller et al. but accuracy is not quantified.
  • ad hoc to paper Diffuse immersed-boundary multi-direct forcing plus artificial two-layer thickening yields an adequate no-slip condition for zero-thickness filaments
    Explicitly introduced to mitigate fluid leakage (Fluid-filament Interaction); the paper acknowledges residual leakage.
  • ad hoc to paper Grid-deposited, FFT-diffused density fields produce acceptable self-avoidance without pairwise collision resolution
    Chosen purely for O(n) scaling (Fields); exact contact dynamics are sacrificed.
  • domain assumption Purcell’s scallop theorem and known low-Re gaits are the correct qualitative benchmarks for physical credibility
    Used as the success criterion for the hand-designed locomotion experiment (Hand-designed Locomotion).
invented entities (1)
  • Microcosmos filament-field engine independent evidence
    purpose: Unified, GPU-native, differentiable substrate that couples elastic filaments to a resolved viscous fluid for ALife evolution
    The concrete software architecture and encoding interface are new; independent evidence is the released code and the four validation experiments.

pith-pipeline@v1.1.0-grok45 · 17062 in / 3157 out tokens · 36447 ms · 2026-07-12T05:51:12.091756+00:00 · methodology

0 comments
read the original abstract

Most artificial life simulators either operate on abstract substrates disconnected from physical reality, or simulate physically grounded worlds that do not scale to the population sizes required for open-ended evolution. We present Microcosmos, a simulation engine in which artificial lifeforms are modeled as elastic filament chains inhabiting a two-dimensional viscous fluid world, designed from the ground up for modern GPU hardware and end-to-end differentiable simulation. We validate the engine through four experiments. Hand-designed locomotion strategies confirm that the fluid coupling respects known physical constraints. Gradient-based optimization of filament folding demonstrates both the full differentiability of the simulator and the expressivity of the filament encodings. Neuroevolution and quality-diversity search produce a wide range of swimming and chemotaxis behaviors automatically. Linear scaling with particle count confirms the engine supports large-scale simulation. Microcosmos is released as an open platform with the long-term goal of supporting large-scale open-ended evolutionary simulations, designed to be physically plausible and computationally scalable.

Figures

Figures reproduced from arXiv: 2607.02954 by Bert Wang-Chak Chan, Ciaran Regan, Grisha Szep, Kenneth O. Stanley, Mark Tensen, Mizuki Oka.

Figure 1
Figure 1. Figure 1: Microcosmos: a scalable simulator for artificial life. Individuals are modeled as flexible filaments in a sim￾ulated fluid environment. (a) The three core physics com￾ponents: (i) Filaments have preferred resting shapes, able to bend and deform elastically. (ii) Self-avoidance and inter￾body repulsion mediated by scalar fields. (iii) Two-way cou￾pling between individuals and the surrounding fluid. (b) Di￾v… view at source ↗
Figure 2
Figure 2. Figure 2: Realistic fluid dynamics of the Microcosmos engine as confirmed by five creatures with hand-designed geometry and locomotion (left) and viscosity sweep (right). Consistent with Purcell’s scallop theorem, time-reversible strategies (e.g. ray) produce zero net displacement at high viscosity, while non-reciprocal strategies (e.g. cilia) main￾tain viable locomotion. See supplementary materials Figure S2 for an… view at source ↗
Figure 3
Figure 3. Figure 3: Differentiability of the Microcosmos engine as demonstrated by MNIST digit folding via stochastic gradient descent (SGD). Each column shows a different target digit (0–9). The first three rows show the filament folding through a single 250- step simulation run (top to bottom: early, mid, and late), with the bottom row showing the corresponding target digit. See supplementary materials Figure S3 for animati… view at source ↗
Figure 5
Figure 5. Figure 5: Wall clock time scales linearly with the number of [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 4
Figure 4. Figure 4: Evolvability of diverse locomotion strategies as demonstrated by QD algorithm MAP-Elites. See [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

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

110 extracted references · 10 linked inside Pith

  1. [1]

    , title =

    Adami, C. , title =. Physics Letters A , year = 1995, volume =

  2. [2]

    and Brown, C

    Adami, C. and Brown, C. T. , title =. Artificial Life IV , pages =

  3. [3]

    and Brown, C.T

    Adami, C. and Brown, C.T. and Haggerty, M.R. , title =. Proceedings of the Third European Conference on Advances in Artificial Life , editor =

  4. [4]

    and Paczuski, M

    Bak, P. and Paczuski, M. , title =. 1996 , abstract =

  5. [5]

    and Sneppen, K

    Bak, P. and Sneppen, K. , title =. Physical Review Letters , year = 1993, volume = 71, pages =

  6. [6]

    and Tang, C

    Bak, P. and Tang, C. and Wiesenfeld, K. , title =. Physical Review Letters , year =

  7. [7]

    and Tang, C

    Bak, P. and Tang, C. and Wiesenfeld, K. , title =. Physical Review A , year =

  8. [8]

    , title =

    Burlando, B. , title =. Journal of Theoretical Biology , year =

  9. [9]

    and Adami, C

    Chu, J. and Adami, C. , title =. Artificial Life V , pages =

  10. [10]

    and Cooper, V.S

    Elena, S.F. and Cooper, V.S. and Lenski, R.E. , title =. Science , year = 1996, volume = 272, pages =

  11. [11]

    and Sornette, D

    Gil, L. and Sornette, D. , title =. Physical Review Letters , year =

  12. [12]

    and Eldredge, N

    Gould, S.J. and Eldredge, N. , title =. Palaeobiology , year = 1977, volume = 3, pages =

  13. [13]

    and Eldredge, N

    Gould, S.J. and Eldredge, N. , title =. Nature , year =

  14. [14]

    and Travisano, M

    Lenski, R. and Travisano, M. , title =. Proceedings of the National Academy of Science USA , year = 1994, volume = 91, pages =

  15. [15]

    , title =

    Newman, M.E.J. , title =. Proceedings of the Royal Society of London, Series B , year = 1996, volume = 263, pages =

  16. [16]

    , title =

    Newman, M.E.J. , title =. Eprint adap-org/9702002 , year =

  17. [17]

    and Fraser, S.M

    Newman, M.E.J. and Fraser, S.M. and Sneppen, K. and Tozier, W.A. , title =. Physics Letters A , year = 1997, volume =

  18. [18]

    and Brown, C.T

    Ofria, C. and Brown, C.T. , title =. 1998 , abstract =

  19. [19]

    , title =

    Raup, D.M. , title =. Paleobiology , year =

  20. [20]

    and Johansen, A

    Sornette, D. and Johansen, A. and Dornic, I. , title =. Journal de Physique I. , year =

  21. [21]

    and Bascompte, J

    Sol\'e, R.V. and Bascompte, J. , title =. Proceedings of the Royal Society of London, Series B , year =

  22. [22]

    , title =

    Willis, J.C. , title =. 1922 , abstract =

  23. [23]

    American Journal of Physics , volume=

    Life at low Reynolds number , author=. American Journal of Physics , volume=

  24. [24]

    Reports on progress in physics , volume=

    The hydrodynamics of swimming microorganisms , author=. Reports on progress in physics , volume=

  25. [25]

    Taylor, G. I. , title =. Proceedings of the Royal Society of London. Series A , volume =. 1951 , abstract =

  26. [26]

    Reports on progress in physics , volume=

    Physics of microswimmers—single particle motion and collective behavior: a review , author=. Reports on progress in physics , volume=. 2015 , publisher=

  27. [27]

    elegans: an integrated neuromechanical model , author=

    Gait modulation in C. elegans: an integrated neuromechanical model , author=. Frontiers in computational neuroscience , volume=. 2012 , publisher=

  28. [28]

    Artificial life , volume=

    Open problems in artificial life , author=. Artificial life , volume=. 2000 , publisher=

  29. [29]

    and Bedau, M

    Packard, N. and Bedau, M. A. and Channon, A. and Ikegami, T. and Rasmussen, S. and Stanley, K. O. and Taylor, T. , title =. Artificial Life , volume =. 2019 , abstract =

  30. [30]

    Peskin, C. S. , title =. Acta Numerica , volume =. 2002 , abstract =

  31. [31]

    , title =

    Succi, S. , title =. 2018 , abstract =

  32. [32]

    Position based dynamics , journal =

    M\". Position based dynamics , journal =. 2007 , abstract =

  33. [33]

    ACM SIGGRAPH 2008 Papers , pages=

    Discrete elastic rods , author=. ACM SIGGRAPH 2008 Papers , pages=. 2008 , publisher=

  34. [34]

    2005 , publisher=

    Nonlinear problems of elasticity , author=. 2005 , publisher=

  35. [35]

    Astrophysics Source Code Library , pages=

    Jax: Autograd and xla , author=. Astrophysics Source Code Library , pages=. 2021 , abstract =

  36. [36]

    arXiv preprint arXiv:1910.00935 , year=

    Difftaichi: Differentiable programming for physical simulation , author=. arXiv preprint arXiv:1910.00935 , year=

  37. [37]

    ACM SIGGRAPH , year =

    Sims, Karl , title =. ACM SIGGRAPH , year =

  38. [38]

    Plantec, Erwan and Hamon, Gautier and Etcheverry, Mayalen and Oudeyer, Pierre-Yves and Moulin-Frier, Cl\'. Flow-. 2023 , doi=

  39. [39]

    ACM Transactions on Graphics (ToG) , volume=

    Diffpd: Differentiable projective dynamics , author=. ACM Transactions on Graphics (ToG) , volume=. 2021 , publisher=

  40. [40]

    and Doolen, G

    Chen, S. and Doolen, G. D. , title =. Annual Review of Fluid Mechanics , volume =. 1998 , abstract =

  41. [41]

    Qian, Y. H. and d'Humi\`. Lattice. Europhysics Letters , volume =. 1992 , abstract =

  42. [42]

    and Zheng, C

    Guo, Z. and Zheng, C. and Shi, B. , title =. Physical Review E , volume =. 2002 , abstract =

  43. [43]

    and Fan, J

    Wang, Z. and Fan, J. and Luo, K. , title =. International Journal of Multiphase Flow , volume =. 2008 , abstract =

  44. [44]

    Physica D: nonlinear phenomena , volume=

    Studying artificial life with cellular automata , author=. Physica D: nonlinear phenomena , volume=. 1986 , publisher=

  45. [45]

    Holland, J. H. , title =. 1975 , abstract =

  46. [46]

    Proceedings of the National Academy of Sciences , volume=

    A scalable pipeline for designing reconfigurable organisms , author=. Proceedings of the National Academy of Sciences , volume=. 2020 , publisher=

  47. [47]

    and Miikkulainen, Risto , title =

    Stanley, Kenneth O. and Miikkulainen, Risto , title =. Evolutionary Computation , volume =. 2002 , doi =

  48. [48]

    , title =

    Stanley, Kenneth O. , title =. Genetic Programming and Evolvable Machines , volume =. 2007 , doi =

  49. [49]

    arXiv preprint arXiv:1504.04909 , year =

    Mouret, Jean-Baptiste and Clune, Jeff , title =. arXiv preprint arXiv:1504.04909 , year =

  50. [50]

    Daniel and Frey, Erik and Raichuk, Anton and Girgin, Sertan and Mordatch, Igor and Bachem, Olivier , title =

    Freeman, C. Daniel and Frey, Erik and Raichuk, Anton and Girgin, Sertan and Mordatch, Igor and Bachem, Olivier , title =. Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track , year =

  51. [51]

    Computer Physics Communications , volume =

    Ataei, Mohammadmehdi and Salehipour, Hesam , title =. Computer Physics Communications , volume =. 2024 , doi =

  52. [52]

    Proceedings of the ACM on Computer Graphics and Interactive Techniques , volume =

    Stuyck, Tuur and Chen, Hsiao-yu , title =. Proceedings of the ACM on Computer Graphics and Interactive Techniques , volume =. 2023 , doi =

  53. [53]

    and Alieva, Ayya and Wang, Qing and Brenner, Michael P

    Kochkov, Dmitrii and Smith, Jamie A. and Alieva, Ayya and Wang, Qing and Brenner, Michael P. and Hoyer, Stephan , title =. Proceedings of the National Academy of Sciences , volume =. 2021 , doi =

  54. [54]

    Nature Communications , volume =

    Zhang, Xiaotian and Chan, Fan Kiat and Parthasarathy, Tejaswin and Gazzola, Mattia , title =. Nature Communications , volume =. 2019 , doi =

  55. [55]

    and Lu, Xi-Yun , title =

    Tian, Fang-Bao and Luo, Haoxiang prestige and Zhu, Luoding and Liao, James C. and Lu, Xi-Yun , title =. Journal of Computational Physics , volume =. 2011 , doi =

  56. [56]

    Proceedings of the 9th International Conference on Motion in Games , year =

    Macklin, Miles and M. Proceedings of the 9th International Conference on Motion in Games , year =. doi:10.1145/2994258.2994272 , abstract =

  57. [57]

    ACM Transactions on Graphics (TOG) , volume=

    Diffaqua: A differentiable computational design pipeline for soft underwater swimmers with shape interpolation , author=. ACM Transactions on Graphics (TOG) , volume=. 2021 , publisher=

  58. [58]

    , title =

    Lehman, Joel and Stanley, Kenneth O. , title =. Evolutionary Computation , volume =. 2011 , doi =

  59. [59]

    Complex Systems , volume =

    Chan, Bert Wang-Chak , title =. Complex Systems , volume =. 2019 , abstract =

  60. [60]

    Scientific Excellence in Supercomputing , year=

    Evolution and optimization of digital organisms , author=. Scientific Excellence in Supercomputing , year=

  61. [61]

    , title =

    Ofria, Charles and Wilke, Claus O. , title =. Artificial Life , volume =. 2004 , doi =

  62. [62]

    ACM SIGEVOlution , volume=

    Unshackling evolution: evolving soft robots with multiple materials and a powerful generative encoding , author=. ACM SIGEVOlution , volume=. 2014 , publisher=

  63. [63]

    and Stanton, Adam and Timperley, Christopher and Vostinar, Anya and Wiser, Michael , title =

    Taylor, Tim and Bedau, Mark and Channon, Alastair and Ackley, David and Banzhaf, Wolfgang and Beslon, Guillaume and Dolson, Emily and Froese, Tom and Hickinbotham, Simon and Ikegami, Takashi and McMullin, Barry and Packard, Norman and Rasmussen, Steen and Virgo, Nathaniel and Agmon, Eran and Clark, Edward and McGregor, Simon and Ofria, Charles and Ropella...

  64. [64]

    and Soros, Lisa B

    Pugh, Justin K. and Soros, Lisa B. and Stanley, Kenneth O. , title =. Frontiers in Robotics and AI , volume =. 2016 , doi =

  65. [65]

    Informatik Spektrum , volume =

    Heinemann, Christian , title =. Informatik Spektrum , volume =. 2008 , doi =

  66. [66]

    2025 , howpublished =

    Whidden, Peter , title =. 2025 , howpublished =

  67. [67]

    2024 , howpublished =

    Heinemann, Christian , title =. 2024 , howpublished =

  68. [68]

    Artificial Life conference proceedings , pages=

    Indefinitely scalable computing= artificial life engineering , author=. Artificial Life conference proceedings , pages=. 2014 , organization=

  69. [69]

    Briefings in bioinformatics , volume=

    High performance cellular level agent-based simulation with FLAME for the GPU , author=. Briefings in bioinformatics , volume=. 2010 , publisher=

  70. [70]

    Software: Practice and Experience , volume=

    FLAME GPU 2: A framework for flexible and performant agent based simulation on GPUs , author=. Software: Practice and Experience , volume=. 2023 , publisher=

  71. [71]

    International Conference on Learning Representations , year =

    Matthews, Michael and Beukman, Michael and Lu, Chris and Foerster, Jakob , title =. International Conference on Learning Representations , year =

  72. [72]

    , title =

    Peskin, Charles S. , title =. Journal of Computational Physics , volume =. 1972 , doi =

  73. [73]

    Royal Society open science , volume=

    Forward and inverse problems in the mechanics of soft filaments , author=. Royal Society open science , volume=. 2018 , publisher=

  74. [74]

    arXiv preprint arXiv:2301.07028 , year=

    Aquarium: A fully differentiable fluid-structure interaction solver for robotics applications , author=. arXiv preprint arXiv:2301.07028 , year=

  75. [75]

    Computer Methods in Applied Mechanics and Engineering , volume=

    Self-propelling, soft, and slender structures in fluids: Cosserat rods immersed in the velocity--vorticity formulation of the incompressible Navier--Stokes equations , author=. Computer Methods in Applied Mechanics and Engineering , volume=. 2025 , publisher=

  76. [76]

    Journal of Computational Physics , volume =

    Fan, Xiantao and Wang, Jian-Xun , title =. Journal of Computational Physics , volume =. 2024 , doi =

  77. [77]

    Computer Methods in Applied Mechanics and Engineering , volume=

    Diff-FlowFSI: A GPU-optimized differentiable CFD platform for high-fidelity turbulence and FSI simulations , author=. Computer Methods in Applied Mechanics and Engineering , volume=. 2026 , publisher=

  78. [78]

    arXiv preprint arXiv:2504.08339 , year=

    TensorNEAT: a GPU-accelerated library for NeuroEvolution of Augmenting Topologies , author=. arXiv preprint arXiv:2504.08339 , year=

  79. [79]

    Proceedings of the companion conference on genetic and evolutionary computation , pages=

    Towards large-scale simulations of open-ended evolution in continuous cellular automata , author=. Proceedings of the companion conference on genetic and evolutionary computation , pages=. 2023 , abstract =

  80. [80]

    Science Advances , volume=

    Discovering sensorimotor agency in cellular automata using diversity search , author=. Science Advances , volume=. 2025 , publisher=

Showing first 80 references.