Pith. sign in

REVIEW 4 major objections 4 minor 1 cited by

Realizing Robotic Swimming with Unified Fluid-Robot Multiphysics

T0 review · 4 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read One least-action principle governs both swimming robot and the fluid around it, and the resulting simulator's gradients plan a C-start turn that works on hardware.

desk verdict A genuinely new differentiable fluid-robot simulator with solid steady-swimming hardware validation; the flagship C-start sim-to-real claim rests on one qualitative trial and needs reinforcement, but the paper deserves a serious referee. read the letter →

arxiv 2506.05012 v2 pith:X4KVTMED submitted 2025-06-05 cs.RO physics.comp-phphysics.flu-dyn

Jeong Hun Lee , Junzhe Hu , Sofia Kwok , Carmel Majidi , Zachary Manchester This is my paper
classification cs.ROphysics.comp-phphysics.flu-dyn MSC 76D0570H2565P1037M15
keywords fluid-robotmultiphysicsvariationalintegratorimmersedboundarymethoddifferentiablesimulationswimmingrobotC-startmaneuversim-to-realtransferNavier-Stokes
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper sets out to show that a swimming robot and the surrounding water can be modeled as a single physical system derived from one least-action principle, rather than as two separate simulators coupled loosely through forces. This yields a differentiable simulator whose gradients can optimize swimming gaits. The authors validate the approach on a six-link eel robot: steady undulating gaits match hardware trajectories more closely than a particle-based baseline, and a C-start escape maneuver optimized in simulation executes a 90-degree turn on the physical robot. The central claim is that strongly coupled fluid-robot simulation is an adequate predictive model for open-loop swimming behaviors.

What carries the argument

The central object is the unified action functional that combines the fluid Lagrangian (kinetic energy minus gravitational potential, with viscous dissipation treated as an external force) and the robot Lagrangian (kinetic minus potential of the articulated bodies), subject to no-slip, incompressibility, boundary, and joint constraints. The argument is carried by three pieces: discrete variational mechanics applied directly to this action, which produces a stable implicit midpoint and leap-frog integrator; an integral-form immersed boundary method whose integrated convolution matrix $\bar{E}$ maps fluid velocities to boundary velocities without node-based singularities; and the implicit function theorem applied to the implicit dynamics, which yields exact derivatives of the coupled system for optimization.

What would settle it

Run the optimized C-start gait on hardware in pools of different water depths, for example 5 cm and 20 cm, and compare the final heading and head trajectory. If the turn angle or path changes substantially with depth, the unmodeled free-surface dynamics are doing real work and the simulator is not predictive in that regime. Alternatively, measure the head-link trajectory for a novel gait never used in optimization; an RMSE above roughly 20 cm over a 4-second rollout would indicate the model is not generally predictive.

Watch

Extended reading notes

Core claim

The paper's central claim is that the coupled rigid-multibody and incompressible Navier-Stokes equations can be derived together as the first-order necessary conditions of one optimization problem: minimize a combined fluid-robot Lagrangian subject to kinematic, incompressibility, boundary, and no-slip constraints. The no-slip condition at the fluid-robot interface is the coupling mechanism, and its Lagrange multiplier produces the interaction force. Discretizing the action directly with discrete variational mechanics gives an implicit, second-order leap-frog integrator over the full state, avoiding the half-time-step mismatch that plagues configuration-only variational integrators. An integral-form immersed boundary method, obtained by integrating the discrete delta function along piecewise-linear boundary segments, enforces no-slip between nodes and removes singularities from overlapping boundary points. Gradients of the fully coupled dynamics are computed with the implicit function theorem, and those gradients drive an L-BFGS optimization of a C-start gait. On hardware, the optimized gait realizes the intended 90-degree turn, and steady swimming matches experimental head-link trajectories substantially better than an SPH baseline, with up to 75 percent error reduction at 40-degree amplitude.

Load-bearing premise

Everything rests on the premise that a two-dimensional incompressible Navier-Stokes model on a 1 cm grid, with fixed nominal physical parameters and no free-surface modeling, is faithful enough to the real three-dimensional hardware pool that an open-loop gait optimized in simulation still turns the robot 90 degrees on hardware.

Editorial extensions

If this is right

  • Gradient-based trajectory optimization can treat the fluid as part of the dynamics, so gaits like the C-start can be designed in simulation and executed open-loop on hardware.
  • Eulerian grid-based fluid models, when strongly coupled to the robot, can outperform particle-based SPH simulators for swimming tasks, especially at high amplitudes.
  • The same least-action derivation can be reused for other coupled multiphysics settings, since coupling forces arise naturally from constraint dual variables.
  • The integral immersed-boundary formulation resolves the node-duplication problem of classical immersed-boundary methods, making grid-based fluid-structure interaction tractable for articulated multibody robots.
  • Because the whole coupled system is differentiable, the same simulator can support closed-loop control, system identification, or learning, all using the same exact gradients.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The success on open-loop gaits suggests the simulator could support closed-loop control or reinforcement learning for underwater robots without large amounts of real-world data, a step the paper does not itself demonstrate.
  • Because the model is two-dimensional and ignores free-surface dynamics, its predictive power likely degrades in shallow water or at higher speeds where three-dimensional and surface effects dominate; a direct test would vary water depth and compare the achieved turn angle.
  • One could test the fluid model directly by comparing simulated vortex fields against particle image velocimetry on the hardware pool, rather than only comparing head-link trajectories.
  • The derivation chain from least action to variational integrator to implicit-function-theorem gradients suggests the same recipe could apply to soft robots or deformable bodies in fluid, where a single Lagrangian can already describe the mechanics.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper presents a differentiable simulator for strongly coupled fluid-robot dynamics. It derives the coupled incompressible Navier-Stokes and manipulator equations from a single least-action principle, discretizes the action with a variational integrator, and introduces an integral-form immersed-boundary method to enforce no-slip at the fluid-robot interface. The simulator is used to realize open-loop undulatory swimming and to gradient-optimize a C-start maneuver for a six-link eel robot. Validation consists of RMSE comparisons against a smoothed-particle-hydrodynamics baseline over 20 hardware trials per amplitude for steady swimming, and a single time-lapse comparison for the C-start. The paper claims successful sim-to-real transfer for both gaits and releases simulation code, hardware data, and schematics.

Significance. If the claims hold, the paper would be a meaningful advance: a variational, differentiable fluid-structure interaction simulator with strong coupling that can optimize dynamic swimming maneuvers and transfer them open-loop to hardware. The steady-swimming validation is a genuine sim-to-real test with repeated trials, and the C-start is a prospective gradient-optimized prediction; the reported 75% RMSE improvement over the SPH baseline at 40 degrees amplitude is notable. The integral-form immersed-boundary method is a plausible remedy for multibody IBM ill-conditioning, and the open release of code and data supports reproducibility. However, the significance is conditional: the C-start validation, which is the flagship demonstration of the gradients, is currently qualitative, and the discrete derivation has unresolved time-indexing issues. These need to be addressed before the paper's central claims can be accepted.

major comments (4)
  1. [Section V-C, Fig. 1] The abstract's claim 'We validate both gaits on physical hardware' rests on a single qualitative C-start trial. The text reports no number of repeated runs, no heading-error or center-of-mass trajectory comparison, no confidence intervals, and no explicit statement that the optimized parameters theta*=[17,18,20,24,36] deg, phi*=-1.16 rad, lambda*=0.57 L_B were executed exactly as computed without adjustment after observing hardware behavior. The same paragraph concedes a systematic undershoot in travel distance. A 90-degree reorientation is a coarse behavioral outcome, so this evidence is too weak to support the claim that gradients from the multiphysics simulator produced a correct dynamic gait. Please provide repeated trials with quantitative heading/COM errors and execution transparency, or restrict the validation claim to steady swimming.
  2. [Section IV-B, Eqs. (25)-(35)] The discrete dual variables have an apparent time-index inconsistency. In the discrete action (25), the constraints are written with step-k variables, and by the convention in (8) the associated multipliers are lambda_k, but the solved constraints (32)-(35) are enforced at k+1. Equations (30)-(31) contain p_k, lambda_3,k, and lambda_6,k, while (32)-(34) enforce G^T v_{k+1}=0, B v_{k+1}=v_bc, and E v_{k+1}=bvr(v_{k+1}). No discrete Euler-Lagrange equation or discrete Legendre transform is given that produces multipliers for the k+1 constraints. As written, the scheme appears to use old-time pressure and coupling multipliers with new-time constraints, which is not the implicit midpoint variational integrator claimed. Please derive the stationarity conditions with consistent time indices and state which multipliers are unknowns in each Newton solve.
  3. [Section IV-A, Eqs. (15)-(22)] The unified least-action derivation is not self-contained. The transition from (19) to (20)-(22) omits the integration-by-parts and functional-derivative steps: the pressure is identified with dot lambda_2, but the boundary terms from c3, the no-slip term c6, and the advection term are not shown; the viscous force in (18) depends on vf while it appears in the action as F(t)^T qf, and its variation with respect to vf is not accounted for. Since the unified derivation is a central contribution, these steps need to be supplied or the derivation cited to a complete reference.
  4. [Section IV-C, Eqs. (36)-(37)] The integral-form immersed-boundary method is not analyzed. The paper claims it properly enforces no-slip between nodes and is unaffected by overlapping nodes, but no convergence test, conditioning comparison, or ablation against the pointwise IBM is reported. This is a listed contribution and a claimed source of improved physical accuracy, so the section needs at least a numerical experiment isolating its effect.
minor comments (4)
  1. [Section III-A and Section IV-A] There are cross-referencing errors: Section III-A refers to 'the optimization problem (5)' before (5) is introduced, and Section IV-A refers to 'the FON conditions of (38)' when the action is (19). Please correct these equation references.
  2. [Section IV-B] The notation q_f is used for both the continuous fluid configuration and the discrete grid state; the acknowledged 'slight abuse of notation' makes the derivation harder to follow. Please use distinct notation for the discrete variables throughout the derivation.
  3. [Section V] No grid-convergence or time-step refinement study is reported for the variational integrator. Given the claim that the scheme is 'physically accurate,' a brief resolution study would strengthen the numerical claims.
  4. [Section VI] The limitations section omits the sim-to-real undershoot discussed in Section V-C and the absence of a grid-convergence study; consider moving these acknowledged limitations into Section VI for a complete assessment.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the least-action derivation is a self-contained reformulation of known coupled physics, and the sim-to-real C-start is an open-loop prediction independent of fitted inputs.

full rationale

The derivation chain is not circular. Equations (20)-(22) are obtained as the first-order necessary conditions of the combined action in Eq. (19), with the no-slip constraint c6 entering as a constraint rather than as an assumed outcome; this is a variational reformulation of standard Navier-Stokes plus manipulator dynamics, and its predictive content is tested against hardware rather than assumed. The integral-form immersed-boundary constraint (37) enforces no-slip by construction, i.e., it is a modeling input, not a predicted quantity. The C-start gait parameters (theta*, phi*, lambda*) were optimized in simulation against the objective (40) and then rolled out open-loop on hardware ('The resulting gaits are rolled out in an open-loop manner'), so the hardware 90-degree turn is a genuine forward prediction; no hardware data are reported as fitting simulator parameters. The paper's own admission that the simulator 'consistently undershoots the travel distance observed in the real world' (Section V-C) is a fidelity limitation and weakens the quantitative sim-to-real claim, but it is not circularity. Self-citations appear for the implicit-function-theorem derivative machinery ([33], Aquarium) and for variational-integrator background ([24], Dojo); these are methodological references and are not used to justify the central empirical claim, which rests on the hardware RMSE trials and the open-loop C-start execution. Because the central validation is prospective and the cited prior work is independently checkable, no circular step is present.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The central claim rests on standard continuum assumptions (incompressible Newtonian fluid, rigid-body robot) and on the numerical scheme's accuracy and conditioning. The hardware validation is the main external anchor, but the modeling of the free surface and 3D body is acknowledged as approximate. No free parameters are fitted to the hardware data, which keeps circularity low. The derivation of the integral-form IBM is the least-supported piece.

assumptions (5)
  • domain assumption The fluid is modeled as incompressible, Newtonian, with constant density and viscosity; free-surface deformation is neglected.
    Section IV-A defines Lf and F via rho and mu; the pool is initially still water with walled boundary conditions (Section V-A); Section VI acknowledges unmodeled free-surface dynamics.
  • domain assumption The robot is modeled as rigid thin plates connected by revolute joints with PD-controlled torques; the 3D hardware body is represented as a 2D body in the fluid.
    Section V-A describes six segments approximated as rigid thin plates with PD controllers; limitations mention thin-plate geometry approximations.
  • standard math The principle of least action with the stated Lagrangians and constraints yields the coupled Navier-Stokes and manipulator equations.
    Section IV-A derives (20)-(22) from (15)-(19); this is a standard variational calculus result, though boundary terms are glossed over.
  • domain assumption The integrated piecewise-linear immersed boundary method produces a convolution matrix E-bar that enforces no-slip along the whole boundary and avoids ill-conditioning from overlapping nodes.
    Section IV-C states this without proof; the conditioning claim is asserted rather than demonstrated.
  • standard math The midpoint variational integrator is stable and second-order accurate for this coupled nonlinear system.
    Section IV-B claims 2nd-order implicit leap-frog; stability is not analyzed for the coupled system.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Realizing Robotic Swimming with Unified Fluid-Robot Multiphysics." pith.science (2026). https://pith.science/paper/X4KVTMED

@misc{pith2026250605012,
  author       = {Pith},
  title        = {Pith review of: Realizing Robotic Swimming with Unified Fluid-Robot Multiphysics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/X4KVTMED}},
  note         = {Machine review of arXiv:2506.05012}
}
read the original abstract

Matching the swimming efficiency and agility of fish has remained an elusive goal in underwater robotics. Such locomotion capabilities rely on complex vortex interactions between the robot's body and the surrounding fluid. However, simulating these dynamics, which are governed by coupled ordinary and partial differential equations, is significantly more difficult than the multi-body dynamics of classical rigid robotic systems. We present a differentiable framework for simulating strongly coupled fluid-robot multiphysics as a unified optimization problem. The coupled manipulator and incompressible Navier-Stokes equations are derived together from a single Lagrangian using the principle of least action. We employ discrete variational mechanics to derive a stable, well-conditioned, and physically accurate scheme for jointly simulating articulated bodies and the surrounding fluid. We leverage the implicit function theorem to compute derivatives of the fully coupled dynamics. Using this simulator and its gradients, we realize undulating swimming gaits and optimize a highly dynamic C-start escape maneuver for a bioinspired eel robot. We validate both gaits on physical hardware, demonstrating successful sim-to-real transfer. Simulation code, hardware data, and schematics for the eel robot can be found here: https://unified-fluid-robot-multiphysics.github.io/

Figures

Figures reproduced from arXiv: 2506.05012 by the authors.

Figure 1
Figure 1. A highly dynamic fish-inspired C-start escape maneuver executed by a multilink eel-robot. The top row (black) shows [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. An overview of the multiphysics problem we address, [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 4
Figure 4. A midpoint variational integrator, where a quadrature [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figures from the paper (4 more)
Figure 6
Figure 6. Figure 6: Our experimental setup, where a bioinspired eel robot [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 8
Figure 8. Figure 8: Average root-mean-square error (RMSE) and associated standard deviations for simulated versus hardware-validated [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Time-lapse sequences showing a multilink eel-robot executing an undulating gait with a [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: Kinematic sequence of the eel robot performing a [PITH_FULL_IMAGE:figures/full_fig_p009_10.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Fire as a Service: Augmenting Robot Simulators with Thermally and Visually Accurate Fire Dynamics

    cs.RO 2026-03 conditional novelty 6.0 of 10

    A co-simulation framework adds physically based fire, smoke, and thermal hazards to Isaac Sim, Gazebo, and MuJoCo robot simulators, enabling thermally aware planning and closed-loop training.

Reference graph

Works this paper leans on

78 extracted references · 59 canonical work pages · cited by 1 Pith paper

  1. [1]

    Lagrange multiplier implementation of Dirichlet boundary conditions in compressible Navier- Stokes finite element methods

    Steven Allmaras. Lagrange multiplier implementation of Dirichlet boundary conditions in compressible Navier- Stokes finite element methods. In17th AIAA Computa- tional Fluid Dynamics Conference, page 4714, 2005

    work page 2005

  2. [2]

    J. M. Anderson, K. Streitlien, D. S. Barrett, and M. S. Triantafyllou. Oscillating foils of high propulsive effi- ciency.Journal of Fluid Mechanics, 360:41–72, 1998. doi: 10.1017/S0022112097008392

    doi:10.1017/s0022112097008392 1998

  3. [3]

    Anderson and Narender K

    Jamie M. Anderson and Narender K. Chhabra. Maneu- vering and stability performance of a robotic tuna.Inte- grative and comparative biology, 42(1):118–126, Febru- ary 2002. ISSN 1540-7063. doi: 10.1093/icb/42.1.118

    work page doi:10.1093/icb/42.1.118 2002

  4. [4]

    Ijspeert

    Emmanouil Angelidis, Jonathan Arreguit, Jan Bender, Patrick Berggold, Ziyuan Liu, Alois Knoll, Alessandro Crespi, and Auke J. Ijspeert. A smoothed particle hy- drodynamics framework for fluid simulation in robotics. Robotics and Autonomous Systems, 185:104885, 2025. ISSN 0921-8890. doi: 10.1016/j.robot.2024.104885

    arXiv 2025

  5. [5]

    Springer, 2009

    Vladimir Igorevich Arnold and Boris A Khesin.Topo- logical Methods in Hydrodynamics, volume 19. Springer, 2009

    work page 2009

  6. [6]

    Linear-time dynamics using Lagrange mul- tipliers

    David Baraff. Linear-time dynamics using Lagrange mul- tipliers. Technical Report CMU-RI-TR-95-44, Carnegie Mellon University, 1996

    work page 1996

  7. [7]

    D. N. Beal, F. S. Hover, M. S. Triantafyllou, J. C. Liao, and G. V . Lauder. Passive propulsion in vortex wakes. Journal of Fluid Mechanics, 549:385–402, 2006. doi: 10.1017/S0022112005007925

    work page doi:10.1017/s0022112005007925 2006

  8. [8]

    Cephalopod-inspired robot capable of cyclic jet propul- sion through shape change.Bioinspiration & biomimet- ics, 16, September 2020

    Caleb Christianson, Yi Cui, Michael Ishida, Xiaobo Bi, Qiang Zhu, Geno Pawlak, and Michael Tolley. Cephalopod-inspired robot capable of cyclic jet propul- sion through shape change.Bioinspiration & biomimet- ics, 16, September 2020. doi: 10.1088/1748-3190/abbc 72

    work page doi:10.1088/1748-3190/abbc 2020

Show all 78 references
  1. [9]

    COMSOL multiphysics® v

    COMSOL AB. COMSOL multiphysics® v. 6.4, 2025

    2025

  2. [10]

    De- sign of a Carangiform Swimming Robot through a Mul- tiphysics Simulation Environment.Biomimetics (Basel, Switzerland), 5(4), September 2020

    Daniele Costa, Giacomo Palmieri, Matteo-Claudio Pal- pacelli, David Scaradozzi, and Massimo Callegari. De- sign of a Carangiform Swimming Robot through a Mul- tiphysics Simulation Environment.Biomimetics (Basel, Switzerland), 5(4), September 2020. ISSN 2313-7673. doi: 10.3390/...

    2020 doi

  3. [11]

    Paolo Domenici and Melina E. Hale. Escape responses of fish: A review of the diversity in motor control, kine- matics and behaviour.Journal of Experimental Biology, 222(18):jeb166009, September 2019. ISSN 0022-0949. doi: 10.1242/jeb.166009

    2019 doi

  4. [12]

    Functional optimization of fluidic devices with differentiable stokes flow.ACM Transactions on Graphics (TOG), 39(6):1–15, 2020

    Tao Du, Kui Wu, Andrew Spielberg, Wojciech Matusik, Bo Zhu, and Eftychios Sifakis. Functional optimization of fluidic devices with differentiable stokes flow.ACM Transactions on Graphics (TOG), 39(6):1–15, 2020

    2020

  5. [13]

    Computational Methods for Fluid Dynamics

    Joel H Ferziger, Milovan Peri ´c, and Robert L Street. Computational Methods for Fluid Dynamics. springer, 2019

    2019

  6. [14]

    Fish and G.V

    F.E. Fish and G.V . Lauder. Passive and Active Flow Control by Swimming Fishes and Mammals.Annual Review of Fluid Mechanics, 38(V olume 38, 2006):193– 224, 2006. ISSN 1545-4479. doi: 10.1146/annurev.fluid .38.050304.092201

    2006

  7. [15]

    Triantafyllou

    Amy Gao and Michael S. Triantafyllou. Independent caudal fin actuation enables high energy extraction and control in two-dimensional fish-like group swimming. Journal of Fluid Mechanics, 850:304–335, 2018. doi: 10.1017/jfm.2018.456

    2018 doi

  8. [16]

    Gazzola, W

    M. Gazzola, W. M. Van Rees, and P. Koumoutsakos. C-start: Optimal start of larval fish.Journal of Fluid Mechanics, 698:5–18, 2012. doi: 10.1017/jfm.2011.558

    2012 doi

  9. [17]

    Genesis: A Universal and Generative Physics Engine for Robotics and Beyond, December 2024

    Genesis Authors. Genesis: A Universal and Generative Physics Engine for Robotics and Beyond, December 2024

    2024

  10. [18]

    J. Gray. Studies in Animal Locomotion: VI. The Propul- sive Powers of the Dolphin.Journal of Experimental Biology, 13(2):192–199, April 1936. ISSN 0022-0949. doi: 10.1242/jeb.13.2.192

    1936 doi

  11. [19]

    Motion planning, design optimization and fabrication of ferromagnetic swimmers

    Jaskaran Grover and Daniel Vedova. Motion planning, design optimization and fabrication of ferromagnetic swimmers. InRobotics Science and Systems, 2019

    2019

  12. [20]

    Geo- metric Motion Planning for a Three-Link Swimmer in a Three-Dimensional low Reynolds-Number Regime

    Jaskaran Grover, Jake Zimmer, Tony Dear, Matthew Travers, Howie Choset, and Scott David Kelly. Geo- metric Motion Planning for a Three-Link Swimmer in a Three-Dimensional low Reynolds-Number Regime. In 2018 Annual American Control Conference (ACC), pages 6067–6074, 2018. doi: ...

    2018

  13. [21]

    J. L. Guermond, P. Minev, and Jie Shen. An overview of projection methods for incompressible flows.Computer Methods in Applied Mechanics and Engineering, 195 (44):6011–6045, 2006. ISSN 0045-7825. doi: 10.101 6/j.cma.2005.10.010

    2006

  14. [22]

    Springer Berlin Heidelberg, Berlin, Heidelberg, 1996

    Ernst Hairer and Gerhard Wanner.Solving Ordinary Differential Equations II, volume 14 ofSpringer Series in Computational Mathematics. Springer Berlin Heidelberg, Berlin, Heidelberg, 1996. ISBN 978-3-642-05220-0 978- 3-642-05221-7. doi: 10.1007/978-3-642-05221-7

    1996 doi

  15. [23]

    Hazel, and Jonathan Boyle

    Matthias Heil, Andrew L. Hazel, and Jonathan Boyle. Solvers for large-displacement fluid–structure inter- action problems: Segregated versus monolithic ap- proaches.Computational Mechanics, 43(1):91–101, De- cember 2008. ISSN 1432-0924. doi: 10.1007/s00466-0 08-0270-6

    2008 doi

  16. [24]

    Dojo: A Differen- tiable Simulator for Robotics

    Taylor A Howell, Simon Le Cleac’, J Zico Kolter, Mac Schwager, and Zachary Manchester. Dojo: A Differen- tiable Simulator for Robotics. 2022

    2022

  17. [25]

    A moving least squares material point method with displacement discontinuity and two-way rigid body coupling.ACM Transactions on Graphics (TOG), 37(4):1–14, 2018

    Yuanming Hu, Yu Fang, Ziheng Ge, Ziyin Qu, Yixin Zhu, Andre Pradhana, and Chenfanfu Jiang. A moving least squares material point method with displacement discontinuity and two-way rigid body coupling.ACM Transactions on Graphics (TOG), 37(4):1–14, 2018

    2018

  18. [26]

    Gomp-fit: Grasp-optimized motion planning for fast inertial transport

    Jeffrey Ichnowski, Yahav Avigal, Yi Liu, and Ken Gold- berg. Gomp-fit: Grasp-optimized motion planning for fast inertial transport. In2022 International Conference on Robotics and Automation (ICRA), pages 5255–5261. IEEE, 2022

    2022

  19. [27]

    Jackson, Kevin Tracy, and Zachary Manchester

    Brian E. Jackson, Kevin Tracy, and Zachary Manchester. Planning With Attitude.IEEE Robotics and Automation Letters, 6(3):5658–5664, 2021. doi: 10.1109/LRA.2021 .3052431

    2021 doi

  20. [28]

    Learning to swim in potential flow.Physical Review Fluids, 6(5), May 2021

    Yusheng Jiao, Feng Ling, Sina Heydari, Nicolas Heess, Josh Merel, and Eva Kanso. Learning to swim in potential flow.Physical Review Fluids, 6(5), May 2021. doi: 10.1103/physrevfluids.6.050505

    2021 doi

  21. [29]

    Katzschmann, Joseph DelPreto, Robert Mac- Curdy, and Daniela Rus

    Robert K. Katzschmann, Joseph DelPreto, Robert Mac- Curdy, and Daniela Rus. Exploration of underwater life with an acoustically controlled soft robotic fish.Science Robotics, 3(16):eaar3449, 2018. doi: 10.1126/scirobotic s.aar3449

    2018 doi

  22. [30]

    RMA: Rapid Motor Adaptation for Legged Robots, 2021

    Ashish Kumar, Zipeng Fu, Deepak Pathak, and Jitendra Malik. RMA: Rapid Motor Adaptation for Legged Robots, 2021

    2021

  23. [31]

    Mallet-Bachelier, 1853

    Joseph Louis Lagrange.M ´ecanique Analytique, vol- ume 1. Mallet-Bachelier, 1853

  24. [32]

    Task-Oriented Active Learning of Model Preconditions for Inaccurate Dynamics Models, 2024

    Alex LaGrassa, Moonyoung Lee, and Oliver Kroemer. Task-Oriented Active Learning of Model Preconditions for Inaccurate Dynamics Models, 2024

    2024

  25. [33]

    Michelis, Robert Katzschmann, and Zachary Manchester

    Jeong Hun Lee, Mike Y . Michelis, Robert Katzschmann, and Zachary Manchester. Aquarium: A Fully Differ- entiable Fluid-Structure Interaction Solver for Robotics Applications. In2023 IEEE International Conference on Robotics and Automation (ICRA), pages 11272–11279,

  26. [34]

    An overview of variational integrators

    Adrian Lew, Jerrold E Marsden, Michael Ortiz, and Matthew West. An overview of variational integrators. Finite element methods, pages 98–115, 1970

    1970

  27. [35]

    Large-amplitude elongated- body theory of fish locomotion.Proceedings of the Royal Society of London

    Michael James Lighthill. Large-amplitude elongated- body theory of fish locomotion.Proceedings of the Royal Society of London. B. Biological Sciences, 179 (1055):125–138, November 1971. ISSN 0080-4649. doi: 10.1098/rspb.1971.0085

    1971

  28. [36]

    Learning Agile Swimming: An End-to-End Approach Without CPGs.IEEE Robotics and Automation Letters, 10(2): 1992–1999, 2025

    Xiaozhu Lin, Xiaopei Liu, and Yang Wang. Learning Agile Swimming: An End-to-End Approach Without CPGs.IEEE Robotics and Automation Letters, 10(2): 1992–1999, 2025. doi: 10.1109/LRA.2025.3527757

    1992

  29. [37]

    FishGym: A High-Performance Physics-based Simulation Framework for Underwater Robot Learning

    Wenji Liu, Kai Bai, Xuming He, Shuran Song, Changxi Zheng, and Xiaopei Liu. FishGym: A High-Performance Physics-based Simulation Framework for Underwater Robot Learning. In2022 International Conference on Robotics and Automation (ICRA), pages 6268–6275,

  30. [38]

    Warp: A High-performance Python Framework for GPU Simulation and Graphics, March 2022

    Miles Macklin. Warp: A High-performance Python Framework for GPU Simulation and Graphics, March 2022

    2022

  31. [39]

    Unified particle physics for real- time applications.ACM Trans

    Miles Macklin, Matthias M ¨uller, Nuttapong Chentanez, and Tae-Yong Kim. Unified particle physics for real- time applications.ACM Trans. Graph., 33(4), July 2014. ISSN 0730-0301. doi: 10.1145/2601097.2601152

    2014

  32. [40]

    Maertens, Amy Gao, and Michael S

    Audrey P. Maertens, Amy Gao, and Michael S. Tri- antafyllou. Optimal undulatory swimming for a single fish-like body and for a pair of interacting swimmers. Journal of Fluid Mechanics, 813:301–345, 2017. doi: 10.1017/jfm.2016.845

    2017 doi

  33. [41]

    Discrete mechanics and variational integrators.Acta Numerica, pages 357–514, 2001

    JE Marsden and M West. Discrete mechanics and variational integrators.Acta Numerica, pages 357–514, 2001

    2001

  34. [42]

    Learning robust perceptive locomotion for quadrupedal robots in the wild.Science Robotics, 7(62):eabk2822, 2022

    Takahiro Miki, Joonho Lee, Jemin Hwangbo, Lorenz Wellhausen, Vladlen Koltun, and Marco Hutter. Learning robust perceptive locomotion for quadrupedal robots in the wild.Science Robotics, 7(62):eabk2822, 2022. doi: 10.1126/scirobotics.abk2822

    2022 doi

  35. [43]

    Don’t be jelly: Exploring effective jellyfish locomotion.arXiv preprint arXiv:1904.09340, 2019

    Jason G Miles and Nicholas A Battista. Don’t be jelly: Exploring effective jellyfish locomotion.arXiv preprint arXiv:1904.09340, 2019

    1904 arXiv

  36. [44]

    Origin and evolution of immersed boundary methods in computational fluid dynamics.Phys

    Rajat Mittal and Jung Hee Seo. Origin and evolution of immersed boundary methods in computational fluid dynamics.Phys. Rev. Fluids, 8(10):100501, October

  37. [45]

    V olume conservation issues in incompressible smoothed particle hydrodynam- ics.Journal of Computational Physics, 297:689–699,

    Prapanch Nair and Gaurav Tomar. V olume conservation issues in incompressible smoothed particle hydrodynam- ics.Journal of Computational Physics, 297:689–699,

  38. [46]

    Fast Aquatic Swimmer Optimization with Differentiable Projective Dynamics and Neural Network Hydrodynamic Models

    Elvis Nava, John Z Zhang, Mike Yan Michelis, Tao Du, Pingchuan Ma, Benjamin F Grewe, Wojciech Matusik, and Robert Kevin Katzschmann. Fast Aquatic Swimmer Optimization with Differentiable Projective Dynamics and Neural Network Hydrodynamic Models. InInterna- tional Conference o...

    2022

  39. [47]

    doi: 10.1103/PhysRevFluids.8.100501

    doi

  40. [48]

    Synchronisation through learning for two self-propelled swimmers.Bioinspiration & biomimetics, 12(3):036001, 2017

    Guido Novati, Siddhartha Verma, Dmitry Alexeev, Diego Rossinelli, Wim M Van Rees, and Petros Koumoutsakos. Synchronisation through learning for two self-propelled swimmers.Bioinspiration & biomimetics, 12(3):036001, 2017

    2017

  41. [49]

    Neural-Fly enables rapid learning for agile flight in strong winds.Science Robotics, 7(66): eabm6597, 2022

    Michael O’Connell, Guanya Shi, Xichen Shi, Kamyar Azizzadenesheli, Anima Anandkumar, Yisong Yue, and Soon-Jo Chung. Neural-Fly enables rapid learning for agile flight in strong winds.Science Robotics, 7(66): eabm6597, 2022. doi: 10.1126/scirobotics.abm6597

    2022 doi

  42. [50]

    AprilTag: A robust and flexible visual fiducial system

    Edwin Olson. AprilTag: A robust and flexible visual fiducial system. In2011 IEEE International Conference on Robotics and Automation, pages 3400–3407, 2011. doi: 10.1109/ICRA.2011.5979561

    2011

  43. [51]

    Wright.Numerical Opti- mization

    Jorge Nocedal and Stephen J. Wright.Numerical Opti- mization. Springer, second edition, 2006

    2006

  44. [52]

    T. C. Papanastasiou, N. Malamataris, and K. Ellwood. A new outflow boundary condition.International Journal for Numerical Methods in Fluids, 14(5):587–608, 1992. doi: 10.1002/fld.1650140506

    1992 doi

  45. [53]

    The immersed boundary method.Acta numerica, 11:479–517, 2002

    Charles S Peskin. The immersed boundary method.Acta numerica, 11:479–517, 2002

    2002

  46. [54]

    Differentiable simulation of soft multi-body sys- tems.Advances in Neural Information Processing Sys- tems, 34:17123–17135, 2021

    Yiling Qiao, Junbang Liang, Vladlen Koltun, and Ming Lin. Differentiable simulation of soft multi-body sys- tems.Advances in Neural Information Processing Sys- tems, 34:17123–17135, 2021

    2021

  47. [55]

    Stanford university unstructured (SU2): Analysis and design technology for turbulent flows

    Francisco Palacios, Thomas D Economon, Aniket Aranake, Sean R Copeland, Amrita K Lonkar, Trent W Lukaczyk, David E Manosalvas, Kedar R Naik, Santiago Padron, Brendan Tracey, et al. Stanford university unstructured (SU2): Analysis and design technology for turbulent flows. In52...

    2014

  48. [56]

    Vari- ational principles in continuum mechanics.Proceedings of the Royal Society of London

    Robert L Seliger and Gerald Beresford Whitham. Vari- ational principles in continuum mechanics.Proceedings of the Royal Society of London. Series A. Mathematical and Physical Sciences, 305(1480):1–25, 1968

    1968

  49. [57]

    Robocook: Long-horizon elasto-plastic object manipulation with diverse tools.arXiv preprint arXiv:2306.14447, 2023

    Haochen Shi, Huazhe Xu, Samuel Clarke, Yunzhu Li, and Jiajun Wu. Robocook: Long-horizon elasto-plastic object manipulation with diverse tools.arXiv preprint arXiv:2306.14447, 2023

    2023 arXiv

  50. [58]

    Pratik Suchde, Christian Leith ¨auser, J ¨org Kuhnert, and St´ephane P. A. Bordas. V olume and Mass Conservation in Lagrangian Meshfree Methods.International Journal for Numerical Methods in Engineering, 126(2):e7657, 2025. doi: 10.1002/nme.7657

    2025 doi

  51. [59]

    The archerfish predictive C-start.Jour- nal of comparative physiology

    Stefan Schuster. The archerfish predictive C-start.Jour- nal of comparative physiology. A, Neuroethology, sen- sory, neural, and behavioral physiology, 209(5):827–837, September 2023. ISSN 1432-1351 0340-7594. doi: 10.1007/s00359-023-01658-2

    2023 doi

  52. [60]

    Mu- JoCo: A physics engine for model-based control

    Emanuel Todorov, Tom Erez, and Yuval Tassa. Mu- JoCo: A physics engine for model-based control. In 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems, pages 5026–5033, 2012. doi: 10.1109/IROS.2012.6386109

    2012

  53. [61]

    M. S. Triantafyllou, G. S. Triantafyllou, and D. K. P. Yue. Hydrodynamics of Fishlike Swimming.Annual Review of Fluid Mechanics, 32(1):33–53, 2000. doi: 10.1146/an nurev.fluid.32.1.33

    doi

  54. [62]

    An Efficient Swimming Machine.Scientific American - SCI AMER, 272:64–70, March 1995

    Michael Triantafyllou and George Triantafyllou. An Efficient Swimming Machine.Scientific American - SCI AMER, 272:64–70, March 1995. doi: 10.1038/scientific american0395-64

    1995 doi

  55. [63]

    The immersed boundary method: A projection approach.Journal of Computational Physics, 225(2):2118–2137, 2007

    Kunihiko Taira and Tim Colonius. The immersed boundary method: A projection approach.Journal of Computational Physics, 225(2):2118–2137, 2007. ISSN 0021-9991. doi: 10.1016/j.jcp.2007.03.005

    2007 doi

  56. [64]

    Immersed Boundary Methods: Histor- ical Perspective and Future Outlook.Annual Review of Fluid Mechanics, 55(V olume 55, 2023):129–155, 2023

    Roberto Verzicco. Immersed Boundary Methods: Histor- ical Perspective and Future Outlook.Annual Review of Fluid Mechanics, 55(V olume 55, 2023):129–155, 2023. ISSN 1545-4479. doi: 10.1146/annurev-fluid-120720-0 22129

    2023 doi

  57. [65]

    P. W. Webb. Fast-Start Performance and Body form In Seven Species of Teleost Fish.Journal of Experimental Biology, 74(1):211–226, June 1978. ISSN 0022-0949. doi: 10.1242/jeb.74.1.211

    1978 doi

  58. [66]

    Tunabot Flex: A tuna-inspired robot with body flexibility improves high-performance swimming.Bioinspiration & Biomimetics, 16(2):026019, March 2021

    Carl H White, George V Lauder, and Hilary Bart-Smith. Tunabot Flex: A tuna-inspired robot with body flexibility improves high-performance swimming.Bioinspiration & Biomimetics, 16(2):026019, March 2021. doi: 10.1088/ 1748-3190/abb86d

    2021

  59. [67]

    Hydrodynamic pressure sensing with an artificial lateral line in steady and unsteady flows.Bioinspi- ration & Biomimetics, 7(3):036004, April 2012

    Roberto Venturelli, Otar Akanyeti, Francesco Visentin, Jaas Je ˇzov, Lily D Chambers, Gert Toming, Jennifer Brown, Maarja Kruusmaa, William M Megill, and Paolo Fiorini. Hydrodynamic pressure sensing with an artificial lateral line in steady and unsteady flows.Bioinspi- ration ...

    2012 doi

  60. [68]

    Stabilizing Reinforcement Learning in Differentiable Multiphysics Simulation, 2024

    Eliot Xing, Vernon Luk, and Jean Oh. Stabilizing Reinforcement Learning in Differentiable Multiphysics Simulation, 2024

    2024

  61. [69]

    DextAIRity: Deformable Manipulation Can be a Breeze, 2022

    Zhenjia Xu, Cheng Chi, Benjamin Burchfiel, Eric Cousineau, Siyuan Feng, and Shuran Song. DextAIRity: Deformable Manipulation Can be a Breeze, 2022

    2022

  62. [70]

    Artificial lateral line with biomimetic neuromasts to emulate fish sens- ing.Bioinspiration & Biomimetics, 5(1):016001, January

    Yingchen Yang, Nam Nguyen, Nannan Chen, Michael Lockwood, Craig Tucker, Huan Hu, Horst Bleckmann, Chang Liu, and Douglas L Jones. Artificial lateral line with biomimetic neuromasts to emulate fish sens- ing.Bioinspiration & Biomimetics, 5(1):016001, January

  63. [71]

    FluidLab: A Differentiable Environment for Benchmark- ing Complex Fluid Manipulation, 2023

    Zhou Xian, Bo Zhu, Zhenjia Xu, Hsiao-Yu Tung, An- tonio Torralba, Katerina Fragkiadaki, and Chuang Gan. FluidLab: A Differentiable Environment for Benchmark- ing Complex Fluid Manipulation, 2023

    2023

  64. [72]

    RoPotter: Toward Robotic Pottery and Deformable Object Manipulation with Structural Pri- ors

    Uksang Yoo, Adam Hung, Jonathan Francis, Jean Oh, and Jeffrey Ichnowski. RoPotter: Toward Robotic Pottery and Deformable Object Manipulation with Structural Pri- ors. In2024 IEEE-RAS 23rd International Conference on Humanoid Robots (Humanoids), pages 843–850. IEEE, 2024

    2024

  65. [73]

    Zhang, Taylor A

    John Z. Zhang, Taylor A. Howell, Zeji Yi, Chaoyi Pan, Guanya Shi, Guannan Qu, Tom Erez, Yuval Tassa, and Zachary Manchester. Whole-Body Model-Predictive Control of Legged Robots with MuJoCo, 2025

    2025

  66. [76]

    Saadatzi, Ruoshi Zhang, Andriy Sherehiy, Danming Wei, Cindy K

    Zhong Yang, Mohammad N. Saadatzi, Ruoshi Zhang, Andriy Sherehiy, Danming Wei, Cindy K. Harnett, and Dan O. Popal. Multiphysics Dynamic Model Validation Methodology for Laser-Driven Microrobots. In2019 IEEE 15th International Conference on Automation Sci- ence and Engineering (...

    2019

  67. [2010]

    doi: 10.1088/1748-3182/5/1/016001

    doi

  68. [2015]

    doi: 10.1016/j.jcp.2015.05.042

    ISSN 0021-9991. doi: 10.1016/j.jcp.2015.05.042

    2015 doi

  69. [2022]

    doi: 10.1109/ICRA46639.2022.9812066

    2022

  70. [2023]

    doi: 10.1109/ICRA48891.2023.10161494

    2023

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.