Pith. sign in

REVIEW 3 major objections 5 minor 60 references

Physics-Grounded Differentiable Simulation for Soft Growing Robots

T0 review · 3 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read New stiffness model makes vine robot simulation match reality

desk verdict A useful batched differentiable simulator for vine robots with a new wrinkling-based stiffness model, but the central validation is curve fitting rather than physics prediction. read the letter →

arxiv 2501.17963 v1 pith:S2RDMX7C submitted 2025-01-29 cs.RO

Lucas Chen , Yitian Gao , Sicheng Wang , Francesco Fuentes , Laura H. Blumenschein , Zachary Kingston This is my paper · ORCID
classification cs.RO
keywords softgrowingrobotsvinedifferentiablesimulationinflatedbeamwrinklingnonlinearstiffnessmodelsim-to-realtransfergradient-basedparameterfittingclosed-formbending
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

This paper tries to establish that a closed-form, first-principles stiffness model for thin-walled inflated tubes, built from the physics of surface wrinkling, can be embedded in a differentiable simulator for soft growing (vine) robots and that the resulting simulator predicts real robot shapes more accurately than the linear or learned stiffness models used in prior work. Vine robots are long pressurized tubes that extend by eversion, and their bending is governed by localized loss of tension on the compressed side of the tube, not by constant or linearly elastic stiffness. The authors derive a bending-angle-dependent moment law from a wrinkling criterion, fit its single parameter to one tube's measured moment data, and then use the simulator's gradients to fit the remaining dynamics parameters to real video demonstrations. On an environment not used for fitting, the proposed model produces lower and more consistent mean-squared error than the alternatives. If correct, this gives planners and control designers a fast, gradient-bearing simulator whose shapes match reality.

What carries the argument

The load-bearing object is the kinematic bridge between wrinkling angle and bending angle, Eq. (2), together with the experimentally fitted wrinkling criterion $\epsilon_{\mathrm{crit}}$. The bridge treats the onset of wrinkling as the point where the overlap of two rigid circular cross-sections implies a length change $\Delta l$ exceeding a threshold $\Delta l_{\mathrm{crit}}$ at some height on the tube surface; geometry then converts that critical height into $\gamma_0$. This converts the classical moment formula $M = \pi P R^3 [\sin 2\gamma_0 + 2\pi - 2\gamma_0] / [4(\sin\gamma_0 + (\pi-\gamma_0)\cos\gamma_0)]$ into a function of bending angle only, with $\epsilon_{\mathrm{crit}}$ fit to experimental moment data and interpolated by a polynomial for use in simulation. The differentiable simulator wraps this stiffness law in a maximal-coordinate rigid-body formulation where feasible velocities are obtained from a differentiable quadratic program, so gradients flow from a shape-matching loss back to physical parameters.

What would settle it

Measure the moment-angle curve for the same tube material at two different radii and pressures, fit $\epsilon_{\mathrm{crit}}$ to one curve, and check whether the same value (or the same non-dimensional scaling) predicts the other curve; if the predicted wrinkling-onset angle $2\sin^{-1}(\epsilon_{\mathrm{crit}})$ and the saturation level do not track the measured curves, the kinematic bridge is not capturing the physics. A direct simulator-level test would compare predicted shapes to skeletonized video in an environment with a single hard obstacle across several approach angles and pressures, looking specifically at the curvature just behind the contact point.

Watch

Extended reading notes

Core claim

The central claim is that vine robot bending can be captured by a nonlinear moment law obtained by inserting a geometric wrinkling-onset condition into the classical wrinkled-beam moment formula. The formula relates the restoring moment $M$ to the wrinkling angle $\gamma_0$, and the paper's bridge equation $\gamma_0 = \cos^{-1}(2\epsilon_{\mathrm{crit}}/\sin(\theta/2)-1)$ connects $\gamma_0$ to the joint bending angle $\theta$ through a single dimensionless wrinkling criterion $\epsilon_{\mathrm{crit}} = \Delta l_{\mathrm{crit}}/2R$. Substituting the bridge into the moment formula yields a closed-form $M(\theta)$ that rises steeply at small angles and saturates near the fully wrinkled moment, matching the measured behavior of inflated LDPE tubes. Integrated into a differentiable rigid-body simulator with variable-length growth and contact complementarity, this stiffness model reproduces non-constant curvature bending when a growing vine contacts obstacles, and parameter fitting by gradient descent on real video data produces lower and more consistent mean-squared error on a held-out environment than both a linear stiffness model and a learned multilayer-perceptron stiffness model.

Load-bearing premise

The whole model leans on the assumption that wrinkling begins and grows according to a single scalar length-change threshold $\epsilon_{\mathrm{crit}}$ fitted to the moment data of one tube, with the shell treated as inextensible and free of shear, so an error in that fitted threshold propagates directly into every simulated bend.

Editorial extensions

If this is right

  • The differentiable simulator can be placed inside gradient-based optimization loops for planning, control, and parameter identification, since gradients flow back through both the quadratic program and the nonlinear stiffness law.
  • Batched parallel rollouts make high-throughput simulation practical, so many launch angles or environment variations can be evaluated simultaneously.
  • The closed-form stiffness model reduces sim-to-real shape error compared with constant-moment and linear-stiffness baselines on unseen environments.
  • The single wrinkling criterion $\epsilon_{\mathrm{crit}}$ can be re-fit by gradient descent on real robot video, avoiding hand-tuned stiffness parameters.
  • The same framework can be extended to other vine robot actuation modes, such as pre-formed welds, pinches, and tensioning cables.

Reading between the lines

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

  • If $\epsilon_{\mathrm{crit}}$ could be predicted from material properties, tube geometry, and pressure rather than fitted per tube, the model would generalize across robot designs without new bending experiments; the paper leaves this derivation open.
  • The wrinkling-bridge idea could transfer to other inflated structures, such as inflatable booms or soft actuators, wherever bending is dominated by tension loss rather than material elasticity.
  • The differentiable coupling between observed motion and stiffness parameters suggests an inverse-problem route: infer operating pressure or material state from video of a deployed vine robot.
  • Replacing the differentiable quadratic-program solver with a learned surrogate or a GPU-accelerated solver, as the paper notes is future work, could remove the current CPU bottleneck and make the simulator useful for reinforcement learning.
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

3 major / 5 minor

Summary. The paper proposes a differentiable simulator for vine robots, building on the prior Jitosho et al. impulse-velocity formulation and extending it with variable-length links, contact complementarity, and a differentiable quadratic programming layer. Its main contribution is a closed-form nonlinear stiffness model that connects the Comer-Levy wrinkling moment (Eq. 1) to a bending angle via a wrinkling criterion and a geometric bridge (Eq. 2). The authors experimentally fit the wrinkling criterion to cantilever moment data for an inflated LDPE tube, fit a polynomial for pressure dependence, and integrate the model into the simulator. They then optimize simulator parameters (mass, inertia, growth rate, damping, and stiffness) against real-robot video data and compare the proposed stiffness model against linear and MLP baselines on a held-out environment.

Significance. If the stiffness model is genuinely predictive, the work would provide a valuable physics-based component for sim-to-real transfer, planning, and control of vine robots, and the open-source differentiable simulator with batched rollouts is a useful engineering contribution. The paper also demonstrates a practical pipeline for fitting simulator parameters through differentiable optimization. However, the validation as presented does not yet establish the main claim that the proposed model 'captures non-constant curvature bending' better than baselines in a physics-predictive sense, because the stiffness parameter is refit on the target task and the moment-curve comparison is in-sample.

major comments (3)
  1. [§III-A, Eq. (2)] The kinematic bridge γ0 = cos⁻¹(2 ε_crit / sin(θ/2) − 1) is derived from the overlap of two rigid circular cross-sections and contains no length scale such as the segment length dsegment. In the calibration setup (Fig. 3c), θ is a global cantilever deflection angle under a point load, where local curvature varies along the beam; in the simulator (Sec. III-B), θ is the relative pin-joint angle between adjacent virtual links separated by dsegment. These are different quantities, so the same physical curvature produces different θ at different discretizations. The authors should either express Eq. (2) in terms of local curvature and dsegment, or demonstrate that the fitted ε_crit is invariant to dsegment. Without this, the improved MSE in Fig. 6 could be explained by a one-parameter curve fit rather than by first-principles physics.
  2. [§III-A, Fig. 3(a)] The 'model prediction' in Fig. 3(a) is obtained by least-squares fitting ε_crit to the same measured moment-angle data (as stated in the text: 'ɛcrit is found via a least-square fitting to the data'). The agreement shown is therefore an in-sample fit, not an independent prediction. In addition, no error bars or repeated-trial statistics are reported for the moment measurements, so the reader cannot assess the scatter or the statistical significance of the mismatch at small angles. The authors should report a hold-out validation (e.g., fit on a subset of pressures or bending angles and predict the rest) and include measurement uncertainty.
  3. [§IV-C, Figs. 6 and 7] The trajectory comparison is not a test of the independently measured stiffness model. The text explicitly states that the stiffness parameter is derived again on a different vine robot using fitting: 'we derive this parameter again on a different vine robot using fitting to demonstrate the capabilities of our model.' Thus, for the proposed model, ε_crit is optimized on the same robot-video training data that is used for all other parameters, and the held-out environment is only unseen after this optimization. This makes Fig. 6 a comparison of curve-fitting capacity on the target task, not a validation that the physics-based model predicts without task-specific tuning. The authors should either fix ε_crit from the Sec. III-A cantilever measurement in the trajectory evaluation, or compare all models with matched parameter counts and report the fitted ε_crit value. The dataset of 10 demonstrations in 4 environments is also too small to support the strong claim about consistency in Fig. 7 without confidence intervals or per-environment breakdowns.
minor comments (5)
  1. [§III-A, after Eq. (2)] The notation for the wrinkling criterion is inconsistent: the text uses 'ε_crit', 'ϵcrit', and Fig. 3(b) uses 'ε'. Please standardize the symbol throughout.
  2. [§III-A, Fig. 3(b)] The third-degree polynomial fit for ε_crit(P) is described as 'heuristic' but the polynomial coefficients or an equation are not given in the text; since the code is open source this may be recoverable, but stating the polynomial would improve reproducibility and make the model self-contained.
  3. [§III-B, Eq. (3)] The notation K(q, ·) is unclear; it would help to define explicitly that K is a stiffness function returning the joint torque as a function of the configuration, and to state the domain of C and the meaning of the damping matrix.
  4. [§IV-B] The video-processing pipeline is described briefly; please provide the key parameters (e.g., blur kernel size, morphological closure size, skeletonization method) or a reference to a repository file so that the data extraction is reproducible.
  5. [§IV-C, optimization] The loss function minimizes positional squared error, but it is not stated whether the rotational degrees of freedom are included in the error or whether the ground-truth link spacing matches the simulator's fixed dsegment. This should be clarified.

Circularity Check

1 steps flagged · score 2.0 of 10

Only minor circularity: Fig. 3(a) calls an in-sample least-squares fit a 'prediction'; the central simulator comparison is tested on an unseen environment.

  1. fitted input called prediction [Section III-A, Fig. 3(a), least-squares fit paragraph]
    "At each input pressure, ϵcrit is found via a least-square fitting to the data, arg minϵcrit(Mmeasured(θ) − Mpredict(θ, ϵcrit))2. Fig. 3(a) shows the experimental results and the best-fit model. The wrinkling criterion-based prediction captures the evolution of the bending moment well..."

    The curve shown as a 'prediction' in Fig. 3(a) is generated with εcrit chosen by least-squares minimization against the very same Mmeasured(θ) data plotted alongside it, so the agreement is an in-sample fit rather than an independent forecast. Calling this a prediction equates the fitted curve with a validation of the model. This is a local issue: the main simulator result in Fig. 6 refits parameters on training trajectories and evaluates on an environment not used for fitting, so that central comparison does not reduce by construction.

full rationale

Aside from the Fig. 3(a) wording, the derivation is self-contained. Eq. (1) comes from the external Comer-Levy model; Eq. (2) is the paper's own geometric bridge; and εcrit is explicitly fitted, not claimed to be derived from material properties. The simulator builds on Jitosho et al. [25], which is not authored by the present authors, and the shape-prediction comparison is performed on a testing environment not used for parameter fitting against linear and MLP baselines. There is no load-bearing self-citation chain or imported uniqueness theorem. The only circular element is the in-sample 'prediction' in Fig. 3(a), which is not the central claim, so the appropriate score is low.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on a new bending model whose key material parameter epsilon_crit is measured and fitted, on a geometric wrinkling criterion that is assumed rather than derived from continuum mechanics, and on a dynamic backbone taken from prior work. No new physical entities are introduced.

free parameters (3)
  • epsilon_crit = fitted per pressure by least squares; shown in Fig. 3(b) as a pressure-dependent curve
    The dimensionless wrinkling criterion is the key parameter in the new bending model. It is fit to the experimental moment data for one LDPE tube and then used in simulation, so the model's moment prediction is not parameter-free.
  • third-degree polynomial coefficients for epsilon_crit(P) = coefficients not given numerically; fitted to the per-pressure epsilon_crit values in Fig. 3(b)
    A cubic polynomial is fit to the measured epsilon_crit values to interpolate the parameter at arbitrary pressures in the simulator. These coefficients are empirical.
  • dynamic fitting parameters (mass m, inertia I, growth rate u, damping C, stiffness parameters) = optimized via AdamW on real robot video data; exact values not reported
    The simulator's dynamics depend on these robot parameters, and they are fitted to the collected trajectories. This is standard system identification, but it means the sim-to-real validation includes several fitted quantities.
assumptions (4)
  • standard math Comer-Levy moment relation for an inflated beam with a wrinkled cross-section (Eq. 1) is taken as given.
    The paper builds directly on [26] without re-deriving the fundamental wrinkled-beam moment formula.
  • domain assumption The geometry in Fig. 2(a): local length change at height y is determined by the overlap of two rigid circular cross-sections, and wrinkling begins when this length change reaches Delta_l_crit = 2R epsilon_crit.
    This is the core physical assumption behind Eq. (2). It reduces a complex material instability to a single geometric length-change threshold and ignores the stress state and material constitutive behavior before wrinkling.
  • ad hoc to paper The vine robot behaves linear-elastically for bending angles below the wrinkling threshold theta < 2 sin^-1(epsilon_crit).
    The authors state this assumption explicitly in Sec. III-A for model completeness, and note that it under- and over-predicts the bending moment at small angles. It is not derived from the material model.
  • domain assumption The vine robot can be modeled as a chain of point-mass rigid bodies with fixed link length, connected by revolute constraints and contact spheres, following Jitosho et al. [25].
    The entire simulator formulation, including the maximal-coordinate representation, Lagrange multipliers, and the QP for feasible velocities, is inherited from [25] and is not independently justified here.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Physics-Grounded Differentiable Simulation for Soft Growing Robots." pith.science (2026). https://pith.science/paper/S2RDMX7C

@misc{pith2026250117963,
  author       = {Pith},
  title        = {Pith review of: Physics-Grounded Differentiable Simulation for Soft Growing Robots},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/S2RDMX7C}},
  note         = {Machine review of arXiv:2501.17963}
}
read the original abstract

Soft-growing robots (i.e., vine robots) are a promising class of soft robots that allow for navigation and growth in tightly confined environments. However, these robots remain challenging to model and control due to the complex interplay of the inflated structure and inextensible materials, which leads to obstacles for autonomous operation and design optimization. Although there exist simulators for these systems that have achieved qualitative and quantitative success in matching high-level behavior, they still often fail to capture realistic vine robot shapes using simplified parameter models and have difficulties in high-throughput simulation necessary for planning and parameter optimization. We propose a differentiable simulator for these systems, enabling the use of the simulator "in-the-loop" of gradient-based optimization approaches to address the issues listed above. With the more complex parameter fitting made possible by this approach, we experimentally validate and integrate a closed-form nonlinear stiffness model for thin-walled inflated tubes based on a first-principles approach to local material wrinkling. Our simulator also takes advantage of data-parallel operations by leveraging existing differentiable computation frameworks, allowing multiple simultaneous rollouts. We demonstrate the feasibility of using a physics-grounded nonlinear stiffness model within our simulator, and how it can be an effective tool in sim-to-real transfer. We provide our implementation open source.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 51 canonical work pages

  1. [1]

    Soft robotic grippers

    J. Shintake, V . Cacucciolo, D. Floreano, and H. Shea, “Soft robotic grippers”, Advanced materials , vol. 30, no. 29, p. 1 707 035, 2018

    work page 2018

  2. [2]

    Fundamentals o f soft robot locomotion

    M. Calisti, G. Picardi, and C . L aschi, “Fundamentals o f soft robot locomotion”, Journal of The Royal Society Interface, vol. 14, no. 130, p. 20 170 101, 2017

    work page 2017

  3. [3]

    A soft robot that navigates its environment through growth

    E. W. Hawkes, L. H. Blumenschein, J. D. Greer, and A. M. Okamura, “A soft robot that navigates its environment through growth”, Science Robotics, vol. 2, no. 8, eaan3028, 2017

    work page 2017

  4. [4]

    Vine robots

    M. M. Coad, L. H. Blumenschein, S. Cutler, J. A. R. Zepeda, N. D. Naclerio, H. El-Hussieny, U. Mehmood, J.-H. Ryu, E. W. Hawkes, and A. M. Okamura, “Vine robots”, IEEE Robo tics & Automation Magazine, vol. 27, no. 3, pp. 120–132, 2019

    work page 2019

  5. [5]

    Model-based con trol of soft r obots: A survey of t he s tate o f t he a rt and open cha llenges

    C. Della San tina, C. Duriez, and D. Rus, “Model-based con trol of soft r obots: A survey of t he s tate o f t he a rt and open cha llenges”, IEEE Control Systems Magazine , vol. 43, no. 3, pp. 30–65, 2023

    work page 2023

  6. [6]

    A review of physics s imulators for r obotic app lications

    J. Collins, S. Chand, A. Vanderkop, and D. Howard, “A review of physics s imulators for r obotic app lications”, IEEE Access, v ol. 9, pp. 51 416–51 431, 2021

    work page 2021

  7. [7]

    Isaac gym: High pe rformance gpu -based ph ysics s imulation for r obot learning

    V .Makoviychuk, L. Wawrzyniak, Y . Guo, M. L u, K. Storey, M. Macklin, D. Hoeller, N. Rudin, A. Allshire, A. Handa, et al., “Isaac gym: High pe rformance gpu -based ph ysics s imulation for r obot learning”, arXiv preprint arXiv:2108.10470 , 2021

    arXiv 2021

  8. [8]

    Contactnets: Learning discon- tinuous contact dynamics with smooth, implicit representations

    S. Pfrommer, M. Halm, and M. Posa, “Contactnets: Learning discon- tinuous contact dynamics with smooth, implicit representations”, in Conference on Robot Learning , PMLR, 2021, pp. 2279–2291

    work page 2021

Show all 60 references
  1. [9]

    Differentiable collision detection for a se t of convex primitives

    K. Tracy, T. A. Howell, and Z. Manchester, “Differentiable collision detection for a se t of convex primitives”, i n 2023 IEEE Interna- tional Conference on Robo tics and Automation (ICRA), IEEE, 2023, pp. 3663–3670

    2023

  2. [10]

    Underwater soft robot modeling and con trol with d ifferentiable s imulation

    T. Du, J. Hughes, S. Wah, W. Matusik, and D. Rus, “Underwater soft robot modeling and con trol with d ifferentiable s imulation”, IEEE Robotics and Automation Letters, vol. 6, no. 3, pp. 4994–5001, 2021

    2021

  3. [11]

    Difftaichi: Differentiable programming for physical simulation

    Y .Hu, L. Anderson, T.-M. Li, Q. Sun, N. Carr, J. Ragan-Kelley, and F. Durand, “Difftaichi: Differentiable programming for physical simulation”, arXiv preprint arXiv:1910.00935 , 2019

    1910 arXiv

  4. [12]

    Chainqueen: A real-time d ifferentiable physical simulator for soft robotics

    Y .Hu, J. Liu, A. Spielberg, J. B. Tenenbaum, W. T. Freeman, J. Wu, D. Rus, and W. Matusik, “ Chainqueen: A real-time d ifferentiable physical simulator for soft robotics”, in 2019 International conference on robotics and automation (ICRA) , IEEE, 2019, pp. 6265–6271

    2019

  5. [13]

    Differentiable simulation for physical system identification

    Q. Le Lidec, I. Kalevatykh, I. L aptev, C. Schmid, and J. Carpentier, “Differentiable simulation for physical system identification”, IEEE Robotics and Automation Letters, vol. 6, no. 2, pp. 3413–3420, 2021

    2021

  6. [14]

    Simulation architectures for reinforcement learning ap- plied to robotics

    D. Ferigo, “Simulation architectures for reinforcement learning ap- plied to robotics”, PhD Thesis, University of Manchester, Jul. 2022

    2022

  7. [15]

    Gradsim: Differentiable simulation for system identification and visuomotor control

    K. M. Jatavallabhula, M. Macklin, F. Golemo, V . Voleti, L. Petrini, M. Weiss, B. Considine, J. Parent-L ´evesque, K. Xie, K. Erleben, et al., “Gradsim: Differentiable simulation for system identification and visuomotor control”, arXiv preprint arXiv:2104.02646 , 2021

    2021 arXiv

  8. [16]

    Accelerated po licy l earning w ith pa rallel differentiable simulation

    J. X u, V . Makoviychuk, Y . Narang, F. Ramos, W. Matusik, A. Garg, and M . Macklin, “ Accelerated po licy l earning w ith pa rallel differentiable simulation”, arXiv preprint arXiv:2204.07137 , 2022

    2022 arXiv

  9. [17]

    Neuralsim: Augmenting d ifferentiable s imulators w ith neu ral net- works

    E. Heiden, D. Millard, E. Coumans, Y .Sheng, and G. S. Sukhatme, “Neuralsim: Augmenting d ifferentiable s imulators w ith neu ral net- works”, i n 2021 I EEE International Conference on Robo tics and Automation (ICRA) , IEEE, 2021, pp. 9474–9481

    2021

  10. [18]

    Finite e lement modeling of pneumatic bending actuators for inflated-beam robots

    C. du Pasqu ier, S. J eong, and A. M. Okamura, “ Finite e lement modeling of pneumatic bending actuators for inflated-beam robots”, IEEE Robotics and Automation Letters , 2023

    2023

  11. [19]

    Towards a physics- based model for steerable eversion growing robots

    Z. Wu, M. D. I. Reyzabal, S. H. Sadati, H. Liu, S. Ourselin, D. Leff, R. K. Katzschmann, K. Rhode, and C. Bergeles, “Towards a physics- based model for steerable eversion growing robots”, IEEE robotics and automation letters , vol. 8, no. 2, pp. 1005–1012, 2023

    2023

  12. [20]

    A bioinspired soft r obot combining the g rowth adap tability of vi ne p lants w ith a coordinated control system

    P. Li, Y . Zhang, G. Zhang, D. Zhou, and L. Li, “ A bioinspired soft r obot combining the g rowth adap tability of vi ne p lants w ith a coordinated control system”, Research, 2021

    2021

  13. [21]

    Geometric so lutions for gen- eral actuator r outing on inflated-beam so ft growing robots

    L. H. Blumenschein, M. Koehler, N. S. Usevitch, E. W. Hawkes, D. C. Rucker, and A. M. Okamura, “Geometric so lutions for gen- eral actuator r outing on inflated-beam so ft growing robots”, IEEE Transactions on Robotics , vol. 38, no. 3, pp. 1820–1840, 2021

    2021

  14. [22]

    An obstacle-interaction planning method for navigation of actuated vine robots

    M. Selvaggio, L. Ramirez, N. D. Naclerio, B. Siciliano, and E. W. Hawkes, “An obstacle-interaction planning method for navigation of actuated vine robots”, i n 2020 I EEE International Conference on Robotics and Automation (ICRA) , IEEE, 2020, pp. 3227–3233

    2020

  15. [23]

    Robust navigation o f a so ft growing robot by exploiting contact with the environment

    J. D. Greer, L. H. Blumenschein, R. Alterovitz, E. W. Hawkes, and A. M. Okamura, “ Robust navigation o f a so ft growing robot by exploiting contact with the environment”, The International J ournal of Robotics Research , vol. 39, no. 14, pp. 1724–1738, 2020

    2020

  16. [24]

    Mapping unknown en viron- ments through passive deformation of soft, growing robots

    F. Fuentes and L. H. Blumenschein, “ Mapping unknown en viron- ments through passive deformation of soft, growing robots”, in 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), IEEE, 2023, pp. 2522–2527

    2023

  17. [25]

    A dy- namics s imulator f or soft growing robots

    R. Jit osho, N. Agharese, A. Okamura, and Z . Manchester, “A dy- namics s imulator f or soft growing robots”, i n 2021 I EEE Interna- tional Conference on Robo tics and Automation (ICRA), IEEE, 2021, pp. 11 775–11 781

    2021

  18. [26]

    Deflections of an inflated circular-cylindrical cantilever beam

    R. Comer and S. Levy, “Deflections of an inflated circular-cylindrical cantilever beam”, AIAA journal, vol. 1, no. 7, pp. 1652–1655, 1963

    1963

  19. [27]

    Anisotropic s tiffness and p rogrammable ac tuation for soft robots enab led b y an inflated rotational j oint

    S. Wang, E. Frias-Miranda, A. A. V aldivia, and L. H. Blumen- schein, “Anisotropic s tiffness and p rogrammable ac tuation for soft robots enab led b y an inflated rotational j oint”, arXiv preprint arXiv:2410.13003, 2024

    2024 arXiv

  20. [28]

    Differentiable con vex optimization layers

    A. Agrawal, B. Amos, S. Barratt, S. Boyd, S. Diamond, and J. Z. Kolter, “ Differentiable con vex optimization layers”, Advances in neural information processing systems , vol. 32, 2019

    2019

  21. [29]

    Operator splitting for a homogeneous embedding of the linear complementarity problem

    B. O’Donoghue, “Operator splitting for a homogeneous embedding of the linear complementarity problem”, SIAM Journal on Optimization, vol. 31, pp. 1999–2023, 3 2021

    1999

  22. [30]

    Pytorch: An imperative style, high-performance deep learning library

    A. Paszke, S. Gross, F. Massa, A. L erer, J. Bradbury, G. Chanan, T. Killeen, Z. Li n, N. Gimelshein, L. Antiga, et al., “ Pytorch: An imperative style, high-performance deep learning library”, Advances in neural information processing systems , vol. 32, 2019

    2019

  23. [31]

    Design, modeling, control, and app lication o f everting vine robots

    L. H. Blumenschein, M. M. Coad, D. A. Haggerty, A. M. Okamura, and E . W. Hawkes, “Design, modeling, control, and app lication o f everting vine robots”, Frontiers in Robotics and AI, vol. 7, p. 548 266, 2020

    2020

  24. [32]

    Modeling of bioinspired ap ical extension in a so ft r obot

    L. H. Blumenschein, A. M. Okamura, and E. W. Hawkes, “Modeling of bioinspired ap ical extension in a so ft r obot”, i n Conference on Biomimetic and Biohybrid Systems , Springer, 2017, pp. 522–531

    2017

  25. [33]

    Retraction o f soft growing robots w ithout buckling

    M. M. Coad, R. P. Thomasson, L. H. Blumenschein, N. S. Usevitch, E. W. Hawkes, and A. M. Okamura, “ Retraction o f soft growing robots w ithout buckling”, IEEE Robo tics and Automation Letters, vol. 5, no. 2, pp. 2115–2122, 2020

    2020

  26. [34]

    Character- izing environmental i nteractions for soft growing robots

    D. A. Haggerty, N. D. Naclerio, and E . W. Hawkes, “ Character- izing environmental i nteractions for soft growing robots”, i n 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), IEEE, 2019, pp. 3335–3342

    2019

  27. [35]

    Collapse of straight soft growing inflated beam robots under their own weight

    C. McFarland and M . M. Coad, “Collapse of straight soft growing inflated beam robots under their own weight”, in 2023 IEEE Interna- tional Conference on So ft Robotics (RoboSoft), IEEE, 2023, pp. 1–8

    2023

  28. [36]

    A soft, steerable continuum robot that grows via tip extension

    J. D. Greer, T. K. Morimoto, A. M. Okamura, and E. W. Hawkes, “A soft, steerable continuum robot that grows via tip extension”, Soft robotics, vol. 6, no. 1, pp. 95–108, 2019

    2019

  29. [37]

    A concise guide to modelling the physics of embodied intelligence in soft robotics

    G. Mengaldo, F. Renda, S. L. Brunton, M. B¨acher, M. Calisti, C. Duriez, G. S. Chirikjian, and C. Laschi, “A concise guide to modelling the physics of embodied intelligence in soft robotics”, Nature Reviews Physics, vol. 4, no. 9, pp. 595–610, 2022

    2022

  30. [38]

    Sofa: A multi-model fr amework for i nteractive ph ysical simulation

    F. Faure, C. Duriez, H. Delingette, J. Allard, B. Gilles, S. Marchesseau, H. Talbot, H. Courtecuisse, G. Bousquet, I. Peterlik, et al., “ Sofa: A multi-model fr amework for i nteractive ph ysical simulation”, Soft tissue biomechanical modeling for computer assisted surgery, pp...

    2012

  31. [39]

    L umped parameter dynamic mode l of an eversion g rowing robot: Analysis, simulation and e xperimental v alidation

    P. Vartholomeos, Z. Wu, S. H. Sadati, and C . Bergeles, “L umped parameter dynamic mode l of an eversion g rowing robot: Analysis, simulation and e xperimental v alidation”, i n 2024 I EEE Interna- tional Conference on Robo tics and Automation (ICRA), IEEE, 2024, pp. 12 734–12 740

    2024

  32. [40]

    Development and evaluation o f an intuitive flexible interface for t eleoperating so ft growing robots

    H. El-Hussieny, U. Mehmood, Z. Mehdi, S.-G. J eong, M. Usman, E. W. Hawkes, A. M. Okarnura, and J.-H. Ryu, “Development and evaluation o f an intuitive flexible interface for t eleoperating so ft growing robots”, i n 2018 IEEE/RSJ I nternational Conference on Intelligent Robot...

    2018

  33. [41]

    Automatic differentiation in machine learning: A survey

    A. G. Baydin, B. A. Pearlmutter, A. A. Radul, and J. M. Siskind, “Automatic differentiation in machine learning: A survey”, Journal of machine learning research , vol. 18, no. 153, pp. 1–43, 2018

    2018

  34. [42]

    Jax: Composable transformations o f python+numpy programs

    J. Bradbury, R. Frostig, P. Hawkins, M. J. J ohnson, C. L eary, D. Maclaurin, G. Necula, A. Paszke, J. V anderPlas, S. Wanderman- Milne, et al., “Jax: Composable transformations o f python+numpy programs”, 2018

    2018

  35. [43]

    Brax–a d ifferentiable ph ysics eng ine for l arge sca le rigid body simulation

    C. D. Freeman, E. Frey, A. Raichuk, S. Girgin, I. Mordatch, and O. Bachem, “ Brax–a d ifferentiable ph ysics eng ine for l arge sca le rigid body simulation”, arXiv preprint arXiv:2106.13281 , 2021

    2021 arXiv

  36. [44]

    Optimization-based inverse model of soft r obots w ith con tact handling

    E. Coevoet, A. Escande, and C. Duriez, “Optimization-based inverse model of soft r obots w ith con tact handling”, IEEE Robo tics and Automation Letters, vol. 2, no. 3, pp. 1413–1419, 2017

    2017

  37. [45]

    Robot model identification and learning: A modern perspective

    T. L ee, J. Kwon, P. M. Wensing, and F . C. Park, “ Robot model identification and learning: A modern perspective”, Annual Review of Control, Robotics, and Autonomous Systems , vol. 7, 2023

    2023

  38. [46]

    Do d iffer- entiable s imulators g ive be tter policy gradients?

    H. J. Suh, M. Simchowitz, K. Zhang, and R . Tedrake, “Do d iffer- entiable s imulators g ive be tter policy gradients?”, i n International Conference on Mach ine Learning, PMLR, 2022, pp. 20 668–20 696

    2022

  39. [47]

    Differentiable physics and s table modes for tool-use and man ipula- tion planning

    M. A. Toussaint, K. R. Allen, K. A. Smith, and J. B. Tenenbaum, “Differentiable physics and s table modes for tool-use and man ipula- tion planning”, 2018

    2018

  40. [48]

    Differentiating through a cone p rogram

    A. Agrawal, S. Barratt, S. Boyd, E. Busseti, and W . M. Moursi, “ Differentiating through a cone p rogram”, arXiv preprint arXiv:1904.09043, 2019

    1904 arXiv

  41. [49]

    Optnet: Differentiable optimization as a layer i n neu ral networks

    B. Amos and J. Z. Kolter, “Optnet: Differentiable optimization as a layer i n neu ral networks”, i n International conference on mach ine learning, PMLR, 2017, pp. 136–145

    2017

  42. [50]

    Dojo: A differentiable s imulator f or r obotics

    T. A. Howell, S. L e C leac’h, J. Z. Kolter, M. Schwager, and Z . Manchester, “ Dojo: A differentiable s imulator f or r obotics”, arXiv preprint arXiv:2203.00806, vol. 9, no. 2, p. 4, 2022

    2022 arXiv

  43. [51]

    Soft robot control with a learned differentiable model

    J. M. Bern, Y .Schnider, P. Banzet, N. Kumar, and S. Coros, “Soft robot control with a learned differentiable model”, in 2020 3rd IEEE International Conference on So ft Robotics (RoboSoft), IEEE, 2020, pp. 417–423

    2020

  44. [52]

    Diffaqua: A differentiable compu tational design pipeline for soft underwater swimmers w ith shape interpolation

    P. Ma, T. Du, J. Z. Zhang, K. Wu, A. Spielberg, R. K. Katzschmann, and W. Matusik, “ Diffaqua: A differentiable compu tational design pipeline for soft underwater swimmers w ith shape interpolation”, ACM Transactions on Graphics (TOG), v ol. 40, no. 4, pp. 1–14, 2021

    2021

  45. [53]

    Learning-in-the-loop optimization: End-to-end control and co-design of soft robots through learned deep latent representations

    A. Spielberg, A. Zhao, Y . Hu, T. Du, W. Matusik, and D. Rus, “Learning-in-the-loop optimization: End-to-end control and co-design of soft robots through learned deep latent representations”, Advances in Neural Information Processing Systems , vol. 32, 2019

    2019

  46. [54]

    Forward-mode automatic differentiation in Julia

    J. Revels, M. Lubin, and T. Papamarkou, “Forward-mode automatic differentiation in Julia”, arXiv:1607.07892 [cs.MS] , 2016

    2016 arXiv

  47. [55]

    Globally convergent type–I Anderson acceleration for non-smooth fixed-point iterations

    J. Zhang, B. O’Donoghue, and S. Boyd, “Globally convergent type–I Anderson acceleration for non-smooth fixed-point iterations”, SIAM Journal on Optimization , vol. 30, no. 4, pp. 3170–3197, 2020

    2020

  48. [56]

    The OpenCV Library

    G. Bradski, “The OpenCV Library”, Dr .Dobb’s Journal of Software Tools, 2000

    2000

  49. [57]

    Fixing weight decay regularization in adam

    I. Loshchilov, F. Hutter, et al., “Fixing weight decay regularization in adam”, arXiv preprint arXiv:1711.05101 , vol. 5, 2017

    2017 arXiv

  50. [58]

    Differen- tiable mpc for end-to-end planning and con trol

    B. Amos, I. Ji menez, J. Sacks, B. Boots, and J. Z. Kolter, “Differen- tiable mpc for end-to-end planning and con trol”, Advances in neu ral information processing systems , vol. 31, 2018

    2018

  51. [59]

    Single-level differentiable contact simulation

    S. Le Cleac’h, M. Schwager, Z. Manchester, V .Sindhwani, P. Flo- rence, and S. Singh, “Single-level differentiable contact simulation”, IEEE Robo tics and Automation Letters, vol. 8, no. 7, pp. 4012–4019, 2023

    2023

  52. [60]

    Rethinking optimization w ith d ifferentiable s imulation from a g lobal perspec- tive

    R. Antonova, J. Yang, K. M. Jatavallabhula, and J. Bohg, “Rethinking optimization w ith d ifferentiable s imulation from a g lobal perspec- tive”, in Conference on Robo t L earning, PMLR, 2023, pp. 276–286

    2023

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.