pith. sign in

arxiv: 2603.16478 · v2 · pith:OKO7RRUPnew · submitted 2026-03-17 · 💻 cs.GR

Fast and Reliable Gradients for Deformables Across Frictional Contact Regimes

Ziqiu Zeng , Gang Yang , Zhenhao Huang , Bingyang Zhou , Yulin Li , Jason Pho , Siyuan Luo , Fan Shi This is my paper

Pith reviewed 2026-05-21 11:29 UTC · model grok-4.3

classification 💻 cs.GR
keywords differentiable simulationfrictional contactdeformable bodiesgradient computationfinite element methodGPU accelerationcontact mechanicsinverse problems
0
0 comments X

The pith

Enforcing strict Markovian dynamics on a position-velocity manifold yields consistent gradients for frictional contact in deformable simulations.

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

The paper aims to show that a differentiable simulator can compute reliable gradients for objects that deform and slide under friction by keeping the underlying dynamics strictly Markovian on a combined position and velocity state. It pairs this with a preconditioner aligned to the object's mass distribution and a softened Fischer-Burmeister function to handle contact forces smoothly, plus a commutation condition that removes singularities in the material model. If these elements work together, optimization routines for inverse problems such as material identification and motion control can run without the gradient vanishing or distorting that appears in many current methods. The result would be more accurate parameter fitting from real-world data and more stable control policies for tasks involving cloth, soft robots, and dexterous handling.

Core claim

The central claim is that a unified GPU-accelerated differentiable simulator establishes mathematical consistency for gradients across frictional contact regimes through long-horizon consistency that enforces strict Markovian dynamics on the coupled position-velocity manifold, unified contact stability achieved with a mass-aligned preconditioner and soft Fischer-Burmeister operator, and resolution of FEM singularities by a derived within-block commutation condition.

What carries the argument

The coupled position-velocity manifold under strict Markovian dynamics, together with the mass-aligned preconditioner and soft Fischer-Burmeister operator.

If this is right

  • Optimization for physical system identification succeeds in contact-rich scenes without gradient distortion.
  • Inverse dynamics control for dexterous manipulation produces low-noise trajectories.
  • Cloth folding simulations maintain gradient fidelity over extended sequences.
  • The sim-to-real gap narrows for deformable objects subject to friction.

Where Pith is reading between the lines

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

  • The same stability properties could support gradient-based design loops for soft mechanisms that experience repeated sliding contacts.
  • Scaling the approach to multi-body systems with many contacts would test whether the preconditioner remains effective without extra parameter adjustment.
  • Embedding the simulator inside a learned policy network might yield controllers that generalize across varying friction coefficients.

Load-bearing premise

The assumption that combining strict Markovian dynamics on the position-velocity manifold with the mass-aligned preconditioner and soft Fischer-Burmeister operator will produce mathematically consistent non-collapsing gradients across all frictional contact regimes without introducing new artifacts.

What would settle it

Measure gradient norms over hundreds of time steps in a long-horizon cloth-folding or grasping sequence; if the norms stay bounded away from zero and downstream optimization converges without manual retuning, the claim holds, while sudden collapse or stagnation would contradict it.

Figures

Figures reproduced from arXiv: 2603.16478 by Bingyang Zhou, Fan Shi, Gang Yang, Jason Pho, Siyuan Luo, Yulin Li, Zhenhao Huang, Ziqiu Zeng.

Figure 1
Figure 1. Figure 1: Taming the Elephant in the Room: We simulate a soft, cable-driven elephant trunk writing "SIGGRAPH" on a wall. Our differentiable physics engine optimizes strongly coupled elastic dynamics involving frictional contact and state-based controls. Using direct gradient descent from a rest state, we optimize the four-cable actuation for each letter in under one hour on a single NVIDIA RTX 5090 GPU. (see [PITH_… view at source ↗
Figure 2
Figure 2. Figure 2: Differentiation graph: gradients are obtained by implicitly differen [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Starting from zero applied force, we optimize a horizontal force to pull [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 3
Figure 3. Figure 3: The smoothed Fischer-Burmeister function transforms a hard com [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: Starting from an initial guess of 𝐸 = 1 × 105 and 𝜈 = 0.2 (left), we perform system identification on a curtain undergoing gravity-induced deformation. Owing to the accuracy and robustness of the proposed gradi￾ents, the optimizer is able to simultaneously recover both parameters and converges to values close to the ground truth (𝐸 = 1 × 104 , 𝜈 = 0.3, right). be symmetric positive definite nor well-condit… view at source ↗
Figure 6
Figure 6. Figure 6: Gradient-based of stiffness (top) and Poisson’s ratio (down) in [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Experimental validation before (top row) and after (bottom row) identification across five different physical scenarios. The top row shows the simulation using initial guesses, while the bottom row utilizes the identified parameters, achieving perfect alignment with the target (pink). The columns from left to right correspond to the identification of: (1) bending constraint, (2) wind force, (3) friction fa… view at source ↗
Figure 8
Figure 8. Figure 8: We launch an elastic rubberduck horizontally onto a frictionless surface, involving frictionless contact dynamics. The optimization objective is the [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Two-parameter system identification under Neo-Hookean hy￾perelasticity. We jointly optimize Young’s modulus 𝐸 and Poisson’s ratio 𝜈 for the hanging-curtain setup. Top: analytical gradients closely match finite-difference references throughout the run, including the coupled tran￾sition around the rapid descent phase. Bottom: the loss (logarithmic scale) decreases smoothly as (𝐸, 𝜈 ) converge to their target… view at source ↗
Figure 10
Figure 10. Figure 10: Gradient unit tests for elasticity and contact. Each panel reports a gradient-based identification task: (a) stiffness, (b) wind force, (c) bending coefficient, (d) Poisson’s ratio, (e) friction factor, (f ) lifting force, (g) pulling force, and initial-velocity identification (h) without friction and (i) with friction. In each panel, the top subplot compares analytical gradients with finite-difference re… view at source ↗
Figure 11
Figure 11. Figure 11: Convergence comparison of different adjoint solvers and preconditioners. We report relative residual error versus iteration count (left) and wall [PITH_FULL_IMAGE:figures/full_fig_p015_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Optimization-based tracking using a 4-cable elastic trunk. We dis [PITH_FULL_IMAGE:figures/full_fig_p016_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Gradient-based control of a cable-driven soft gripper for object [PITH_FULL_IMAGE:figures/full_fig_p016_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Friction behavior at time step 0 (left) and 3000 (right): blue block [PITH_FULL_IMAGE:figures/full_fig_p018_14.png] view at source ↗
read the original abstract

Differentiable simulation establishes the mathematical foundation for solving challenging inverse problems in computer graphics and robotics, such as physical system identification and inverse dynamics control. However, rigor in frictional contact remains the "elephant in the room." Current frameworks often avoid contact singularities via non-Markovian position approximations or heuristic gradients. This lack of mathematical consistency distorts gradients, causing optimization stagnation or failure in complex frictional contact and large-deformation scenarios. We introduce our unified fully GPU-accelerated differentiable simulator, which establishes a rigorous theoretical paradigm through: Long-Horizon Consistency: enforcing strict Markovian dynamics on a coupled position-velocity manifold to prevent gradient collapse; Unified Contact Stability: employing a mass-aligned preconditioner and soft Fischer--Burmeister operator for smooth frictional optimization; Robust Material Identification: resolving FEM singularities via a derived "Within-block Commutation" condition. Our experiments demonstrate our solver efficacy in bridging the Sim-to-Real gap, delivering precise, low-noise gradients in contact-rich tasks like dexterous manipulation and cloth folding. By mitigating the gradient instability issues common in conventional approaches, our framework significantly enhances the fidelity of physical system identification and control.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The manuscript introduces a unified fully GPU-accelerated differentiable simulator for deformable objects with frictional contacts. It establishes a theoretical paradigm via three components: Long-Horizon Consistency through strict Markovian dynamics on a coupled position-velocity manifold to avoid gradient collapse; Unified Contact Stability via a mass-aligned preconditioner and soft Fischer-Burmeister operator for smooth optimization; and Robust Material Identification by resolving FEM singularities with a derived Within-block Commutation condition. Experiments demonstrate efficacy in contact-rich tasks including dexterous manipulation and cloth folding, with claims of precise low-noise gradients that bridge the Sim-to-Real gap.

Significance. If the central claims hold, the work would meaningfully advance differentiable simulation for inverse problems in computer graphics and robotics by providing a mathematically consistent treatment of frictional contact singularities. The GPU implementation and focus on non-collapsing gradients across regimes could improve reliability in system identification and control tasks.

major comments (1)
  1. The weakest assumption—that the combination of strict Markovian dynamics on the position-velocity manifold, mass-aligned preconditioner, and soft Fischer-Burmeister operator yields mathematically consistent non-collapsing gradients across all frictional regimes without new artifacts or post-hoc tuning—requires explicit derivation and error analysis to confirm it is load-bearing and independent of prior heuristics.
minor comments (2)
  1. Abstract: the description of the soft Fischer-Burmeister operator would benefit from a short parenthetical clarification or reference for readers unfamiliar with the variant.
  2. Experiments section: quantitative metrics (e.g., gradient noise norms or optimization success rates versus baselines) should be added to substantiate the 'precise, low-noise' claim.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive and detailed review. We address the single major comment below and have revised the manuscript to strengthen the theoretical presentation as suggested.

read point-by-point responses

  1. Referee: The weakest assumption—that the combination of strict Markovian dynamics on the position-velocity manifold, mass-aligned preconditioner, and soft Fischer-Burmeister operator yields mathematically consistent non-collapsing gradients across all frictional regimes without new artifacts or post-hoc tuning—requires explicit derivation and error analysis to confirm it is load-bearing and independent of prior heuristics.

    Authors: We thank the referee for identifying this as the central theoretical claim requiring further substantiation. The manuscript derives Long-Horizon Consistency from the strict Markovian structure on the coupled position-velocity manifold (Section 3.1), shows that the mass-aligned preconditioner removes inertial-contact misalignment that otherwise produces singular Jacobians, and establishes that the soft Fischer-Burmeister operator yields a differentiable complementarity condition whose Jacobian remains well-conditioned. These steps are presented as a unified argument that the combination prevents gradient collapse without additional heuristics. Nevertheless, we acknowledge that the current exposition would benefit from a more self-contained derivation and quantitative error analysis. In the revision we will add a dedicated subsection (and appendix) that (i) provides an explicit chain-rule derivation of the composite gradient operator, (ii) derives an a-priori bound on the deviation from ideal non-singular behavior across the stick-slip and separation regimes, and (iii) includes numerical verification on a set of controlled contact problems demonstrating that no new artifacts appear and that the method does not rely on post-hoc parameter tuning. This material will make the load-bearing character of the three components explicit. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation presented as independent

full rationale

The provided abstract and context introduce three main contributions—strict Markovian dynamics on a position-velocity manifold, a mass-aligned preconditioner with soft Fischer-Burmeister operator, and a derived Within-block Commutation condition—without any equations, fitted parameters, or self-citations visible in the text. Claims of 'derived' conditions and 'enforcing' dynamics are stated at a high level with no reduction to inputs by construction or load-bearing self-citation chains. The framework is described as establishing a rigorous paradigm through these elements, but absent explicit derivations or prior-author uniqueness theorems in the given material, the central claims remain self-contained and do not collapse to tautological fits or renamings.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available, so no concrete free parameters, axioms, or invented entities can be extracted. The approach appears to rest on standard domain assumptions of differentiable contact mechanics and GPU-accelerated FEM without additional detail.

pith-pipeline@v0.9.0 · 5748 in / 1290 out tokens · 71017 ms · 2026-05-21T11:29:38.201075+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

41 extracted references · 41 canonical work pages

  1. [1]

    Yinoussa Adagolodjo, Laurent Goffin, Michel De Mathelin, and Hadrien Courte- cuisse. 2019. Robotic Insertion of Flexible Needle in Deformable Structures Using Inverse Finite-Element Simulation.IEEE Transactions on Robotics35, 3 (2019), 697–708. doi:10.1109/TRO.2019.2897858

    work page doi:10.1109/tro.2019.2897858 2019

  2. [2]

    Atilim Gunes Baydin, Barak A Pearlmutter, Alexey Andreyevich Radul, and Jeffrey Mark Siskind. 2018. Automatic differentiation in machine learning: a survey.Journal of machine learning research18, 153 (2018), 1–43

    work page 2018

  3. [3]

    Sofien Bouaziz, Sebastian Martin, Tiantian Liu, Ladislav Kavan, and Mark Pauly

    work page

  4. [4]

    In Seminal Graphics Papers: Pushing the Boundaries, Volume 2

    Projective dynamics: Fusing constraint projections for fast simulation. In Seminal Graphics Papers: Pushing the Boundaries, Volume 2. 787–797

    work page

  5. [5]

    Coevoet, T

    E. Coevoet, T. Morales-Bieze, F. Largilliere, Z. Zhang, M. Thieffry, M. Sanz-Lopez, B. Carrez, D. Marchal, O. Goury, J. Dequidt, and C. Duriez. 2017. Software toolkit for modeling, simulation, and control of soft robots.Advanced Robotics31, 22 (Nov. 2017), 1208–1224. doi:10.1080/01691864.2017.1395362 Publisher: Taylor & Francis

    work page doi:10.1080/01691864.2017.1395362 2017

  6. [6]

    Jonas Degrave, Michiel Hermans, Joni Dambre, and Francis Wyffels. 2019. A dif- ferentiable physics engine for deep learning in robotics.Frontiers in neurorobotics 13 (2019), 6

    work page 2019

  7. [7]

    Tao Du, Kui Wu, Pingchuan Ma, Sebastien Wah, Andrew Spielberg, Daniela Rus, and Wojciech Matusik. 2021. Diffpd: Differentiable projective dynamics.ACM Transactions on Graphics (ToG)41, 2 (2021), 1–21

    work page 2021

  8. [8]

    Kenny Erleben. 2018. Methodology for Assessing Mesh-Based Contact Point Methods.ACM Trans. Graph.37, 3 (June 2018), 1–30. doi:10.1145/3096239 Fast and Reliable Gradients for Deformables Across Frictional Contact Regimes•17

    work page doi:10.1145/3096239 2018

  9. [9]

    Moritz Geilinger, David Hahn, Jonas Zehnder, Moritz Bächer, Bernhard Thomaszewski, and Stelian Coros. 2020. Add: Analytically differentiable dy- namics for multi-body systems with frictional contact.ACM Transactions on Graphics (TOG)39, 6 (2020), 1–15

    work page 2020

  10. [10]

    Eric Heiden, David Millard, Erwin Coumans, and Gaurav S Sukhatme. 2020. Augmenting differentiable simulators with neural networks to close the sim2real gap.arXiv preprint arXiv:2007.06045(2020)

    work page arXiv 2020

  11. [11]

    Eric Heiden, David Millard, Hejia Zhang, and Gaurav S Sukhatme. 2019. Interac- tive differentiable simulation.arXiv preprint arXiv:1905.10706(2019)

    work page arXiv 2019

  12. [12]

    Philipp Holl, Vladlen Koltun, and Nils Thuerey. 2020. Learning to control pdes with differentiable physics.arXiv preprint arXiv:2001.07457(2020)

    work page arXiv 2020

  13. [13]

    Yuanming Hu, Luke Anderson, Tzu-Mao Li, Qi Sun, Nathan Carr, Jonathan Ragan- Kelley, and Frédo Durand. 2019. Difftaichi: Differentiable programming for physical simulation.arXiv preprint arXiv:1910.00935(2019)

    work page arXiv 2019

  14. [14]

    Krishna Murthy Jatavallabhula, Miles Macklin, Florian Golemo, Vikram Voleti, Linda Petrini, Martin Weiss, Breandan Considine, Jérôme Parent-Lévesque, Kevin Xie, Kenny Erleben, et al. 2021. gradsim: Differentiable simulation for system identification and visuomotor control.arXiv preprint arXiv:2104.02646(2021)

    work page arXiv 2021

  15. [15]

    Quentin Le Lidec, Igor Kalevatykh, Ivan Laptev, Cordelia Schmid, and Justin Carpentier. 2021. Differentiable simulation for physical system identification. IEEE Robotics and Automation Letters6, 2 (2021), 3413–3420

    work page 2021

  16. [16]

    Yifei Li, Tao Du, Kui Wu, Jie Xu, and Wojciech Matusik. 2022. Diffcloth: Differen- tiable cloth simulation with dry frictional contact.ACM Transactions on Graphics (TOG)42, 1 (2022), 1–20

    work page 2022

  17. [17]

    Junbang Liang, Ming Lin, and Vladlen Koltun. 2019. Differentiable cloth simula- tion for inverse problems.Advances in neural information processing systems32 (2019)

    work page 2019

  18. [18]

    Jun Lv, Yunhai Feng, Cheng Zhang, Shuang Zhao, Lin Shao, and Cewu Lu. 2025. Sam-rl: Sensing-aware model-based reinforcement learning via differentiable physics-based simulation and rendering.The International Journal of Robotics Research44, 10-11 (2025), 1767–1783

    work page 2025

  19. [19]

    Miles Macklin. 2024. Warp: Differentiable Spatial Computing for Python. InACM SIGGRAPH 2024 Courses. 1–147

    work page 2024

  20. [20]

    Miles Macklin, Kenny Erleben, Matthias Müller, Nuttapong Chentanez, Stefan Jeschke, and Viktor Makoviychuk. 2019. Non-smooth newton methods for de- formable multi-body dynamics.ACM Transactions on Graphics (TOG)38, 5 (2019), 1–20

    work page 2019

  21. [21]

    Miles Macklin, Matthias Müller, and Nuttapong Chentanez. 2016. XPBD: position- based simulation of compliant constrained dynamics. InProceedings of the 9th International Conference on Motion in Games. 49–54

    work page 2016

  22. [22]

    Sebastian Martin, Bernhard Thomaszewski, Eitan Grinspun, and Markus Gross

    work page

  23. [23]

    InACM SIGGRApH 2011 papers

    Example-based elastic materials. InACM SIGGRApH 2011 papers. 1–8

    work page 2011

  24. [24]

    Etienne Ménager, Louis Montaut, Quentin Le Lidec, and Justin Carpentier. 2025. Differentiable Simulation of Soft Robots with Frictional Contacts. In2025 IEEE 8th International Conference on Soft Robotics (RoboSoft). IEEE, 1–8

    work page 2025

  25. [25]

    Etienne Ménager, Louis Montaut, Quentin Le Lidec, and Justin Carpentier

    work page

  26. [26]

    In2025 IEEE 8th International Conference on Soft Robotics (RoboSoft)

    Differentiable Simulation of Soft Robots with Frictional Contacts. In2025 IEEE 8th International Conference on Soft Robotics (RoboSoft). 1–8. doi:10.1109/ RoboSoft63089.2025.11020844

    work page arXiv 2025

  27. [27]

    Rhys Newbury, Jack Collins, Kerry He, Jiahe Pan, Ingmar Posner, David Howard, and Akansel Cosgun. 2024. A review of differentiable simulators.IEEE Access (2024)

    work page 2024

  28. [28]

    Brown, Jie Li, and Rahul Narain

    Matthew Overby, George E. Brown, Jie Li, and Rahul Narain. 2017. ADMM Projective Dynamics: Fast Simulation of Hyperelastic Models with Dynamic Constraints.IEEE Transactions on Visualization and Computer Graphics23, 10 (Oct. 2017), 2222–2234. doi:10.1109/TVCG.2017.2730875

    work page doi:10.1109/tvcg.2017.2730875 2017

  29. [29]

    Tao Pang and Russ Tedrake. 2021. A convex quasistatic time-stepping scheme for rigid multibody systems with contact and friction. In2021 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 6614–6620

    work page 2021

  30. [30]

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. 2019. Pytorch: An imperative style, high-performance deep learning library.Advances in neural information processing systems32 (2019)

    work page 2019

  31. [31]

    Yi-Ling Qiao, Junbang Liang, Vladlen Koltun, and Ming C Lin. 2021. Efficient differentiable simulation of articulated bodies. InInternational Conference on Machine Learning. PMLR, 8661–8671

    work page 2021

  32. [32]

    Samuel Schoenholz and Ekin Dogus Cubuk. 2020. Jax md: a framework for differentiable physics.Advances in Neural Information Processing Systems33 (2020), 11428–11441

    work page 2020

  33. [33]

    Changkyu Song and Abdeslam Boularias. 2020. Learning to slide unknown objects with differentiable physics simulations.arXiv preprint arXiv:2005.05456(2020)

    work page arXiv 2020

  34. [34]

    Tuur Stuyck and Hsiao-yu Chen. 2023. Diffxpbd: Differentiable position-based simulation of compliant constraint dynamics.Proceedings of the ACM on Computer Graphics and Interactive Techniques6, 3 (2023), 1–14

    work page 2023

  35. [35]

    Dylan Turpin, Liquan Wang, Eric Heiden, Yun-Chun Chen, Miles Macklin, Stavros Tsogkas, Sven Dickinson, and Animesh Garg. 2022. Grasp’d: Differentiable contact-rich grasp synthesis for multi-fingered hands. InEuropean Conference on Computer Vision. Springer, 201–221

    work page 2022

  36. [36]

    Gang Yang, Siyuan Luo, Yunhai Feng, Zhixin Sun, Chenrui Tie, and Lin Shao. 2024. Jade: A differentiable physics engine for articulated rigid bodies with intersection- free frictional contact. In2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 16915–16922

    work page 2024

  37. [37]

    Xinyuan Yu, Siheng Zhao, Siyuan Luo, Gang Yang, and Lin Shao. 2023. Diffclothai: Differentiable cloth simulation with intersection-free frictional contact and dif- ferentiable two-way coupling with articulated rigid bodies. In2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 400–407

    work page 2023

  38. [38]

    Ziqiu Zeng, Siyuan Luo, Fan Shi, and Zhongkai Zhang. 2025. Fast But Accurate: A Real-Time Hyperelastic Simulator with Robust Frictional Contact. 44, 4, Article 88 (July 2025), 19 pages. doi:10.1145/3730834

    work page doi:10.1145/3730834 2025

  39. [39]

    Yaofeng Desmond Zhong, Jiequn Han, and Georgia Olympia Brikis. 2022. Differ- entiable physics simulations with contacts: Do they have correct gradients wrt position, velocity and control?arXiv preprint arXiv:2207.05060(2022)

    work page arXiv 2022

  40. [40]

    Yaofeng Desmond Zhong, Jiequn Han, Biswadip Dey, and Georgia Olympia Brikis

    work page

  41. [41]

    InLearning for dynamics and control conference

    Improving gradient computation for differentiable physics simulation with contacts. InLearning for dynamics and control conference. PMLR, 128–141. A Frictional Contact A.1 Construct Local Transformation When both∥𝜹 𝑓 ∥and∥𝝀 𝑓 ∥are nonzero, we define R= 1 ∥𝜹 𝑓 ∥  ∥𝜹 𝑓 ∥0 0 0𝛿 𝑓1 𝛿 𝑓2 0−𝛿 𝑓2 𝛿 𝑓1  = 1 ∥𝝀 𝑓 ∥  ∥𝝀 𝑓 ∥0 0 0−𝜆 𝑓1 −𝜆𝑓2 0𝜆 𝑓2 −𝜆...

    work page