Pith. sign in

REVIEW 3 major objections 5 minor 37 references

Geometric Formulation of Unified Force-Impedance Control on SE(3) for Robotic Manipulators

T0 review · 3 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read A geometric unified force-impedance controller on the SE(3) manifold tracks a desired force and a trajectory simultaneously while keeping the closed loop passive.

desk verdict A useful SE(3) extension of unified force-impedance control with code and sensible simulations, but the passivity proof as written is missing the derivative of the tank-modification factor, so the central guarantee is not yet established. read the letter →

arxiv 2504.17080 v3 pith:QH3OY5K6 submitted 2025-04-23 cs.RO cs.SYeess.SY

Joohwan Seo , Nikhil Potu Surya Prakash , Soomi Lee , Arvind Kruthiventy , Megan Teng , Jongeun Choi , Roberto Horowitz This is my paper · ORCID
classification cs.ROcs.SYeess.SY MSC 93C8570E6093D05
keywords geometricunifiedforce-impedancecontrolSE(3)manifoldenergytankaugmentationpassivityforcetrackingimpedancevelocityfieldequivariantmanipulation
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 proposes a geometric unified force-impedance control (GUFIC) law for a robot manipulator whose end effector moves on SE(3), the space of rigid positions and orientations. The authors aim to show that their controller can track both a desired end-effector path and a desired interaction force while keeping the closed-loop system passive with respect to external wrenches. Passivity is the property that makes contact with unknown, stiff surfaces safe: a passive robot cannot inject more energy into the environment than it receives. The paper establishes this as Theorem 1 under a full-rank Jacobian and reachability assumption, and supports it with simulations of force- and trajectory-tracking tasks.

What carries the argument

The load-bearing mechanism is energy-tank augmentation on two sub-ports of the interaction: one tank, defined by equations (17)-(18), absorbs energy when the force-tracking wrench would violate passivity, and another tank, defined by equations (19)-(20), absorbs energy from the impedance and feedforward port and dissipates it through damping. The desired velocity is modified to $(V_d^*)' = (\gamma_i + \alpha_i(1-\gamma_i))V_d^*$, and the trajectory is generated from a velocity field $V_d^*(t,g)$ of the form (22), which defines the desired motion as a field depending on both time and current pose. This field structure is what makes the velocity modification causal. The proof of Theorem 1 then cancels the impedance-tank flow against the storage-function derivative, leaving only non-positive terms plus the supply rate $(V_b)^T F_e$.

What would settle it

Recompute inequality (35) from equations (31)-(34) with $\alpha_i = 0$, $\gamma_i = 0$, and $F_f = 0$, so only the impedance tank is active and its port is in the passivity-violating direction. With the printed equations, the term $(V_b)^T F_e - (V_d^*)^T F_e$ remains while the correct error velocity is $e'_V = V_b$ because $(V_d^*)' = 0$; the claimed non-positive grouping in (35) therefore has an unmatched $-(V_d^*)^T F_e$ contribution and the inequality fails. Replacing $V_d^*$ by $(V_d^*)'$ in the storage derivative removes the unmatched term and the passivity bound follows.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central claim is Theorem 1: under Assumption 1, the GUFIC closed loop with control law (29) is passive for the port $(V_b, F_e)$ with storage function (32), which sums a geometric SE(3) kinetic-plus-potential energy with two energy-tank levels. The controller splits the input wrench into a geometrically consistent impedance term and a force-tracking term, modifies both through energy tanks, and generates the desired configuration from a time-and-pose-dependent velocity field rather than from a fixed trajectory. If the theorem is correct, a single controller can hold a desired force against a surface while tracking a six-degree-of-freedom trajectory, and it does so without the non-causal modified-desired-velocity integration that plagued the earlier unified force-impedance control formulation.

Load-bearing premise

The load-bearing premise of the proof is that the cancellation in the passivity algebra is legitimate: the derivation in equations (20) and (31) needs the modified desired velocity $(V_d^*)' = (\gamma_i + \alpha_i(1-\gamma_i))V_d^*$, but the printed equations use the unmodified $V_d^*$; if that mismatch is literal rather than a typographical slip, the claimed inequality (35) does not hold whenever $\alpha_i < 1$, which is exactly the regime where the impedance tank is being protected.

Editorial extensions

If this is right

  • The closed-loop manipulator is passive from external wrench to body velocity, so it can press against a stiff or unknown surface without driving the interaction unstable.
  • The controller can track a desired force and an SE(3) trajectory at the same time, with the energy tank storing energy produced in the passivity-violating direction and releasing it to keep the interaction passive.
  • Because the desired configuration comes from a velocity field rather than a pre-integrated trajectory, modifications to the force or velocity field never require future information, resolving the causality problem in earlier unified force-impedance control.
  • The left-invariant, SE(3)-formulated structure makes learned versions of the controller equivariant, so learning algorithms that output velocity or force fields can inherit symmetry priors and require fewer demonstrations.
  • The controller's passivity guarantee holds while gains, feedforward velocity, and desired force are changed online, which is exactly the regime where classical variable impedance control loses its stability certificate.

Reading between the lines

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

  • Editorial inference: if the apparent notation inconsistency in equations (20) and (31) is resolved by using the modified desired velocity $(V_d^*)'$ wherever the storage derivative is expanded, the passivity theorem goes through as written; the printed text would need a one-line correction to make the cancellation explicit.
  • Editorial inference: the velocity-field and force-field formulation suggests a natural learning target in which an equivariant policy directly outputs $V_d^*(t,g)$ and $F_d(t,g)$, and the energy tanks preserve passivity no matter what content the learned fields have.
  • Editorial inference: the same tank-plus-field construction should extend to redundant or mobile manipulators, since the energy argument does not depend on the number of joints once an operational-space Jacobian and an SE(3) error are defined.
  • Editorial inference: when the impedance tank depletes, the modified desired velocity becomes zero and task progress is deliberately sacrificed to preserve passivity; an experimental study of when this happens under stiff contacts or aggressive force tracking could quantify the safety-performance tradeoff.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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 GUFIC, a geometric unified force-impedance controller formulated on SE(3). The controller combines the authors' prior geometric impedance control (GIC) with a force-tracking PID term, and uses energy tanks for both the force-tracking port and the impedance port to enforce passivity with respect to the external wrench. A time- and pose-dependent velocity field is introduced to avoid the non-causal trajectory update of the earlier UFIC formulation. The main theoretical result is Theorem 1, which claims passivity of the closed-loop system for the channel (Vb, Fe) with storage function (32). The paper also claims SE(3) invariance/equivariance, stability and force convergence in remarks, and presents MuJoCo simulations for two contact tasks with the Indy7 robot. The code is publicly available.

Significance. If Theorem 1 is correct, the paper gives a systematic way to combine force tracking with impedance control on SE(3) while preserving passivity, which is directly relevant to contact-rich manipulation and to learning-based policies that rely on low-level passive controllers. The velocity-field reformulation is a genuine improvement over the causality issues in the original UFIC, and the explicit differential-geometric formulation is a useful step. The release of simulation code supports reproducibility, and the equivariance argument, if fully justified, is a plausible route to improved sample efficiency in learning manipulation tasks. However, the main passivity proof currently has a load-bearing notational inconsistency and an unaddressed derivative term; these need to be fixed before the central claim is supported.

major comments (3)
  1. [Section III-B.4, Theorem 1 proof, Eqs. (20), (31), (34)-(35)] The passivity proof is internally inconsistent in its use of the modified desired velocity. Equation (31) expands the storage derivative in terms of e'_V, which is defined as e'_V = Vb - (V*_d)' with (V*_d)' = (gamma_i + alpha_i(1-gamma_i)) V*_d, but Eqs. (20) and (34) use the unprimed (V*_d)^T in the impedance tank port and in the cross term. Read literally, the cancellation that produces Eq. (35) fails whenever alpha_i < 1, which is exactly the tank-depleting regime that the design is meant to handle. For example, when alpha_i = 0, (V*_d)' = 0 while (V*_d)^T(F'_f + Fe) remains in the expansion of ˙S, and the tank term proportional to alpha_i(1-gamma_i) vanishes, so the claimed inequality (37) is not established. If this is a typographical slip, the proof must be corrected by replacing every occurrence of (V*_d)^T in Eqs. (20) and (34) with ((V*_d)')^T or by redefining the tank port consistently with Eq. (21).
  2. [Section III-B.3, Eq. (27), Appendix A] The paper never defines what (˙V*_d)' means when the modification law (21) is applied to the velocity field. Since gamma_i and alpha_i are time-varying and nonsmooth, the total time derivative of (V*_d)' = mu_i V*_d includes the term mu_dot_i V*_d in addition to mu_i ˙V*_d. Appendix A derives ˙V*_d only for the unmodified field. If Eq. (27) is implemented as mu_i ˙V*_d, then the error dynamics (30) acquire an extra forcing term -tilde M mu_dot_i V*_d, and the storage derivative acquires -mu_dot_i (e'_V)^T tilde M V*_d. This indefinite-sign contribution is not bounded by either tank dynamics (17) or (20), so the claimed passivity inequality (37) is not established. The authors should specify the derivative convention explicitly and either compensate mu_dot_i V*_d in the tank design or restrict gamma_i and alpha_i to be constant on intervals where the extra term is nonpositive.
  3. [Remark 1, Section III-B.4] The remark claims asymptotic convergence to the modified setpoint g'_d and force convergence to Fd = -Fe, but no proof is given. Passivity inequality (37) alone does not imply these convergence statements, and the reference to previous works [20], [21], [32] does not supply the needed dissipation and coupling conditions for the tank states and the force-tracking integrator. Since force tracking is a stated contribution, this is a gap: either provide a Lyapunov argument or clearly label the convergence claims as beyond the scope of the paper.
minor comments (5)
  1. [Eq. (21)] The subscript in alpha is missing: Eq. (21) writes (gamma_i + alpha(1-gamma_i)) V*_d, while the surrounding text uses alpha_i.
  2. [Eq. (26) and surrounding text] The notation (Vb_d)' is ambiguous: it is used both as the body-frame velocity of the modified trajectory and as the adjoint-transported modification Ad_{g_ed^{-1}} (V*_d)', and the discrete update formula is only sketched.
  3. [Remark 3] The controller shaping function rho is said to be implemented, but no details are given and the simulations do not state whether rho is active in the reported results; please clarify its role in the experiments.
  4. [Section IV] The simulations keep both tanks above their lower thresholds, so the proposed behavior in the tank-depleting regime, where the modified velocity and force are set to zero, is not exercised; a scenario with an initially low tank would better validate the claimed safety property.
  5. [Appendix A] The time-derivative expressions use a mixture of ̅V*_d and V*_d notation without explicit definitions, which makes it hard to verify the algebra against Eq. (22).

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity in the passivity derivation; GUFIC extends prior GIC lemmas rather than restating fitted inputs.

full rationale

Theorem 1's passivity claim is derived algebraically from the closed-loop error dynamics and the tank dynamics: Eq. (30) follows by substituting the control law (27), Eq. (31) expands the storage derivative using e'_V = Vb - (V*_d)', and the tank update laws (17), (20) are explicitly constructed to cancel the indefinite cross terms, leaving (35), whose sign is fixed by the definitions of gamma, beta, and alpha. This is a new derivation, not a renamed input or a fitted parameter called a prediction; the simulation gains are hand-set and the controller is compared against GIC. The paper does rely on the authors' earlier GIC results, specifically the identity dot P = f_G^T e_V from [21], [22] and the equivariance Lemma 1 of [23], but these are parameter-free mathematical facts from prior published derivations that do not assume the GUFIC passivity theorem, so under the stated rules they count as independent support rather than circularity. A separate technical gap exists and should be weighed as a correctness issue, not circularity: Eq. (27) uses (dot V*_d)' while Appendix A computes only dot V*_d, so the total derivative of (gamma_i + alpha_i(1 - gamma_i)) V*_d, including dot gamma_i and dot alpha_i, is not specified; this missing-derivative issue does not amount to a circular reduction of the passivity claim.

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

The core passivity theorem rests on standard rigid-body dynamics, the GIC potential-gradient identity dot P = f_G^T e_V (from the authors' prior publications), the ideal-sensor assumption, and standard energy-tank switching. The only simulation-specific numbers are gains and tank bounds, none of which are needed for the theorem. The printed proof has a port-velocity inconsistency between Eqs. (20), (31) and (35) that must be resolved; this is the main burden.

free parameters (4)
  • zeta (velocity field gain) = 5 (both scenarios)
    Hand-tuned positive scalar in velocity field (22); affects convergence, not the passivity theorem.
  • Force PID gains (kp, ki, kd) = Scenario 1: (1.0, 0.5, 4.0); Scenario 2: (1.5, 0.75, 6.0)
    Chosen for simulation; not required by the passivity proof.
  • Energy tank parameters T(0), T_l, T_u, delta_T = Scenario 1: Tf(0)=Ti(0)=10, T_l=0.1, T_u=20, delta_T=0.5; Scenario 2: Ti(0)=90, T_u,i=100
    Hand-selected so tanks do not deplete; the paper notes Ti(0) must be high in Scenario 2, indicating performance sensitivity.
  • Impedance gains Kp, KR, Kd = Scenario 1: Kp=blkdiag(2000,2000,10), KR=blkdiag(2000,2000,2000), Kd=500I
    Design gains; positive definiteness suffices for the theorem.
assumptions (4)
  • domain assumption Assumption 1: Jacobian Jb is full rank; end-effector and desired trajectory lie in reachable set; desired trajectory continuously differentiable.
    Stated in Section II-C; required for the GIC-based formulation and for the velocity field to be well-defined.
  • standard math The SE(3) potential P in (8) satisfies dot P = f_G^T e_V along the error dynamics, and the generalized inertia/coriolis matrices satisfy 1/2 dot Mtilde - Ctilde skew-symmetric.
    Taken from prior GIC papers [21],[22]; used in the storage derivative (15) and (31). These are published derivations, not re-proven here.
  • domain assumption Force sensor output bar F_e equals the true external wrench F_e.
    Section III-A states: 'we use bar F_e ... without discerning it from Fe the actual external wrench'. Sensor noise, delay, or calibration error would break the passivity argument.
  • domain assumption Environment is passive so passivity of the robot implies closed-loop stability.
    Standard in passivity-based robot control, invoked implicitly in the introduction; not part of the theorem statement.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Geometric Formulation of Unified Force-Impedance Control on SE(3) for Robotic Manipulators." pith.science (2026). https://pith.science/paper/QH3OY5K6

@misc{pith2026250417080,
  author       = {Pith},
  title        = {Pith review of: Geometric Formulation of Unified Force-Impedance Control on SE(3) for Robotic Manipulators},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QH3OY5K6}},
  note         = {Machine review of arXiv:2504.17080}
}
read the original abstract

In this paper, we present an impedance control framework on the SE(3) manifold, which enables force tracking while guaranteeing passivity. Building upon the unified force-impedance control (UFIC) and our previous work on geometric impedance control (GIC), we develop the geometric unified force impedance control (GUFIC) to account for the SE(3) manifold structure in the controller formulation using a differential geometric perspective. As in the case of the UFIC, the GUFIC utilizes energy tank augmentation for both force-tracking and impedance control to guarantee the manipulator's passivity relative to external forces. This ensures that the end effector maintains safe contact interaction with uncertain environments and tracks a desired interaction force. Moreover, we resolve a non-causal implementation problem in the UFIC formulation by introducing velocity and force fields. Due to its formulation on SE(3), the proposed GUFIC inherits the desirable SE(3) invariance and equivariance properties of the GIC, which helps increase sample efficiency in machine learning applications where a learning algorithm is incorporated into the control law. The proposed control law is validated in a simulation environment under scenarios requiring tracking an SE(3) trajectory, incorporating both position and orientation, while exerting a force on a surface. The codes are available at https://github.com/Joohwan-Seo/GUFIC_mujoco.

Figures

Figures reproduced from arXiv: 2504.17080 by the authors.

Figure 1
Figure 1. Implemented Mujoco Simulation Environment with [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 3
Figure 3. Force tracking result of GUFIC and GIC for Scenario [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Energy Tanks Tf and Ti of GUFIC for Scenario 1 [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figures from the paper (4 more)
Figure 5
Figure 5. Figure 5: Trajectory tracking results of GUFIC for Scenario [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Error Function Ψ (23) of GUFIC for Scenario 2. 0 2 4 6 8 10 0 5 10 15 Time (s) fz (N) GUFIC GIC Desired [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Force tracking result of GUFIC for Scenario [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Energy Tanks [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 26 canonical work pages

  1. [20]

    Unified force-impedance control,

    S. Haddadin and E. Shahriari, “Unified force-impedance control,” The International Journal of Robotics Research, vol. 43, no. 13, pp. 2112– 2141, 2024

    2024

  2. [21]

    Geometric impedance control on SE(3) for robotic manipulators,

    J. Seo et al. , “Geometric impedance control on SE(3) for robotic manipulators,” IFAC-PapersOnLine, vol. 56, no. 2, pp. 276–283, 2023

    work page 2023

  3. [32]

    Tracking for fully actuated mechanical systems: a geometric framework,

    F. Bullo and R. M. Murray, “Tracking for fully actuated mechanical systems: a geometric framework,” Automatica, vol. 35, no. 1, pp. 17– 34, 1999

    1999

  4. [1]

    Impedance control: An approach to manipulation: Part ii—implementation,

    N. Hogan, “Impedance control: An approach to manipulation: Part ii—implementation,” 1985

    1985

  5. [2]

    A unified approach for motion and force control of robot manipulators: The operational space formulation,

    O. Khatib, “A unified approach for motion and force control of robot manipulators: The operational space formulation,” IEEE Journal on Robotics and Automation , vol. 3, no. 1, pp. 43–53, 1987

    1987

  6. [3]

    Learning variable impedance control for contact sensitive tasks,

    M. Bogdanovic et al. , “Learning variable impedance control for contact sensitive tasks,” IEEE Robotics and Automation Letters, vol. 5, no. 4, pp. 6129–6136, 2020

    work page 2020

  7. [4]

    Behavior transformers: Cloning k modes with one stone,

    N. M. Shafiullah et al. , “Behavior transformers: Cloning k modes with one stone,” Advances in neural information processing systems , vol. 35, pp. 22 955–22 968, 2022

    work page 2022

  8. [5]

    Diffusion policy: Visuomotor policy learning via action diffusion,

    C. Chi et al. , “Diffusion policy: Visuomotor policy learning via action diffusion,” The International Journal of Robotics Research , p. 02783649241273668, 2023

    work page 2023

Show all 37 references
  1. [6]

    Mobile aloha: Learning bimanual mobile manip- ulation with low-cost whole-body teleoperation,

    Z. Fu et al. , “Mobile aloha: Learning bimanual mobile manip- ulation with low-cost whole-body teleoperation,” arXiv preprint arXiv:2401.02117, 2024

    2024 arXiv

  2. [7]

    Admittance visuomotor policy learning for general- purpose contact-rich manipulations,

    B. Zhou et al. , “Admittance visuomotor policy learning for general- purpose contact-rich manipulations,” arXiv preprint arXiv:2409.14440, 2024

    2024 arXiv

  3. [8]

    Learning variable compliance control from a few demonstrations for bimanual robot with haptic feedback teleoperation system,

    T. Kamijo et al. , “Learning variable compliance control from a few demonstrations for bimanual robot with haptic feedback teleoperation system,” in 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . IEEE, 2024, pp. 12 663–12 670

    2024

  4. [9]

    Tacdiffusion: Force-domain diffusion policy for precise tactile manipulation,

    Y . Wu et al., “Tacdiffusion: Force-domain diffusion policy for precise tactile manipulation,” arXiv preprint arXiv:2409.11047 , 2024

    2024 arXiv

  5. [10]

    A survey of robot manipulation in contact,

    M. Suomalainen et al. , “A survey of robot manipulation in contact,” Robotics and Autonomous Systems , vol. 156, p. 104224, 2022

    2022

  6. [11]

    Stability considerations for variable impedance control,

    K. Kronander and A. Billard, “Stability considerations for variable impedance control,” IEEE Transactions on Robotics , vol. 32, no. 5, pp. 1298–1305, 2016

    2016

  7. [12]

    Passive velocity field control of mechanical manipulators,

    P. Y . Li and R. Horowitz, “Passive velocity field control of mechanical manipulators,” IEEE Transactions on robotics and automation, vol. 15, no. 4, pp. 751–763, 2002

    2002

  8. [13]

    Passive velocity field control of (pvfc), part i. geometry and robustness,

    ——, “Passive velocity field control of (pvfc), part i. geometry and robustness,” IEEE Transactions on Automatic Control , vol. 46, no. 9, pp. 1346–1359, 2001

    2001

  9. [14]

    Passive velocity field control of (pvfc), part ii. application to contour following,

    ——, “Passive velocity field control of (pvfc), part ii. application to contour following,” IEEE Transactions on Automatic Control, vol. 46, no. 9, pp. 1360–1371, 2001

    2001

  10. [15]

    Control of smart exercise machines. i. problem formulation and nonadaptive control,

    ——, “Control of smart exercise machines. i. problem formulation and nonadaptive control,” IEEE IEEE/ASME Transactions On Mechatron- ics, vol. 2, no. 4, pp. 237–247, 1997

    1997

  11. [16]

    Control of smart exercise machines. ii. self-optimizing control,

    ——, “Control of smart exercise machines. ii. self-optimizing control,” IEEE IEEE/ASME Transactions On Mechatronics , vol. 2, no. 4, pp. 248–258, 1997

    1997

  12. [17]

    An energy tank-based interactive control ar- chitecture for autonomous and teleoperated robotic surgery,

    F. Ferraguti et al. , “An energy tank-based interactive control ar- chitecture for autonomous and teleoperated robotic surgery,” IEEE Transactions on Robotics , vol. 31, no. 5, pp. 1073–1088, 2015

    2015

  13. [18]

    Energy tank-based wrench/impedance control of a fully-actuated hexarotor: A geometric port-hamiltonian approach,

    R. Rashad et al., “Energy tank-based wrench/impedance control of a fully-actuated hexarotor: A geometric port-hamiltonian approach,” in 2019 International Conference on Robotics and Automation (ICRA) . IEEE, 2019, pp. 6418–6424

    2019

  14. [19]

    Passivity-based variable impedance control for redundant manipulators,

    Y . Michel et al. , “Passivity-based variable impedance control for redundant manipulators,” IFAC-PapersOnLine, vol. 53, no. 2, pp. 9865–9872, 2020

    2020

  15. [22]

    A comparison between lie group-and lie algebra-based po- tential functions for geometric impedance control,

    ——, “A comparison between lie group-and lie algebra-based po- tential functions for geometric impedance control,” arXiv preprint arXiv:2401.13190, 2024

    2024 arXiv

  16. [23]

    Contact-rich SE(3)-equivariant robot manipulation task learn- ing via geometric impedance control,

    ——, “Contact-rich SE(3)-equivariant robot manipulation task learn- ing via geometric impedance control,” IEEE Robotics and Automation Letters, 2023

    2023

  17. [24]

    Equivariant descriptor fields: SE(3)-equivariant energy- based models for end-to-end visual robotic manipulation learning,

    H. Ryu et al., “Equivariant descriptor fields: SE(3)-equivariant energy- based models for end-to-end visual robotic manipulation learning,” in The Eleventh International Conference on Learning Representations (ICLR), 2023

    2023

  18. [25]

    Diffusion-EDFs: Bi-equivariant denoising generative mod- eling on SE(3) for visual robotic manipulation,

    ——, “Diffusion-EDFs: Bi-equivariant denoising generative mod- eling on SE(3) for visual robotic manipulation,” arXiv preprint arXiv:2309.02685, 2023

    2023 arXiv

  19. [26]

    Fourier transporter: Bi-equivariant robotic manipu- lation in 3d,

    H. Huang et al., “Fourier transporter: Bi-equivariant robotic manipu- lation in 3d,” arXiv preprint arXiv:2401.12046 , 2024

    2024 arXiv

  20. [27]

    SE(3)-equivariant robot learning and control: A tutorial survey,

    J. Seo et al., “SE(3)-equivariant robot learning and control: A tutorial survey,” arXiv preprint arXiv:2503.09829 , 2025

    2025 arXiv

  21. [28]

    R. M. Murray, Z. Li, and S. S. Sastry, A mathematical introduction to robotic manipulation. CRC press, 1994

    1994

  22. [29]

    K. M. Lynch and F. C. Park, Modern robotics. Cambridge University Press, 2017

    2017

  23. [30]

    A tank-based approach to impedance control with variable stiffness,

    F. Ferraguti et al., “A tank-based approach to impedance control with variable stiffness,” in 2013 IEEE international conference on robotics and automation. IEEE, 2013, pp. 4948–4953

    2013

  24. [31]

    Force tracking impedance control with variable target stiffness,

    K.-k. Lee and M. Buss, “Force tracking impedance control with variable target stiffness,” IFAC Proceedings Volumes, vol. 41, no. 2, pp. 6751–6756, 2008

    2008

  25. [33]

    Deep geometric potential functions for tracking on manifolds,

    N. P. S. Prakash et al. , “Deep geometric potential functions for tracking on manifolds,” in 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . IEEE, 2024, pp. i–viii

    2024

  26. [34]

    Variable impedance control using deep geometric potential fields,

    ——, “Variable impedance control using deep geometric potential fields,” IFAC-PapersOnLine, vol. 58, no. 28, pp. 480–485, 2024

    2024

  27. [35]

    Geometric tracking control of a quadrotor uav on se (3),

    T. Lee et al. , “Geometric tracking control of a quadrotor uav on se (3),” in 49th IEEE conference on decision and control (CDC) . IEEE, 2010, pp. 5420–5425

    2010

  28. [36]

    Mujoco: A physics engine for model-based control,

    E. Todorov et al. , “Mujoco: A physics engine for model-based control,” in 2012 IEEE/RSJ international conference on intelligent robots and systems . IEEE, 2012, pp. 5026–5033

    2012

  29. [37]

    A two-step method for dynamic parameter iden- tification of indy7 collaborative robot manipulator,

    M. Tadese et al. , “A two-step method for dynamic parameter iden- tification of indy7 collaborative robot manipulator,” Sensors, vol. 22, no. 24, p. 9708, 2022

    2022

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.