Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Motor current and joint states alone can teach robot hands when to grip gently or firmly, without adding tactile sensors.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-12 01:50 UTC pith:24ZHKZ4D

load-bearing objection Solid systems paper: skip wrench estimation, learn a PD-compatible compliance reference from raw motor current; real multi-hand ablations support the claim, with the main soft spot being human CRP labels and hardware-specific current informativeness. the 3 major comments →

arxiv 2607.03529 v1 pith:24ZHKZ4D submitted 2026-07-03 cs.RO

Current as Touch: Proprioceptive Contact Feedback for Compliant Dexterous Manipulation

Chenyang Ma , Yunchao Yao , Zhenyu Wei , Ruogu Li , Daniel Szafir , Mingyu Ding This is my paper
classification cs.RO
keywords dexterous manipulationcomplianceproprioceptive sensingmotor currentcompliance reference positionteleoperationpolicy learningsensorless contact
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Dexterous robot hands usually need extra force or touch sensors to handle fragile, thin, or changing objects without crushing or dropping them. This paper argues that the current already flowing through the hand's motors, together with joint positions, carries enough contact information to replace those sensors. Instead of estimating force and then commanding torque, the method predicts a compliance reference position: a joint target for the ordinary PD controller whose position error produces the right grasping force. Human teleoperation demonstrations supply the training targets, because the operator's corrected commands already encode successful contact. The same interface works for assisted teleoperation and for learned policies. Across stacking foam cups, wiping a board, picking a single card, and holding a bottle while water is poured into it, the approach yields safer teleoperation, fewer failures, and more stable grasps on multiple low-cost hands.

Core claim

Motor current and joint states are a learnable proprioceptive contact signal: a model trained on them can predict compliance reference positions that let a standard PD controller generate appropriate contact force, enabling tactile-free compliant grasping, safer teleoperation, and stronger policy learning on multiple dexterous hands.

What carries the argument

Compliance reference position (CRP): the ideal joint-position target sent to the fixed-gain PD controller; the induced tracking error produces the desired contact torque. The CRP is predicted from a short history of joint states, raw motor current, and either user-intent velocity or task pose.

Load-bearing premise

The method assumes that human teleoperation commands are good enough labels for the right grip targets, and that motor current on the particular hand carries repeatable contact information rather than being drowned by friction, cable stretch, or noise.

What would settle it

On the same hands and tasks, replace the current-conditioned CRP predictor with an otherwise identical model that never sees motor current; if teleoperation safety and policy success rates stay the same, the claim that current supplies necessary contact cues fails.

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

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes a proprioception-driven compliance framework for dexterous hands that predicts a compliance reference position (CRP) from motor current and joint states, rather than estimating external wrenches or commanding torque. The CRP is the ideal joint-position target for a standard PD controller (Eq. 1), so that the induced tracking error produces appropriate contact force. Supervision comes from human teleoperation commands treated as noisy but task-valid CRP labels (Eq. 2), with user intent represented only as command velocity to avoid target leakage. The same interface is used for assisted teleoperation and ACT-style policy learning, with raw current at inference and an auxiliary smoothed-current loss. Real-robot experiments on LEAP Hand and Unitree Dex3 cover foam-cup stacking, whiteboard wiping, single-card picking, and dynamic bottle holding, with ablations of retargeting, current-free CRP prediction, and full current-conditioned prediction.

Significance. If the result holds, the work offers a practical route to contact-aware compliance on low-cost position-controlled hands without external tactile or force/torque sensors. The position-reference formulation is a genuine systems contribution: it matches the interfaces used by mainstream teleoperation and imitation-learning pipelines, unlike torque-command or model-based admittance methods. Strengths include multi-hand, multi-task real-robot evaluation; a clear w/o-current ablation; free-space vs. contact demonstration design; and an explicit physical discussion of current-to-torque limitations (Appendix C). The motivating force-regression measurements (Fig. 3) and qualitative failure modes further support that current carries usable contact information. These are concrete, deployable advances for contact-rich dexterous manipulation.

major comments (3)
  1. [§3.2, Eq. (2); Tables 1–4] §3.2 and Eq. (2): CRP supervision is human teleoperation targets q_cmd. Tables 1–4 show large gains of w/ Current over w/o Current, but the residual improvement could still partly reflect better reconstruction of the demonstrated correction manifold rather than current-driven force regulation. A load-bearing control is missing: hold the human label distribution fixed while ablating contact informativeness of current (e.g., free-space-only current statistics, temporally scrambled current, or contact-masked current) and report whether the teleop/policy gains collapse. Without that, the claim that motor current supplies the contact cue for compliance remains only partially isolated.
  2. [§4.2–4.3; Tables 1–4] §4.2–4.3 and Tables 1–4: Trial counts are modest (15 teleop trials per user/method, 12 bottle-holding trials per load, 52 card-picking trials) and no uncertainty measures (std, CI, or significance tests) are reported. The central multi-task claim depends on these effect sizes remaining stable; at minimum, report variability and, where feasible, increase replications or add a third operator for teleop transfer. This is fixable but currently under-supports the strength of the empirical conclusions.
  3. [Limitations; Appendix C] Limitations and Appendix C correctly note that current mixes contact with friction, bias, and dynamics, and that load sensing benefits from LEAP abduction joints. The paper should more clearly bound when the method is expected to fail (hands without informative shear/load joints; transmissions with large non-repeatable friction) and, ideally, add one negative or weaker-signal hardware/task case. Otherwise the “without external sensors” claim risks over-generalization beyond the two evaluated hands.
minor comments (5)
  1. [§4.2; Table 1] Table 1 caption/text: the foam-cup discussion in §4.2 briefly drifts into wiping language (“sustained wiping requires contact force regulation”); align the prose with the cup-stacking task.
  2. [§4.1; Appendix A.5] Appendix A.5 Table 7 reports M=3 as best for bottle holding, while the main text (§4.1) says averaging the first two predictions; reconcile the default execution aggregation used in the main results.
  3. [Fig. 3] Fig. 3 reports force-regression RMSE/R² but does not state the force units consistently with the “g” scale used in the text; clarify units and sensor calibration briefly.
  4. [§2] Related Work could more sharply contrast Minimalist Compliance Control [41] and Zhao et al. [40] on the position-interface vs. wrench/admittance distinction already claimed in the introduction.
  5. [§3; Appendix A] Notation: q^c_ref vs. q_cmd vs. q_exec appears in multiple places; a short symbol table in the appendix would help.

Circularity Check

0 steps flagged

No derivation circularity: CRP supervision is external human commands; gains are empirical ablations, not identities by construction.

full rationale

The paper’s load-bearing chain is: (i) motor current correlates with contact (motivating Fig. 3 vs external F/T), (ii) compliance is cast as predicting a joint-position reference for a fixed PD law (Eq. 1), (iii) that reference is supervised by human teleop targets q_cmd (Eq. 2) with intent only as velocity so the label is not leaked (Eq. 3), and (iv) a current-conditioned model is compared to retargeting and a current-free CRP baseline on held-out users/tasks (Tables 1–4). None of these steps reduces a claimed prediction to its own inputs by definition. Eq. 2 is ordinary imitation-learning supervision, not a self-definition of the model’s output; the paper explicitly treats labels as noisy human corrections, not analytical force optima. The central performance claim is an empirical residual of w/ Current over w/o Current and retargeting, plus free-space demos to separate contact from internal currents—independent checks, not fitted identities renamed as predictions. Related-work citations (including Minimalist Compliance and other current-based methods) are contrastive, not load-bearing uniqueness theorems from the same authors. Residual dependence on demonstration quality is a standard IL limitation (Limitations; Appendix C), not circular derivation. Score 0.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 1 invented entities

The central claim rests on standard PD control, the empirical correlation of motor current with contact, and the decision to treat human teleop commands as CRP labels. Free parameters are ordinary training and execution hyperparameters; no new physical constants are fitted. The only invented construct is the CRP itself as the prediction target.

free parameters (5)
  • auxiliary current loss weight λ_cur = 0.1
    Set to 0.1 by default; balances reference prediction against smoothed-current reconstruction and is not derived from first principles.
  • observation/action horizons H, K = 10 / 10
    Chosen as 10 frames after informal testing of 8–16; affects temporal context available to the model.
  • KL weight λ_max_KL and annealing length = 1e-5 / 100 epochs
    Default 1e-5 annealed over 100 epochs; ablations show teleop success is sensitive to this choice.
  • execution aggregation window M and EMA weights = M≈2–3
    Number of recent CRP predictions averaged at runtime (ablated 1–4); trades jitter against load responsiveness.
  • offline current smoother (median + moving average windows) = 5 / 5
    km=ku=5 used only for auxiliary labels; affects what the latent is regularized to preserve.
axioms (4)
  • domain assumption Motor current is sufficiently correlated with actuator torque and contact resistance to serve as a learnable proprioceptive contact cue on the tested hands.
    Stated in §1–3 and Appendix C; supported by motivating force-regression plots but hardware-dependent (Limitations).
  • ad hoc to paper Human teleoperation target trajectories q_cmd are valid (noisy) supervision for the ideal compliance reference position.
    Eq. (2) and §3.2; the paper explicitly treats operator-corrected commands as CRP labels rather than analytical force optima.
  • domain assumption A standard joint-level PD controller converts position error into interaction torque that can realize compliant grasps when the reference is correctly chosen.
    Eq. (1); standard low-level interface assumed for low-cost hands.
  • ad hoc to paper ACT-style sequence modeling with raw-current encoding and auxiliary smoothed-current loss is an adequate function class for CRP prediction.
    §3.3–3.4 and Appendix A; architectural choice not derived from theory.
invented entities (1)
  • Compliance Reference Position (CRP) no independent evidence
    purpose: The joint-position target predicted from proprioception so that PD tracking error yields appropriate contact force without torque commands or external sensors.
    Core modeling construct introduced in §3.1; not a physical particle or force but a new control interface the paper defines and learns.

pith-pipeline@v1.1.0-grok45 · 19822 in / 3045 out tokens · 30564 ms · 2026-07-12T01:50:17.125251+00:00 · methodology

0 comments
read the original abstract

Compliance is essential for dexterous manipulation, yet existing solutions often rely on external tactile or force sensors that are costly, fragile, and difficult to deploy on low-cost robot hands. We propose a proprioception-driven framework that learns contact-aware compliance cues from motor current and joint states. Since motor current is closely related to actuator torque, it provides an intrinsic signal for perceiving contact force, object resistance, and grasp stability without additional sensing hardware. Rather than estimating external wrenches or commanding torque, our method predicts a compliance reference position: an ideal joint-position target for a standard PD controller whose induced position error generates appropriate grasping force. This position-based formulation is compatible with mainstream teleoperation and policy-learning pipelines, while enabling the robot to adapt interaction forces from real-time proprioceptive feedback. Thus, motor current serves not only as a force proxy but also as a learnable proprioceptive contact signal for compliance reference prediction. Experiments on multiple dexterous hands and contact-rich tasks, including fragile object handling, sustained surface contact, thin-object retrieval, and dynamic load adaptation, show stable compliant grasping, safer and more efficient teleoperation, and improved downstream policy learning without external tactile or force sensors.

Figures

Figures reproduced from arXiv: 2607.03529 by Chenyang Ma, Daniel Szafir, Mingyu Ding, Ruogu Li, Yunchao Yao, Zhenyu Wei.

Figure 1
Figure 1. Figure 1: Current as Touch for contact-rich dexterous manipulation. Our proprioception-driven compliance framework uses motor current and joint states as tactile-like contact cues, enabling com￾pliant manipulation without external tactile sensors. Across teleoperation and policy-learning set￾tings, the same compliance-reference prediction interface supports fragile object handling, sustained surface contact, thin-ob… view at source ↗
Figure 2
Figure 2. Figure 2: Motivation for Current as Touch. Rigid position control may generate excessive contact forces and damage fragile objects when target positions are inaccurate. Motor current and joint states offer built-in tactile-like proprioceptive feedback about contact and object resistance. We use this feedback to predict a compliance reference position (CRP), allowing standard PD position control to produce compliant … view at source ↗
Figure 3
Figure 3. Figure 3: Motivating measurement of motor current as proprioceptive contact feedback. On Unitree Dex3 [8] and LEAP Hand [9], motor current and joint states vary consistently with contact force measured by an external force/torque sensor. This observation motivates using intrinsic actu￾ator signals as contact feedback without treating them as explicit force estimates. However, the goal of compliant manipulation is no… view at source ↗
Figure 4
Figure 4. Figure 4: Teleoperation data manifold for learning CRPs. Demonstrations couple user intent with contact responses: command slopes encode grasp/release intent, while current changes relative to joint motion, e.g., ∆I/∆q, provide stiffness-related contact cues. The model learns CRPs from this human-corrected manifold. force and slip. The desired CRP therefore depends on the current hand state, contact condition, objec… view at source ↗
Figure 5
Figure 5. Figure 5: Current-conditioned prediction pipeline. The model predicts CRPs from proprioceptive history. In teleoperation, inputs include robot state, raw motor current, and user intent velocity; in policy mode, user intent is replaced by task-level object or goal pose. where fθ denotes the learned CRP predictor parameterized by θ, qt specifies the hand configuration, It provides contact-dependent feedback, and vinte… view at source ↗
Figure 6
Figure 6. Figure 6: Object set for teleoperation training and evaluation. We collect grasping demonstra￾tions on everyday objects with diverse shapes and stiffness, together with free-space hand-motion trajectories, to expose the model to both contact-induced and internal hand-motion currents. The full training objective, with λcur balancing the auxiliary current-prediction loss, is: L = Lref + λcurLcur. (9) This auxiliary lo… view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative results across teleoperation and policy tasks. Without current-conditioned CRP prediction, direct retargeting or behavior cloning can deform cups, lose wiping contact, draw multiple cards, or drop a dynamically loaded bottle. Motor-current feedback predicts contact-aware CRPs for fragile object handling, sustained surface contact, thin-object retrieval, and dynamic load adaptation. 4.2 Teleoper… view at source ↗

discussion (0)

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

Forward citations

Cited by 1 Pith paper

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

  1. DexVerse: A Modular Benchmark for Multi-Task, Multi-Embodiment Dexterous Manipulation

    cs.RO 2026-07 conditional novelty 6.0

    A modular benchmark of 100 dexterous manipulation tasks across 3 arms and 6 hands with 3,180 demonstrations reveals that current policies (Diffusion Policy, DP3, OpenVLA, π0.5) achieve only 34% mean success, exposing ...

Reference graph

Works this paper leans on

45 extracted references · 6 canonical work pages · cited by 1 Pith paper

  1. [1]

    Okamura, N

    A. Okamura, N. Smaby, and M. Cutkosky. An overview of dexterous manipulation. InPro- ceedings 2000 ICRA. Millennium Conference. IEEE International Conference on Robotics and Automation. Symposia Proceedings (Cat. No.00CH37065), volume 1, pages 255–262 vol.1,

    2000

  2. [2]

    doi:10.1109/ROBOT.2000.844067

    work page doi:10.1109/robot.2000.844067 2000

  3. [3]

    Z. Wei, Z. Xu, J. Guo, Y . Hou, C. Gao, Z. Cai, J. Luo, and L. Shao.D(R,O)grasp: A unified representation of robot and object interaction for cross-embodiment dexterous grasping. In 2025 IEEE International Conference on Robotics and Automation (ICRA), pages 4982–4988,

    2025

  4. [4]

    doi:10.1109/ICRA55743.2025.11127754

    doi:10.1109/icra55743.2025.11127754 2025

  5. [5]

    W. Yuan, S. Dong, and E. H. Adelson. Gelsight: High-resolution robot tactile sensors for estimating geometry and force.Sensors, 17(12):2762, 2017

    2017

  6. [6]

    Lambeta, P.-W

    M. Lambeta, P.-W. Chou, S. Tian, B. Yang, B. Maloon, V . R. Most, D. Stroud, R. Santos, A. Byagowi, G. Kammerer, et al. Digit: A novel design for a low-cost compact high-resolution tactile sensor with application to in-hand manipulation.IEEE Robotics and Automation Letters, 5(3):3838–3845, 2020

    2020

  7. [7]

    Bhirangi, T

    R. Bhirangi, T. Hellebrekers, C. Majidi, and A. Gupta. Reskin: versatile, replaceable, lasting tactile skins.arXiv preprint arXiv:2111.00071, 2021

    Pith/arXiv arXiv 2021

  8. [8]

    Bhirangi, V

    R. Bhirangi, V . Pattabiraman, E. Erciyes, Y . Cao, T. Hellebrekers, and L. Pinto. Anyskin: Plug- and-play skin sensing for robotic touch. In2025 IEEE International Conference on Robotics and Automation (ICRA), pages 16563–16570. IEEE, 2025

    2025

  9. [9]

    H. Choi, J. E. Low, T. M. Huh, S. Hong, G. A. Uribe, K. A. Hoffmann, J. Di, T. G. Chen, A. A. Stanley, and M. R. Cutkosky. Coinft: A coin-sized, capacitive 6-axis force torque sensor for robotic applications.arXiv preprint arXiv:2503.19225, 2025

    arXiv 2025

  10. [10]

    Unitree dex3-1 power control dexterous hand.https://www.unitree

    Unitree Robotics. Unitree dex3-1 power control dexterous hand.https://www.unitree. com/Dex3-1. Accessed: 2026-05-26

    2026

  11. [11]

    K. Shaw, A. Agarwal, and D. Pathak. Leap hand: Low-cost, efficient, and anthropomorphic hand for robot learning.Robotics: Science and Systems (RSS), 2023

    2023

  12. [12]

    Z. Wei, Y . Yao, and M. Ding. One hand to rule them all: Canonical representations for unified dexterous manipulation.arXiv preprint arXiv:2602.16712, 2026

    Pith/arXiv arXiv 2026

  13. [13]

    Liang, Y

    Z. Liang, Y . Mu, Y . Wang, T. Chen, W. Shao, W. Zhan, M. Tomizuka, P. Luo, and M. Ding. Dexhanddiff: Interaction-aware diffusion planning for adaptive dexterous manipulation. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 1745–1755, 2025

    2025

  14. [14]

    C. Bao, H. Xu, Y . Qin, and X. Wang. DexArt: Benchmarking generalizable dexterous ma- nipulation with articulated objects. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21190–21200, 2023

    2023

  15. [15]

    R. Wang, J. Zhang, J. Chen, Y . Xu, P. Li, T. Liu, and H. Wang. Dexgraspnet: A large-scale robotic dexterous grasp dataset for general objects based on simulation. In2023 IEEE Inter- national Conference on Robotics and Automation, pages 11359–11366, 2023

    2023

  16. [16]

    Y . Xu, W. Wan, J. Zhang, H. Liu, Z. Shan, H. Shen, R. Wang, H. Geng, Y . Weng, J. Chen, T. Liu, L. Yi, and H. Wang. UniDexGrasp: Universal robotic dexterous grasping via learning diverse proposal generation and goal-conditioned policy. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 4737–4746, 2023. 16

    2023

  17. [17]

    W. Wan, H. Geng, Y . Liu, Z. Shan, Y . Yang, L. Yi, and H. Wang. UniDexGrasp++: Improving dexterous grasping policy learning via geometry-aware curriculum and iterative generalist- specialist learning. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 3891–3902, 2023

    2023

  18. [18]

    Zhang, H

    J. Zhang, H. Liu, D. Li, X. Yu, H. Geng, Y . Ding, J. Chen, and H. Wang. Dexgraspnet 2.0: Learning generative dexterous grasping in large-scale synthetic cluttered scenes. InProceed- ings of the 8th Conference on Robot Learning, volume 270 ofProceedings of Machine Learn- ing Research. PMLR, 2025

    2025

  19. [19]

    Zhang, Q

    G. Zhang, Q. Xu, H. Zhang, J. Ma, L. He, Y . Bao, Z. Ping, Z. Yuan, C. Lu, C. Yuan, et al. Unidex: A robot foundation suite for universal dexterous hand control from egocentric hu- man videos. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1841–1852, 2026

    2026

  20. [20]

    S. Zhao, X. Zhu, Y . Chen, C. Li, Y . Xie, X. Zhang, M. Ding, and M. Tomizuka. Dexh2r: Task-oriented dexterous manipulation from human to robots.IEEE/ASME Transactions on Mechatronics, 2025

    2025

  21. [21]

    Huang, Z

    D. Huang, Z. Wei, Z. Xu, Y . Yao, S. Li, and M. Ding. Dexcompose: Reusing dexterous policies for multi-task manipulation with a single hand.arXiv preprint arXiv:2606.28323, 2026

    Pith/arXiv arXiv 2026

  22. [22]

    S. Li, S. Li, Z. Wei, Y . Yao, C. Li, and M. Ding. Coordex: Coordinating body and hand pri- ors for continuous dexterous humanoid loco-manipulation.arXiv preprint arXiv:2606.23680, 2026

    Pith/arXiv arXiv 2026

  23. [23]

    Romero, F

    B. Romero, F. Veiga, and E. Adelson. Soft, round, high resolution tactile fingertip sensors for dexterous robotic manipulation. In2020 IEEE international conference on robotics and automation (ICRA), pages 4796–4802. IEEE, 2020

    2020

  24. [24]

    C. Lin, Z. Lin, S. Wang, and H. Xu. Dtact: A vision-based tactile sensor that measures high- resolution 3d geometry directly from darkness.arXiv preprint arXiv:2209.13916, 2022

    Pith/arXiv arXiv 2022

  25. [25]

    C. Lin, H. Zhang, J. Xu, L. Wu, and H. Xu. 9dtact: A compact vision-based tactile sensor for accurate 3d shape reconstruction and generalizable 6d force estimation.IEEE Robotics and Automation Letters, 9(2):923–930, 2023

    2023

  26. [26]

    Z. Zhao, J. Tang, J. Yuan, Y . Li, Y . Dai, J. Yao, Q. Zhang, S. Ding, T. Li, R. Zhang, et al. Large-scale integrated flexible tactile sensor array for sensitive smart robotic touch.ACS nano, 16(10):16784–16795, 2022

    2022

  27. [27]

    Huang, Y

    B. Huang, Y . Wang, X. Yang, Y . Luo, and Y . Li. 3d-vitac: Learning fine-grained manipulation with visuo-tactile sensing.arXiv preprint arXiv:2410.24091, 2024

    Pith/arXiv arXiv 2024

  28. [28]

    J. Zhao, N. Kuppuswamy, S. Feng, B. Burchfiel, and E. Adelson. Polytouch: A robust multi- modal tactile sensor for contact-rich manipulation using tactile-diffusion policies. In2025 IEEE International Conference on Robotics and Automation (ICRA), pages 104–110. IEEE, 2025

    2025

  29. [29]

    N. Gu, K. Kosuge, and M. Hayashibe. Tactilealoha: Learning bimanual manipulation with tactile sensing.IEEE Robotics and Automation Letters, 10(8):8348–8355, 2025. doi:10.1109/ LRA.2025.3585396

    arXiv 2025

  30. [30]

    H. Xue, J. Ren, W. Chen, G. Zhang, Y . Fang, G. Gu, H. Xu, and C. Lu. Reactive diffusion policy: Slow-fast visual-tactile policy learning for contact-rich manipulation.arXiv preprint arXiv:2503.02881, 2025

    Pith/arXiv arXiv 2025

  31. [31]

    H. Choi, Y . Hou, C. Pan, S. Hong, A. Patel, X. Xu, M. R. Cutkosky, and S. Song. In-the-wild compliant manipulation with umi-ft.arXiv preprint arXiv:2601.09988, 2026. 17

    arXiv 2026

  32. [32]

    M. H. Raibert and J. J. Craig. Hybrid position/force control of manipulators.Journal of Dynamic Systems, Measurement, and Control, 103(2):126–133, June 1981. doi:10.1115/1. 3139652

    doi:10.1115/1 1981

  33. [33]

    N. Hogan. Impedance control: An approach to manipulation. In1984 American Control Conference, pages 304–313, 1984. doi:10.23919/ACC.1984.4788393

    doi:10.23919/acc.1984.4788393 1984

  34. [34]

    Anderson and M

    R. Anderson and M. Spong. Hybrid impedance control of robotic manipulators.IEEE Journal on Robotics and Automation, 4(5):549–556, 1988. doi:10.1109/56.20440

    work page doi:10.1109/56.20440 1988

  35. [35]

    Siciliano and L

    B. Siciliano and L. Villani.Robot F orce Control, volume 540 ofThe Springer International Series in Engineering and Computer Science. Springer New York, New York, NY , 1 edition,

  36. [36]

    doi:10.1007/978-1-4615-4431-9

    ISBN 978-1-4615-4431-9. doi:10.1007/978-1-4615-4431-9

    work page doi:10.1007/978-1-4615-4431-9

  37. [37]

    F. J. Abu-Dakka and M. Saveriano. Variable impedance control and learning—a review.Fron- tiers in Robotics and AI, 7:590681, 2020

    2020

  38. [38]

    Wahrburg, J

    A. Wahrburg, J. B ¨os, K. D. Listmann, F. Dai, B. Matthias, and H. Ding. Motor-current-based estimation of cartesian contact forces and torques for robotic manipulators and its application to force control.IEEE Transactions on Automation Science and Engineering, 15(2):879–886,

  39. [39]

    doi:10.1109/TASE.2017.2691136

    work page doi:10.1109/tase.2017.2691136 2017

  40. [40]

    S. Liu, L. Wang, and X. V . Wang. Sensorless force estimation for industrial robots using disturbance observer and neural learning of friction approximation.Robotics and Computer- Integrated Manufacturing, 71:102168, Oct. 2021. doi:10.1016/j.rcim.2021.102168

    work page doi:10.1016/j.rcim.2021.102168 2021

  41. [41]

    L. Han, J. Mao, P. Cao, Y . Gan, and S. Li. Toward sensorless interaction force estimation for industrial robots using high-order finite-time observers.IEEE Transactions on Industrial Electronics, 69(7):7275–7284, 2022. doi:10.1109/TIE.2021.3095820

    work page doi:10.1109/tie.2021.3095820 2022

  42. [42]

    S. Chen, J. Bohg, and C. K. Liu. Springgrasp: Synthesizing compliant, dexterous grasps under shape uncertainty.arXiv preprint arXiv:2404.13532, 2024

    Pith/arXiv arXiv 2024

  43. [43]

    Y . Hou, Z. Liu, C. Chi, E. Cousineau, N. Kuppuswamy, S. Feng, B. Burchfiel, and S. Song. Adaptive compliance policy: Learning approximate compliance for diffusion guided control. In2025 IEEE International Conference on Robotics and Automation (ICRA), pages 4829–

  44. [44]

    Z. Zhao, H. Dong, Z. He, Y . Li, X. Yi, and Z. Li. Closing the reality gap: Zero-shot sim-to-real deployment for dexterous force-based grasping and manipulation.arXiv preprint arXiv:2601.02778, 2026

    arXiv 2026

  45. [45]

    H. Shi, S. Hu, Y . Hou, W. Wang, K. Liu, and S. Song. Minimalist compliance control.arXiv preprint arXiv:2603.00913, 2026. 18

    arXiv 2026