Pith. sign in

REVIEW 2 major objections 5 minor 46 references

CoorGrasp: Coordinated Contact Control for Adaptive Dexterous Grasping Under Uncertainty

T0 review · 2 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read A tactile-driven model-predictive controller that coordinates all fingertip contacts during both approaching and grasping phases can execute planned dexterous grasps under shape and position uncertainty while minimizing undesired in-hand ob

desk verdict Solid engineering contribution with a real caveat: the force model assumes no object motion, and the paper never directly validates the moving-object regime, but the empirical results suggest the approximation is good enough. read the letter →

arxiv 2607.03557 v2 pith:KQLDVFEN submitted 2026-07-03 cs.RO

Mingrui Yu , Yongpeng Jiang , Yongyi Jia , Yi Ren , Xiang Li This is my paper
classification cs.RO
keywords dexterousgraspingtactilefeedbackmodelpredictivecontrolcontactcoordinationgraspexecutionwrenchbalanceuncertaintymulti-fingermanipulation
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 targets a gap in dexterous grasping: planning grasps is well-studied, but executing them under real-world uncertainty is not. The authors claim that the failure mode — multiple fingers landing at different times and pushing the object — can be tamed by coordinating contacts from the very first touch. They propose a tactile-driven model-predictive controller that separates approach from grasp based on whether the current contacts can apply a balanced wrench, uses arm motions to compensate for position errors, and reallocates fingertip forces online. If correct, this gives a plug-in execution layer for any existing grasp planner, requiring only tactile feedback, and it makes delicate tabletop grasping practical for fragile or easily tipped objects.

What carries the argument

The load-bearing identity is the motion-contact model δf = K_s J(q_d) δq_d, derived under quasi-static, linear-elastic point-contact assumptions, which lets the controller predict how a commanded joint motion will change every contact force before it happens. Around this, the controller builds two mechanisms: a phase-transition criterion that decides approach is over once the measured contacts can form a balanced wrench (normalized wrench below a threshold), and a force-allocation step that re-solves wrench balance online so new contacts are absorbed without unbalancing the grasp.

What would settle it

Attach a motion-tracking target to the object and record its six-degree-of-freedom pose throughout a real grasp with the proposed controller. If the object translates more than a few millimeters or rotates more than a degree before the phase-transition criterion declares balanced contacts, the constant-pose assumption of the motion-contact model is violated and the claimed minimization of undesired motion cannot hold.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that coordinated multi-contact execution — not better planning — is what suppresses unwanted object motion during dexterous grasping. The controller tracks the planned finger path only until its measured contacts can exert a balanced wrench; then it increases forces while continuously re-solving a wrench-balance objective. An analytical motion-contact model, δf = K_s J δq_d, links commanded joint motions to predicted force changes, and the arm is steered to absorb object-position errors that would otherwise force individual fingers into large deviations. The authors validate this in 15k simulated grasps on 478 objects with three hands and in

Load-bearing premise

The controller's predictions assume the object stays perfectly still while the fingers push on it; if the first contacts nudge or tilt the object before all fingers are in place, the force model is wrong, and the balance constraints are computed against an incorrect state.

Editorial extensions

If this is right

  • Grasp execution becomes a plug-in: the same controller can follow poses from any existing grasp generator, because it only needs tactile contact measurement, not a grasp-type-specific model.
  • Tabletop objects are disturbed far less during approach, which should make dexterous grasping practical for fragile or easily tipped objects.
  • The controller naturally handles different contact counts and locations, including contacts beyond fingertips, since the wrench-balance formulation is agnostic to where contacts are.
  • The phase-transition criterion gives a principled, online decision for when to switch from 'reach and touch' to 'squeeze', replacing hand-tuned timing or per-finger contact detection.

Reading between the lines

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

  • Our inference: the constant-pose assumption in the motion-contact model is the weakest link; replacing it with an online pose estimate from the very tactile contacts the controller measures would make the controller's predictions valid even when the object starts to move.
  • Our inference: the wrench-balance criterion is a general coordination primitive; it could be lifted out of grasping and applied to any multi-contact manipulation task where balanced interaction forces matter (e.g., assembly or in-hand reorientation).
  • Our inference: the reported performance drop at high tactile noise (0.5 scale) suggests that the practical ceiling of this approach is set by tactile sensing accuracy, not by the controller; improving tangential force estimation would likely yield bigger gains than further controller tuning.
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

2 major / 5 minor

Summary. The paper proposes CoorGrasp, a tactile-driven model predictive controller for executing planned dexterous grasps under shape and position uncertainty. The controller separates approaching and grasping phases based on a wrench-balance criterion, coordinates arm and hand motions during approaching to compensate for position errors, and adaptively allocates contact forces during grasping to maintain balance. A linear motion-contact model (Eq. 8) predicts force changes from commanded joint motions and is embedded in the MPC. The approach is evaluated in large-scale MuJoCo simulation (15k grasps, 478 objects, three hands) and in real-world experiments on eight objects with a LEAP Hand and Tac3D tactile sensors. Compared with open-loop execution, a per-finger feedback-control baseline, and two ablations, the method reports higher success rates and lower object position/rotation errors and normalized wrenches.

Significance. If the claims hold, this is a useful contribution to dexterous grasping execution: it provides a general, tactile-driven control formulation that works with arbitrary planned grasp poses and contact configurations, and it addresses coordination across fingers and arm motions in a principled way. The large-scale simulation study (15k grasps, three hands) and the real-world experiments on position-perturbed objects are notable strengths, as are the ablations isolating the contributions of arm-hand coordination and adaptive force coordination. The paper is generally clearly written and the method is reproducible in principle given the provided implementation details.

major comments (2)
  1. [§IV-B, Eq. (8); Appendix II] The force-prediction model is derived under the assumption dp_o=0 (temporally constant object pose). Under the position/shape uncertainty the paper targets, the first contacting finger can push or tilt the object before balanced contacts are established, so the omitted term -K_s dp_o and the resulting coupling among contacts are precisely the phenomena the controller aims to suppress. Constraints (10) and (14), the phase-transition optimization (20), and the wrench-balance cost all use the model-predicted f_{t+1}; if the model is biased in the intended operating regime, the MPC may command motions that continue pushing the object or fail to detect unbalanced states. The only direct validation (Appendix II, Fig. 10) uses a single finger against a fixed object and does not exercise the multi-contact, object-moving case. Since the central robustness claim depends on this approximation, plea
  2. [§V-B, Table III; Abstract] The real-world experiments do not support the abstract's unqualified claim of 'higher grasp success rates': Table III shows 40/40 and 20/20 success for all methods under both shape and position uncertainty. The real-world evidence supports reduced object movement and wrench, not improved success rate. The success-rate improvement is demonstrated only in simulation (Table II). Please qualify the claim accordingly, e.g., 'higher success rates in simulation and reduced object movement in real-world experiments.'
minor comments (5)
  1. [Table II] The LEAP Hand 'Wrench' entry for 'Ours' reports 0.18±1.15, while the corresponding rows for the other hands are around 0.17±0.17. This appears to be a typographical error (likely 0.18±0.17); please check.
  2. [Appendix I-A] The text says 'apply the linear-squares method'; this should be 'least-squares method.'
  3. [§V-B, Fig. 7 and Table III] The real-world SR column, being saturated at 100% for all methods, gives no discriminative information. Consider reporting per-object results or a different metric (e.g., lift success under perturbation) to make the real-world comparison more informative.
  4. [§IV-B, Eq. (6)] The notation p_o,d is not explicitly defined before it is used in Eq. (6). Defining p_o,d = p_d - p_o (and its derivative) would remove ambiguity in the derivation.
  5. [Appendix I-D] The paper states that tangential force magnitudes are 'reasonable' but 'not quantitatively evaluated.' Since the wrench-balance cost and phase-transition criterion rely on tangential forces, a sentence on how this limitation could affect the quantitative results would be helpful.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the motion-contact model is derived from compliance and wrench balance, the controller parameters are physical or hand-set rather than fit to the evaluation metrics, and self-citations are contextual only.

full rationale

The paper's derivation chain is self-contained. The motion-contact model (Eqs. 3-8) follows from the linear-elastic contact law f = K_o(p_e-p_o), the quasi-static torque balance J^T f = K_p(q_d-q), and forward kinematics; the reduction to Eq. (8) explicitly states the temporary constant-object-pose assumption (dp_o=0). The joint stiffness K_p is identified in Appendix I from quasi-static pressing data as an independent physical parameter, while K_o, F_appr, epsilon_b, and the W weights are manually set hyperparameters; none are fitted to the Section V success rates or object-motion metrics. The MPC constraints (10), (14) use the model only as a predictive input, and the costs (9), (13) minimize guiding-path deviation and wrench imbalance, not the evaluation outcomes themselves. The main claims are supported by comparisons with open-loop, feedback-control, no-arm-motion, and independent-force baselines using success rate and independently measured position/rotation errors. Self-citations [2], [4] are contextual references to dexterous manipulation and carry no load-bearing justification; no uniqueness theorem or ansatz is imported from the authors' prior work. Appendix II's admission that the quasi-static and constant-object-pose assumptions 'are not fully satisfied in real-world scenarios' identifies a model-accuracy limitation under the targeted uncertainty, i.e., a correctness risk, not a circular reduction: the model is an input to the controller, not a restatement of the measured outcomes. The normalized-wrench evaluation metric aligns with the controller's wrench-balance objective, but it is reported alongside independent success and object-motion metrics and is not a fitted parameter disguised as a prediction. No circular step can be exhibited, so the score is 0.

Assumptions & free parameters 8 free parameters · 8 assumptions · 0 invented entities

The central claim rests on a linearized quasi-static contact model and tactile observability assumptions, all acknowledged by the authors as approximations. The hand-tuned hyperparameters and identified K_p are inputs rather than quantities fitted to evaluation outcomes. No new physical entities are introduced.

free parameters (8)
  • F_appr = 0.2 N
    Threshold limiting total normal force during the approaching phase; chosen by hand (Appendix I-B).
  • epsilon_b = 0.2
    Phase-transition threshold on normalized wrench; chosen by hand (Appendix I-B).
  • mu_hat = 0.3 (sim) / 0.7 (real)
    Estimated friction coefficient used in friction-cone constraints; set conservatively below actual values.
  • K_o = 1e5 (sim) / 1e4 (real)
    Object stiffness used in the motion-contact model; chosen by hand (Appendix I-B).
  • K_p = ~0.8 (real, all joints); ground truth in MuJoCo XML (sim)
    Joint position stiffness identified via least-squares from pressing data in Appendix I-A; an input model parameter, not fit to evaluation outcomes.
  • F_ub = not determined by the paper
    Pre-defined final grasp force; the paper explicitly does not solve how to set it. For fair comparison, baselines matched the averaged final force observed in open-loop execution.
  • Pre-grasp extrapolation factor = 3
    Pre-grasp pose is recomputed as a linear extrapolation three times the original distance to reduce collision risk under position uncertainty (Appendix I-C).
  • MPC cost weights (W_g, W_p, W_t) = W_q=1.0/0; W_p=diag(0,0,100,10,10,10); W_t=diag(0,1000,1000)
    Weights in J_guide, J_palm, and J_tm; hand-selected values given in Appendix I-B.
assumptions (8)
  • domain assumption All hand-object contacts are point contacts with friction (Section III, Assumption 1).
    Standard grasp-modeling assumption but restricts validity for soft or large-area contacts.
  • domain assumption Location, normal, and force of each contact can be measured by tactile sensors in real time (Section III, Assumption 2).
    In the real system only fingertip pads have Tac3D sensors, so this is only partially satisfied and limits real experiments to fingertip grasps.
  • domain assumption Fingers are driven by joint-space PD controllers with known stiffness K_p (Section III, Assumption 3; Appendix I-A).
    The LEAP Hand's current-based PID control is only approximately equivalent to a linear position controller with constant K_p, as the authors note.
  • domain assumption The manipulation process is quasi-static and contacts are linear-elastic (Section IV-B; Appendix II).
    The motion-contact model Eq. (3)-(8) relies on this; dynamic effects and nonlinear material behavior are neglected.
  • domain assumption The object pose is temporally constant during force prediction, dp_o = 0 (Eq. 8; Appendix II).
    This is the most fragile assumption because the method targets uncertainty that can cause exactly the object motion the model ignores.
  • domain assumption Object stiffness is much larger than robot joint stiffness, so dq is much smaller than dq_d (Section IV-B).
    Used to approximate d p_e,d using J(q_d), simplifying K_s in Eq. (7)-(8).
  • domain assumption The planned pre-grasp, grasp, and squeeze poses are valid, and fingers required for grasp contact the object at the planned squeeze pose (Section III, Assumption 4).
    The controller guides toward these poses; if the planner's poses are badly wrong, the method inherits the error.
  • ad hoc to paper Real LEAP Hand motors can be approximated as linear position-controlled joints with a constant identified K_p, despite nonlinear current-torque behavior (Appendix I-A).
    The authors acknowledge this is an approximation of the actual motor mechanism and that it affects force tracking.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CoorGrasp: Coordinated Contact Control for Adaptive Dexterous Grasping Under Uncertainty." pith.science (2026). https://pith.science/paper/KQLDVFEN

@misc{pith2026260703557,
  author       = {Pith},
  title        = {Pith review of: CoorGrasp: Coordinated Contact Control for Adaptive Dexterous Grasping Under Uncertainty},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KQLDVFEN}},
  note         = {Machine review of arXiv:2607.03557}
}
read the original abstract

While recent research has focused heavily on dexterous grasp pose generation, less attention has been devoted to the execution of planned grasps. Under shape and position uncertainty, open-loop execution often yields uncoordinated contacts, causing undesired in-hand object motion and even grasp failures. To address this, this paper proposes a tactile-driven model predictive controller for adaptive and delicate execution of diverse dexterous grasps. Our approach emphasizes multi-contact coordination across both approaching and grasping phases, with three key novelties: (i) coordination-aware phase separation, (ii) arm-hand coordination to compensate for position errors, and (iii) adaptive force coordination to increase contact forces in a balanced manner. An analytical model is employed to relate contact forces to robot joint motions for predictive control. Our formulation imposes no restrictions on grasp types or contact configurations and integrates seamlessly with state-of-the-art grasp pose generation methods. We validate the approach through large-scale simulations involving 15k grasps across 478 objects on three robotic hands, and real-world experiments on 8 objects. Results demonstrate that our method achieves higher grasp success rates and reduced undesired object movements.

Figures

Figures reproduced from arXiv: 2607.03557 by the authors.

Figure 1
Figure 1. Execution of planned dexterous grasps. Due to uncertainty, open￾loop execution of imperfect planned grasp poses can cause unintended in￾hand object movements or grasp failures. This work proposes a tactile-driven control approach that coordinates multiple contacts during both approaching and grasping, reducing undesired object motion and enabling adaptive, delicate execution of diverse dexterous grasps. geometry; an… view at source ↗
Figure 2
Figure 2. Overview of our tactile-driven coordinated contact control method for adaptive execution of planned grasp poses generated from observations with uncertainty. Our method employs a coordination-aware separation of the approaching and grasping phases, using the criteria of wrench balance. During the approaching phase, the fingers make contact with the object using gentle forces, while coordinated arm motions compensate… view at source ↗
Figure 3
Figure 3. Two cases to illustrate the issues when only using finger motions to compensate for object position errors. The baseline strategy is to stop the first-contacting fingers and actuate the remaining fingers until all fingers make contact, without involving arm motions (top-right figures of each case). It may lead to issues that (a) the fingers reach an unstable grasp configuration; or (b) the remaining fingers fail to … view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Illustration of the mathematical formulation of contacts. The undeformed, actual, and desired positions of the i th contact as well as the contact normals and forces of the j th contact are shown [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Visualization of some grasps in our large-scale simulation evaluation. The evaluation totally involves 15k grasps on 478 objects and three dexterous hands. These figures show the grasp states after lifting from a table, achieved using our approach. TABLE II: Large-scal…
Figure 6
Figure 6. Figure 6: Simulation evaluation results under object position uncertainty, averaged across the three hands. “0cm" refers to no position errors, and “2cm" refers to perturbing the positions by 2 cm along eight uniformly distributed planar directions respectively. Baselines: We ch…
Figure 7
Figure 7. Figure 7: Real-world experiments include comparison of our method with baselines on eight different everyday objects and planned grasps. The AprilTag markers are only used for quantitative evaluation of undesired object movements. TABLE III: Quantitative results of the real-worl…
Figure 8
Figure 8. Figure 8: Comparison between the methods under position uncertainty. 4) unbalanced forces from multiple contacts lead to large undesired object movements, even worse than open-loop executions that preserves kinematic-level finger coordination. 2) Evaluation with Position Uncerta…
Figure 9
Figure 9. Figure 9: Hardware setup for real-world experiments. We use Tac3D, a vision-based tactile sensor, to measure the contact states. position, and the global contact position p i is obtained through the robot’s forward kinematics. We experimentally find that the sensed normal force …
Figure 10
Figure 10. Figure 10: Real-world test of the motion-contact model. The X-axis represents the normal force, while the Y- and Z-axes represent the friction forces [PITH_FULL_IMAGE:figures/full_fig_p009_10.png]
Figure 11
Figure 11. Figure 11: Manipulation processes for grasping the position-perturbed mosquito repellent bottle ( [PITH_FULL_IMAGE:figures/full_fig_p011_11.png]

Review history (2 revisions) →

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

46 extracted references · 4 linked inside Pith

  1. [1]

    Robotic manipulation of hand tools: The case of screwdriving,

    L. Tang, Y .-B. Jia, and Y . Xue, “Robotic manipulation of hand tools: The case of screwdriving,” inIEEE Int. Conf. Robot. Autom., 2024, pp. 13 883–13 890

    2024

  2. [2]

    Contact-implicit model predictive control for dexterous in-hand manipulation: A long- horizon and robust approach,

    Y . Jiang, M. Yu, X. Zhu, M. Tomizuka, and X. Li, “Contact-implicit model predictive control for dexterous in-hand manipulation: A long- horizon and robust approach,” inIEEE/RSJ Int. Conf. Intell. Robots Syst., 2024, pp. 5260–5266

    2024

  3. [3]

    Multi-finger manipulation via trajectory optimization with differentiable rolling and geometric constraints,

    F. Yang, T. Power, S. A. Marinovic, S. Iba, R. S. Zarrin, and D. Berenson, “Multi-finger manipulation via trajectory optimization with differentiable rolling and geometric constraints,”IEEE Robot. Autom. Lett., 2025

    2025

  4. [4]

    Robotic in-hand manipulation for large-range precise object movement: The rgmc champion solution,

    M. Yu, Y . Jiang, C. Chen, Y . Jia, and X. Li, “Robotic in-hand manipulation for large-range precise object movement: The rgmc champion solution,”IEEE Robot. Autom. Lett., vol. 10, no. 5, pp. 4738–4745, 2025

    2025

  5. [5]

    Adaptive grasping by multi- fingered hand with tactile sensor based on robust force and position control,

    T. Takahashi, T. Tsuboi, T. Kishida, Y . Kawanami, S. Shimizu, M. Iribe, T. Fukushima, and M. Fujita, “Adaptive grasping by multi- fingered hand with tactile sensor based on robust force and position control,” inIEEE Int. Conf. Robot. Autom., 2008, pp. 264–271

    2008

  6. [6]

    Dexterous grasping under shape uncertainty,

    M. Li, K. Hang, D. Kragic, and A. Billard, “Dexterous grasping under shape uncertainty,”Robot. Auton. Syst., vol. 75, pp. 352–364, 2016

    2016

  7. [7]

    Grasping force control of multi-fingered robotic hands through tactile sensing for object stabilization,

    Z. Deng, Y . Jonetzko, L. Zhang, and J. Zhang, “Grasping force control of multi-fingered robotic hands through tactile sensing for object stabilization,”Sensors, vol. 20, no. 4, p. 1050, 2020

    2020

  8. [8]

    Tactile-driven gentle grasping for human-robot collaborative tasks,

    C. J. Ford, H. Li, J. Lloyd, M. G. Catalano, M. Bianchi, E. Pso- mopoulou, and N. F. Lepora, “Tactile-driven gentle grasping for human-robot collaborative tasks,” inIEEE Int. Conf. Robot. Autom., 2023, pp. 10 394–10 400

    2023

Show all 46 references
  1. [9]

    Shear-based grasp control for multi-fingered underactuated tactile robotic hands,

    C. J. Ford, H. Li, M. G. Catalano, M. Bianchi, E. Psomopoulou, and N. F. Lepora, “Shear-based grasp control for multi-fingered underactuated tactile robotic hands,”IEEE Trans. Robot., 2025

    2025

  2. [10]

    An adaptive compliant multi-finger approach- to-grasp strategy for objects with position uncertainties,

    Z. Chen, T. Wimböck, M. A. Roa, B. Pleintinger, M. Neves, C. Ott, C. Borst, and N. Y . Lii, “An adaptive compliant multi-finger approach- to-grasp strategy for objects with position uncertainties,” inIEEE Int. Conf. Robot. Autom., 2015, pp. 4911–4918

    2015

  3. [11]

    Multi-fingered tactile servoing for grasping adjustment under partial observation,

    H. Liu, B. Huang, Q. Li, Y . Zheng, Y . Ling, W. Lee, Y . Liu, Y .-Y . Tsai, and C. Yang, “Multi-fingered tactile servoing for grasping adjustment under partial observation,” inIEEE/RSJ Int. Conf. Intell. Robots Syst., 2022, pp. 7781–7788

    2022

  4. [12]

    Multifingered grasping based on multimodal reinforcement learning,

    H. Liang, L. Cong, N. Hendrich, S. Li, F. Sun, and J. Zhang, “Multifingered grasping based on multimodal reinforcement learning,” IEEE Robot. Autom. Lett., vol. 7, no. 2, pp. 1174–1181, 2021

    2021

  5. [13]

    Learning adaptive grasping from human demonstrations,

    S. Wang, W. Hu, L. Sun, X. Wang, and Z. Li, “Learning adaptive grasping from human demonstrations,”IEEE/ASME Trans. Mechatron- ics, vol. 27, no. 5, pp. 3865–3873, 2022

    2022

  6. [14]

    Synthesizing diverse and physically stable grasps with arbitrary hand structures using differentiable force closure estimator,

    T. Liu, Z. Liu, Z. Jiao, Y . Zhu, and S.-C. Zhu, “Synthesizing diverse and physically stable grasps with arbitrary hand structures using differentiable force closure estimator,”IEEE Robot. Autom. Lett., vol. 7, no. 1, pp. 470–477, 2021

    2021

  7. [15]

    Dexgraspnet: A large-scale robotic dexterous grasp dataset for general objects based on simulation,

    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,” inIEEE Int. Conf. Robot. Autom., 2023, pp. 11 359–11 366

    2023

  8. [16]

    Gendexgrasp: Generalizable dexterous grasping,

    P. Li, T. Liu, Y . Li, Y . Geng, Y . Zhu, Y . Yang, and S. Huang, “Gendexgrasp: Generalizable dexterous grasping,” inIEEE Int. Conf. Robot. Autom., 2023, pp. 8068–8074

    2023

  9. [17]

    Bodex: Scalable and efficient robotic dexterous grasp synthesis using bilevel optimization,

    J. Chen, Y . Ke, and H. Wang, “Bodex: Scalable and efficient robotic dexterous grasp synthesis using bilevel optimization,”arXiv preprint arXiv:2412.16490, 2024

    2024 arXiv

  10. [18]

    Dexonomy: Synthesiz- ing all dexterous grasp types in a grasp taxonomy,

    J. Chen, Y . Ke, L. Peng, and H. Wang, “Dexonomy: Synthesiz- ing all dexterous grasp types in a grasp taxonomy,”arXiv preprint arXiv:2504.18829, 2025

    2025 arXiv

  11. [19]

    Unidexgrasp: Universal robotic dexterous grasping via learning diverse proposal generation and goal- conditioned policy,

    Y . Xu, W. Wan, J. Zhang, H. Liu, Z. Shan, H. Shen, R. Wang, H. Geng, Y . Weng, J. Chenet al., “Unidexgrasp: Universal robotic dexterous grasping via learning diverse proposal generation and goal- conditioned policy,” inProc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit., 2023...

    2023

  12. [20]

    Learning human-like functional grasping for multi-finger hands from few demonstrations,

    W. Wei, P. Wang, S. Wang, Y . Luo, W. Li, D. Li, Y . Huang, and H. Duan, “Learning human-like functional grasping for multi-finger hands from few demonstrations,”IEEE Trans. Robot., 2024

    2024

  13. [21]

    D (r, o) grasp: A unified representation of robot and object interaction for cross-embodiment dexterous grasping,

    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,” inIEEE Int. Conf. Robot. Autom., 2025

    2025

  14. [22]

    Springgrasp: Synthesizing com- pliant, dexterous grasps under shape uncertainty,

    S. Chen, J. Bohg, and C. K. Liu, “Springgrasp: Synthesizing com- pliant, dexterous grasps under shape uncertainty,”arXiv preprint arXiv:2404.13532, 2024

    2024 arXiv

  15. [23]

    Unidexgrasp++: Improving dexterous grasping policy learning via geometry-aware curriculum and iterative generalist-specialist learn- ing,

    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 learn- ing,” inProc. IEEE/CVF Int. Conf. Comput. Vis., 2023, pp. 3891– 3902

    2023

  16. [24]

    Dexpoint: Gener- alizable point cloud reinforcement learning for sim-to-real dexterous manipulation,

    Y . Qin, B. Huang, Z.-H. Yin, H. Su, and X. Wang, “Dexpoint: Gener- alizable point cloud reinforcement learning for sim-to-real dexterous manipulation,” inConf. Robot. Learn., 2023, pp. 594–605

    2023

  17. [25]

    Dexrepnet: Learning dexterous robotic grasping network with geo- metric and spatial hand-object representations,

    Q. Liu, Y . Cui, Q. Ye, Z. Sun, H. Li, G. Li, L. Shao, and J. Chen, “Dexrepnet: Learning dexterous robotic grasping network with geo- metric and spatial hand-object representations,” inIEEE/RSJ Int. Conf. Intell. Robots Syst., 2023, pp. 3153–3160

    2023

  18. [26]

    Grasp adjustment on novel objects using tactile experience from similar local geometry,

    H. Dang and P. K. Allen, “Grasp adjustment on novel objects using tactile experience from similar local geometry,” inIEEE/RSJ Int. Conf. Intell. Robots Syst., 2013, pp. 4007–4012

    2013

  19. [27]

    Avita: Adaptive visual-tactile dexter- ous grasping,

    H. Yan, H.-S. Fang, and C. Lu, “Avita: Adaptive visual-tactile dexter- ous grasping,”IEEE Robot. Autom. Lett., 2024

    2024

  20. [28]

    Learning of grasp adaptation through experience and tactile sensing,

    M. Li, Y . Bekiroglu, D. Kragic, and A. Billard, “Learning of grasp adaptation through experience and tactile sensing,” inIEEE/RSJ Int. Conf. Intell. Robots Syst., 2014, pp. 3339–3346

    2014

  21. [29]

    Hierarchical fingertip space: A unified framework for grasp planning and in-hand grasp adaptation,

    K. Hang, M. Li, J. A. Stork, Y . Bekiroglu, F. T. Pokorny, A. Billard, and D. Kragic, “Hierarchical fingertip space: A unified framework for grasp planning and in-hand grasp adaptation,”IEEE Trans. Robot., vol. 32, no. 4, pp. 960–972, 2016

    2016

  22. [30]

    Adaptive fingers coordination for robust grasp and in-hand manipulation under disturbances and unknown dynamics,

    F. Khadivar and A. Billard, “Adaptive fingers coordination for robust grasp and in-hand manipulation under disturbances and unknown dynamics,”IEEE Trans. Robot., vol. 39, no. 5, pp. 3350–3367, 2023

    2023

  23. [31]

    A learning-based con- troller for multi-contact grasps on unknown objects with a dexterous hand,

    D. Winkelbauer, R. Triebel, and B. Bäuml, “A learning-based con- troller for multi-contact grasps on unknown objects with a dexterous hand,” inIEEE/RSJ Int. Conf. Intell. Robots Syst., 2024, pp. 7331– 7337

    2024

  24. [32]

    GraspXL: Generating grasping motions for diverse objects at scale,

    H. Zhang, S. Christen, Z. Fan, O. Hilliges, and J. Song, “GraspXL: Generating grasping motions for diverse objects at scale,” inEuropean Conference on Computer Vision (ECCV), 2024

    2024

  25. [33]

    RobustDex- Grasp: Robust dexterous grasping of general objects,

    H. Zhang, Z. Wu, L. Huang, S. Christen, and J. Song, “RobustDex- Grasp: Robust dexterous grasping of general objects,” inConference on Robot Learning (CoRL), 2025

    2025

  26. [34]

    Planning optimal grasps,

    C. Ferrari, J. Cannyet al., “Planning optimal grasps,” inIEEE Int. Conf. Robot. Autom., vol. 3, 1992, pp. 2290–2295

    1992

  27. [35]

    Model predictive interaction control for robotic manipulation tasks,

    T. Gold, A. Völz, and K. Graichen, “Model predictive interaction control for robotic manipulation tasks,”IEEE Trans. Robot., vol. 39, no. 1, pp. 76–89, 2022

    2022

  28. [36]

    Active stiffness control of a manipulator in cartesian coordinates,

    J. K. Salisbury, “Active stiffness control of a manipulator in cartesian coordinates,” inIEEE Conf. Decis. Control, 1980, pp. 95–100

    1980

  29. [37]

    Force estimation and slip detec- tion/classification for grip control using a biomimetic tactile sensor,

    Z. Su, K. Hausman, Y . Chebotar, A. Molchanov, G. E. Loeb, G. S. Sukhatme, and S. Schaal, “Force estimation and slip detec- tion/classification for grip control using a biomimetic tactile sensor,” inIEEE-RAS Int. Conf. Humanoid Robots, 2015, pp. 297–303

    2015

  30. [38]

    Mujoco: A physics engine for model-based control,

    E. Todorov, T. Erez, and Y . Tassa, “Mujoco: A physics engine for model-based control,” inIEEE/RSJ Int. Conf. Intell. Robots Syst., 2012, pp. 5026–5033

    2012

  31. [39]

    Leap hand: Low-cost, efficient, and anthropomorphic hand for robot learning,

    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

  32. [40]

    Tac3d: A novel vision-based tactile sensor for measuring force distribution and estimating friction coefficient distribution,

    L. Zhang, Y . Wang, and Y . Jiang, “Tac3d: A novel vision-based tactile sensor for measuring force distribution and estimating friction coefficient distribution,”arXiv preprint arXiv:2202.06211, 2022. APPENDIXI IMPLEMENTATIONDETAILS A. Identification of Robot Joint Stiffness I...

    2022 arXiv

  33. [41]

    The manipulation process is quasi-static

  34. [42]

    0 2 4 6 8 10 12 Time (s) −0.4 −0.2 0.0 0.2 0.4 Change of contact force (N) X (sensed) X (pred) Y (sensed) Y (pred) Z (sensed) Z (pred) Fig

    The object pose is assumed to be temporally constant. 0 2 4 6 8 10 12 Time (s) −0.4 −0.2 0.0 0.2 0.4 Change of contact force (N) X (sensed) X (pred) Y (sensed) Y (pred) Z (sensed) Z (pred) Fig. 10:Real-world test of the motion-contact model.The X-axis represents the normal for...

  35. [43]

    Contacts are modeled as point contacts with friction, using a first-order analysis that ignores second-order contact curvatures and frictional moments

  36. [44]

    No slippage occurs temporally

  37. [45]

    Robot joint positions are controlled by low-level PD controllers, with joint torques as the control inputs

  38. [46]

    These assumptions are not fully satisfied in real-world sce- narios, making the model a coarse approximation of actual contact dynamics

    The model of robot kinematics is accurate. These assumptions are not fully satisfied in real-world sce- narios, making the model a coarse approximation of actual contact dynamics. Nevertheless, we find that its accuracy is acceptable to enable model predictive grasping control...

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.