Pith. sign in

REVIEW 3 major objections 5 minor 58 references

AnyDexRT maps human fingertip motion to diverse robot hands without calibration, using self-supervised shape matching plus a few human anchors.

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

2026-07-10 09:04 UTC pith:ZGCRZI45

load-bearing objection Solid engineering fix for GeoRT-style global matching: few-shot anchors + partial Chamfer + local motion give real multi-hand LMC gains and usable teleop, with scope limited mainly to one real hand. the 3 major comments →

arxiv 2607.08341 v1 pith:ZGCRZI45 submitted 2026-07-09 cs.RO

AnyDexRT: Calibration-Free Dexterous Hand Retargeting with Few-Shot Human Guidance

Chenxi Wang , Ying Feng , Hongjie Fang , Shangning Xia , Lixin Yang , Chuan Wen , Cewu Lu This is my paper
classification cs.RO
keywords dexterous hand retargetingteleoperationfingertip correspondencefew-shot human guidancecalibration-freecontact classifiermotion consistency
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.

Controlling a multi-finger robot hand by teleoperation only works if human hand motion is turned into robot joint commands that feel natural and stay feasible. Prior retargeting either needs hand-crafted objectives and careful calibration, or forces a global shape match that can drag the mapping into redundant robot regions with no human counterpart. AnyDexRT instead treats fingertip trajectories as manifolds to be matched in one direction: it learns a fingertip map with partial Chamfer, distance-preservation, and local-motion losses, then anchors ambiguous regions with a few paired human–robot gestures the operator can collect in under two minutes. A small contact classifier further corrects pinch poses when glove sensors fail to register contact. Across seven human-like hands in simulation and real teleoperation on one of them, the method raises local motion consistency, cuts hyperparameters, resists frame misalignment, and shortens task times while raising pinch success. The claim is that this combination gives intuitive, calibration-free control for data collection and real-time teleoperation without hand-specific redesign.

Core claim

AnyDexRT shows that self-supervised one-way fingertip shape correspondence, stabilized by few-shot human–robot anchors and optional contact refinement, produces more intuitive and stable retargeting across human-like dexterous hands than optimization-based task-vector methods or global Chamfer alignment, without precise coordinate calibration.

What carries the argument

The fingertip mapper fm, trained with partial Chamfer, pairwise distance preservation, local-frame motion consistency, and few-shot anchor alignment losses; optionally refined at inference by a contact classifier that snaps mapped pinch poses to nearby contact templates.

Load-bearing premise

The method assumes that for human-like hands, fingertip targets plus natural joint synergies are enough to uniquely determine usable robot joint commands, so improving the fingertip map is what mainly improves teleoperation.

What would settle it

On a human-like hand with substantial joint redundancy or under-actuation, measure whether high local motion consistency of AnyDexRT still yields faster, more successful teleoperation than baselines when operators must execute contact-rich tasks; if operators report ambiguous or uncontrollable poses despite high LMC, the fingertip-map premise 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. AnyDexRT proposes a calibration-free kinematic retargeting pipeline for human-like dexterous hands. It learns a per-finger fingertip map fm via self-supervised partial Chamfer, pairwise distance preservation, and local-frame motion consistency, anchors the map with few-shot paired human–robot gesture anchors, and optionally refines pinch poses with a contact classifier before converting fingertip targets to joints (NNS/IK). The paper claims this yields more intuitive, stable, and low-tuning teleoperation than optimization-based retargeting and GeoRT, with multi-seed simulation on seven morphologies (GMC/LMC, calibration-rotation stress, stability) and a real multi-operator teleop study on Wuji Hand (four tasks, completion time and pinch success).

Significance. If the results hold, the work is a useful systems contribution for dexterous teleoperation and demonstration collection: it reduces hand-specific calibration and hyperparameter burden while improving local motion consistency and real-task efficiency on a high-DoF hand. Strengths include multi-seed evaluation across seven morphologies (Table 1, Fig. 6), an explicit calibration-rotation stress test (Fig. 5), ablations of each loss term (Table 3 / Fig. 3), and a multi-operator real teleop study with pinch success (Table 4). The design choices (partial rather than full Chamfer; local rather than global motion preservation; sparse anchors) are well motivated against GeoRT’s failure modes and are of practical interest to the teleoperation and imitation-learning communities.

major comments (3)
  1. §3.1 (A1) and §4.2: Generality claim (R3) is only partially supported. Simulation covers seven hands, but real teleoperation (Table 4, Fig. 8) is only on Wuji Hand with NNS-based fs. Underactuation, joint coupling, or non-unique IK can break the assumption that LMC gains on fm translate to controllable teleop. The paper should either (i) add real teleop on at least one additional morphology, or (ii) clearly scope R3 to simulation-plus-one-hand and discuss when A1 fails.
  2. §3.2 Eq. (3) and §4.1 metrics: LMC is both the primary evaluation metric and closely aligned with the optimized L_motion objective (local directional consistency). GMC is secondary and sometimes lower for AnyDexRT (e.g., Leap Hand GMC 54.5 vs GeoRT 73.4 in Table 1). The paper should report an independent held-out proxy of intuitiveness (e.g., operator preference / NASA-TLX, or task success under fixed time budget without contact snap-in) so that gains are not largely explained by optimizing the reported metric.
  3. §3.4 and Table 4 Pick-10: Pinch success (62%) is a main real-world claim, but the contact classifier + template nearest-neighbor snap-in is a discrete post-process, not a pure geometric map. Ablate or report Pick-10 without fc (mapper-only) so readers can separate correspondence quality from contact refinement; otherwise the comparison to GeoRT/optimization on pinch is confounded.
minor comments (5)
  1. §3.2: Clarify how local frames T(x) are defined for human samples and how nearest-neighbor robot rotations are assigned when CR is sparse; a short pseudocode block would help reproducibility.
  2. Appendix B: Loss is written as unweighted sum of four terms with no sensitivity study; a brief note on whether reweighting is needed across hands would strengthen the “3 hyperparameters / rarely tuned” claim in Table 2.
  3. Table 1: Report units/normalization for GMC/LMC more explicitly in the caption (values are ×10−2); also state whether human test trajectories are held out from training samples.
  4. Fig. 3 and Fig. 8: Qualitative figures are informative but would benefit from consistent color legends and a short description of which finger is shown when multi-finger spaces are plotted.
  5. §5 Limitations: The call for downstream imitation learning is appropriate; even a small policy-learning pilot on collected demos would strengthen the data-collection motivation stated in the introduction.

Circularity Check

1 steps flagged

No significant circularity: empirical method paper whose losses are designed regularizers; LMC is closely related to L_motion but independent real-world and GMC metrics remain.

specific steps
  1. fitted input called prediction [§3.2 Eq. (3) L_motion; §4.1 LMC definition and Tab. 1]
    "We define the local motion loss as L_motion(C^{H,i}) = −1/|C^{H,i}| ∑ ⟨T^{-1}(x^{H,i}_j) Δx/∥Δx∥ , T^{-1}(f^i_m(x^{H,i}_j)) Δf^i_m / ∥Δf∥⟩ ... AnyDexRT achieves strong performance across different hand embodiments, improving the average local motion consistency from 59.8% to 90.2%."

    L_motion maximizes average local directional cosine similarity of fingertip displacements. LMC is defined as the corresponding local-frame motion consistency. Training therefore directly optimizes the quantity later reported as the headline quality metric; the large LMC gains are statistically forced by the objective rather than an independent out-of-sample prediction of a distinct physical or kinematic quantity. (Mitigated by GMC, real-world task metrics, and ablations that remain informative.)

full rationale

AnyDexRT is a standard robotics/ML methods paper. It defines geometric losses (partial Chamfer, distance preservation, local motion, few-shot alignment) plus a contact classifier, trains fingertip mappers, and evaluates on motion-consistency metrics, seed stability, calibration perturbations, hyperparameter count, and multi-operator real-world task times/pinch success. Nothing is claimed as a first-principles derivation or uniqueness theorem. The only mild alignment is that L_motion (Eq. 3) directly optimizes local directional cosine similarity, which is essentially what LMC reports; high LMC after training is therefore partly by construction of the objective. This is ordinary supervised design, not a definitional loop or fitted-parameter-as-prediction of a distinct quantity. GMC (not directly optimized), ablations (Tab. 3), cross-hand seed variance (Fig. 6), rotation robustness (Fig. 5), and independent teleoperation times/pinch rates (Tab. 4) supply external content. No self-citation is load-bearing for a uniqueness claim, no ansatz is smuggled, and no known empirical pattern is merely renamed. Score remains low (1) because the central claims of improved teleoperation quality and reduced tuning rest on those independent measurements rather than on redefinition.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 2 invented entities

Load-bearing content is mostly standard robotics geometry plus two domain assumptions about human-like hands, sparse operator anchors, and several hand-chosen training/inference constants. No new physical entities are postulated; the ‘entities’ are learned maps and a contact head. Free parameters are few and mostly defaults the authors claim transfer across hands.

free parameters (5)
  • Number of human-guided anchors (K0=5 per type; interpolated K=50 lateral / K=100 bending)
    Sparse supervision budget chosen by authors; anchors pin scale/translation and resolve multi-modal maps. Central stability claims depend on this few-shot set.
  • Bending synergy ratio λ=2 (β1=β2=λ β3)
    Hand-chosen constant for collecting bending anchors (authors note literature also uses 3/2); affects which human poses are paired to robot anchors.
  • Contact classifier threshold 0.5 and pinch template nearest-neighbor search
    Inference rule that converts classifier scores into refined pinch poses; directly affects Pick-10 success rates.
  • MLP widths/depths and unweighted sum of four mapping losses
    Architecture and equal loss sum are fixed design choices; authors state weights are not tuned, but the equal-sum choice still shapes the learned map.
  • Local-frame assignment via nearest robot sample for L_motion
    Heuristic for defining robot local rotations when fm outputs only positions; affects the local motion objective that drives LMC gains.
axioms (5)
  • domain assumption (A1) Human-like robot hands have stable finger coupling/synergies so fingertip targets sufficiently constrain IK (fs approximately one-to-one).
    Stated in §3.1; justifies focusing almost all learning on fm and using NNS/IK for joints.
  • domain assumption (A2) After a suitable geometric transform, human fingertip motion space is covered by the robot fingertip space (one-way coverage only).
    §3.1; motivates partial Chamfer instead of bidirectional global matching used by GeoRT.
  • ad hoc to paper Partial Chamfer, pairwise distance preservation, and local directional consistency are appropriate self-supervised proxies for intuitive teleoperation.
    §3.2 design; supported by ablations but not derived from a formal human-control model.
  • ad hoc to paper Few paired gesture anchors collected in <2 min adequately disambiguate redundant robot reachable regions for task-relevant teleop.
    §3.3 and Appendix D; core of the ‘calibration-free / few-shot’ claim.
  • standard math Standard point-set / Chamfer geometry and MLP function approximation apply to fingertip clouds.
    Background used throughout §3.2 without novel mathematical claims.
invented entities (2)
  • AnyDexRT fingertip mapper fm (per-finger MLPs trained with partial Chamfer + distance + local motion + align) no independent evidence
    purpose: Learned human-to-robot fingertip correspondence without full calibration or global coverage.
    Central algorithmic object; evaluated only inside this paper’s metrics and tasks, though falsifiable by reimplementation on other hands.
  • Pinch contact classifier fc with template snap-in no independent evidence
    purpose: Recover intended fingertip contacts when glove measurements miss pinches.
    Optional refinement module; evidence is the Pick-10 success lift on one platform.

pith-pipeline@v1.1.0-grok45 · 18045 in / 3685 out tokens · 60669 ms · 2026-07-10T09:04:39.840466+00:00 · methodology

0 comments
read the original abstract

Teleoperation is a key interface for controlling dexterous robotic hands and collecting demonstrations for imitation learning. Its effectiveness largely depends on kinematic retargeting, which maps operator hand motions to feasible and intuitive robot hand motions. Existing methods often require hand-crafted objectives, precise calibration, or global shape matching between human and robot hand spaces, making them sensitive to hand-specific tuning and less reliable across different dexterous hands. We propose AnyDexRT, a calibration-free retargeting method for intuitive dexterous teleoperation across human-like dexterous hands. AnyDexRT combines self-supervised fingertip correspondence learning with few-shot human guidance to anchor the mapping in task-relevant regions, and further refines pinch-related poses using a contact classifier. Experiments on diverse dexterous hands and real-world teleoperation tasks show that AnyDexRT improves retargeting quality, reduces manual tuning, and provides more intuitive and efficient control than prior retargeting methods. Project website: https://chenxi-wang.github.io/projects/anydexrt

Figures

Figures reproduced from arXiv: 2607.08341 by Cewu Lu, Chenxi Wang, Chuan Wen, Hongjie Fang, Lixin Yang, Shangning Xia, Ying Feng.

Figure 1
Figure 1. Figure 1: AnyDexRT System. AnyDexRT is a calibration-free dexterous hand retargeting system. With few￾shot human guidance, our system achieves high-quality retargeting across diverse human-like hands. Abstract: Teleoperation is a key interface for controlling dexterous robotic hands and collecting demonstrations for imitation learning. Its effectiveness largely de￾pends on kinematic retargeting, which maps operator … view at source ↗
Figure 2
Figure 2. Figure 2: Overview of AnyDexRT. (Left) During training, AnyDexRT learns a fingertip mapper from collected human fingertip samples with few paired human-robot anchors as guidance, and trains a contact classifier using collected or generated pinch/non-pinch poses. (Right) During deployment, the fingertip mapper produces retargeted targets, which are refined by the contact classifier and converted into robot joint comm… view at source ↗
Figure 3
Figure 3. Figure 3: Retargeting Objectives. (a) Combining all objectives produces a geometrically consistent and intuitive mapping from the human fingertip space to the robot fingertip space. (b) Full Chamfer loss can force unnatural coverage of redundant robot fingertip regions. (c) Distance loss preserves the geometric structure of the retargeted space and reduces mapping distortion. (d) Local motion preservation encourages… view at source ↗
Figure 4
Figure 4. Figure 4: Sensor Failure dur￾ing Pinches. The measured data fail to capture hand pinches. Grasping tiny objects requires higher retargeting accuracy, as even small positional errors can lead to task failure. This is especially critical for pinch motions, where precise fingertip contact is re￾quired. As shown in [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Calibration Sensitivity. AnyDexRT maintains consistent re￾targeting quality, demonstrating its calibration-free capability. 10 3 10 2 GeoRT 10 3 10 2 AnyDexRT Inspire Ability XHand Wuji Allegro Leap Shadow [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: illustrates real-world qualitative results of AnyDexRT on Wuji Hand [45]. We can observe that AnyDexRT provides precise and intuitive hand retargeting on multiple types of gestures [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Analysis of Real-World Teleoperation. We compare the retargeted pinch poses and power grasps output by AnyDexRT and GeoRT [55] in [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Anchor Types and Collection. The anchors can be defined according to the op￾erators’ habits. In [PITH_FULL_IMAGE:figures/full_fig_p013_9.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

58 extracted references · 58 canonical work pages · 3 internal anchors

  1. [1]

    May 2026.URL:https://www.allegrohand.com/sub/product/ p.php?idx=1

    Allegro.Allegro Hand V4. May 2026.URL:https://www.allegrohand.com/sub/product/ p.php?idx=1

    work page 2026

  2. [2]

    Dexterous Manipulation Through Imitation Learning: A Survey

    Shan An et al. “Dexterous Manipulation Through Imitation Learning: A Survey”. In:IEEE Transactions on Automation Science and Engineering23 (2025), pp. 1760–1792

    work page 2025

  3. [3]

    Task-Oriented Hand Motion Retar- geting for Dexterous Manipulation Imitation

    Dafni Antotsiou, Guillermo Garcia-Hernando, and Tae-Kyun Kim. “Task-Oriented Hand Motion Retar- geting for Dexterous Manipulation Imitation”. In:European Conference on Computer Vision Workshops. 2018

    work page 2018

  4. [4]

    Parametric Correspondence and Chamfer Matching: Two New Techniques for Image Matching

    Harry G. Barrow et al. “Parametric Correspondence and Chamfer Matching: Two New Techniques for Image Matching”. In:International Joint Conference on Artificial Intelligence. William Kaufmann, 1977, pp. 659–663

    work page 1977

  5. [5]

    Visual Dexterity: In-Hand Reorientation of Novel and Complex Object Shapes

    Tao Chen et al. “Visual Dexterity: In-Hand Reorientation of Novel and Complex Object Shapes”. In: Science Robotics8.84 (2023), eadc9244

    work page 2023

  6. [6]

    Vegetable Peeling: A Case Study in Constrained Dexterous Manipulation

    Tao Chen et al. “Vegetable Peeling: A Case Study in Constrained Dexterous Manipulation”. In:IEEE International Conference on Robotics and Automation. 2025, pp. 4542–4550

    work page 2025

  7. [7]

    Open-TeleVision: Teleoperation with Immersive Active Visual Feedback

    Xuxin Cheng et al. “Open-TeleVision: Teleoperation with Immersive Active Visual Feedback”. In:Con- ference on Robot Learning. 2024

    work page 2024

  8. [8]

    Universal Manipulation Interface: In-The-Wild Robot Teaching Without In-The-Wild Robots

    Cheng Chi et al. “Universal Manipulation Interface: In-The-Wild Robot Teaching Without In-The-Wild Robots”. In:Robotics: Science and Systems. 2024

    work page 2024

  9. [9]

    A Learning-Based Harmonic Mapping: Frame- work, Assessment, and Case Study of Human-to-Robot Hand Pose Mapping

    Eunsuk Chong, Lionel Zhang, and Veronica J Santos. “A Learning-Based Harmonic Mapping: Frame- work, Assessment, and Case Study of Human-to-Robot Hand Pose Mapping”. In:The International Journal of Robotics Research40.2-3 (2021), pp. 534–557

    work page 2021

  10. [10]

    Bunny-VisionPro: Real-Time Bimanual Dexterous Teleoperation for Imitation Learn- ing

    Runyu Ding et al. “Bunny-VisionPro: Real-Time Bimanual Dexterous Teleoperation for Imitation Learn- ing”. In:IEEE/RSJ International Conference on Intelligent Robots and Systems. IEEE. 2025, pp. 12248– 12255

    work page 2025

  11. [11]

    A Closed-Form Solution for Human Finger Positioning

    Roel Duits, Arjan Egges, and A. Frank van der Stappen. “A Closed-Form Solution for Human Finger Positioning”. In:Proceedings of the 8th ACM SIGGRAPH Conference on Motion in Games, MIG 2015, Paris, France, November 16-18, 2015. ACM, 2015, pp. 73–78

    work page 2015

  12. [12]

    A Point Set Generation Network for 3D Object Re- construction from a Single Image

    Haoqiang Fan, Hao Su, and Leonidas J. Guibas. “A Point Set Generation Network for 3D Object Re- construction from a Single Image”. In:CVPR. IEEE Computer Society, 2017, pp. 2463–2471

    work page 2017

  13. [13]

    AnyDexGrasp: General Dexterous Grasping for Different Hands with Human-level Learning Efficiency

    Hao-Shu Fang et al. “AnyDexGrasp: General Dexterous Grasping for Different Hands with Human- Level Learning Efficiency”. In:arXiv preprint arXiv:2502.16420(2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  14. [14]

    DEXOP: A Device for Robotic Transfer of Dexterous Human Manipulation

    Hao-Shu Fang et al. “DEXOP: A Device for Robotic Transfer of Dexterous Human Manipulation”. In: arXiv preprint arXiv:2509.04441(2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  15. [15]

    AirExo: Low-Cost Exoskeletons for Learning Whole-Arm Manipulation in the Wild

    Hongjie Fang et al. “AirExo: Low-Cost Exoskeletons for Learning Whole-Arm Manipulation in the Wild”. In:IEEE International Conference on Robotics and Automation. 2024

    work page 2024

  16. [16]

    AirExo-2: Scaling up Generalizable Robotic Imitation Learning with Low-Cost Exoskeletons

    Hongjie Fang et al. “AirExo-2: Scaling up Generalizable Robotic Imitation Learning with Low-Cost Exoskeletons”. In:Conference on Robot Learning. 2025

    work page 2025

  17. [17]

    Learning Dexterous Manipulation with Quantized Hand State

    Ying Feng et al. “Learning Dexterous Manipulation with Quantized Hand State”. In:IEEE International Conference on Robotics and Automation. 2026

    work page 2026

  18. [18]

    DexPilot: Vision-Based Teleoperation of Dexterous Robotic Hand-Arm System

    Ankur Handa et al. “DexPilot: Vision-Based Teleoperation of Dexterous Robotic Hand-Arm System”. In:IEEE International Conference on Robotics and Automation. 2020

    work page 2020

  19. [19]

    2026.URL:https://github.com/wuji- technology/wuji-retargeting

    Guanqi He and Wentao Zhang.WujiHand Retargeting. 2026.URL:https://github.com/wuji- technology/wuji-retargeting

    work page 2026

  20. [20]

    HumDex: Humanoid Dexterous Manipulation Made Easy

    Liang Heng et al. “HumDex: Humanoid Dexterous Manipulation Made Easy”. In:arXiv preprint arXiv:2603.12260(2026)

    work page arXiv 2026

  21. [21]

    Cross-Hand Latent Representation for Vision-Language-Action Models

    Guangqi Jiang et al. “Cross-Hand Latent Representation for Vision-Language-Action Models”. In:arXiv preprint arXiv:2603.10158(2026)

    work page arXiv 2026

  22. [22]

    Adaptive Visuo-Tactile Fusion with Predictive Force Attention for Dexterous Manip- ulation

    Jinzhou Li et al. “Adaptive Visuo-Tactile Fusion with Predictive Force Attention for Dexterous Manip- ulation”. In:IEEE/RSJ International Conference on Intelligent Robots and Systems. 2025, pp. 3232– 3239. 9

    work page 2025

  23. [23]

    Vision-Based Teleoperation of Shadow Dexterous Hand Using End-to-End Deep Neu- ral Network

    Shuang Li et al. “Vision-Based Teleoperation of Shadow Dexterous Hand Using End-to-End Deep Neu- ral Network”. In:IEEE International Conference on Robotics and Automation. 2019, pp. 416–422

    work page 2019

  24. [24]

    Learning Visuotactile Skills With Two Multifingered Hands

    Toru Lin et al. “Learning Visuotactile Skills With Two Multifingered Hands”. In:IEEE International Conference on Robotics and Automation. 2025, pp. 5637–5643

    work page 2025

  25. [25]

    TypeTele: Releasing Dexterity in Teleoperation by Dexterous Manipulation Types

    Yuhao Lin et al. “TypeTele: Releasing Dexterity in Teleoperation by Dexterous Manipulation Types”. In:Conference on Robot Learning. 2025, pp. 4975–4993

    work page 2025

  26. [26]

    A Glove-Based System for Studying Hand-Object Manipulation via Joint Pose and Force Sensing

    Hangxin Liu et al. “A Glove-Based System for Studying Hand-Object Manipulation via Joint Pose and Force Sensing”. In:IEEE/RSJ International Conference on Intelligent Robots and Systems. 2017, pp. 6617–6624

    work page 2017

  27. [27]

    High-Fidelity Grasping in Virtual Reality using a Glove-Based System

    Hangxin Liu et al. “High-Fidelity Grasping in Virtual Reality using a Glove-Based System”. In:IEEE International Conference on Robotics and Automation. 2019, pp. 5180–5186

    work page 2019

  28. [28]

    May 2026.URL:https://www.manus- meta.com/ products/quantum-metagloves

    MANUS.MANUS Quantum Metagloves. May 2026.URL:https://www.manus- meta.com/ products/quantum-metagloves

    work page 2026

  29. [29]

    Human to Robot Hand Motion Mapping Methods: Review and Classification

    Roberto Meattini et al. “Human to Robot Hand Motion Mapping Methods: Review and Classification”. In:IEEE Transactions on Robotics39.2 (2022), pp. 842–861

    work page 2022

  30. [30]

    An Overview of Dexterous Manipulation

    Allison M. Okamura, Niels Smaby, and Mark R. Cutkosky. “An Overview of Dexterous Manipulation”. In:IEEE International Conference on Robotics and Automation. 2000

    work page 2000

  31. [31]

    May 2026.URL:https://www.psyonic.io/ability-hand

    Psyonic.Ability Hand. May 2026.URL:https://www.psyonic.io/ability-hand

    work page 2026

  32. [32]

    DexMV: Imitation Learning for Dexterous Manipulation from Human Videos

    Yuzhe Qin et al. “DexMV: Imitation Learning for Dexterous Manipulation from Human Videos”. In: European Conference on Computer Vision. 2022, pp. 570–587

    work page 2022

  33. [33]

    AnyTeleop: A General Vision-Based Dexterous Robot Arm-Hand Teleoperation Sys- tem

    Yuzhe Qin et al. “AnyTeleop: A General Vision-Based Dexterous Robot Arm-Hand Teleoperation Sys- tem”. In:Robotics: Science and Systems. 2023

    work page 2023

  34. [34]

    Computer animation of knowledge-based human grasping

    Hans Rijpkema and Michael Girard. “Computer animation of knowledge-based human grasping”. In: Proceedings of the 18th Annual Conference on Computer Graphics and Interactive Techniques, SIG- GRAPH 1991, Providence, RI, USA, April 27-30, 1991. ACM, 1991, pp. 339–348

    work page 1991

  35. [35]

    May 2026.URL:https://shadowrobot.com/ dexterous-hand-series/

    Shadow Robot.Shadow Dexterous Hand Series. May 2026.URL:https://shadowrobot.com/ dexterous-hand-series/

    work page 2026

  36. [36]

    May 2026.URL:https://www.robotera.com/en/goods1/4

    RobotEra.RobotEra XHand1. May 2026.URL:https://www.robotera.com/en/goods1/4. html#/product/XHAND

    work page 2026

  37. [37]

    May 2026.URL:https://www.flexiv.com/product/ rizon

    Flexiv Robotics.Flexiv Rizon Arm. May 2026.URL:https://www.flexiv.com/product/ rizon

    work page 2026

  38. [38]

    May 2026.URL:https://en.inspire-robots.com/ product/rh56bfx

    Inspire Robots.Inspire Hand RH56BFX. May 2026.URL:https://en.inspire-robots.com/ product/rh56bfx

    work page 2026

  39. [39]

    Postural Hand Synergies for Tool Use

    Marco Santello, Martha Flanders, and John F. Soechting. “Postural Hand Synergies for Tool Use”. In: Journal of Neuroscience18.23 (1998), pp. 10105–10115

    work page 1998

  40. [40]

    LEAP Hand: Low-Cost, Efficient, and Anthro- pomorphic Hand for Robot Learning

    Kenneth Shaw, Ananye Agarwal, and Deepak Pathak. “LEAP Hand: Low-Cost, Efficient, and Anthro- pomorphic Hand for Robot Learning”. In:Robotics: Science and Systems. 2023

    work page 2023

  41. [41]

    Bimanual Dexterity for Complex Tasks

    Kenneth Shaw et al. “Bimanual Dexterity for Complex Tasks”. In:Conference on Robot Learning. 2024

    work page 2024

  42. [42]

    Tilde: Teleoperation for Dexterous In-Hand Manipulation Learning with a DeltaHand

    Zilin Si et al. “Tilde: Teleoperation for Dexterous In-Hand Manipulation Learning with a DeltaHand”. In:Robotics: Science and Systems. 2024

    work page 2024

  43. [43]

    Robotic Telekinesis: Learning a Robotic Hand Imitator by Watching Humans on Youtube

    Aravind Sivakumar, Kenneth Shaw, and Deepak Pathak. “Robotic Telekinesis: Learning a Robotic Hand Imitator by Watching Humans on Youtube”. In:Robotics: Science and Systems. 2022

    work page 2022

  44. [44]

    Dexterous Contact-Rich Manipulation via the Contact Trust Region

    HJ Terry Suh et al. “Dexterous Contact-Rich Manipulation via the Contact Trust Region”. In:The Inter- national Journal of Robotics Research(2025), p. 02783649251398875

    work page 2025

  45. [45]

    May 2026.URL:https://wuji.tech/en/hand

    Wuji Technology.Wuji Hand. May 2026.URL:https://wuji.tech/en/hand

    work page 2026

  46. [46]

    May 2026.URL:https://www.vive.com/us/accessory/ tracker3/

    HTC Vive.HTC Vive Tracker 3.0. May 2026.URL:https://www.vive.com/us/accessory/ tracker3/

    work page 2026

  47. [47]

    DexCap: Scalable and Portable Mocap Data Collection System for Dexterous Manip- ulation

    Chen Wang et al. “DexCap: Scalable and Portable Mocap Data Collection System for Dexterous Manip- ulation”. In:Robotics: Science and Systems. 2024

    work page 2024

  48. [48]

    One Hand to Rule Them All: Canonical Representations for Unified Dexterous Manipulation

    Zhenyu Wei, Yunchao Yao, and Mingyu Ding. “One Hand to Rule Them All: Canonical Representations for Unified Dexterous Manipulation”. In:Robotics: Science and Systems. 2026

    work page 2026

  49. [49]

    Dexterous Teleoperation of 20-DoF ByteDexter Hand via Human Motion Retargeting

    Ruoshi Wen et al. “Dexterous Teleoperation of 20-DoF ByteDexter Hand via Human Motion Retarget- ing”. In:arXiv preprint arXiv:2507.03227(2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  50. [50]

    Cycle4Completion: Unpaired Point Cloud Completion Using Cycle Transformation With Missing Region Coding

    Xin Wen et al. “Cycle4Completion: Unpaired Point Cloud Completion Using Cycle Transformation With Missing Region Coding”. In:CVPR. Computer Vision Foundation / IEEE, 2021, pp. 13080–13089

    work page 2021

  51. [51]

    Analyzing Key Objectives in Human-to-Robot Retargeting for Dexterous Manip- ulation

    Chendong Xin et al. “Analyzing Key Objectives in Human-to-Robot Retargeting for Dexterous Manip- ulation”. In:IEEE Robotics and Automation Practice(2026)

    work page 2026

  52. [52]

    DexUMI: Using Human Hand as the Universal Manipulation Interface for Dexterous Manipulation

    Mengda Xu et al. “DexUMI: Using Human Hand as the Universal Manipulation Interface for Dexterous Manipulation”. In:Conference on Robot Learning. 2025. 10

    work page 2025

  53. [53]

    UniDexGrasp: Universal Robotic Dexterous Grasping via Learning Diverse Proposal Generation and Goal-Conditioned Policy

    Yinzhen Xu et al. “UniDexGrasp: Universal Robotic Dexterous Grasping via Learning Diverse Proposal Generation and Goal-Conditioned Policy”. In:IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2023, pp. 4737–4746

    work page 2023

  54. [54]

    ACE: A Cross-Platform and Visual-Exoskeletons System for Low-Cost Dexterous Teleoperation

    Shiqi Yang et al. “ACE: A Cross-Platform and Visual-Exoskeletons System for Low-Cost Dexterous Teleoperation”. In:Conference on Robot Learning. PMLR. 2024, pp. 4895–4911

    work page 2024

  55. [55]

    Geometric Retargeting: A Principled, Ultrafast Neural Hand Retargeting Algo- rithm

    Zhao-Heng Yin et al. “Geometric Retargeting: A Principled, Ultrafast Neural Hand Retargeting Algo- rithm”. In:IEEE/RSJ International Conference on Intelligent Robots and Systems. 2025, pp. 17376– 17382

    work page 2025

  56. [56]

    KineDex: Learning Tactile-Informed Visuomotor Policies via Kinesthetic Teaching for Dexterous Manipulation

    Di Zhang et al. “KineDex: Learning Tactile-Informed Visuomotor Policies via Kinesthetic Teaching for Dexterous Manipulation”. In:Conference on Robot Learning. 2025, pp. 4123–4138

    work page 2025

  57. [57]

    UniDex: A Robot Foundation Suite for Universal Dexterous Hand Control from Ego- centric Human Videos

    Gu Zhang et al. “UniDex: A Robot Foundation Suite for Universal Dexterous Hand Control from Ego- centric Human Videos”. In:arXiv preprint arXiv:2603.22264(2026)

    work page arXiv 2026

  58. [58]

    DOGlove: Dexterous Manipulation with a Low-Cost Open-Source Haptic Force Feed- back Glove

    Han Zhang et al. “DOGlove: Dexterous Manipulation with a Low-Cost Open-Source Haptic Force Feed- back Glove”. In:Robotics: Science and Systems. 2025. 11 Appendix A Qualitative Results Fig. 7 illustrates real-world qualitative results of AnyDexRT on Wuji Hand [45]. We can observe that AnyDexRT provides precise and intuitive hand retargeting on multiple typ...

    work page 2025