Pith. sign in

REVIEW 3 major objections 5 minor 37 references

DexCtrl: Towards Sim-to-Real Dexterity with Adaptive Controller Learning

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

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

Pith's one-line read DexCtrl claims that jointly predicting actions and controller parameters from recent joint history closes more of the sim-to-real gap than fixed-gain tuning.

desk verdict Jointly predicting PD gains with actions is a plausible and apparently effective sim-to-real idea, but the evidence is thinly reported and the gain module has an unaddressed train/test mismatch. read the letter →

arxiv 2505.00991 v1 pith:ONF66BC2 submitted 2025-05-02 cs.RO cs.SYeess.SY

Shuqi Zhao , Ke Yang , Yuxin Chen , Chenran Li , Yichen Xie , Xiang Zhang , Changhao Wang , Masayoshi Tomizuka This is my paper · ORCID
classification cs.ROcs.SYeess.SY
keywords dexterousmanipulationsim-to-realtransferadaptivecontrolcontrollerparameterpredictionin-handrotationobjectflippingproprioceptionreinforcementlearning
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 identifies the low-level torque controller as a neglected source of the sim-to-real gap in dexterous manipulation: the same desired joint trajectory produces different contact forces when gains differ. It proposes DexCtrl, which on every control step outputs not only the next desired joint position but also the diagonal PD gains $K_P$ and $K_D$, using the past ten steps of desired and actual joint trajectories plus previous gains. A policy trained this way adapts stiffness and damping during execution, so the sim-to-real mismatch is addressed without manual gain tuning or heavy controller randomization. In real-world tests on a 16-DOF hand, DexCtrl's average rotation speed was 11.04 compared with 2.43 for a manually tuned fixed-gain baseline and 4.99 for an action-only variant, and the method also generalized to a table-flipping task.

What carries the argument

The load-bearing mechanism is a per-step diagonal PD torque controller, $\tau = K_P(q_d - q_c) + K_D(\dot{q}_d - \dot{q}_c)$, whose gain vector $K = \{K_P, K_D\} \in \mathbb{R}^{32}$ is predicted at every time step alongside the desired joint action. A self-attention module maps the last ten steps of desired and actual joint trajectories to the next action, and a cross-attention module uses the predicted action as query against the same history to produce the next gains. During training, an oracle PPO policy with access to object properties collects data in simulation, and two student modules are distilled from that data in open loop; at deployment the modules run closed loop with real sensor readings, with gains linearly mapped from simulation to the real system using only approximate upper and lower bounds. This design also puts controller parameters in the observation, so force interactions can be reasoned about from proprioception alone.

What would settle it

Run the same DexCtrl action sequence on the real hand with the predicted gains frozen at their time-averaged values. If rotation speed stays near the adaptive value, per-step gain variation is not the causal driver; if it collapses toward the action-only baseline, the adaptive gain schedule is doing the work.

Watch

Extended reading notes

Core claim

The central claim is that controller parameters should be treated as part of the policy's output and observation, not as fixed training hyperparameters. By jointly predicting actions and gains from proprioceptive history, the policy can estimate the manipulated object's mass and friction from ten steps of joint data alone, without vision or force sensing, and can adjust contact forces to keep the object stably rotating or flipping. The paper shows the learned stiffness responds systematically to object properties: predicted $K_P$ rises with object mass, while friction produces task- and joint-specific stiffness patterns. On hardware with unseen objects, this adaptive gain schedule substantially outperforms both manual tuning and an action-only variant, and the gap is larger in the real world than in simulation, which the paper takes as evidence that per-step gain adjustment is exactly what the sim-to-real gap requires.

Load-bearing premise

The approach assumes that a diagonal PD controller with only two gains per joint describes the real hand's low-level dynamics well enough that gains learned in simulation remain meaningful after a rough linear mapping, and that ten steps of joint history can infer the object's mass and friction without vision or tactile feedback.

Editorial extensions

If this is right

  • In zero-shot real-world transfer, DexCtrl's average RotR is 11.04, versus 2.43 for manual tuning and 4.99 for the action-only variant.
  • DexCtrl improves over the manual-tuning baseline even when there is no controller gap in simulation, so adaptive gains help stability and speed beyond domain adaptation.
  • Including controller parameters in the observation lets the policy reason about contact forces without vision, force, or tactile sensing.
  • Only approximate upper and lower gain bounds are needed for the sim-to-real mapping, reducing the human tuning burden.
  • The same method transfers to a flipping task, where contact with the table makes the task more sensitive to gain variation.

Reading between the lines

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

  • The predicted gain schedule can be read as an implicit mass-and-friction estimator: because stiffness separates systematically by object mass, the cross-attention module could be reused for proprioceptive system identification.
  • If the gain prediction module is as transferable as the action module, the paper's future plan of sharing one controller-parameter module across tasks is likely to work; a concrete test is training the module on rotation and evaluating it on flipping without retraining.
  • With force or tactile feedback added, the rough linear gain mapping could be replaced by online fine-tuning, which would likely improve contact-rich cases where ten steps of history cannot fully disambiguate state.
  • The underlying principle, making controller gains a learned output rather than a randomized or hand-tuned fixed value, should apply to other impedance-controlled contact tasks such as assembly or insertion, where the same PD torque law governs behavior.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes DexCtrl, a sim-to-real transfer framework for dexterous manipulation in which a student policy is distilled from a simulation-trained RL oracle into two modules: an action prediction module and a control-parameter prediction module. At each step the action module outputs desired joint positions from a ten-step history of proprioception, and the control-parameter module outputs diagonal PD gains K = {K_P, K_D} using the current predicted action as a cross-attention query. The authors argue that jointly predicting actions and controller parameters adaptively reduces the low-level controller gap, avoiding manual tuning and heavy randomization. The method is evaluated on two contact-rich tasks, in-hand rotation and flipping, on the LEAP hand, both in simulation and on real hardware, with comparisons against a manually tuned baseline and an action-only ablation (Ours w/o PD). The main empirical result is a real-world rotation advantage: DexCtrl achieves an average RotR of 11.04 versus 4.99 for Ours w/o PD and 2.43 for manual tuning (Table 3).

Significance. If the central claim holds, DexCtrl identifies controller-parameter mismatch as a distinct and addressable component of the sim-to-real gap, offering a practical alternative to manual gain tuning and aggressive domain randomization. The paper has genuine strengths: the real-world rotation evaluation uses unseen objects and is a real transfer test; the ablation Ours w/o PD isolates the contribution of adaptive gains; and the analysis in Section 4.5 relating learned stiffness to object mass and friction is a useful first step toward interpreting the mechanism. The paper also explicitly acknowledges its limitations, including the absence of force/tactile sensing and the restriction to a single hardware platform. However, the current evidence is not yet conclusive: the train/test distribution shift in the control-parameter module is unaddressed, the quantitative claims lack statistical support, and the real-world flipping evaluation is only qualitative. These issues are fixable within the manuscript's scope, so the appropriate decision is major revision rather than rejection.

major comments (3)
  1. [§3.2, Fig. 2] The control-parameter module is trained open-loop with oracle actions as the cross-attention query and oracle K values in the historical input, but at deployment the query is the student-predicted action â_t and the history contains previously predicted K values. The Gaussian noise added to current trajectory values during student training does not cover either of these shifted inputs. Because the paper's central claim is that jointly predicting actions and controller parameters improves transfer, the real-world gain of DexCtrl over Ours w/o PD (Table 3, RotR 11.04 vs 4.99) could reflect behavior under this out-of-distribution query rather than a genuinely learned action–K relationship. Please add an experiment that isolates this effect, for example evaluating the K module at inference with oracle actions as queries versus student-predicted actions, or fine-tuning with closed-loop student rollouts in a DAgger-style procedure, and report both RotR and K-module prediction error.
  2. [§4.3, Tables 3–4] All quantitative tables report only mean values, with no standard deviations, per-trial distributions, or trial counts, despite repeated claims of 'significant' improvement in Sections 4.2–4.4. In addition, the text states that twelve real-world objects were used for rotation, but Table 3 lists only five objects (Cube, Bottle, Apple, Yogurt, Baseball), and the real-world flipping claim rests on a visualization (Figure 3) rather than a quantitative table. Please provide per-trial data or at least standard deviations and trial counts for every table, clarify the twelve-object discrepancy, and either add quantitative flipping results or temper the corresponding claim.
  3. [§4.2, Tables 1–2] The simulation experiments in Tables 1 and 2 also report only means, and the claim that DexCtrl 'significantly outperforms' the baselines is not supported by any uncertainty quantification or significance test. Since these tables are used to support the fundamental claim that adaptive controller parameters improve performance even without a controller gap, please report standard deviations or confidence intervals over the 1024 randomized initial poses, and state whether the differences are statistically significant.
minor comments (5)
  1. [§4.3] The text says 'twelve different real-world unseen objects' but Table 3 has five object columns; please make the object count and selection procedure consistent between text and table.
  2. [§4.1] The baseline name 'Ours w/o PD' is potentially misleading: the ablation removes the adaptive PD parameter prediction module but retains a PD torque controller. A name such as 'Ours w/o adaptive K' or 'Ours w/ fixed K' would be clearer.
  3. [§4.4] The sentence 'the lightest and smoothest objects exhibit the highest speed and lowest stability, respectively' is not fully supported by Table 4: the friction-small RotR is 10.414, friction-medium is 9.386, and friction-large is 10.681, so the friction trend is not monotonic. Please either qualify this statement or provide additional evidence.
  4. [§2] The notation [q_d, q̇_d] ∈ R^32 and [q_c, q̇_c] ∈ R^32 is slightly confusing because q_d and q̇_d are each 16-dimensional; consider stating this explicitly to avoid an apparent dimension mismatch in Eq. (1).
  5. [Global] The hardware name is written inconsistently as 'LEAP hand' in Section 2 and 'LeapHand' in Section 7; please unify the capitalization.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: DexCtrl's adaptive controller prediction is validated on an external real-world benchmark and does not reduce to its own fitted inputs.

full rationale

The paper's derivation chain is self-contained rather than circular. The oracle policy is trained in simulation with PPO and outputs both actions and controller parameters (a_t, K_t); the student modules are then distilled to predict these externally generated targets from a fixed dataset. The K-prediction module's regression target K_t comes from the oracle's reward optimization, not from the student's own predictions, so no target is defined in terms of the output. The central empirical claim is supported by real-world rotation results (RotR 11.04 for DexCtrl versus 2.43 for manual tuning and 4.99 for the action-only variant in Table 3), and those numbers come from physical robot experiments, not from the paper's equations. The manual-tuning and Ours-w/o-PD baselines are evaluated under the same protocol, so the comparison is external even if the baselines were constructed by the same team; baseline construction is a comparison-quality issue, not circularity. The self-citations in the paper ([20], [21], [32], [34], [36]) appear only in related-work or contextual references and are not load-bearing for the central contribution. The cross-attention module's training/inference mismatch—trained with oracle actions as queries but deployed with student-predicted actions—is a genuine distribution-shift and robustness concern, but it is not circularity: even if this mismatch degrades performance, the paper's claim is not true by construction. The linear sim-to-real mapping of controller parameters using approximate upper and lower bounds is an explicit calibration step rather than a hidden reuse of the target result. No equation in the paper reduces to its fitted outputs, and no uniqueness theorem or author-imported ansatz is invoked to force the method. Accordingly, no significant circularity is present.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim depends on several domain assumptions about the controller model and sim-to-real fidelity, plus a set of hand-chosen numbers (reward weights, gain bounds, history length, noise level) that are not reported in the v1. No new physical entities are introduced.

free parameters (5)
  • Reward weights for rrotation, rcontact, rsmoothness, rterminate = not reported in v1 (appendix absent)
    The oracle policy behavior is shaped by these hand-chosen weights; the paper defers details to an appendix not present in the v1.
  • Sim-to-real controller parameter bounds for K_P and K_D = not reported
    Section 3.2 states K values are linearly mapped to the real system with only an approximate estimate of upper and lower bounds; this mapping determines whether predicted gains are physically meaningful on hardware.
  • History length for student input = 10 steps
    Section 3.2 selects the last ten steps as historical information; the student's ability to infer object properties relies on this window.
  • Gaussian noise standard deviation for current trajectory observations = not reported
    Section 3.2 states that adding Gaussian noise during student training is sufficient for sim-to-real transfer, but the noise level is not given.
  • PPO hyperparameters and network sizes = not reported
    Required for training the oracle and student modules, but absent from the v1 text.
assumptions (4)
  • domain assumption The real robot's low-level controller is accurately described by the linear diagonal PD law tau = K_P(qd-qc) + K_D(qdot_d-qdot_c) with desired velocities zero.
    Section 2, Eq. 1; the whole adaptive-gain mechanism assumes the predicted K changes torques on hardware in the same way as in simulation.
  • domain assumption Object physical properties (mass, friction, scale) can be inferred from the last ten steps of joint positions, desired positions, and previous controller parameters.
    Section 3.2 uses historical proprioception to replace direct object information in deployment; Section 7 notes no force or tactile sensing is used.
  • domain assumption Simulator contact dynamics are close enough to reality that a policy distilled from a sim oracle transfers with only observation noise and approximate linear gain-range mapping.
    This is the sim-to-real premise of the whole method, asserted in Sections 3.2 and 4.3 rather than proven.
  • domain assumption Reward design in Eq. 2 induces a desirable oracle policy for both tasks.
    Section 3.1; if the oracle policy is suboptimal or reward-shaped, the distilled student inherits those limitations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DexCtrl: Towards Sim-to-Real Dexterity with Adaptive Controller Learning." pith.science (2026). https://pith.science/paper/ONF66BC2

@misc{pith2026250500991,
  author       = {Pith},
  title        = {Pith review of: DexCtrl: Towards Sim-to-Real Dexterity with Adaptive Controller Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ONF66BC2}},
  note         = {Machine review of arXiv:2505.00991}
}
read the original abstract

Dexterous manipulation has seen remarkable progress in recent years, with policies capable of executing many complex and contact-rich tasks in simulation. However, transferring these policies from simulation to real world remains a significant challenge. One important issue is the mismatch in low-level controller dynamics, where identical trajectories can lead to vastly different contact forces and behaviors when control parameters vary. Existing approaches often rely on manual tuning or controller randomization, which can be labor-intensive, task-specific, and introduce significant training difficulty. In this work, we propose a framework that jointly learns actions and controller parameters based on the historical information of both trajectory and controller. This adaptive controller adjustment mechanism allows the policy to automatically tune control parameters during execution, thereby mitigating the sim-to-real gap without extensive manual tuning or excessive randomization. Moreover, by explicitly providing controller parameters as part of the observation, our approach facilitates better reasoning over force interactions and improves robustness in real-world scenarios. Experimental results demonstrate that our method achieves improved transfer performance across a variety of dexterous tasks involving variable force conditions.

Figures

Figures reproduced from arXiv: 2505.00991 by the authors.

Figure 1
Figure 1. Compared to previous work with only action prediction (upper left), DexCtrl (lower left) [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview framework of DexCtrl, where aˆt and Kˆ t mean predicted joint position actions and predicted control parameters, respectively. As shown in Eq. 1, the torque output is directly modulated by the choice of K, which necessitates careful tuning of these parameters. In particular, besides control parameters K that directly deter￾mine the actual torque values, increasing stiffness KP reduces steady-state error but… view at source ↗
Figure 3
Figure 3. Flipping task performance in simulation and real world. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Real-world results of object rotation with different physical parameters. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Visualization for same-shape objects rotation with different masses and frictions. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Average stiffness curve under mass (left) and friction change (middle and right). [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Stiffness over time under varying mass and friction. [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 11 canonical work pages

  1. [1]

    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. In8th Annual Conference on Robot Learning, 2024

    2024

  2. [2]

    Z.-H. Yin, C. Wang, L. Pineda, K. Bodduluri, T. Wu, P. Abbeel, and M. Mukadam. Geo- metric retargeting: A principled, ultrafast neural hand retargeting algorithm. arXiv preprint arXiv:2503.07541, 2025

    arXiv 2025

  3. [3]

    J. Ye, J. Wang, B. Huang, Y . Qin, and X. Wang. Learning continuous grasping function with a dexterous hand from human demonstrations. IEEE Robotics and Automation Letters , 8(5): 2882–2889, 2023

    2023

  4. [4]

    M. Yang, C. Lu, A. Church, Y . Lin, C. Ford, H. Li, E. Psomopoulou, D. A. Barton, and N. F. Lepora. Anyrotate: Gravity-invariant in-hand object rotation with sim-to-real touch. arXiv preprint arXiv:2405.07391, 2024

    arXiv 2024

  5. [5]

    T. Chen, M. Tippur, S. Wu, V . Kumar, E. Adelson, and P. Agrawal. Visual dexterity: In- hand reorientation of novel and complex object shapes. Science Robotics , 8(84):eadc9244,

    work page

  6. [6]

    H. Qi, A. Kumar, R. Calandra, Y . Ma, and J. Malik. In-hand object rotation via rapid motor adaptation. In Conference on Robot Learning, pages 1722–1732. PMLR, 2023

    2023

  7. [7]

    H. Qi, B. Yi, S. Suresh, M. Lambeta, Y . Ma, R. Calandra, and J. Malik. General in-hand object rotation with vision and touch. In Conference on Robot Learning , pages 2549–2564. PMLR, 2023

    work page 2023

  8. [8]

    J. Wang, Y . Yuan, H. Che, H. Qi, Y . Ma, J. Malik, and X. Wang. Lessons from learning to spin” pens”. arXiv preprint arXiv:2407.18902, 2024

    arXiv 2024

Show all 37 references
  1. [9]

    H. Qi, B. Yi, M. Lambeta, Y . Ma, R. Calandra, and J. Malik. From simple to complex skills: The case of in-hand object reorientation. arXiv preprint arXiv:2501.05439, 2025

    2025 arXiv

  2. [10]

    Zakka, P

    K. Zakka, P. Wu, L. Smith, N. Gileadi, T. Howell, X. B. Peng, S. Singh, Y . Tassa, P. Florence, A. Zeng, et al. Robopianist: Dexterous piano playing with deep reinforcement learning. arXiv preprint arXiv:2304.04150, 2023

    2023 arXiv

  3. [11]

    C. Qian, J. Urain, K. Zakka, and J. Peters. Pianomime: Learning a generalist, dexterous piano player from internet demonstrations. arXiv preprint arXiv:2407.18178, 2024

    2024 arXiv

  4. [12]

    Z.-H. Yin, C. Wang, L. Pineda, F. Hogan, K. Bodduluri, A. Sharma, P. Lancaster, I. Prasad, M. Kalakrishnan, J. Malik, et al. Dexteritygen: Foundation controller for unprecedented dex- terity. arXiv preprint arXiv:2502.04307, 2025. 9

    2025 arXiv

  5. [13]

    X. Liu, J. Adalibieke, Q. Han, Y . Qin, and L. Yi. Dextrack: Towards generalizable neu- ral tracking control for dexterous manipulation from human references. arXiv preprint arXiv:2502.09614, 2025

    2025 arXiv

  6. [14]

    M. Yu, B. Liang, X. Zhang, X. Zhu, L. Sun, C. Wang, S. Song, X. Li, and M. Tomizuka. In-hand following of deformable linear objects using dexterous fingers with tactile sensing. In 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages 13518–13...

    2024

  7. [15]

    Lin, Z.-H

    T. Lin, Z.-H. Yin, H. Qi, P. Abbeel, and J. Malik. Twisting lids off with two hands. arXiv preprint arXiv:2403.02338, 2024

    2024 arXiv

  8. [16]

    Akkaya, M

    I. Akkaya, M. Andrychowicz, M. Chociej, M. Litwin, B. McGrew, A. Petron, A. Paino, M. Plappert, G. Powell, R. Ribas, et al. Solving rubik’s cube with a robot hand. arXiv preprint arXiv:1910.07113, 2019

    1910 arXiv

  9. [17]

    O. M. Andrychowicz, B. Baker, M. Chociej, R. Jozefowicz, B. McGrew, J. Pachocki, A. Petron, M. Plappert, G. Powell, A. Ray, et al. Learning dexterous in-hand manipulation. The International Journal of Robotics Research , 39(1):3–20, 2020

    2020

  10. [18]

    C. B. Teeple, B. Aktas ¸, M. C. Yuen, G. R. Kim, R. D. Howe, and R. J. Wood. Controlling palm-object interactions via friction for enhanced in-hand manipulation. IEEE Robotics and Automation Letters, 7(2):2258–2265, 2022

    2022

  11. [19]

    Z.-H. Yin, B. Huang, Y . Qin, Q. Chen, and X. Wang. Rotating without seeing: Towards in-hand dexterity through touch. arXiv preprint arXiv:2303.10880, 2023

    2023 arXiv

  12. [20]

    D. Guo, Y . Xiang, S. Zhao, X. Zhu, M. Tomizuka, M. Ding, and W. Zhan. Phygrasp: gen- eralizing robotic grasping with physics-informed large multimodal models. arXiv preprint arXiv:2402.16836, 2024

    2024 arXiv

  13. [21]

    S. Zhao, X. Zhu, Y . Chen, C. Li, X. Zhang, M. Ding, and M. Tomizuka. Dexh2r: Task-oriented dexterous manipulation from human to robots. arXiv preprint arXiv:2411.04428, 2024

    2024

  14. [22]

    F. Lan, S. Wang, Y . Zhang, H. Xu, O. Oseni, Z. Zhang, Y . Gao, and T. Zhang. Dexcatch: Learning to catch arbitrary objects with dexterous hands. arXiv preprint arXiv:2310.08809 , 2023

    2023 arXiv

  15. [23]

    C. Wang, H. Shi, W. Wang, R. Zhang, L. Fei-Fei, and C. K. Liu. Dexcap: Scalable and portable mocap data collection system for dexterous manipulation. arXiv preprint arXiv:2403.07788 , 2024

    2024 arXiv

  16. [24]

    K. Shaw, Y . Li, J. Yang, M. K. Srirama, R. Liu, H. Xiong, R. Mendonca, and D. Pathak. Bimanual dexterity for complex tasks. arXiv preprint arXiv:2411.13677, 2024

    2024 arXiv

  17. [25]

    Z. Fu, T. Z. Zhao, and C. Finn. Mobile aloha: Learning bimanual mobile manipulation with low-cost whole-body teleoperation. arXiv preprint arXiv:2401.02117, 2024

    2024 arXiv

  18. [26]

    S. P. Arunachalam, S. Silwal, B. Evans, and L. Pinto. Dexterous imitation made easy: A learning-based framework for efficient dexterous manipulation. In 2023 ieee international conference on robotics and automation (icra) , pages 5954–5961. IEEE, 2023

    2023

  19. [27]

    Qin, Y .-H

    Y . Qin, Y .-H. Wu, S. Liu, H. Jiang, R. Yang, Y . Fu, and X. Wang. Dexmv: Imitation learning for dexterous manipulation from human videos. In European Conference on Computer Vision, pages 570–587. Springer, 2022

    2022

  20. [28]

    T. Z. Zhao, J. Tompson, D. Driess, P. Florence, K. Ghasemipour, C. Finn, and A. Wahid. Aloha unleashed: A simple recipe for robot dexterity. arXiv preprint arXiv:2410.13126, 2024. 10

    2024 arXiv

  21. [29]

    K. Xu, S. Zhao, Z. Zhou, Z. Li, H. Pi, Y . Zhu, Y . Wang, and R. Xiong. A joint modeling of vision-language-action for target-oriented grasping in clutter. In 2023 IEEE International Conference on Robotics and Automation (ICRA) , pages 11597–11604. IEEE, 2023

    2023

  22. [30]

    X. Chen, J. Hu, C. Jin, L. Li, and L. Wang. Understanding domain randomization for sim-to- real transfer. arXiv preprint arXiv:2110.03239, 2021

    2021 arXiv

  23. [31]

    Mart ´ın-Mart´ın, M

    R. Mart ´ın-Mart´ın, M. A. Lee, R. Gardner, S. Savarese, J. Bohg, and A. Garg. Variable impedance control in end-effector space: An action space for reinforcement learning in contact-rich tasks. In 2019 IEEE/RSJ international conference on intelligent robots and sys- tems (IRO...

    2019

  24. [32]

    C. Wang, X. Zhang, Z. Kuang, and M. Tomizuka. Safe online gain optimization for cartesian space variable impedance control. In 2022 IEEE 18th International Conference on Automation Science and Engineering (CASE) , pages 751–757. IEEE, 2022

    2022

  25. [33]

    Buchli, F

    J. Buchli, F. Stulp, E. Theodorou, and S. Schaal. Learning variable impedance control. The International Journal of Robotics Research, 30(7):820–833, 2011

    2011

  26. [34]

    Zhang, C

    X. Zhang, C. Wang, L. Sun, Z. Wu, X. Zhu, and M. Tomizuka. Efficient sim-to-real transfer of contact-rich manipulation skills with online admittance residual learning. In Conference on Robot Learning, pages 1621–1639. PMLR, 2023

    2023

  27. [35]

    C. C. Beltran-Hernandez, D. Petit, I. G. Ramirez-Alpizar, and K. Harada. Variable compliance control for robotic peg-in-hole assembly: A deep-reinforcement-learning approach. Applied Sciences, 10(19):6923, 2020

    2020

  28. [36]

    Zhang, M

    X. Zhang, M. Tomizuka, and H. Li. Bridging the sim-to-real gap with dynamic compliance tuning for industrial insertion. In 2024 IEEE International Conference on Robotics and Au- tomation (ICRA), pages 4356–4363. IEEE, 2024. 11

    2024

  29. [2023]

    URL https://www.science.org/doi/abs/10

    doi:10.1126/scirobotics.adc9244. URL https://www.science.org/doi/abs/10. 1126/scirobotics.adc9244

    doi

Pith tools

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