Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

A single gate turns two position loops into force-feedback teleoperation

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 · deepseek-v4-flash

2026-08-04 20:58 UTC pith:AF4SRFV6

load-bearing objection A simple, practical gating law for low-cost bilateral teleoperation; the experiments are convincing but the performance comparisons are qualitative and the theoretical advantage partly rests on the low-rate high-level loop. the 4 major comments →

arxiv 2509.08226 v1 pith:AF4SRFV6 submitted 2025-09-10 cs.RO

Input-gated Bilateral Teleoperation: An Easy-to-implement Force Feedback Teleoperation Method for Low-cost Hardware

Yoshiki Kanai , Akira Kanazawa , Hideyuki Ichiwara , Hiroshi Ito , Naoaki Noguchi , Tetsuya Ogata This is my paper
classification cs.RO
keywords input-gated bilateral teleoperationforce-sensorless teleoperationlow-cost robot hardwareleader-follower manipulationcontact-rich manipulationimitation learningposition controlcontact stability
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.

This paper proposes Input-Gated Bilateral Teleoperation (IGBT), a force-feedback teleoperation method built for low-cost leader-follower robot arms. The authors aim to show that a simple position-control loop, plus a limiter that clamps the leader's control input to the absolute value of the follower's control input, gives operators a useful sense of contact force without any force sensor. In numerical simulation and real experiments, the authors report that IGBT matches the operability of force-reflection teleoperation and the contact stability of symmetric position teleoperation, while needing only the same two position-controller gains as the simplest baseline. The method also tolerates slow high-level cycle rates, so it works even when the host computer communicates with the actuators at 100 Hz instead of 1 kHz. If correct, this lowers the practical barrier to collecting high-quality contact-rich demonstrations for imitation learning.

Core claim

IGBT is a bilateral teleoperation scheme: both the leader and follower run standard position controllers, and a limiter gates the leader's control input to the absolute value of the follower's control input, after optional scaling by force and position factors. During free motion, the follower's control input is small, so the leader's allowed torque or current is tightly clamped and the operator moves freely; during contact, the follower's control input rises, widening the clamp and keeping the leader's position loop active, which preserves contact stability. The paper argues that this mechanism makes a contact-rich force-feedback teleoperator implementable on low-cost hardware with effectiv

What carries the argument

The central mechanism is the input gate, equation (6): the leader's control input is clamped to the absolute value of the follower's control input after feedforward compensation and scaling. This absolute-value clamp is what turns two independent position loops into a bilateral force-feedback system: it automatically distinguishes free motion from contact based on the follower's reaction, without explicit state estimation or gain switching. The follower's control input doubles as the force proxy, so no torque or force sensor is required.

Load-bearing premise

The load-bearing premise is that the follower's control input accurately reflects external contact force, which requires high backdrivability and low friction in the follower; otherwise the gate cannot tell free motion from contact.

What would settle it

Concretely, run IGBT on a follower with a high-ratio gearbox so currents no longer represent external force. If the leader stays stiff in free motion or contact causes visible bouncing even at default gains, the central claim fails for low-backdrivability hardware. A more quantitative test: vary the environmental stiffness over a wide range at a 100 Hz high-level cycle and record the maximum contact force overshoot; the paper's own simulations show overshoot growing, so data on this curve would settle the claimed robustness.

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

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Demonstration data for contact-rich tasks can be collected with haptic feedback on low-cost robot arms without adding force or torque sensors.
  • A teleoperation setup can keep its default actuator position gains across different host computers and cycle rates, removing manual retuning.
  • Input gating can be layered onto actuator modules that already expose current limiting, so no high-level feedback controller is required.
  • The alpha and beta scaling factors let a smaller or weaker follower be paired with a different leader, extending IGBT to asymmetric robot pairs.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same absolute-value gate could be transplanted onto other position-controlled teleoperation schemes; any pair of actuators with a torque or current limit function might get the free-motion operability of FRBT and the contact stability of SPBT for the cost of one extra clamp.
  • The gate's sensitivity to the follower's control input suggests a stress test the paper does not run: use a follower with substantial static friction or a geared transmission, where free-motion currents do not drop to near zero, to see whether the clamp loses its ability to improve operability.
  • For imitation learning, one could compare policy success on data collected with IGBT against data from force-sensor-based teleoperation to see whether the cheaper hardware yields equivalent task performance.
  • The observed contact overshoot could be reduced by filtering or softening the gate, but that would reintroduce the tuning burden IGBT is designed to avoid; testing such variants would map the method's robustness boundary.

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

4 major / 5 minor

Summary. The paper proposes Input-Gated Bilateral Teleoperation (IGBT), a force-feedback teleoperation method that requires no force sensors and uses only low-level position controllers. The leader's control input is clamped to the absolute value of the follower's control effort, intended to reduce leader resistance in free motion while preserving contact stability. The method is evaluated in numerical simulation, on a pair of Dynamixel actuators, and on two low-cost multi-joint platforms (OpenMANIPULATOR-X and ALOHA) in USB insertion tasks. The authors claim that IGBT achieves operability comparable to FRBT and contact stability comparable to SPBT, with minimal parameter tuning and robustness to low high-level cycle rates.

Significance. The proposed method is simple and practically relevant for low-cost teleoperation and imitation learning. The idea of using the follower's control signal as a dynamic gate is intuitive and appears easy to implement on commercial actuator modules, avoiding force sensors and complex observer-based controllers. The paper includes both simulation and real-hardware validation, including multi-joint systems with different actuator types. If the trade-off claims are rigorously established, the method could meaningfully lower the implementation barrier for force-feedback teleoperation. The strengths are the use of default position-controller gains, the absence of force sensors, and demonstration on two commercially available platforms.

major comments (4)
  1. [Section III, Eq. (6); Section IV-A] The mechanism underlying the claimed operability advantage is not adequately analyzed. In the ideal symmetric continuous-time case (identical P, C, α=β=1, u_FF=0), u_f = C(y_l - y_f) = -u_l, so the limiter in Eq. (6) is never active and IGBT reduces exactly to SPBT. The operability improvement seen in Fig. 3 therefore depends on the discrete-time, zero-order-hold asynchrony between the high-level cycle (F1) and edge controllers (F2). The paper does not provide any analysis or bound characterizing when |u_f| < |u_l|, nor does it test fast free-motion maneuvers where the tracking effort u_f can become large and the gate inactive. Without such support, the general claim that 'IGBT achieves operability equivalent to FRBT' (stated after Fig. 3) is not established. Please provide a theoretical analysis or define the operating envelope, and include experiments or simulations with fast, dynamic
  2. [Section IV-A and IV-B, Figs. 3 and 5] The core comparisons are qualitative. Operability is inferred from the slope of y_f in simulation and the time for the follower to reach the obstacle in the actuator experiment; contact stability is inferred from convergence to a 1 mrad error or from the apparent action-reaction current relationship. No quantitative metrics (e.g., tracking error, velocity, overshoot, settling time, transparency) are reported, and no repeated trials or statistical analysis are provided. Since the central contribution is a specific trade-off claim (FRBT-like operability with SPBT-like stability), the evaluation needs defined metrics and multiple runs, especially for the single-actuator experiment where the operator applies force by hand and the force profile is not standardized.
  3. [Section IV-C and Abstract] The claim that the method requires 'zero parameter adjustments' is contradicted by the implementation details. For OpenMANIPULATOR-X, u_FF_f is computed from a joint-angle-dependent gravity model; for ALOHA, α is set from nominal maximum torque ratios, β is set for the gripper opening width, and u_FF_f is again computed from a model. These are parameters requiring model knowledge and per-hardware calibration, even though the position-controller gains are left at defaults. The current wording in the Abstract and Section IV-C is misleading; please either soften the claim or clearly separate 'control gains' (untuned) from 'model-based feedforward/scaling parameters' (tuned).
  4. [Section V] The limitations section acknowledges low backdrivability, but the more fundamental issue is that u_f conflates the follower's tracking effort with the external contact force. In free motion, u_f is the effort required to track the leader, including inertial and damping terms, and can be large during fast maneuvers. In contact, u_f also contains tracking error that exists even without contact. Therefore the gate can over- or under-estimate the environmental force. The paper's experiments use slow, quasi-static motions, which may not expose this failure mode; this issue is not limited to low-backdrivability hardware. Please discuss this limitation explicitly and, if possible, provide an experimental or simulation case with fast free motion to bound the effect.
minor comments (5)
  1. [Throughout] There are inconsistent uses of 'IBGT' and 'IGBT' (e.g., end of Section III and Section IV-C). Please correct the spelling.
  2. [Section IV-A, Table I] The text says 'simulations were conducted with set to 100 Hz and 1 kHz' — the variable F1 is missing. Please add the F1 values to Table I or to the text.
  3. [Fig. 8 caption] The caption contains typos: (b) should be 'Positions of right gripper', (c) should be 'Currents of left joint 3', and (d) should be 'Currents of right gripper' (currently repeated as '(a) Currents of right gripper').
  4. [References [24] and [31]] The word 'Available' is misspelled as 'Abailable' in both references. Also, several references lack DOIs or page ranges; please check the reference formatting.
  5. [Section III] The phrase 'it is sufficient for the leader system to follow the position displacement of the follower system' is unclear; in SPBT the leader follows the follower and vice versa. Please rephrase for clarity.

Circularity Check

0 steps flagged

No significant circularity: IGBT is a new control law validated by independent simulation and experiments; no fitted-input prediction, no load-bearing self-citation, and no definitional equivalence.

full rationale

The paper proposes a new bilateral teleoperation controller, IGBT, defined by Eqs. (1)-(6). The core gating law (Eq. 6) is a design choice, not a fitted parameter or a renamed known result. The claimed performance—operability equivalent to FRBT and contact stability equivalent to SPBT—is not derived analytically from the definition; it is established through numerical simulation (Section IV-A) and physical experiments (Sections IV-B and IV-C) that implement the method on real low-cost hardware. There is no parameter fitted to the outcome: the simulation uses fixed PD gains and the hardware experiments use default Dynamixel parameters with zero tuning. The only mild point is that the simulations use the same IGBT equations as the proposed method, but that is standard model-based validation, not circularity. The ideal symmetric continuous-time limit of IGBT reduces to SPBT, but the paper explicitly relies on the asynchronous high-level exchange at cycle rate F1 to make the limiter active during free motion; this is a stated design mechanism, not a hidden equivalence that forces the result. The limitations acknowledged in Section V—untested low-backdrivability hardware, unclear advantage over ABC, and lack of high-end validation—are honest scope restrictions, not evidence of circularity. References to prior work are background or independent support and are not used as a self-citation chain to justify the central claim. Therefore the derivation is self-contained and no circular step is present.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

The method is a feedback control law with a saturation element; it introduces no new physical entities. The ledger lists the hand-set scaling factors, gravity compensation, and controller gains that the authors needed to specify beyond the abstract rule, plus the modeling assumptions in the simulation.

free parameters (4)
  • Position controller gains Kp, Kd = Simulation: Kp=10, Kd=2; experiments: Dynamixel default gains
    The method's behavior depends on these gains; they are not tuned in a fitting sense, but the claim of zero tuning assumes the defaults are adequate for any platform.
  • Force scaling factor α = 1 for identical actuators; for ALOHA, ratio of nominal max torque of leader and follower actuators
    Set by hand to map force between actuators with different output capabilities; affects the gating threshold.
  • Position scaling factor β = 1 for identical actuators; for ALOHA gripper, set to adjust opening width
    Set by hand to map position between leader and follower, especially for the gripper on ALOHA.
  • Gravity feedforward u_FF_f = Sequentially computed from joint angles using current-torque model from reference [24]
    Compensates self-weight; requires a robot model and calibration of current to torque, so it is platform-specific.
axioms (4)
  • domain assumption The plant is a linear second-order system with moment of inertia J and viscous damping D (Eq. 7), and current control is ideal so current is proportional to torque.
    Used in the numerical simulation to compare methods; real actuators have additional friction, saturation, and current dynamics.
  • domain assumption The environment contact is modeled as a one-sided linear spring with stiffness K_env (Eq. 10).
    Contact in the simulation; real contacts are often nonlinear and dissipative.
  • domain assumption The follower's control input uf is a reliable real-time indicator of the external force ff.
    This is the load-bearing assumption of the gating law; it holds only for backdrivable actuators with low friction, as acknowledged in Section V.
  • domain assumption The high-level system operates at a lower cycle rate than the edge controller, and a zero-order hold is used for commands.
    Used in the simulation and matches the actuator-module architecture; the method's robustness claim depends on this separation.

pith-pipeline@v1.3.0-alltime-deepseek · 13101 in / 11176 out tokens · 121979 ms · 2026-08-04T20:58:57.278711+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Input-gated Bilateral Teleoperation: An Easy-to-implement Force Feedback Teleoperation Method for Low-cost Hardware." pith.science (2026). https://pith.science/paper/AF4SRFV6

@misc{pith2026250908226,
  author       = {Pith},
  title        = {Pith review of: Input-gated Bilateral Teleoperation: An Easy-to-implement Force Feedback Teleoperation Method for Low-cost Hardware},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AF4SRFV6}},
  note         = {Machine review of arXiv:2509.08226}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Effective data collection in contact-rich manipulation requires force feedback during teleoperation, as accurate perception of contact is crucial for stable control. However, such technology remains uncommon, largely because bilateral teleoperation systems are complex and difficult to implement. To overcome this, we propose a bilateral teleoperation method that relies only on a simple feedback controller and does not require force sensors. The approach is designed for leader-follower setups using low-cost hardware, making it broadly applicable. Through numerical simulations and real-world experiments, we demonstrate that the method requires minimal parameter tuning, yet achieves both high operability and contact stability, outperforming conventional approaches. Furthermore, we show its high robustness: even at low communication cycle rates between leader and follower, control performance degradation is minimal compared to high-speed operation. We also prove our method can be implemented on two types of commercially available low-cost hardware with zero parameter adjustments. This highlights its high ease of implementation and versatility. We expect this method will expand the use of force feedback teleoperation systems on low-cost hardware. This will contribute to advancing contact-rich task autonomy in imitation learning.

Figures

Figures reproduced from arXiv: 2509.08226 by Akira Kanazawa, Hideyuki Ichiwara, Hiroshi Ito, Naoaki Noguchi, Tetsuya Ogata, Yoshiki Kanai.

Figure 1
Figure 1. Figure 1: Block diagram of the generalized IGBT. Both the [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Configuration diagram of bilateral teleoperation us [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Numerical simulation results. The time responses of [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: experimental environment. A pair of Dynamixel [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Results for Experiments on actuator modules.(a) Pos [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: experimental environment for low-cost hardware. (a [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: experimental results of the USB Insertion Task [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗

discussion (0)

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

Forward citations

Cited by 1 Pith paper

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

  1. Refinement of Accelerated Demonstrations via Incremental Iterative Reference Learning Control for Fast Contact-Rich Imitation Learning

    cs.RO 2026-04 unverdicted novelty 6.0

    Incremental Iterative Reference Learning Control refines accelerated demonstrations to achieve up to 10x faster execution in contact-rich imitation learning with 22.5% better trajectory similarity than direct IRLC and...

Reference graph

Works this paper leans on

31 extracted references · 27 canonical work pages · cited by 1 Pith paper

  1. [1]

    Imitation Learning: A Survey of Learn ing Meth- ods,

    A. Hussein et al., “Imitation Learning: A Survey of Learn ing Meth- ods,” ACM Computing Survey , 50(2), pp. 1–35, 2017

    work page 2017

  2. [2]

    Human-to-Robot Imitation in the Wild,

    S. Bahl et al., “Human-to-Robot Imitation in the Wild,” i n Proc. Robotics: Science and Systems (RSS) , 2022

    work page 2022

  3. [3]

    Learning Fine-Grained Bimanual Manip ulation with Low-Cost Hardware,

    T. Z. Zhao et al., “Learning Fine-Grained Bimanual Manip ulation with Low-Cost Hardware,” in Proc. Robotics: Science and Systems (RSS) , 2023

    work page 2023

  4. [4]

    GELLO: A General, Low-Cost, and Intuitive T eleopera- tion Framework for Robot Manipulators,

    P . Wu et al., “GELLO: A General, Low-Cost, and Intuitive T eleopera- tion Framework for Robot Manipulators,” in Proc. IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS) , pp. 12156-12163, 2024

    work page 2024

  5. [5]

    Open X-Embodiment: Robotic Learning Datasets and RT-X Models,

    A. O’Neill et al., “Open X-Embodiment: Robotic Learning Datasets and RT-X Models,” in Proc. IEEE Int. Conf. Robotics Automation (ICRA), pp. 6892–6903, 2024

    work page 2024

  6. [6]

    π0: A Vision-Language-Action Flow Model for General Robot Control,

    K. Black et al., “ π0: A Vision-Language-Action Flow Model for General Robot Control,” arXiv preprint arXiv:2410.24164 , 2024

    Pith/arXiv arXiv 2024

  7. [7]

    Using human gaze to improve robustness ag ainst irrelevant objects in robot manipulation tasks,

    H. Kim, et al., “Using human gaze to improve robustness ag ainst irrelevant objects in robot manipulation tasks,” IEEE Robotics and Automation Letters, vol. 5, no. 3, pp. 4415–4422, 2020

    work page 2020

  8. [8]

    Bilateral control of master-slave ma- nipulators for ideal kinesthetic coupling-formulation an d experiment,

    Y . Y okokohji and T. Y oshikawa, “Bilateral control of master-slave ma- nipulators for ideal kinesthetic coupling-formulation an d experiment,” IEEE Trans. on Robotics and Automation , vol. 10, no. 5, pp. 605–620, 1994

    work page 1994

  9. [9]

    Experimental and simulat ion studies of hard contact in force reflecting teleoperation,

    B. Hannaford and R. Anderson, “Experimental and simulat ion studies of hard contact in force reflecting teleoperation,” in Proc. IEEE Int. Conf. Robotics Automation (ICRA) , vol. 1, pp. 584–589, 1988

    work page 1988

  10. [10]

    Imitation Learning for Nonprehensi le Manipulation Through Self-Supervised Learning Considering Motion Spee d,

    Y . Saigusa et al., “Imitation Learning for Nonprehensi le Manipulation Through Self-Supervised Learning Considering Motion Spee d,” IEEE Access, V ol. 10, pp. 68291-68306, 2022

    work page 2022

  11. [11]

    ALPHA- α and Bi-ACT Are All Y ou Need: Importance of Position and Force Control and Information in Imitation Learning for Unimanual and Bimanual Robotic Manipulation,

    M. Kobayashi et al., “ALPHA- α and Bi-ACT Are All Y ou Need: Importance of Position and Force Control and Information in Imitation Learning for Unimanual and Bimanual Robotic Manipulation, ” IEEE Access, vol. 13, pp. 29886-29899, 2025

    work page 2025

  12. [12]

    FACTR: Force-attending curriculum tr aining for contact-rich policy learning,

    J. J. Liuet al., “FACTR: Force-attending curriculum tr aining for contact-rich policy learning,” arXiv preprint arXiv:2502.17432 , 2025

    Pith/arXiv arXiv 2025

  13. [13]

    A Survey on Imitation Learning for Cont act-Rich Tasks in Robotics,

    T. Tsuji et al., “A Survey on Imitation Learning for Cont act-Rich Tasks in Robotics,” arXiv preprint arXiv:2506.13498 , 2025

    Pith/arXiv arXiv 2025

  14. [14]

    Reproducibility and Operationality in Bilateral Teleop- eration,

    W. Iida et al., “Reproducibility and Operationality in Bilateral Teleop- eration,” in proc. IEEE Int. W ork. on Advanced Motion Control (AMC), 2004

    work page 2004

  15. [15]

    Real haptics and its appli cations,

    K. Ohnishi and T. Mizoguchi, “Real haptics and its appli cations,” IEEJ Trans. on Electrical and Electronic Engineering , vol. 12, no. 6, pp. 803–808, 2017

    work page 2017

  16. [16]

    Stability and Robustness of Dist urbance-Observer- Based Motion Control Systems,

    E. Sariyildiz et al., “Stability and Robustness of Dist urbance-Observer- Based Motion Control Systems,” IEEE Trans. on Industrial Electron- ics, V ol. 62, No. 1, pp. 414–422, 2015

    work page 2015

  17. [17]

    Digital control of dynamic syste ms,

    G. F. Franklinet al., “Digital control of dynamic syste ms,” V ol. 3. Menlo Park: Addison-wesley, 1998

    work page 1998

  18. [18]

    Mobile aloha: Learning bimanual mobile ma nipulation with low-cost wholebody teleoperation,

    Z. Fu et al., “Mobile aloha: Learning bimanual mobile ma nipulation with low-cost wholebody teleoperation,” in Proc. 8th Annual Conf. on Robot Learning (CoRL) , 2024

    work page 2024

  19. [19]

    Imitation Learning Based on Bilate ral Control for Human–Robot Cooperation,

    A. Sasagawa et al., “Imitation Learning Based on Bilate ral Control for Human–Robot Cooperation,” IEEE Robotics and Automation Letters , vol. 5, no. 4, pp. 6169–6176, 2020

    work page 2020

  20. [20]

    Universal manipulation interface: In-t he-wild robot teaching without in-the-wild robots,

    C. Chi et al., “Universal manipulation interface: In-t he-wild robot teaching without in-the-wild robots,” in Proc. Robotics: Science and Systems (RSS) , 2024

    work page 2024

  21. [21]

    ForceMimic: Force-Centric Imitation Le arning with Force-Motion,

    W. Liu et al., “ForceMimic: Force-Centric Imitation Le arning with Force-Motion,” arXiv preprint arXiv:2410.07554, 2024

    Pith/arXiv arXiv 2024

  22. [22]

    Fang et al., ”AirExo: Low-Cost Exoskeletons for Lear ning Whole- Arm Manipulation in the Wild,” in Proc

    H. Fang et al., ”AirExo: Low-Cost Exoskeletons for Lear ning Whole- Arm Manipulation in the Wild,” in Proc. IEEE Int. Conf. Robotics Automation (ICRA) , pp. 15031–15038, 2024

    work page 2024

  23. [23]

    ACE: A Cross-Platform Visual-Exoskele tons System for Low-Cost Dexterous Teleoperation,

    S. Y ang et al., “ACE: A Cross-Platform Visual-Exoskele tons System for Low-Cost Dexterous Teleoperation,” in Proc. 8th Annual Conf. on Robot Learning (CoRL) , 2024

    work page 2024

  24. [24]

    Dynamixel,

    ROBOTIS, “Dynamixel,” [Online]. Abailable: https://emanual. robotis.com/docs/en/dxl/

    work page

  25. [25]

    Proprioceptive actuator design i n the mit cheetah: Impact mitigation and high-bandwidth physical in teraction for dynamic legged robots,

    P . M. Wensing et al., “Proprioceptive actuator design i n the mit cheetah: Impact mitigation and high-bandwidth physical in teraction for dynamic legged robots,” IEEE Trans. on Robotics , V ol. 33, No. 3, pp. 509–522, 2017

    work page 2017

  26. [26]

    Quasi-direct drive for low-cost com pliant robotic manipulation,

    D. V . Gealy et al., “Quasi-direct drive for low-cost com pliant robotic manipulation,” in Proc. IEEE Int. Conf. Robotics Automation (ICRA) , pp. 437–443, 2019

    work page 2019

  27. [27]

    Design and Experimental Cha racterisation of a Novel Quasi-Direct Drive Actuator for Highly Dynamic Ro botic Applications,

    C. A. P´ erez-D´ ıaz et al., “Design and Experimental Cha racterisation of a Novel Quasi-Direct Drive Actuator for Highly Dynamic Ro botic Applications,” in Proc. IEEE Int. Conf. Robotics Automation (ICRA) , pp. 183–189, 2024

    work page 2024

  28. [28]

    Motion Control for Advanced Mechatr onics,

    K. Ohnishi et al., “Motion Control for Advanced Mechatr onics,” IEEE/ASME Trans. on Mechatronics , V ol. 1, No. 1, pp. 56–67, 1996

    work page 1996

  29. [29]

    Torque sensorless control in Multi degree-of- Freedom Manipulator,

    T. Murakami et al., “Torque sensorless control in Multi degree-of- Freedom Manipulator,” IEEE Trans. on Industrial Electronics , V ol. 40, No. 2, pp. 259–265, 1993

    work page 1993

  30. [30]

    Multi-DOF Micro-Macro Bilateral Co ntroller Using Oblique Coordinate Control,

    S. Sakaino et al., “Multi-DOF Micro-Macro Bilateral Co ntroller Using Oblique Coordinate Control,” IEEE Trans. on Industrial Informatics , V ol. 7, No. 3, pp. 446-454, 2011

    work page 2011

  31. [31]

    OpenMANIPULA TOR-X,

    ROBOTIS, “OpenMANIPULA TOR-X,” [Online]. Abailable: https://emanual.robotis.com/docs/en/platform/ openmanipulator_x/overview/

    work page