Pith. sign in

REVIEW 4 major objections 5 minor 35 references

This paper claims that a single framework can turn a therapist's RGB-D video demonstration into safe, self-paced robot-assisted rehabilitation across passive, active-assisted, and active-resistive modes, using body-centric trajectory encodi

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

2026-08-02 18:10 UTC pith:RYD2FAJN

load-bearing objection A useful, honest systems integration paper whose central 6-DoF claim is only partially backed by position-only error metrics; worth refereeing but needs repeated trials and orientation validation. the 4 major comments →

arxiv 2603.14160 v2 pith:RYD2FAJN submitted 2026-03-15 cs.RO

See, Learn, Assist: Safe and Self-Paced Robotic Rehabilitation via Video-Based Learning from Demonstration

Ali Alabbas , Camillo Murgia , Joanne Regan , Philip Long This is my paper
classification cs.RO
keywords robot-assisted rehabilitationlearning from demonstrationdynamic movement primitivesGaussian mixture regressionvirtual tunneleffort-modulated controlbody-centric encodingtrajectory reversal
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.

The paper aims to let a therapist teach a collaborative robot an upper-limb rehabilitation exercise simply by recording an RGB-D video, and to have the robot then deliver that exercise safely to a patient. The proposed pipeline encodes the demonstration as a body-centric 6-degree-of-freedom trajectory using Cartesian Dynamic Movement Primitives, so the same motion can be scaled to patients of different limb lengths and postures. Execution is governed by a hybrid controller that keeps the arm inside a virtual spatial tunnel and couples the exercise's speed to how hard the patient pushes tangentially, yielding passive, active-assisted, and active-resistive modes in one framework. A personalized Gaussian Mixture Regression model monitors interaction force and triggers a smooth reversal if the force leaves a learned safe corridor. The paper reports average trajectory reproduction error of 3.7 cm, ROM error of 5.5 degrees, and reversal reaction time of about 0.3 seconds.

Core claim

The central claim is that one closed-loop framework can take a markerless RGB-D video demonstration of a therapist performing an exercise, encode it as a body-centric 6-DoF Cartesian DMP, and execute it through a decoupled controller that separates spatial path enforcement from temporal progression. The effort-modulated phase law, τ dot s = -max(0, ε + γ f_t) s, ties the DMP phase advance to the patient's tangential force; by tuning the gains γ and ε, the same law produces passive, active-assisted, and active-resistive modalities. A Gaussian Mixture Regression model learned on-the-fly from the patient's own limb predicts a force corridor, and any violation triggers a smooth reversal along th

What carries the argument

The two central mechanisms are body-centric Cartesian Dynamic Movement Primitives (DMPs), which encode the therapist's motion relative to the patient's shoulder and allow limb-length scaling, and the effort-modulated phase law that couples exercise progress to measured tangential force. Around these, the controller builds a stiff virtual tunnel for orthogonal compliance and a personalized Gaussian Mixture Regression (GMR) force corridor that triggers smooth trajectory reversal on anomaly. The DMP canonical phase, rather than clock time, is the shared variable linking the spatial trajectory, the effort-based dilation, and the safety model.

Load-bearing premise

The whole 6-DoF reproduction claim rests on the markerless vision pipeline giving trustworthy end-effector orientation—especially axial forearm rotation—which the paper itself says is hard and does not validate with orientation error numbers.

What would settle it

Measure the reproduced 6-DoF trajectory against marker-based motion capture during a continuous pronation/supination exercise and report orientation RMSE. If the orientation error is large while position RMSE stays at 3.7 cm, the claim that the system reproduces 6-DoF therapeutic motion fails even though the position metric passes.

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

If this is right

  • One recorded video can produce a robot program that preserves the exercise's spatial shape for patients with different arm lengths and seated postures; the reported data show normalized reach-ratio deviations under 4% after scaling.
  • A single controller can switch between passive, active-assisted, and active-resistive therapy by changing two gains, so recovery-stage transitions require no new hardware or reprogramming.
  • Because progression is tied to tangential force, the exercise self-paces: if a patient stops pushing, the robot slows to a baseline; if a patient resists, the phase stalls rather than forcing motion.
  • Abnormal interaction forces—spasms, pain, or range limits—can be detected from the patient's own limb data and answered by reversing along the safe portion of the trajectory within roughly 0.3 s, rather than stopping abruptly.
  • Body-centric encoding plus limb-length scaling could allow the same demonstration to serve patients of very different body sizes, which is a precondition for remote, video-based therapy prescription.

Where Pith is reading between the lines

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

  • The effort-modulated phase variable could be repurposed as a continuous measure of patient engagement, giving clinicians a quantitative record of active participation per repetition without any extra sensors.
  • If continuous axial-orientation estimation is solved, the same pipeline would cover nearly all common upper-limb exercises; currently the paper validates position strongly but does not report orientation error for continuous rotations, so the 6-DoF claim in its strongest form is only partially supported.
  • A natural extension beyond the paper is to run the framework with the therapist and patient in different locations connected by a video link, which would quantify how much clinical workflow is actually saved by video-based learning from demonstration.
  • The paper's proposed integration of vision-language models to set control parameters could remove manual gain tuning and let a therapist give high-level instructions such as 'more resistance now,' turning the framework into a more autonomous therapy assistant.

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 a video-based Learning-from-Demonstration framework for upper-limb robotic rehabilitation. A therapist's exercise is captured by RGB-D cameras, converted to skeleton keypoints, expressed in a body-centric frame, and encoded as a 6-DoF Cartesian DMP. The robot executes the DMP through a decoupled hybrid controller: a stiff admittance in the orthogonal axes provides a virtual tunnel, while the DMP phase is modulated by the patient's tangential force. By tuning two gains, passive, active-assisted, and active-resistive modes are obtained. A GMR model, learned from the patient's own limb forces, defines a safety corridor that triggers trajectory reversal. Experiments on a UR5e with three exercises report position reproduction RMSE below 5 cm, a mean ROM error of 5.5°, and a reversal reaction time of about 0.3 s.

Significance. The paper addresses a practically important problem: enabling remote, unobtrusive therapist demonstrations for robot-assisted rehabilitation. The hardware experiments are real, include independent mocap ground truth, and demonstrate that the full pipeline—video capture, body-centric scaling, DMP execution, virtual-tunnel compliance, and force-triggered reversal—can run at 100 Hz on a cobot. These are meaningful feasibility results. The main contributions are the integration of existing components (YOLO, DMP, GMR, admittance control) into a single system and the explicit coupling of patient effort to DMP phase. If the validation were strengthened, the framework would be a useful step toward practical remote programming of rehabilitation robots. However, as presented, the 6-DoF orientation claim, the safety guarantee, and the statistical basis of the headline numbers are not yet established.

major comments (4)
  1. [Abstract; Section III-C, Eq. (1)] The claim that the system 'guarantee[s] safety' is not supported. Eq. (1) is an ad hoc phase-modulation law with no passivity/stability analysis; there is no proof that the coupled human-robot system remains stable, that the phase variable stays bounded for all force profiles, or that the orthogonal admittance cannot inject energy. The GMR corridor is a heuristic (±5σ) rather than a formal safety boundary. Please either provide a rigorous guarantee under stated assumptions or replace 'guarantee' with an appropriate weaker claim, and report the safety mechanism's false-trigger and missed-trigger behavior.
  2. [Section III-A; Table I; Section V] The 6-DoF reproduction claim is not validated. The orientation is reconstructed from keypoints plus a knuckle-vector heuristic (Section III-A), and Section V admits the pipeline 'struggles to capture continuous axial joint rotations.' Table I reports only translational RMSE (cm); Fig. 3 shows only flexion/extension and abduction/adduction and omits the axial shoulder internal–external rotation angle; Section IV-A3 reports a goniometer tracking error but no end-to-end orientation reproduction error. Thus the claim that the system faithfully reproduces 6-DoF therapeutic motion is unsupported. Please report orientation RMSE per axis, especially for internal–external rotation, or explicitly revise the contribution to 3-D position with discrete grasp-orientation recognition.
  3. [Section IV-A, Table I, Table II] The experiments lack repetitions and error bars. The baseline reproduction uses a single healthy subject (Table I), the anatomical-scaling trial uses four subjects but no repeated trajectories, and the posture-invariance test (Section IV-A1) is reported without repeated trials. Consequently the headline numbers '3.7 cm' and '5.5°' have no variance estimate and cannot be statistically assessed. Please repeat each condition at least 3–5 times and report mean ± standard deviation and per-trial data. This is essential before drawing quantitative conclusions.
  4. [Section IV-C, Fig. 5] Figure 5 is a demonstration of the control law itself, not an independent test of the contribution. Since Eq. (1) defines ṡ = −max(0, ε+γ f_t) s, the monotonic relationship between applied force and exercise progress is true by construction; observing it in Fig. 5 confirms implementation, not therapeutic efficacy. Please frame Fig. 5 as controller verification and add a quantitative check (e.g., measured phase rate versus the predicted ε+γ f_t, or a goodness-of-fit metric) to make the validation informative.
minor comments (5)
  1. [Eq. (1)] The canonical-system equation should state the domain of s (e.g., s ∈ [1, 0]), the initial condition, and how 'progress' in Fig. 5 is computed from s. As written, the reader cannot map s to the percentage axis.
  2. [Section IV-A3] The text says the low reflectivity of the OptiTrack suit degraded visual tracking. Please clarify which camera/system this refers to and how it affects the claimed ground truth, since the suit is also the mocap target.
  3. [Fig. 4 caption] The caption says 'energy-based temporal dilation,' but the mechanism in Eq. (1) is force-based, not energy-based. Use consistent terminology.
  4. [Section III-C1] The hybrid controller is described for translational axes only; it is unclear whether orientation is also governed by a virtual tunnel or only the position. Please specify the orientation-control law and how it interacts with the force-based phase modulation.
  5. [Section IV-D] Minor editing: 'measured from the moment the force violation occurs to the start of the reversal motion' would read better as 'measured from the moment the force violation occurs to the start of the reversal motion.'

Circularity Check

0 steps flagged

No significant circularity: the core reproduction, ROM, spatial-adherence, and reaction-time claims are benchmarked against independent ground truth; the effort-based progression law is a design specification, not a fitted prediction.

full rationale

The paper's main quantitative claims are verified against external references: reproduction RMSE is computed against OptiTrack motion capture (Table I), ROM deviation against mocap joint angles (Fig. 3, Sec. IV-A3), spatial adherence against a recorded reference path (Fig. 4), and reversal reaction time as a measured latency (Sec. IV-D). These are independent measurements, not consequences of the control equations. The effort-modulated phase law in Eq. (1) is a proposed controller: progress is defined to depend on f_t, so Fig. 5 shows the implemented robot following its own specification rather than confirming an independently derived prediction. That is an engineering verification, not circular derivation. Similarly, the body-centric frame and limb-length scaling are algebraic constructions; the reported RMSE values mainly quantify perception noise and implementation fidelity, not a discovered empirical law. No self-citation chain is load-bearing: [28] and [32] are external prior works. Section V's admitted limitation that 'the vision pipeline struggles to capture continuous axial joint rotations' and the absence of orientation error in Table I are genuine external-validity/correctness concerns for the 6-DoF claim, but they are not circularity: they weaken the claim's evidence without making the derivation self-referential. Overall, the derivation chain is self-contained against external benchmarks, so the circularity score is 0.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

The central claims rest on standard robotics tools (DMPs, GMR, admittance control) plus hand-tuned control gains and several domain assumptions about body fixation and force measurement. No new physical entities are introduced.

free parameters (4)
  • γ (effort sensitivity gain) = 0.04 (tunnel validation), 0.08 (active-assisted), 0.005 (active-resistive)
    Hand-tuned in Sections IV-B/C to set how strongly tangential force modulates the DMP phase; directly determines modality responsiveness.
  • ε (baseline progression rate) = 0.001 in active modes; 0.5 in reference trajectory
    Hand-tuned baseline phase speed; keeps the robot moving slowly when the patient stops pushing.
  • A_stiff (orthogonal admittance gain) = 0.005 I_{3×3}
    Hand-tuned stiffness of the virtual tunnel; controls how much sideways motion is permitted.
  • GMR safety corridor width (±5σ) = ±5 standard deviations
    Chosen design margin inspired by [28], not derived from a risk analysis.
axioms (5)
  • standard math Cartesian DMPs can encode and spatially generalize 6-DoF trajectories including orientation.
    The learning module relies on [32]; the paper does not independently verify orientation generalization.
  • domain assumption The patient's trunk and non-involved body segments remain fixed during exercise.
    Explicitly acknowledged in the Conclusions as an adopted assumption; it affects body-centric scaling and ROM interpretation.
  • domain assumption The wrist force sensor reading f_ex, projected onto the DMP reference tangent, is a clean measure of voluntary tangential effort.
    Eq. (1) uses f_t to pace therapy; no gravity/dynamics compensation or sensor-noise analysis is reported.
  • domain assumption A GMM/GMR fitted to forces collected during a gravity-compensated baseline on the patient's own limb is representative of normal interaction forces.
    The ±5σ corridor is built from this model; if the baseline includes pathological forces, detection will fail.
  • ad hoc to paper The phase-modulation law τṡ = −max(0, ε+γ f_t)s is stable and safe for all interaction conditions.
    Introduced without stability or passivity proof; behavior is only demonstrated in pilot trials.

pith-pipeline@v1.3.0-alltime-deepseek · 10065 in / 12645 out tokens · 122840 ms · 2026-08-02T18:10:19.249414+00:00 · methodology

0 comments
read the original abstract

In this paper, we propose a novel framework that allows therapists to teach robot-assisted rehabilitation exercises remotely via RGB-D video. Our system encodes demonstrations as 6-DoF body-centric trajectories using Cartesian Dynamic Movement Primitives (DMPs), ensuring accurate posture-independent spatial generalisation across diverse patient anatomies. Crucially, we execute these trajectories through a decoupled hybrid control architecture that constructs a spatially compliant virtual tunnel, paired with an effort-based temporal dilation mechanism. This architecture is applied to three distinct rehabilitation modalities: Passive, Active-Assisted, and Active-Resistive, by dynamically linking the exercise's execution phase to the patient's tangential force contribution. To guarantee safety, a Gaussian Mixture Regression (GMR) model is learned on-the-fly from the patient's own limb. This allows the detection of abnormal interaction forces and, if necessary, reverses the trajectory to prevent injury. Experimental validation demonstrates the system's precision, achieving an average trajectory reproduction error of 3.7cm and a range of motion (ROM) error of 5.5 degrees. Furthermore, dynamic interaction trials confirm that the controller successfully enforces effort-based progression while maintaining strict spatial path adherence against human disturbances.

Figures

Figures reproduced from arXiv: 2603.14160 by Ali Alabbas, Camillo Murgia, Joanne Regan, Philip Long.

Figure 1
Figure 1. Figure 1: Hardware architecture of the system (left) and body [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: A GMR model learned from a subject’s limb during [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of time-normalized joint angle trajecto [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: Exercise progress vs patient’s delivered force in the [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Force monitoring and adaptive reversibility during the three rehabilitation exercises. The green line indicates exercise [PITH_FULL_IMAGE:figures/full_fig_p006_6.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

35 extracted references · 3 linked inside Pith

  1. [1]

    Ageing europe - statictics on population developments,

    “Ageing europe - statictics on population developments,”

  2. [2]

    The use of sports rehabilitation robotics to assist in the recovery of physical abilities in elderly patients with degenerative diseases: A literature review,

    F. Ju, Y . Wang, B. Xie, Y . Mi, M. Zhao, and J. Cao, “The use of sports rehabilitation robotics to assist in the recovery of physical abilities in elderly patients with degenerative diseases: A literature review,” Healthcare (Basel), vol. 11, no. 3, p. 326, 2023

    2023

  3. [3]

    Robotics in physical rehabilitation: Systematic review,

    A. D. Banyai and C. Bris ,an, “Robotics in physical rehabilitation: Systematic review,” inHealthcare, vol. 12, no. 17. MDPI, 2024, p. 1720

    2024

  4. [4]

    S. H. Lee, G. Park, D. Y . Cho, H. Y . Kim, J.-Y . Lee, S. Kim, S.-B. Park, and J.-H. Shin, “Comparisons between end-effector and exoskeleton rehabilitation robots regarding upper extremity function among chronic stroke patients with moderate-to-severe upper limb impairment,”Scientific reports, vol. 10, no. 1, p. 1806, 2020

    2020

  5. [5]

    Soft robotics in upper limb neurorehabilitation and assistance: current clinical evidence and recommendations,

    N. Tanczak, A. Yurkewich, F. Missiroli, S. K. Wee, S. Kager, H. Choi, K.-J. Cho, H. K. Yap, C. Piazza, L. Masiaet al., “Soft robotics in upper limb neurorehabilitation and assistance: current clinical evidence and recommendations,”Soft robotics, vol. 12, no. 3, pp. 303–314, 2025

    2025

  6. [6]

    Morris, R

    L. Morris, R. S. Diteesawat, N. Rahman, A. Turton, M. Cramp, and J. Rossiter, “The-state-of-the-art of soft robotics to assist mobility: a review of physiotherapist and patient identified limitations of current lower-limb exoskeletons and the potential soft-robotic solutions,” Journal of neuroengineering and rehabilitation, vol. 20, no. 1, p. 18, 2023

    2023

  7. [7]

    Robotic rehabilitation for end-effector device and botulinum toxin in upper limb rehabilitation in chronic post-stroke patients: an integrated rehabilitative approach,

    T. Paolucci, F. Agostini, M. Mangone, A. Bernetti, L. Pezzi, V . Liotti, E. Recubini, C. Cantarella, R. G. Bellomo, C. D’Aurizioet al., “Robotic rehabilitation for end-effector device and botulinum toxin in upper limb rehabilitation in chronic post-stroke patients: an integrated rehabilitative approach,”Neurological Sciences, vol. 42, no. 12, pp. 5219–5229, 2021

    2021

  8. [8]

    Mit-manus: a workstation for manual therapy and training. i,

    N. Hogan, H. I. Krebs, J. Charnnarong, P. Srikrishna, and A. Sharon, “Mit-manus: a workstation for manual therapy and training. i,” in

  9. [9]

    A compact motorized end-effector for ankle rehabilitation training,

    R. Wu, M. Luo, J. Fan, J. Ma, N. Zhang, J. Li, Q. Li, F. Gao, and G. Dan, “A compact motorized end-effector for ankle rehabilitation training,”Frontiers in Robotics and AI, vol. 11, p. 1453097, 2024

    2024

  10. [10]

    Test–retest reliability of biodex system 4 pro for isometric ankle-eversion and-inversion measurement,

    G. Tankevicius, D. Lankaite, and A. Krisciunas, “Test–retest reliability of biodex system 4 pro for isometric ankle-eversion and-inversion measurement,”Journal of sport rehabilitation, vol. 22, no. 3, pp. 212– 215, 2013

    2013

  11. [11]

    Robot-assisted upper limb rehabilitation using imitation learning,

    I. A. Auta, A. Fares, H. Iwata, and H. El-Hussieny, “Robot-assisted upper limb rehabilitation using imitation learning,”Journal of Robotics and Control (JRC), vol. 6, no. 1, pp. 89–100, 2025

    2025

  12. [12]

    Mimicplay: Long-horizon imitation learning by watching human play,

    C. Wang, L. Fan, J. Sun, R. Zhang, L. Fei-Fei, D. Xu, Y . Zhu, and A. Anandkumar, “Mimicplay: Long-horizon imitation learning by watching human play,”arXiv preprint arXiv:2302.12422, 2023

    Pith/arXiv arXiv 2023

  13. [13]

    Okami: Teaching humanoid robots manipulation skills through single video imitation,

    J. Li, Y . Zhu, Y . Xie, Z. Jiang, M. Seo, G. Pavlakos, and Y . Zhu, “Okami: Teaching humanoid robots manipulation skills through single video imitation,”arXiv preprint arXiv:2410.11792, 2024

    Pith/arXiv arXiv 2024

  14. [14]

    Dexmv: Imitation learning for dexterous manipulation from human videos,

    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,” inEuropean Conference on Computer Vision. Springer, 2022, pp. 570–587

    2022

  15. [15]

    Research on adaptive impedance control technology of upper limb rehabilitation robot based on impedance parameter prediction,

    Y . Zhang, T. Li, H. Tao, F. Liu, B. Hu, M. Wu, and H. Yu, “Research on adaptive impedance control technology of upper limb rehabilitation robot based on impedance parameter prediction,”Frontiers in Bioengi- neering and Biotechnology, vol. 11, p. 1332689, 2024

    2024

  16. [16]

    User-adaptive variable impedance control using bayesian optimization for robot-aided ankle rehabilitation,

    G. Manoharan and H. Lee, “User-adaptive variable impedance control using bayesian optimization for robot-aided ankle rehabilitation,”IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2025

    2025

  17. [17]

    Adaptive patient-cooperative compliant control of lower limb rehabilitation robot,

    L. Chen, J. Huang, Y . Wang, S. Guo, M. Wang, and X. Guo, “Adaptive patient-cooperative compliant control of lower limb rehabilitation robot,”Biomimetic Intelligence and Robotics, vol. 4, no. 2, p. 100155, 2024

    2024

  18. [18]

    Impedance control for robotic rehabilita- tion: a robust markovian approach,

    A. L. Jutinico, J. C. Jaimes, F. M. Escalante, J. C. Perez-Ibarra, M. H. Terra, and A. A. Siqueira, “Impedance control for robotic rehabilita- tion: a robust markovian approach,”Frontiers in neurorobotics, vol. 11, p. 43, 2017

    2017

  19. [19]

    A rehabilitation robot control framework with adaptation of training tasks and robotic assistance,

    J. Xu, K. Huang, T. Zhang, K. Cao, A. Ji, L. Xu, and Y . Li, “A rehabilitation robot control framework with adaptation of training tasks and robotic assistance,”Frontiers in bioengineering and biotechnology, vol. 11, p. 1244550, 2023

    2023

  20. [20]

    Human–robot variable-impedance skill transfer learning based on dynamic movement primitives and a vision system,

    H. Zhang, F. Peng, and M. Cai, “Human–robot variable-impedance skill transfer learning based on dynamic movement primitives and a vision system,”Sensors, vol. 25, no. 18, p. 5630, 2025

    2025

  21. [21]

    Visual imitation learning from one- shot demonstration for multi-step robot pick and place tasks,

    S. Lu, C. H ¨ardtlein, and J. Schilp, “Visual imitation learning from one- shot demonstration for multi-step robot pick and place tasks,”Scientific Reports, 2025

    2025

  22. [22]

    Emergence of human to robot transfer in vision- language-action models,

    S. Kareer, K. Pertsch, J. Darpinian, J. Hoffman, D. Xu, S. Levine, C. Finn, and S. Nair, “Emergence of human to robot transfer in vision- language-action models,”arXiv preprint arXiv:2512.22414, 2025

    arXiv 2025

  23. [23]

    Physical human- robot interaction mediates the association of motor impairment and kinematic performance for poststroke arm rehabilitation,

    Z.-J. Chen, Y . Chen, J. Xu, X.-L. Huang, and C. He, “Physical human- robot interaction mediates the association of motor impairment and kinematic performance for poststroke arm rehabilitation,”BMC Sports Science, Medicine and Rehabilitation, vol. 17, no. 1, p. 310, 2025

    2025

  24. [24]

    Integrating subject- specific workspace constraint and performance-based control strategy in robot-assisted rehabilitation,

    Q. Miao, S. Min, C. Wang, and Y .-F. Chen, “Integrating subject- specific workspace constraint and performance-based control strategy in robot-assisted rehabilitation,”Frontiers in neuroscience, vol. 18, p. 1473755, 2024

    2024

  25. [25]

    Research of intent recognition in rehabilitation robots: a systematic review,

    S. Luo, Q. Meng, S. Li, and H. Yu, “Research of intent recognition in rehabilitation robots: a systematic review,”Disability and Rehabil- itation: Assistive Technology, vol. 19, no. 4, pp. 1307–1318, 2024

    2024

  26. [26]

    An intelligent rehabilitation robot with passive and active direct switching training: improving intelligence and security of human–robot interaction systems,

    P. Sun, R. Shan, and S. Wang, “An intelligent rehabilitation robot with passive and active direct switching training: improving intelligence and security of human–robot interaction systems,”IEEE Robotics & Automation Magazine, vol. 30, no. 1, pp. 72–83, 2022

    2022

  27. [27]

    Safety supervisory strategy for an upper-limb rehabilitation robot based on impedance control,

    L. Pan, A. Song, G. Xu, H. Li, H. Zeng, and B. Xu, “Safety supervisory strategy for an upper-limb rehabilitation robot based on impedance control,”International Journal of Advanced Robotic Systems, vol. 10, no. 2, p. 127, 2013

    2013

  28. [28]

    Imitation learning-based system for the exe- cution of self-paced robotic-assisted passive rehabilitation exercises,

    R. J. Escarabajal, J. L. Pulloquinga, P. Zamora-Ortiz, ´A. Valera, V . Mata, and M. Vall´es, “Imitation learning-based system for the exe- cution of self-paced robotic-assisted passive rehabilitation exercises,” IEEE Robotics and Automation Letters, vol. 8, no. 7, pp. 4283–4290, 2023

    2023

  29. [29]

    Ultralytics yolov8,

    G. Jocher, A. Chaurasia, and J. Qiu, “Ultralytics yolov8,” 2023. [Online]. Available: https://github.com/ultralytics/ultralytics

    2023

  30. [30]

    Yolo-pose: Enhancing yolo for multi person pose estimation using object keypoint similarity loss,

    D. Maji, S. Nagori, M. Mathew, and D. Poddar, “Yolo-pose: Enhancing yolo for multi person pose estimation using object keypoint similarity loss,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 2637–2646

    2022

  31. [31]

    Mediapipe: A framework for building perception pipelines,

    C. Lugaresi, J. Tang, H. Nash, C. McClanahan, E. Uboweja, M. Hays, F. Zhang, C.-L. Chang, M. G. Yong, J. Leeet al., “Mediapipe: A framework for building perception pipelines,”arXiv preprint arXiv:1906.08172, 2019

    Pith/arXiv arXiv 1906

  32. [32]

    Orientation in carte- sian space dynamic movement primitives,

    A. Ude, B. Nemec, T. Petri ´c, and J. Morimoto, “Orientation in carte- sian space dynamic movement primitives,” in2014 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2014, pp. 2997–3004

    2014

  33. [33]

    The effect of proficiency level on measurement error of range of motion,

    K. Akizuki, K. Yamaguchi, Y . Morita, and Y . Ohashi, “The effect of proficiency level on measurement error of range of motion,”Journal of physical therapy science, vol. 28, no. 9, pp. 2644–2651, 2016

    2016

  34. [1992]

    IEEE, 1992, pp

    Proceedings IEEE International Workshop on Robot and Human Communication. IEEE, 1992, pp. 161–165

    1992

  35. [2024]

    Available: https://ec.europa.eu/eurostat/ statistics-explained/index.php?title=Ageing Europe - statistics on population developments

    [Online]. Available: https://ec.europa.eu/eurostat/ statistics-explained/index.php?title=Ageing Europe - statistics on population developments