arxiv: 2604.09431 · v1 · submitted 2026-04-10 · 💻 cs.RO

Recognition: no theorem link

Musculoskeletal Motion Imitation for Learning Personalized Exoskeleton Control Policy in Impaired Gait

Itak Choi , Ilseung Park , Eni Halilaj , Inseung Kang

Authors on Pith no claims yet

Pith reviewed 2026-05-10 17:07 UTC · model grok-4.3

classification 💻 cs.RO

keywords exoskeleton controlmusculoskeletal simulationreinforcement learningimpaired gaitpersonalized assistancemetabolic costassistive torquegait symmetry

0 comments

The pith

Physiologically plausible musculoskeletal simulation with reinforcement learning learns personalized exoskeleton control policies for both able-bodied and impaired gait without task-specific tuning.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper aims to establish that combining musculoskeletal simulation with reinforcement learning creates a scalable framework for designing exoskeleton assistance tailored to individual users, including those with gait impairments. This approach would matter because traditional methods demand extensive data collection or repeated optimization that limit access for clinical populations. The policies generate realistic locomotion patterns and replicate how people compensate for muscle deficits. If the claim holds, exoskeleton assistance can be generated that matches validated human designs, reduces energy expenditure across speeds, and restores symmetry in impaired models without prescribing a specific gait target.

Core claim

By training within physiologically plausible musculoskeletal simulations, reinforcement learning produces device-agnostic exoskeleton control policies that generate natural locomotion dynamics, capture clinically observed compensatory strategies under targeted muscular deficits, and deliver hip and ankle assistance that aligns with state-of-the-art profiles while reducing metabolic cost across walking speeds. For simulated impaired-gait models the same policies generate asymmetric, deficit-specific torque that improves energetic efficiency and bilateral kinematic symmetry without any explicit target gait pattern or additional tuning.

What carries the argument

The device-agnostic framework that integrates physiologically plausible musculoskeletal simulation with reinforcement learning to imitate natural motion and learn control policies.

If this is right

Assistive torque profiles at the hip and ankle align with state-of-the-art profiles validated in human experiments without task-specific tuning.
Metabolic cost is consistently reduced across multiple walking speeds for able-bodied models.
For impaired-gait models the policies produce asymmetric, deficit-specific assistance that improves energetic efficiency and bilateral kinematic symmetry.
The method provides a unified computational model of healthy and pathological gait that captures compensatory strategies.
Extensive physical trials are eliminated because the simulation serves as a scalable foundation for personalized control.

Where Pith is reading between the lines

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

The approach could lower barriers for developing exoskeletons for clinical populations by reducing the amount of human testing required during design.
Similar simulation-plus-learning pipelines might extend to other wearable devices such as prosthetics if the underlying musculoskeletal models can be adapted.
The ability to explore many impairment scenarios in simulation before physical testing could speed up identification of effective assistance strategies for rare gait deficits.

Load-bearing premise

The musculoskeletal simulation must accurately capture real human movement strategies and compensatory behaviors, and the resulting policies must transfer directly to physical exoskeletons without further tuning or domain changes.

What would settle it

Applying the learned policies to physical exoskeletons worn by human subjects with and without gait impairments, then measuring whether the observed metabolic cost reductions, hip and ankle torque profiles, and improvements in gait symmetry match the simulation predictions.

Figures

Figures reproduced from arXiv: 2604.09431 by Eni Halilaj, Ilseung Park, Inseung Kang, Itak Choi.

**Figure 1.** Figure 1: Overview of the musculoskeletal simulation and learning framework. (a) Reference kinematics and joint moments are extracted from a biomechanics [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 2.** Figure 2: Kinematic and kinetic tracking performance of the baseline policy across five locomotor speeds (walking and running). Blue and gray lines [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** Figure 3: Generated exoskeleton assistive torque profiles (Nm/kg, normalized [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

**Figure 4.** Figure 4: Gross metabolic cost (W/kg) of the generated exoskeleton control [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

**Figure 5.** Figure 5: Kinematic and kinetic deviations of simulated impaired gaits (orange) relative to the able-bodied baseline (blue). Top row: joint angles and moments [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗

**Figure 6.** Figure 6: Exoskeleton assistive torque profiles for the affected and non [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗

read the original abstract

Designing generalizable control policies for lower-limb exoskeletons remains fundamentally constrained by exhaustive data collection or iterative optimization procedures, which limit accessibility to clinical populations. To address this challenge, we introduce a device-agnostic framework that combines physiologically plausible musculoskeletal simulation with reinforcement learning to enable scalable personalized exoskeleton assistance for both able-bodied and clinical populations. Our control policies not only generate physiologically plausible locomotion dynamics but also capture clinically observed compensatory strategies under targeted muscular deficits, providing a unified computational model of both healthy and pathological gait. Without task-specific tuning, the resulting exoskeleton control policies produce assistive torque profiles at the hip and ankle that align with state-of-the-art profiles validated in human experiments, while consistently reducing metabolic cost across walking speeds. For simulated impaired-gait models, the learned control policies yield asymmetric, deficit-specific exoskeleton assistance that improves both energetic efficiency and bilateral kinematic symmetry without explicit prescription of the target gait pattern. These results demonstrate that physiologically plausible musculoskeletal simulation via reinforcement learning can serve as a scalable foundation for personalized exoskeleton control across both able-bodied and clinical populations, eliminating the need for extensive physical trials.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This simulation-RL pipeline for exoskeleton policies in impaired gait is a reasonable proof-of-concept idea but rests on unvalidated simulation claims with no shown transfer or quantitative human data.

read the letter

The main thing to know is that the paper combines musculoskeletal simulation with reinforcement learning to generate exoskeleton control policies for both healthy and impaired gait, claiming the policies produce hip and ankle torques that match human-validated profiles and reduce metabolic cost without task-specific tuning. For impaired models it generates asymmetric assistance that improves efficiency and symmetry on its own. That framing avoids prescribing target gaits explicitly, which is a practical shift from some prior work. The device-agnostic setup is also straightforward and could apply across different hardware. The approach does a clean job of showing how RL on a physiologically plausible model can reproduce compensatory strategies under simulated muscle deficits, which is a useful computational angle for reducing reliance on exhaustive human trials. The abstract presents this as scalable for clinical populations, and the core pipeline logic holds together on its own terms. The soft spots are more substantial. The abstract makes direct claims about alignment with state-of-the-art profiles and consistent metabolic reductions, yet supplies no numbers, error bars, or details on how those outcomes were measured. All results stay in simulation; there is no quantitative match to real impaired-gait kinematics, kinetics, or EMG, and no physical exoskeleton deployment or domain-randomized transfer test is described. The assumption that the model faithfully captures real compensatory mechanisms therefore stays untested, which undercuts the transfer story. This work is aimed at robotics and biomechanics researchers who build simulation-driven controllers for wearable devices. Readers who want early-stage ideas on deficit-specific assistance without heavy data collection could find value in the framing. It is coherent enough on its own terms to deserve a serious referee who can examine the full methods, reward design, and any hidden parameter choices, even if the current evidence level is preliminary.

Referee Report

2 major / 1 minor

Summary. The manuscript introduces a device-agnostic framework combining physiologically plausible musculoskeletal simulation with reinforcement learning to learn personalized lower-limb exoskeleton control policies for able-bodied and impaired gait. It claims that the resulting policies generate plausible locomotion, capture clinically observed compensatory strategies under muscular deficits, produce hip and ankle assistive torque profiles aligning with state-of-the-art human-validated profiles without task-specific tuning, consistently reduce metabolic cost across walking speeds, and for simulated impaired models yield asymmetric deficit-specific assistance that improves energetic efficiency and bilateral kinematic symmetry.

Significance. If the simulation-to-real transfer and model fidelity claims hold, the work would offer a scalable computational route to personalized exoskeleton assistance that avoids exhaustive human data collection or iterative physical optimization, with potential impact on clinical accessibility. The approach of using RL on musculoskeletal models to reproduce both healthy and pathological gait patterns is a constructive direction for bridging simulation and robotics control.

major comments (2)

[Abstract] Abstract: The claims that policies produce SOTA-aligned torque profiles and reduce metabolic cost are presented without any quantitative metrics, error bars, simulation fidelity measures, or statistical details, which is load-bearing for assessing whether the central results support the assertion of alignment with human-validated profiles.
[Abstract and Results] Abstract and Results: The manuscript reports only simulation outcomes and asserts direct applicability to clinical populations and physical exoskeletons without additional tuning, yet provides no quantitative comparisons of simulated vs. real human kinematics, kinetics, GRFs, EMG, or metabolic data for deficit-specific impaired gait, nor any physical deployment or domain-randomization experiments. This is load-bearing for the claim that the framework eliminates the need for extensive physical trials.

minor comments (1)

[Abstract] The abstract would be strengthened by including at least one concrete quantitative result (e.g., percentage metabolic reduction or torque correlation value) to ground the claims.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive review and recognition of the potential impact of our simulation-based framework. We address each major comment point by point below, with clarifications on the scope of the work and specific revisions planned.

read point-by-point responses

Referee: [Abstract] Abstract: The claims that policies produce SOTA-aligned torque profiles and reduce metabolic cost are presented without any quantitative metrics, error bars, simulation fidelity measures, or statistical details, which is load-bearing for assessing whether the central results support the assertion of alignment with human-validated profiles.

Authors: We agree that the abstract, being a high-level summary, omits specific quantitative details present in the results section. The full manuscript includes quantitative torque alignment metrics (e.g., mean absolute errors against reference profiles), metabolic cost reductions with standard deviations across speeds, and simulation fidelity checks. In the revised manuscript, we will update the abstract to incorporate key quantitative highlights, such as average alignment errors and percentage metabolic reductions, while maintaining brevity and directing readers to the detailed results. revision: yes
Referee: [Abstract and Results] Abstract and Results: The manuscript reports only simulation outcomes and asserts direct applicability to clinical populations and physical exoskeletons without additional tuning, yet provides no quantitative comparisons of simulated vs. real human kinematics, kinetics, GRFs, EMG, or metabolic data for deficit-specific impaired gait, nor any physical deployment or domain-randomization experiments. This is load-bearing for the claim that the framework eliminates the need for extensive physical trials.

Authors: The manuscript is explicitly a simulation study demonstrating the musculoskeletal RL framework's ability to generate plausible policies and capture compensatory strategies. We do not present or claim completed sim-to-real transfer, physical deployments, or direct quantitative matches to real impaired-gait datasets in this work; the assertion regarding reduced need for physical trials is forward-looking based on the framework's design. We will revise the abstract and add explicit language in the discussion to clarify the simulation-only scope, acknowledge the absence of real-world validation as a current limitation, and outline planned future steps including domain randomization. This maintains the contribution while addressing the concern. revision: partial

Circularity Check

0 steps flagged

No circularity: simulation outcomes presented as independent results

full rationale

The paper's core derivation uses physiologically plausible musculoskeletal simulation combined with reinforcement learning to generate exoskeleton policies for healthy and impaired gait. Claims of torque profile alignment with state-of-the-art human-validated results, metabolic cost reduction, and improved symmetry are explicitly described as emergent outcomes of the learned policies rather than fitted inputs or self-referential definitions. No equations or steps reduce by construction to the inputs (e.g., no reward terms directly encoding the target torques or costs). No load-bearing self-citations or uniqueness theorems from prior author work are invoked to force the results. The approach is self-contained against external benchmarks in simulation, with no evidence of renaming known results or smuggling ansatzes via citation.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides insufficient detail to enumerate specific free parameters or axioms; typical musculoskeletal models contain many scaling factors and the RL component likely includes reward weights that are not disclosed.

pith-pipeline@v0.9.0 · 5503 in / 1189 out tokens · 46214 ms · 2026-05-10T17:07:02.915683+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

41 extracted references · 4 canonical work pages

[1]

The exoskeleton expansion: improving walking and running economy,

G. S. Sawicki, O. N. Beck, I. Kang, and A. J. Young, “The exoskeleton expansion: improving walking and running economy,”Journal of neuroengineering and rehabilitation, vol. 17, no. 1, p. 25, 2020

2020
[2]

Opportunities and challenges in the development of exoskeletons for locomotor assistance,

C. Siviy, L. M. Baker, B. T. Quinlivan, F. Porciuncula, K. Swami- nathan, L. N. Awad, and C. J. Walsh, “Opportunities and challenges in the development of exoskeletons for locomotor assistance,”Nature biomedical engineering, vol. 7, no. 4, pp. 456–472, 2023

2023
[3]

Wearable technologies for assisted mobility in the real world,

S. Gao, J. Chen, Y . Xia, X. Li, W. Ma, H. Yang, J. Li, X. Zhou, T. Jia, Y . Xu,et al., “Wearable technologies for assisted mobility in the real world,”Nature Communications, 2025

2025
[4]

A lower-extremity exoskeleton improves knee extension in children with crouch gait from cerebral palsy,

Z. F. Lerner, D. L. Damiano, and T. C. Bulea, “A lower-extremity exoskeleton improves knee extension in children with crouch gait from cerebral palsy,”Science translational medicine, vol. 9, no. 404, p. eaam9145, 2017

2017
[5]

A soft robotic exosuit improves walking in patients after stroke,

L. N. Awad, J. Bae, K. O’donnell, S. M. De Rossi, K. Hendron, L. H. Sloot, P. Kudzia, S. Allen, K. G. Holt, T. D. Ellis,et al., “A soft robotic exosuit improves walking in patients after stroke,”Science translational medicine, vol. 9, no. 400, p. eaai9084, 2017

2017
[6]

The role of user preference in the customized control of robotic exoskeletons,

K. A. Ingraham, C. D. Remy, and E. J. Rouse, “The role of user preference in the customized control of robotic exoskeletons,”Science robotics, vol. 7, no. 64, p. eabj3487, 2022

2022
[7]

Soft robotic apparel to avert freezing of gait in parkinson’s disease,

J. Kim, F. Porciuncula, H. D. Yang, N. Wendel, T. Baker, A. Chin, T. D. Ellis, and C. J. Walsh, “Soft robotic apparel to avert freezing of gait in parkinson’s disease,”Nature medicine, vol. 30, no. 1, pp. 177– 185, 2024

2024
[8]

Powered knee exoskeleton improves sit-to- stand transitions in stroke patients using electromyographic control,

A. J. Gunnell, S. V . Sarkisian, H. A. Hayes, K. B. Foreman, L. Gabert, and T. Lenzi, “Powered knee exoskeleton improves sit-to- stand transitions in stroke patients using electromyographic control,” Communications Engineering, vol. 4, no. 1, p. 104, 2025

2025
[9]

Online adaptation framework enables personaliza- tion of exoskeleton assistance during locomotion in patients affected by stroke,

I. Kang, D. D. Molinaro, D. Park, D. Lee, P. Kunapuli, K. R. Herrin, and A. J. Young, “Online adaptation framework enables personaliza- tion of exoskeleton assistance during locomotion in patients affected by stroke,”IEEE Transactions on Robotics, 2025

2025
[10]

Portable hip exoskeleton improves walking economy for stroke survivors,

K. Pruyn, R. Murray, L. Gabert, K. B. Foreman, and T. Lenzi, “Portable hip exoskeleton improves walking economy for stroke survivors,”Nature Communications, 2026

2026
[11]

Human-in-the-loop optimization of exoskeleton assistance during walking,

J. Zhang, P. Fiers, K. A. Witte, R. W. Jackson, K. L. Poggensee, C. G. Atkeson, and S. H. Collins, “Human-in-the-loop optimization of exoskeleton assistance during walking,”Science, vol. 356, no. 6344, pp. 1280–1284, 2017

2017
[12]

Human-in-the- loop optimization of hip assistance with a soft exosuit during walking,

Y . Ding, M. Kim, S. Kuindersma, and C. J. Walsh, “Human-in-the- loop optimization of hip assistance with a soft exosuit during walking,” Science robotics, vol. 3, no. 15, p. eaar5438, 2018

2018
[13]

Im- proving the energy economy of human running with powered and unpowered ankle exoskeleton assistance,

K. A. Witte, P. Fiers, A. L. Sheets-Singer, and S. H. Collins, “Im- proving the energy economy of human running with powered and unpowered ankle exoskeleton assistance,”Science Robotics, vol. 5, no. 40, p. eaay9108, 2020

2020
[14]

Reducing the energy cost of walking with low assistance levels through opti- mized hip flexion assistance from a soft exosuit,

J. Kim, B. T. Quinlivan, L.-A. Deprey, D. Arumukhom Revi, A. Eckert-Erdheim, P. Murphy, D. Orzel, and C. J. Walsh, “Reducing the energy cost of walking with low assistance levels through opti- mized hip flexion assistance from a soft exosuit,”Scientific reports, vol. 12, no. 1, p. 11004, 2022

2022
[15]

On human-in-the-loop optimization of human–robot interaction,

P. Slade, C. Atkeson, J. M. Donelan, H. Houdijk, K. A. Ingraham, M. Kim, K. Kong, K. L. Poggensee, R. Riener, M. Steinert,et al., “On human-in-the-loop optimization of human–robot interaction,”Nature, vol. 633, no. 8031, pp. 779–788, 2024

2024
[16]

Estimating human joint moments unifies exoskeleton control, reducing user effort,

D. D. Molinaro, I. Kang, and A. J. Young, “Estimating human joint moments unifies exoskeleton control, reducing user effort,”Science robotics, vol. 9, no. 88, p. eadi8852, 2024

2024
[17]

Task-agnostic exoskeleton control via biological joint moment estimation,

D. D. Molinaro, K. L. Scherpereel, E. B. Schonhaut, G. Evangelopou- los, M. K. Shepherd, and A. J. Young, “Task-agnostic exoskeleton control via biological joint moment estimation,”Nature, vol. 635, no. 8038, pp. 337–344, 2024

2024
[18]

Exo-plore: Exploring exoskeleton control space through human-aligned simulation,

G. Leem, J. Lee, J. Lee, S. Song, and J. Won, “Exo-plore: Exploring exoskeleton control space through human-aligned simulation,”arXiv preprint arXiv:2601.22550, 2026

work page arXiv 2026
[19]

Smat: Staged multi- agent training for co-adaptive exoskeleton control,

Y . Yuan, G. Androwis, and X. Zhou, “Smat: Staged multi- agent training for co-adaptive exoskeleton control,”arXiv preprint arXiv:2603.07618, 2026

work page arXiv 2026
[20]

Learning hip exoskeleton control policy via predictive neuromusculoskeletal simulation,

I. Park, C. Song, and I. Kang, “Learning hip exoskeleton control policy via predictive neuromusculoskeletal simulation,”arXiv preprint arXiv:2603.04166, 2026

work page arXiv 2026
[21]

A human lower-limb biomechanics and wearable sensors dataset during cyclic and non-cyclic activities,

K. Scherpereel, D. Molinaro, O. Inan, M. Shepherd, and A. Young, “A human lower-limb biomechanics and wearable sensors dataset during cyclic and non-cyclic activities,”Scientific Data, vol. 10, no. 1, p. 924, 2023

2023
[22]

Opensim: open-source soft- ware to create and analyze dynamic simulations of movement,

S. L. Delp, F. C. Anderson, A. S. Arnold, P. Loan, A. Habib, C. T. John, E. Guendelman, and D. G. Thelen, “Opensim: open-source soft- ware to create and analyze dynamic simulations of movement,”IEEE transactions on biomedical engineering, vol. 54, no. 11, pp. 1940– 1950, 2007

1940
[23]

Reinforcement learning- based motion imitation for physiologically plausible musculoskeletal motor control,

M. Simos, A. Silvio Chiappa, and A. Mathis, “Reinforcement learning- based motion imitation for physiologically plausible musculoskeletal motor control,”arXiv e-prints, pp. arXiv–2503, 2025

2025
[24]

Musclevae: Model-based controllers of muscle-actuated characters,

Y . Feng, X. Xu, and L. Liu, “Musclevae: Model-based controllers of muscle-actuated characters,” inSIGGRAPH Asia 2023 Conference Papers, pp. 1–11, 2023

2023
[25]

Magnet: Muscle activation generation networks for diverse human movement,

J. Park, E. Jung, J. Lee, and J. Won, “Magnet: Muscle activation generation networks for diverse human movement,” inProceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers, pp. 1–11, 2025

2025
[26]

Towards embodied ai with musclemimic: Unlocking full-body musculoskeletal motor learning at scale,

C. Li, C. Wang, B. Ziliotto, M. Simos, J. Kovecses, G. Durandau, and A. Mathis, “Towards embodied ai with musclemimic: Unlocking full-body musculoskeletal motor learning at scale,”arXiv preprint arXiv:2603.25544, 2026

work page arXiv 2026
[27]

The Hyfydy simulation software,

T. Geijtenbeek, “The Hyfydy simulation software,” 11 2021.https: //hyfydy.com

2021
[28]

Scone: Open source software for predictive simula- tion of biological motion,

T. Geijtenbeek, “Scone: Open source software for predictive simula- tion of biological motion,”Journal of Open Source Software, vol. 4, no. 38, p. 1421, 2019

2019
[29]

Emergence of natural and robust bipedal walking by learning from biologically plausible objectives,

P. Schumacher, T. Geijtenbeek, V . Caggiano, V . Kumar, S. Schmitt, G. Martius, and D. F. Haeufle, “Emergence of natural and robust bipedal walking by learning from biologically plausible objectives,” iScience, vol. 28, no. 4, 2025

2025
[30]

Openexo: An open-source modular exoskeleton to augment human function,

J. R. Williams, C. F. Cuddeback, S. Fang, D. Colley, N. Enlow, P. Cox, P. Pridham, and Z. F. Lerner, “Openexo: An open-source modular exoskeleton to augment human function,”Science Robotics, vol. 10, no. 103, p. eadt1591, 2025

2025
[31]

Deepmimic: Example-guided deep reinforcement learning of physics-based char- acter skills,

X. B. Peng, P. Abbeel, S. Levine, and M. Van de Panne, “Deepmimic: Example-guided deep reinforcement learning of physics-based char- acter skills,”ACM Transactions On Graphics (TOG), vol. 37, no. 4, pp. 1–14, 2018

2018
[32]

A model of human muscle energy expenditure,

B. R. Umberger, K. G. Gerritsen, and P. E. Martin, “A model of human muscle energy expenditure,”Computer methods in biomechanics and biomedical engineering, vol. 6, no. 2, pp. 99–111, 2003

2003
[33]

Stretching your energetic budget: how tendon compliance affects the metabolic cost of running,

T. K. Uchida, J. L. Hicks, C. L. Dembia, and S. L. Delp, “Stretching your energetic budget: how tendon compliance affects the metabolic cost of running,”PloS one, vol. 11, no. 3, p. e0150378, 2016

2016
[34]

A physiological model for the evaluation of muscular forces in human locomotion: theoretical aspects,

M. R. Pierrynowski and J. B. Morrison, “A physiological model for the evaluation of muscular forces in human locomotion: theoretical aspects,”Mathematical Biosciences, vol. 75, no. 1, pp. 69–101, 1985

1985
[35]

Data on the distribution of fibre types in thirty-six human muscles: an autopsy study,

M. Johnson, J. Polgar, D. Weightman, and D. Appleton, “Data on the distribution of fibre types in thirty-six human muscles: an autopsy study,”Journal of the neurological sciences, vol. 18, no. 1, pp. 111– 129, 1973

1973
[36]

Fibre types in human abdominal muscles,

T. H ¨aggmark and A. Thorstensson, “Fibre types in human abdominal muscles,”Acta Physiologica Scandinavica, vol. 107, no. 4, pp. 319– 325, 1979

1979
[37]

Soft actor-critic: Off- policy maximum entropy deep reinforcement learning with a stochastic actor,

T. Haarnoja, A. Zhou, P. Abbeel, and S. Levine, “Soft actor-critic: Off- policy maximum entropy deep reinforcement learning with a stochastic actor,” inInternational conference on machine learning, pp. 1861– 1870, Pmlr, 2018

2018
[38]

Stable-baselines3: Reliable reinforcement learning im- plementations,

A. Raffin, A. Hill, A. Gleave, A. Kanervisto, M. Ernestus, and N. Dormann, “Stable-baselines3: Reliable reinforcement learning im- plementations,”Journal of machine learning research, vol. 22, no. 268, pp. 1–8, 2021

2021
[39]

Optimized hip–knee–ankle exoskeleton assistance at a range of walking speeds,

G. M. Bryan, P. W. Franks, S. Song, A. S. V oloshina, R. Reyes, M. P. O’Donovan, K. N. Gregorczyk, and S. H. Collins, “Optimized hip–knee–ankle exoskeleton assistance at a range of walking speeds,” Journal of neuroengineering and rehabilitation, vol. 18, no. 1, p. 152, 2021

2021
[40]

Reducing the metabolic rate of walking and running with a versatile, portable exosuit,

J. Kim, G. Lee, R. Heimgartner, D. A. Revi, N. Karavas, D. Nathanson, I. Galiana, A. Eckert-Erdheim, P. Murphy, D. Perry, N. Menard, D. K. Choe, P. Malcolm, and C. J. Walsh, “Reducing the metabolic rate of walking and running with a versatile, portable exosuit,”Science, vol. 365, no. 6454, pp. 668–672, 2019

2019
[41]

Hip hiking and circumduction,

D. C. Kerrigan, E. P. Frates, S. Rogan, and P. O. Riley, “Hip hiking and circumduction,”American Journal of Physical Medicine & Rehabilitation, vol. 79, no. 3, p. 247–252, 2000

2000