pith. sign in

arxiv: 2604.08882 · v1 · submitted 2026-04-10 · 💻 cs.RO

Simulation of Adaptive Running with Flexible Sports Prosthesis using Reinforcement Learning of Hybrid-link System

Yuta Shimane , Ko Yamamoto This is my paper

Pith reviewed 2026-05-10 18:19 UTC · model grok-4.3

classification 💻 cs.RO
keywords reinforcement learningsports prosthesisamputee runninghybrid-link systemmetabolic cost of transportpiece-wise constant straintranstibial amputeeflexible deformation
0
0 comments X

The pith

A reinforcement learning system with a hybrid-link model simulates amputee running motions and shows that prosthetic stiffness changes metabolic energy cost.

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

The paper sets out to demonstrate that whole-body running simulations for a unilateral transtibial amputee can be generated by training reinforcement learning policies on a hybrid-link system that treats the leaf-spring prosthesis as a flexible body. The method combines imitation of motion-capture data with explicit computation of prosthetic deformation under the Piece-wise Constant Strain model, then varies the virtual stiffness of the prosthesis across trials and extracts the resulting metabolic cost of transport. If the simulations are faithful, the approach would allow designers to explore user-specific prosthesis properties in software rather than through repeated physical trials. This matters because current selection of sports prostheses still relies on trial-and-error fitting that exposes athletes to risk and time cost. The work therefore positions simulation as a practical route to performance analysis under conditions that cannot be tested safely in the real world.

Core claim

The central claim is that a reinforcement-learning pipeline built on the hybrid-link system produces adaptive running motions for a unilateral transtibial amputee; when the same pipeline is run with different virtual stiffness values for the leaf-spring prosthesis, the computed metabolic cost of transport changes, indicating that stiffness directly influences running performance.

What carries the argument

The hybrid-link system, which augments rigid-body dynamics with the Piece-wise Constant Strain model to compute continuous flexible deformation of the prosthesis while the reinforcement-learning controller generates whole-body joint torques from motion-capture imitation.

If this is right

  • Varying virtual stiffness in the simulation can identify candidate stiffness values before any physical prosthesis is fabricated.
  • Metabolic cost of transport becomes a quantitative output that can be compared across prosthesis designs without instrumented human trials.
  • Whole-body interaction between residual-limb motion and prosthetic bending can be analyzed at every instant of the gait cycle.
  • The same imitation-plus-RL pipeline can be reused to test motions that differ from the captured reference, such as faster or more asymmetric running.
  • User-specific tuning becomes feasible by substituting an individual's motion-capture data into the same training loop.

Where Pith is reading between the lines

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

  • If the simulation proves predictive, it could shorten the design cycle for sports prostheses by replacing early physical prototypes with virtual sweeps.
  • The framework might be extended to other high-impact activities such as jumping or cutting, where prosthetic compliance also matters.
  • Real-time control laws for powered or variable-stiffness prostheses could be pre-tuned inside the same hybrid-link environment before hardware deployment.
  • The approach invites direct validation against force-plate and oxygen-consumption data collected on the same athletes whose motion was captured.

Load-bearing premise

The hybrid-link system with the Piece-wise Constant Strain model, combined with imitation learning from motion capture, produces whole-body dynamics that are sufficiently realistic to support conclusions about how real prosthetic stiffness affects metabolic cost and performance.

What would settle it

A side-by-side experiment in which actual transtibial amputee runners wearing leaf-spring prostheses of known stiffness show metabolic costs that systematically diverge from the costs predicted by the simulation under matching stiffness settings.

Figures

Figures reproduced from arXiv: 2604.08882 by Ko Yamamoto, Yuta Shimane.

Figure 1
Figure 1. Figure 1: Overview of the prosthesis integrated skeletal mode [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Reinforcement learning with the hybrid-link system [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Forward dynamics simulation of the prosthetic motio [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Learning return on prosthetic running: (Left) Retur [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Learning returns comparing walking and running: (Le [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 5
Figure 5. Figure 5: Running simulation results: (a) angle of hip and knee [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 8
Figure 8. Figure 8: Metabolic costs of transport obtained from running s [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗
Figure 7
Figure 7. Figure 7: Simulation results under different stiffness condi [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
read the original abstract

This study proposes a reinforcement learning-based adaptive running motion simulation for a unilateral transtibial amputee with the flexibility of a leaf-spring-type sports prosthesis using hybrid-link system. The design and selection of sports prostheses often rely on trial and error. A comprehensive whole-body dynamics analysis that considers the interaction between human motion and prosthetic deformation could provide valuable insights for user-specific design and selection. The hybrid-link system facilitates whole-body dynamics analysis by incorporating the Piece-wise Constant Strain model to represent the flexible deformation of the prosthesis. Based on this system, the simulation methodology generates whole-body dynamic motions of a unilateral transtibial amputee through a reinforcement learning-based approach, which combines imitation learning from motion capture data with accurate prosthetic dynamics computation. We simulated running motions under different virtual prosthetic stiffness conditions and analyzed the metabolic cost of transport obtained from the simulations, suggesting that variations in stiffness influence running performance. Our findings demonstrate the potential of this approach for simulation and analysis under virtual conditions that differ from real conditions.

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.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes a reinforcement learning framework for simulating adaptive running motions in a unilateral transtibial amputee equipped with a flexible leaf-spring sports prosthesis. It models the prosthesis via a hybrid-link system incorporating the Piece-wise Constant Strain formulation, combines imitation learning from motion-capture data with explicit prosthetic deformation dynamics, generates whole-body running trajectories under multiple virtual stiffness conditions, computes metabolic cost of transport from the resulting motions, and concludes that stiffness variations influence running performance.

Significance. If the simulated whole-body dynamics and derived CoT values prove predictive of real physiological outcomes, the approach would offer a valuable virtual testbed for prosthesis design and selection, enabling systematic exploration of stiffness effects without repeated physical trials and potentially accelerating user-specific optimization.

major comments (2)
  1. [Abstract] Abstract: the central claim that simulated CoT trends under varying stiffness 'suggest that variations in stiffness influence running performance' is presented without any quantitative results, error metrics, baseline comparisons, or description of the metabolic-cost computation method (e.g., which energy model or equations convert simulated joint powers and kinematics into CoT). This absence leaves the claim unsupported by visible evidence and makes it impossible to judge whether observed differences arise from genuine biomechanical effects or model artifacts.
  2. [Methods / Results] The hybrid-link + Piece-wise Constant Strain model is asserted to produce sufficiently realistic whole-body dynamics for conclusions about real amputee performance, yet the manuscript supplies no direct validation (comparison of simulated kinematics, joint powers, or CoT against measured oxygen consumption or ground-truth data from the same subjects across stiffness conditions). Without such anchoring, differences in simulated CoT cannot be confidently attributed to prosthetic stiffness rather than modeling assumptions.
minor comments (1)
  1. [Methods] Notation for the hybrid-link system and Piece-wise Constant Strain parameters should be introduced with explicit definitions and units at first use to improve readability for readers outside the immediate subfield.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which have helped us improve the clarity and scope of our simulation study. We have revised the abstract to include quantitative summaries and a brief description of the metabolic cost computation. For the validation concern, we have added explicit discussion of modeling assumptions and literature comparisons while clarifying that the work is simulation-focused.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that simulated CoT trends under varying stiffness 'suggest that variations in stiffness influence running performance' is presented without any quantitative results, error metrics, baseline comparisons, or description of the metabolic-cost computation method (e.g., which energy model or equations convert simulated joint powers and kinematics into CoT). This absence leaves the claim unsupported by visible evidence and makes it impossible to judge whether observed differences arise from genuine biomechanical effects or model artifacts.

    Authors: We agree the abstract was too concise on these points. In the revised version, we now include a summary of the observed CoT trends (e.g., relative changes across the tested stiffness values), note that metabolic cost is computed via integration of positive joint powers using a standard model from the literature (detailed in the Methods), and reference baseline comparisons to the original motion-capture trajectories. Full quantitative results, including RL convergence metrics and CoT values, remain in the Results section. These additions make the central claim traceable to the evidence without altering the abstract's length substantially. revision: yes

  2. Referee: [Methods / Results] The hybrid-link + Piece-wise Constant Strain model is asserted to produce sufficiently realistic whole-body dynamics for conclusions about real amputee performance, yet the manuscript supplies no direct validation (comparison of simulated kinematics, joint powers, or CoT against measured oxygen consumption or ground-truth data from the same subjects across stiffness conditions). Without such anchoring, differences in simulated CoT cannot be confidently attributed to prosthetic stiffness rather than modeling assumptions.

    Authors: We acknowledge that direct experimental validation against oxygen-consumption data from the same subjects would provide stronger grounding. This study develops and demonstrates a simulation framework rather than claiming predictive accuracy for real physiology; we therefore lack paired real-world CoT measurements across stiffness conditions. In revision we have added a dedicated paragraph in the Discussion that (i) states the key assumptions of the Piece-wise Constant Strain formulation and hybrid-link dynamics, (ii) compares simulated kinematic trends and CoT sensitivity to published experimental findings on leaf-spring prostheses, and (iii) explicitly frames the results as simulation-derived insights to inform future user-specific experiments. The conclusions have been tempered accordingly. revision: partial

Circularity Check

0 steps flagged

No significant circularity detected; derivation relies on external motion capture and independent dynamics computation

full rationale

The paper describes a simulation pipeline that combines reinforcement learning, imitation learning from motion capture data, and a hybrid-link model with Piece-wise Constant Strain for prosthesis deformation. Running motions are generated for varying virtual stiffness values, after which metabolic cost of transport is computed from the resulting dynamics. No equations or procedures in the provided text reduce a claimed prediction or result to its own fitted inputs by construction. No self-citations are invoked as load-bearing uniqueness theorems, no ansatzes are smuggled via prior work, and no known empirical patterns are merely renamed. The chain is self-contained: external motion data and standard RL training supply the inputs, while the stiffness variations and CoT estimates are downstream computations that do not loop back definitionally to those inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, additional axioms, or invented entities are stated beyond the modeling choice described.

axioms (1)
  • domain assumption The Piece-wise Constant Strain model accurately represents flexible deformation of the leaf-spring prosthesis within the hybrid-link system.
    Invoked to enable whole-body dynamics analysis of prosthesis-human interaction.

pith-pipeline@v0.9.0 · 5467 in / 1247 out tokens · 43754 ms · 2026-05-10T18:19:56.799791+00:00 · methodology

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

30 extracted references · 30 canonical work pages · 1 internal anchor

  1. [1]

    Somatosensory computation for man-machine in- terface from motion-capture data and musculoskeletal huma n model,

    Y . Nakamura, et al., “Somatosensory computation for man-machine in- terface from motion-capture data and musculoskeletal huma n model,” IEEE Trans. Robotics , vol. 21, no. 1, pp. 58–66, 2005

    work page 2005

  2. [2]

    Opensim: Open-source software to create and an- alyze dynamic simulations of movement,

    S. Delp, et al. , “Opensim: Open-source software to create and an- alyze dynamic simulations of movement,” IEEE Trans. Biomedical Engineering, vol. 54, pp. 1940 – 1950, 12 2007

    work page 1940

  3. [3]

    Can forward dynamics simulation with simple model estimate complex phenomena?: Case study on sprinting using running- specific prosthesis,

    A. Murai, et al., “Can forward dynamics simulation with simple model estimate complex phenomena?: Case study on sprinting using running- specific prosthesis,” ROBOMECH J., vol. 5, no. 10, 2018

    work page 2018

  4. [4]

    A prosthesis-specific multi-link segment model of lower-limb amputee sprinting,

    S. M. Rigney, et al. , “A prosthesis-specific multi-link segment model of lower-limb amputee sprinting,” J. Biomechanics , vol. 49, no. 14, pp. 3185–3193, 2016

    work page 2016

  5. [5]

    Sofa: A multi-model framework for interactive phys- ical simulation,

    F. Faure, et al., “Sofa: A multi-model framework for interactive phys- ical simulation,” in Soft Tissue Biomechanical Modeling for Computer Assisted Surgery. Springer, 2012, pp. 283–321

    work page 2012

  6. [6]

    Mechanical characterization and comparison of en- ergy storage and return prostheses,

    S. Rigney, et al. , “Mechanical characterization and comparison of en- ergy storage and return prostheses,” Medical Engineering and Physics , vol. 41, pp. 90–96, 2017

    work page 2017

  7. [7]

    Designing custom mechanics in running- specific prosthetic feet via shape optimization,

    M. K. Shepherd, et al. , “Designing custom mechanics in running- specific prosthetic feet via shape optimization,” IEEE Trans. Biomed- ical Engineering , vol. 70, no. 2, pp. 747–755, 2023

    work page 2023

  8. [8]

    Parametric modeling of sports prostheses based on the flat spring design formulas,

    K. Hase, et al. , “Parametric modeling of sports prostheses based on the flat spring design formulas,” J. Biomechanical Science and Engineering, vol. 15, no. 1, pp. 19–00 446, 2020

    work page 2020

  9. [9]

    Flexibility control of el astic robotic arms and its application to contouring control,

    T. Fukuda and Y . Kuribayashi, “Flexibility control of el astic robotic arms and its application to contouring control,” in IEEE Int. Conf. Robotics and Automation , vol. 1, 1984, pp. 540–545

    work page 1984

  10. [10]

    Modeling of a flexible manipul ator dy- namics based upon holzer’s model,

    A. Konno and M. Uchiyama, “Modeling of a flexible manipul ator dy- namics based upon holzer’s model,” in IEEE/RSJ Int. Conf. Intelligent Robots and Systems , vol. 1, 1996, pp. 223–229 vol.1

    work page 1996

  11. [11]

    Design and kinematic m odeling of constant curvature continuum robots: A review,

    R. J. Webster III and B. A. Jones, “Design and kinematic m odeling of constant curvature continuum robots: A review,” Int. J. Robotics Research, vol. 29, no. 13, pp. 1661–1683, 2010

    work page 2010

  12. [12]

    Discrete cosserat approach for multisection soft manipulator dynamics,

    F. Renda, et al. , “Discrete cosserat approach for multisection soft manipulator dynamics,” IEEE Trans. Robotics , vol. 34, no. 6, pp. 1518–1533, 2018

    work page 2018

  13. [13]

    Dynamics computation of a hybrid multi-link humanoid robot integrating rigid and soft bodie s,

    T. Ishigaki and K. Y amamoto, “Dynamics computation of a hybrid multi-link humanoid robot integrating rigid and soft bodie s,” in IEEE/RSJ Int. Conf. Intelligent Robots and Systems , 2021, pp. 2816– 2821

    work page 2021

  14. [14]

    Sorosim: A matlab toolbox for hybrid rigid- soft robots based on the geometric variable-strain approac h,

    A. T. Mathew, et al. , “Sorosim: A matlab toolbox for hybrid rigid- soft robots based on the geometric variable-strain approac h,” IEEE Robotics and Automation Magazine , vol. 30, pp. 106–122, 9 2023

    work page 2023

  15. [15]

    Inverse kinematics of hybrid multi-link system and its application to motion capture for athlete wearing sports pr osthesis,

    S. Kim, et al., “Inverse kinematics of hybrid multi-link system and its application to motion capture for athlete wearing sports pr osthesis,” in IEEE-RAS 21st Int. Conf. Humanoid Robots , 2022, pp. 837–842

    work page 2022

  16. [16]

    Viscoelasticity estimation of sports prosthesis by energy-minimizing inverse kinematics and its validation b y forward dynamics,

    Y . Shimane, et al. , “Viscoelasticity estimation of sports prosthesis by energy-minimizing inverse kinematics and its validation b y forward dynamics,” Advanced Robotics, vol. 38, no. 21, pp. 1515–1527, 2024

    work page 2024

  17. [17]

    A model of the neuro-musculo-skeletal sys tem for human locomotion,

    T. Gentaro, “A model of the neuro-musculo-skeletal sys tem for human locomotion,” Biological Cybernetics , vol. 73, p. 97–111, 1995

    work page 1995

  18. [18]

    Modeling and optimal contro l of human- like running,

    G. Schultz and K. Mombaur, “Modeling and optimal contro l of human- like running,” IEEE/ASME Trans. Mechatronics , vol. 15, no. 5, pp. 783–792, 2010

    work page 2010

  19. [19]

    Human gait cost function varies with walking speed: An inverse optimal control study,

    J. Weng, et al. , “Human gait cost function varies with walking speed: An inverse optimal control study,” IEEE Robotics and Automation Letters, vol. 8, no. 8, pp. 4777–4784, 2023

    work page 2023

  20. [20]

    Natural walking with musculoskelet al models using deep reinforcement learning,

    J. Weng and et al., “Natural walking with musculoskelet al models using deep reinforcement learning,” IEEE Robotics and Automation Letters, vol. PP , pp. 1–1, 03 2021

    work page 2021

  21. [21]

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

    X. Peng, et al. , “Deepmimic: Example-guided deep reinforcement learning of physics-based character skills,” ACM Trans. Graphics , vol. 37, 04 2018

    work page 2018

  22. [22]

    Policy design for an ankle-foot orthosis using simulated physical human–robot interaction via deep reinf orcement learning,

    J. I. Han, et al. , “Policy design for an ankle-foot orthosis using simulated physical human–robot interaction via deep reinf orcement learning,” IEEE Trans. Neural Systems and Rehabilitation Engineer- ing, vol. 30, pp. 2186–2197, 2022

    work page 2022

  23. [23]

    Self model for embodied intelligence: Modeling full- body human musculoskeletal system and locomotion control w ith hierarchical low-dimensional representation,

    C. Zuo, et al. , “Self model for embodied intelligence: Modeling full- body human musculoskeletal system and locomotion control w ith hierarchical low-dimensional representation,” in IEEE Int. Conf. on Robotics and Automation , 2024, pp. 13 062–13 069

    work page 2024

  24. [24]

    Application of piece-wise constant strain model to flexible deformation calculation of sports prosthesis an d stiffness estimation,

    Y . Shimane, et al. , “Application of piece-wise constant strain model to flexible deformation calculation of sports prosthesis an d stiffness estimation,” in IEEE/RSJ Int. Conf. Intelligent Robots and Systems , 2022, pp. 12 758–12 763

    work page 2022

  25. [25]

    S. S. Antman, Nonlinear Problems of Elasticity , ser. Applied Mathe- matical Sciences. Springer, 2005, vol. 107

    work page 2005

  26. [26]

    Unified framework of gradient computation for hybrid-link system and its dynamical simulation by implici t method,

    T. Ishigaki, et al. , “Unified framework of gradient computation for hybrid-link system and its dynamical simulation by implici t method,” IEEE Robotics and Automation Letters , vol. 10, no. 5, pp. 4188–4195, 2025

    work page 2025

  27. [27]

    Proximal Policy Optimization Algorithms

    J. Schulman, et al., “Proximal policy optimization algorithms,” ArXiv, vol. abs/1707.06347, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  28. [28]

    Deeploco: dynamic locomotion skills using hierarchi- cal deep reinforcement learning,

    X. Peng, et al., “Deeploco: dynamic locomotion skills using hierarchi- cal deep reinforcement learning,” ACM Trans. on Graphics , vol. 36, pp. 1–13, 07 2017

    work page 2017

  29. [29]

    Prosthetic model, but not stiffness or height, affects the metabolic cost of running for athletes with unilateral t ranstibial amputations,

    O. Beck, et al. , “Prosthetic model, but not stiffness or height, affects the metabolic cost of running for athletes with unilateral t ranstibial amputations,” Journal of Applied Physiology , vol. 123(1), pp. 38–48, 07 2017

    work page 2017

  30. [30]

    Mechanical work accounts for mos t of the energetic cost in human running,

    R. Riddick and A. Kuo, “Mechanical work accounts for mos t of the energetic cost in human running,” Scientific Reports , vol. 12, 01 2022

    work page 2022