pith. sign in

arxiv: 2503.00794 · v2 · pith:HHOZJJMTnew · submitted 2025-03-02 · 💻 cs.RO

Detecting Heel Strike and toe off Events Using Kinematic Methods and LSTM Models

Longbin Zhang , Zhizhang Li , Xinyi Fu , Yi Xie , Xiaoyue Yan , Suiyuan Wang , Te Zhang , Hui Zhang
show 5 more authors
Kailun Yang Tsung-Lin Wu Prayook Jatesiktat Ananda Sidarta Wei Tech Ang
This is my paper

Pith reviewed 2026-05-23 01:09 UTC · model grok-4.3

classification 💻 cs.RO
keywords gait event detectionheel striketoe-offLSTMkinematic methodsrehabilitationexoskeleton control
0
0 comments X

The pith

LSTM model matches the accuracy of the leading kinematic method for heel strike and toe-off detection without systematic bias.

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

The paper evaluates seven kinematics-based techniques and an LSTM model on 4363 gait cycles from 588 able-bodied subjects to identify heel strike and toe-off events. The LSTM performs at the level of the strongest traditional method, Zeni et al., while avoiding the biases or tuning demands seen in the others. Accurate event detection matters for controlling exoskeletons and supporting gait analysis in rehabilitation. The work positions the LSTM as a data-driven option but notes that results are limited to healthy walkers so far.

Core claim

The LSTM model performed comparably to the Zeni et al. method in accuracy for heel strike and toe-off detection across the dataset, providing a data-driven alternative without the systematic biases exhibited by other kinematics-based approaches.

What carries the argument

LSTM neural network trained on kinematic signals to classify heel strike and toe-off events, benchmarked directly against seven traditional kinematic detection rules including the Zeni method.

If this is right

  • The LSTM supplies a data-driven option that does not require dataset-specific tuning for gait event detection.
  • Deep learning models can serve as practical alternatives to rule-based kinematic methods in exoskeleton control.
  • Further testing in pathological gait is needed before clinical deployment.

Where Pith is reading between the lines

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

  • The bias-free nature of the LSTM could support more consistent long-term tracking of gait changes across sessions.
  • Integration with additional sensor inputs might extend reliable detection to outdoor or variable walking surfaces.

Load-bearing premise

Results measured on able-bodied subjects will translate to clinical populations and varied gait conditions without retraining or retuning.

What would settle it

Running the LSTM on gait data from post-stroke or knee osteoarthritis patients and finding accuracy or bias levels that fall below the Zeni et al. method.

Figures

Figures reproduced from arXiv: 2503.00794 by Ananda Sidarta, Hui Zhang, Kailun Yang, Longbin Zhang, Prayook Jatesiktat, Suiyuan Wang, Te Zhang, Tsung-Lin Wu, Wei Tech Ang, Xiaoyue Yan, Xinyi Fu, Yi Xie, Zhizhang Li.

Figure 1
Figure 1. Figure 1: Schematic diagram of gait event identification using kinematics-based or machine learning models. The inputs for these models consist of key [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The boxplot of prediction errors (time differences) between the measured and calculated heel strike (left) and toe off (right) events using various [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The histogram of time differences between the measured and cal [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
read the original abstract

Accurate gait event detection is crucial for gait analysis, rehabilitation, and assistive technology, particularly in exoskeleton control, where precise identification of stance and swing phases is essential. This study evaluated the performance of seven kinematics-based methods and a Long Short-Term Memory (LSTM) model for detecting heel strike and toe-off events across 4363 gait cycles from 588 able-bodied subjects. The results indicated that while the Zeni et al. method achieved the highest accuracy among kinematics-based approaches, other methods exhibited systematic biases or required dataset-specific tuning. The LSTM model performed comparably to Zeni et al., providing a data-driven alternative without systematic bias. These findings highlight the potential of deep learning-based approaches for gait event detection while emphasizing the need for further validation in clinical populations and across diverse gait conditions. Future research will explore the generalizability of these methods in pathological populations, such as individuals with post-stroke conditions and knee osteoarthritis, as well as their robustness across varied gait conditions and data collection settings to enhance their applicability in rehabilitation and exoskeleton control.

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

0 major / 3 minor

Summary. The manuscript evaluates seven kinematics-based methods and an LSTM model for detecting heel-strike and toe-off events on a dataset of 4363 gait cycles from 588 able-bodied subjects. It reports that the Zeni et al. method achieves the highest accuracy among the kinematic approaches, while other kinematic methods show systematic biases or require per-dataset tuning; the LSTM model performs comparably to Zeni et al. without systematic bias and is positioned as a data-driven alternative. The abstract explicitly limits scope to able-bodied subjects and flags the need for future validation in clinical populations such as post-stroke and knee osteoarthritis.

Significance. The large sample size (4363 cycles, 588 subjects) lends statistical weight to the empirical comparisons. If the results hold under the reported labeling and evaluation protocol, the work supplies a reproducible benchmark showing that an LSTM can match the best kinematic method while avoiding the biases documented for the other six kinematic approaches. The explicit scoping to able-bodied gait and the call for clinical validation are strengths that keep the central claim proportionate to the evidence presented.

minor comments (3)
  1. [Abstract] Abstract: the claim that the LSTM 'performed comparably' would be strengthened by reporting the specific mean absolute errors or percentage accuracies (with standard deviations) for both the LSTM and Zeni method rather than a qualitative statement.
  2. [Methods] Methods section: the description of the LSTM architecture, hyperparameter selection procedure, and train/validation/test split should explicitly confirm that hyperparameter tuning was performed only on the training portion to rule out any leakage into the reported test-set performance.
  3. [Results] Results: Table or figure presenting per-method bias (mean signed error) would make the 'systematic bias' claim for the non-Zeni kinematic methods directly verifiable; currently the abstract states the finding but the quantitative support is not summarized in the provided text.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive evaluation of our manuscript, including recognition of the large sample size, the benchmark value of the comparisons, and the appropriate scoping to able-bodied gait with a call for future clinical validation. We are pleased that the recommendation is for minor revision and will incorporate any editorial suggestions in the revised version.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper is an empirical comparison of seven kinematics-based methods and an LSTM model on 4363 gait cycles from 588 able-bodied subjects. Performance is measured directly against measured heel-strike and toe-off events; no derivation chain, fitted parameter renamed as prediction, or self-citation load-bearing step exists. The Zeni et al. reference is an external benchmark, and the abstract explicitly flags the need for future clinical validation rather than claiming generalization by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that kinematic signals contain sufficient information to label heel-strike and toe-off events accurately and that the able-bodied dataset is representative enough for method ranking.

axioms (1)
  • domain assumption Kinematic data alone suffices to define ground-truth gait events for method evaluation
    All seven methods and the LSTM are scored against events derived from the same kinematic recordings.

pith-pipeline@v0.9.0 · 5757 in / 1184 out tokens · 35759 ms · 2026-05-23T01:09:32.911825+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

31 extracted references · 31 canonical work pages

  1. [1]

    Physical human– robot interaction of a robotic exoskeleton by admittance control,

    Z. Li, B. Huang, Z. Ye, M. Deng, and C. Yang, “Physical human– robot interaction of a robotic exoskeleton by admittance control,”IEEE Transactions on Industrial Electronics, vol. 65, no. 12, pp. 9614–9624, 2018

    work page 2018

  2. [2]

    Asymmetric bimanual control of dual-arm exoskeletons for human- cooperative manipulations,

    Z. Li, B. Huang, A. Ajoudani, C. Yang, C.-Y . Su, and A. Bicchi, “Asymmetric bimanual control of dual-arm exoskeletons for human- cooperative manipulations,” IEEE Transactions on Robotics , vol. 34, no. 1, pp. 264–271, 2017

    work page 2017

  3. [3]

    Ankle joint torque prediction using an nms solver informed-ann model and transfer learning,

    L. Zhang, X. Zhu, E. M. Gutierrez-Farewik, and R. Wang, “Ankle joint torque prediction using an nms solver informed-ann model and transfer learning,”IEEE Journal of Biomedical and Health Informatics, vol. 26, no. 12, pp. 5895–5906, 2022

    work page 2022

  4. [4]

    Hybrid brain/muscle signals powered wearable walking exoskeleton enhancing motor ability in climbing stairs activity,

    Z. Li, Y . Yuan, L. Luo, W. Su, K. Zhao, C. Xu, J. Huang, and M. Pi, “Hybrid brain/muscle signals powered wearable walking exoskeleton enhancing motor ability in climbing stairs activity,”IEEE Transactions on Medical Robotics and Bionics , vol. 1, no. 4, pp. 218–227, 2019

    work page 2019

  5. [5]

    Trajectory generation and control of a lower limb exoskeleton for gait assistance,

    L. Luo, M. J. Foo, M. Ramanathan, J. K. Er, C. H. Chiam, L. Li, W. Y . Yau, and W. T. Ang, “Trajectory generation and control of a lower limb exoskeleton for gait assistance,” Journal of Intelligent & Robotic Systems, vol. 106, no. 3, p. 64, 2022

    work page 2022

  6. [6]

    Adaptive trajectory tracking control of a parallel ankle rehabilitation robot with joint-space force distribution,

    M. Zhang, A. McDaid, A. J. Veale, Y . Peng, and S. Q. Xie, “Adaptive trajectory tracking control of a parallel ankle rehabilitation robot with joint-space force distribution,”IEEE Access, vol. 7, pp. 85 812–85 820, 2019

    work page 2019

  7. [7]

    A deep-learning approach for automatically detecting gait-events based on foot-marker kinematics in children with cerebral palsy—which markers work best for which gait patterns?

    Y . K. Kim, R. M. Visscher, E. Viehweger, N. B. Singh, W. R. Taylor, and F. V ogl, “A deep-learning approach for automatically detecting gait-events based on foot-marker kinematics in children with cerebral palsy—which markers work best for which gait patterns?” Plos One, vol. 17, no. 10, pp. 1–16, 2022

    work page 2022

  8. [8]

    Robust deep learning-based gait event detection across various pathologies,

    B. Dumphart, D. Slijepcevic, M. Zeppelzauer, A. Kranzl, F. Unglaube, A. Baca, and B. Horsak, “Robust deep learning-based gait event detection across various pathologies,” Plos One , vol. 18, no. 8, pp. 1–16, 2023

    work page 2023

  9. [9]

    Two simple methods for determining gait events during treadmill and overground walking using kinematic data,

    J. Zeni Jr, J. Richards, and J. Higginson, “Two simple methods for determining gait events during treadmill and overground walking using kinematic data,” Gait & Posture, vol. 27, no. 4, pp. 710–714, 2008

    work page 2008

  10. [10]

    Baker and H

    R. Baker and H. M. Hart, Measuring walking: a handbook of clinical gait analysis. Mac Keith Press London, 2013, vol. 1

    work page 2013

  11. [11]

    Quantification of human motion: gait analysis—benefits and limitations to its application to clinical problems,

    S. R. Simon, “Quantification of human motion: gait analysis—benefits and limitations to its application to clinical problems,” Journal of Biomechanics, vol. 37, no. 12, pp. 1869–1880, 2004

    work page 2004

  12. [12]

    The evolution of clinical gait analysis: Part ii kinematics,

    D. H. Sutherland, “The evolution of clinical gait analysis: Part ii kinematics,” Gait & Posture, vol. 16, no. 2, pp. 159–179, 2002

    work page 2002

  13. [13]

    Modeling and simulation of a human knee exoskeleton’s assistive strategies and interaction,

    L. Zhang, Y . Liu, R. Wang, C. Smith, and E. M. Gutierrez-Farewik, “Modeling and simulation of a human knee exoskeleton’s assistive strategies and interaction,” Frontiers in Neurorobotics, vol. 15, pp. 1– 12, 2021

    work page 2021

  14. [14]

    Identifying gait events without a force plate during running: A comparison of methods,

    J. Leitch, J. Stebbins, G. Paolini, and A. B. Zavatsky, “Identifying gait events without a force plate during running: A comparison of methods,” Gait & posture , vol. 33, no. 1, pp. 130–132, 2011

    work page 2011

  15. [15]

    An evaluation of three kinematic methods for gait event detection compared to the kinetic-based ‘gold standard’,

    N. Zahradka, K. Verma, A. Behboodi, B. Bodt, H. Wright, and S. C. Lee, “An evaluation of three kinematic methods for gait event detection compared to the kinetic-based ‘gold standard’,” Sensors, vol. 20, no. 18, pp. 1–15, 2020

    work page 2020

  16. [16]

    Identifying heel contact and toe-off using forceplate thresholds with a range of digital-filter cutoff frequencies,

    O. Tirosh and W. Sparrow, “Identifying heel contact and toe-off using forceplate thresholds with a range of digital-filter cutoff frequencies,” Journal of Applied Biomechanics , vol. 19, no. 2, pp. 178–184, 2003

    work page 2003

  17. [17]

    Automated event detection algo- rithms in pathological gait,

    D. A. Bruening and S. T. Ridge, “Automated event detection algo- rithms in pathological gait,” Gait & Posture, vol. 39, no. 1, pp. 472– 477, 2014

    work page 2014

  18. [18]

    A new deep learning-based method for the detection of gait events in children with gait disorders: Proof- of-concept and concurrent validity,

    M. Lempereur, F. Rousseau, O. R ´emy-N´eris, C. Pons, L. Houx, G. Quellec, and S. Brochard, “A new deep learning-based method for the detection of gait events in children with gait disorders: Proof- of-concept and concurrent validity,” Journal of Biomechanics, vol. 98, pp. 1–4, 2020

    work page 2020

  19. [19]

    An algorithm for accurate marker-based gait event detection in healthy and pathological populations during complex motor tasks,

    T. Bonci, F. Salis, K. Scott, L. Alcock, C. Becker, S. Bertuletti, E. Buckley, M. Caruso, A. Cereatti, S. Del Dinet al., “An algorithm for accurate marker-based gait event detection in healthy and pathological populations during complex motor tasks,” Frontiers in Bioengineering and Biotechnology, vol. 10, pp. 1–14, 2022

    work page 2022

  20. [20]

    A simple method for determination of gait events,

    A. H. Hansen, D. S. Childress, and M. R. Meier, “A simple method for determination of gait events,” Journal of Biomechanics , vol. 35, no. 1, pp. 135–138, 2002

    work page 2002

  21. [21]

    Foot contact event detection using kinematic data in cerebral palsy children and normal adults gait,

    E. Desailly, Y . Daniel, P. Sardain, and P. Lacouture, “Foot contact event detection using kinematic data in cerebral palsy children and normal adults gait,” Gait & Posture, vol. 29, no. 1, pp. 76–80, 2009

    work page 2009

  22. [22]

    Assessment and validation of a simple automated method for the detection of gait events and intervals,

    S. Ghoussayni, C. Stevens, S. Durham, and D. Ewins, “Assessment and validation of a simple automated method for the detection of gait events and intervals,” Gait & Posture , vol. 20, no. 3, pp. 266–272, 2004

    work page 2004

  23. [23]

    Automatic detection of gait events using kinematic data,

    C. M. O’Connor, S. K. Thorpe, M. J. O’Malley, and C. L. Vaughan, “Automatic detection of gait events using kinematic data,” Gait & Posture, vol. 25, no. 3, pp. 469–474, 2007

    work page 2007

  24. [24]

    Gait timing event determination using kinematic data for the toe walking children with cerebreal palsy,

    B. Hsue, F. Miller, F. Su, J. Henley, and C. Church, “Gait timing event determination using kinematic data for the toe walking children with cerebreal palsy,” Journal of Biomechanics , vol. 40, no. 2, p. S529, 2007

    work page 2007

  25. [25]

    Algorithms to determine event timing during normal walking using kinematic data,

    A. Hreljac and R. N. Marshall, “Algorithms to determine event timing during normal walking using kinematic data,” Journal of Biomechan- ics, vol. 33, no. 6, pp. 783–786, 2000

    work page 2000

  26. [26]

    Comparison of three kine- matic gait event detection methods during overground and treadmill walking for individuals post stroke,

    M. A. French, C. Koller, and E. S. Arch, “Comparison of three kine- matic gait event detection methods during overground and treadmill walking for individuals post stroke,” Journal of Biomechanics, vol. 99, pp. 1–7, 2020

    work page 2020

  27. [27]

    Automatic gait events detection with inertial measurement units: healthy subjects and moderate to severe impaired patients,

    C. V oisard, N. de l’Escalopier, D. Ricard, and L. Oudre, “Automatic gait events detection with inertial measurement units: healthy subjects and moderate to severe impaired patients,” Journal of NeuroEngineer- ing and Rehabilitation , vol. 21, no. 1, pp. 1–17, 2024

    work page 2024

  28. [28]

    A data-driven approach for detecting gait events during turning in people with parkinson’s disease and freezing of gait,

    B. Filtjens, A. Nieuwboer, N. D’cruz, J. Spildooren, P. Slaets, and B. Vanrumste, “A data-driven approach for detecting gait events during turning in people with parkinson’s disease and freezing of gait,” Gait & Posture, vol. 80, pp. 130–136, 2020

    work page 2020

  29. [29]

    Lower- limb joint torque prediction using lstm neural networks and transfer learning,

    L. Zhang, D. Soselia, R. Wang, and E. M. Gutierrez-Farewik, “Lower- limb joint torque prediction using lstm neural networks and transfer learning,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 30, pp. 600–609, 2022

    work page 2022

  30. [30]

    Prediction of achilles tendon force dur- ing common motor tasks from markerless video,

    Z. Xia, B. M. Cornish, D. Devaprakash, R. S. Barrett, D. G. Lloyd, A. H. Hams, and C. Pizzolato, “Prediction of achilles tendon force dur- ing common motor tasks from markerless video,” IEEE Transactions on Neural Systems and Rehabilitation Engineering , 2024

    work page 2024

  31. [31]

    Integrating an lstm framework for predicting ankle joint biomechan- ics during gait using inertial sensors,

    L. Xiang, Y . Gu, Z. Gao, P. Yu, V . Shim, A. Wang, and J. Fernandez, “Integrating an lstm framework for predicting ankle joint biomechan- ics during gait using inertial sensors,” Computers in Biology and Medicine, vol. 170, p. 108016, 2024

    work page 2024