Human sensory-musculoskeletal modeling and control of whole-body movements

Chenhui Zuo; Chen Zhang; Guohao Lin; Shanning Zhuang; Yanan Sui

REVIEW 5 major objections 6 minor 51 references

Human sensory-musculoskeletal modeling and control of whole-body movements

T0 review · 5 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A whole-body digital human with 1,266 muscle-tendon units, driven by multisensory closed-loop control, reproduces walking, vision-guided manipulation, and cycling, and exposes muscle activations that cannot be measured directly.

desk verdict A serious, anatomically detailed whole-body musculoskeletal platform that is currently oversold as validated: the walking and cycling results are largely in-sample, and the unmeasurable-dynamics claims rest on unvalidated parameters. read the letter →

arxiv 2506.00071 v1 pith:US3KF4A2 submitted 2025-05-29 q-bio.NC cs.AIcs.RO

Chenhui Zuo , Guohao Lin , Chen Zhang , Shanning Zhuang , Yanan Sui This is my paper

classification q-bio.NCcs.AIcs.RO

keywords humanmusculoskeletalmodelwhole-bodysimulationdeepreinforcementlearningmultisensoryintegrationbipedallocomotionvision-guidedmanipulationcyclingmuscleactivationprediction

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper reports a whole-body human model—called SMS-Human—built from 206 bones, 278 joints, and 1,266 muscle-tendon units, with binocular vision, vestibular, proprioceptive, and tactile sensing. The authors' central claim is that a stage-wise hierarchical reinforcement-learning controller can drive this high-dimensional musculoskeletal body to reproduce natural bipedal walking, vision-guided object manipulation, and cycling, and that the resulting movements closely match human motion-capture and surface-EMG data. If the claim holds, the model becomes a way to inspect muscle activations and joint dynamics that cannot be measured experimentally in behaving humans, such as deep spinal and toe-flexor muscles during gait. That matters because it would give rehabilitation, prosthetics, and humanoid-robot design a testable digital subject in which every muscle state is observable.

What carries the argument

The central object is SMS-Human, a physics-based digital body whose 1,266 muscle-tendon units, 278 joints, and 175 articulated segments are driven by muscle excitation rather than joint torques. The central mechanism is the hierarchical controller: a high-level group-action network issues shared commands to anatomically related muscle groups, a low-level unit-action network supplies state-dependent adjustment weights for each individual muscle-tendon unit, and an action-refinement module multiplies the two before the critic updates. Training proceeds stage-wise, gradually tightening pose-tracking constraints and task demands, which is what makes the 1,266-dimensional action space tractable. Together these components convert raw visual, vestibular, proprioceptive, and tactile observations into physiologically plausible muscle activations.

What would settle it

Record fine-wire intramuscular EMG from a deep muscle the model predicts, such as flexor digitorum longus or internal oblique, during treadmill walking at the reported speed, and compare onset/offset timing and relative amplitude to the simulated activation traces; repeated out-of-phase or absent activations would show the parameter estimates or Hill-type muscle model do not faithfully reproduce the unmeasured dynamics.

Watch

Extended reading notes

Core claim

On its own terms, the paper's discovery is that anatomical completeness plus closed-loop multisensory feedback is enough to produce naturalistic whole-body motor behaviour. The authors constructed SMS-Human with all major skeletal muscles represented as muscle-tendon units, connected it to visual, vestibular, proprioceptive, and tactile sensors, and trained one controller for three tasks. Simulated walking reached 1.22 m/s with a 1.04 s gait cycle, and eight representative joint-angle trajectories tracked reference motion-capture data over three gait cycles, while eight lower-limb muscle activity patterns largely matched surface EMG from the same subject. In the manipulation task the model kept the held bottle within 0.15 ± 0.25 cm of a moving target and produced direction-specific extraocular muscle activations; in cycling it followed reference keypoint trajectories with a maximum error of 2.38 ± 0.27 cm. The paper treats these matches as evidence that its predictions for muscles that cannot be measured directly—for example flexor digitorum longus and internal oblique during walking—are informative.

Load-bearing premise

The load-bearing premise is that every one of the 1,266 muscle-tendon units, including those with no published anatomical measurements, has sufficiently accurate architectural parameters for the forces and activations the model computes to be trustworthy.

Editorial extensions

If this is right

Simulated walking reproduces natural joint kinematics and muscle-timing patterns, so the model can be used to predict joint moments and muscle forces across the full gait cycle without invasive measurement.
Because the manipulation controller learns from egocentric binocular images and never receives explicit object coordinates, the same training pipeline should transfer to other visually guided reaching and grasping tasks in the model.
The model predicts direction-dependent extraocular and limb muscle synergies during object tracking, giving a mechanistic account of eye-head-hand coordination that can be compared with human recordings.
Deep-muscle activation profiles (e.g., flexor digitorum longus during push-off, internal oblique during stance) are concrete, falsifiable predictions about unmeasured physiology.
The stage-wise hierarchical learning scheme generalizes the control of a 1,266-muscle body beyond the three demonstrated tasks, suggesting it can scale to other whole-body movements.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Beyond the paper's claims, the same simulation could serve as a synthetic lesion subject: remove the vestibular or tactile channel and measure how gait stability and muscle coordination degrade, yielding predictions that could be checked against patient populations with sensory loss.
Beyond the paper's claims, the deep-muscle predictions are only indirect inferences from a Hill-type muscle model and estimated parameters; a stringent test would be to record fine-wire EMG from one predicted deep muscle during the exact reported task and compare activation phases.
Beyond the paper's claims, the control pipeline could in principle transfer to a physical humanoid, but the sim-to-real gap for 1,266 muscle-like actuators is large; a nearer-term use is to learn reduced-order muscle synergies from the simulator and test whether human EMG exhibits the same low-dimensional structure.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

First thing to know: this is a build-and-demonstrate paper, not a validation paper. The SMS-Human model is genuinely impressive as a modeling effort—175 segments, 278 joints, 1,266 muscle-tendon units, vision/vestibular/proprioceptive/tactile sensors, all in MuJoCo, with a staged hierarchical SAC controller that actually produces walking, visually guided manipulation, and cycling. That integration is new relative to the authors' own ICRA 2024 model and to MyoSuite/OpenSim work. The anatomical detail, especially extraocular, hand, foot, and deep trunk muscles, goes beyond prior whole-body musculoskeletal simulators. The paper is worth a serious referee simply because a platform like this, if released, could be broadly used.

The soft spots are exactly where the reader's report says they are, and I agree with most of it. The walking result is largely in-sample: the controller is trained on the same adult-male mocap trajectories it is then compared against, so the joint-angle match in Fig. 4b is partly a report of reward optimization, not evidence of naturalness. The EMG comparison in Fig. 4c is more independent, but it covers only eight muscles, shows qualitative agreement with some phase shifts, and has no error metric. The bicycling reference is synthetic, so that task can show control competence but not human resemblance. The manipulation task is not trained on human data, which is good for independence, but its success metric is the distance in the reward function; again, that demonstrates task achievement, not biological fidelity.

The deepest issue is the claim about unmeasurable dynamics. With 1,266 actuators driving a highly redundant system, many activation patterns produce the same kinematics. The deep-muscle predictions (flexor digitorum longus, internal oblique) are therefore underdetermined by the motion data; they depend on the learned policy and on muscle parameters that, for units without published values, were estimated by the authors. The paper itself admits Hill-type muscle simplifications and limited EMG data. So the unmeasurable dynamics are plausible but not established. The authors should recalibrate claims, run held-out predictions (e.g., leave-one-out motion trials, or predict EMG for muscles not used in training), and publish code and model parameters.

On the citation pattern: they cite the relevant OpenSim, MyoSuite, anatomical, and EMG-cycling literature, and they are honest about their own ICRA work. Self-citation is fine here because the ICRA model is the direct ancestor.

Who is this for: biomechanics modelers, embodied intelligence researchers, and DRL people working on high-dimensional actuation. It deserves a serious referee, but my recommendation is conditional acceptance at best—the claim of close resemblance and valid unmeasurable dynamics needs held-out evidence, baselines, and artifact release before the platform can be used as a trustworthy prediction tool. I would not cite it as a validated model yet, but I would cite it as a modeling benchmark once artifacts are available.

Referee Report

5 major / 6 minor

Summary. The manuscript introduces SMS-Human, a whole-body human sensory-musculoskeletal model implemented in MuJoCo, featuring 175 rigid segments, 278 joints, and 1,266 muscle-tendon units, together with visual, vestibular, proprioceptive, and tactile sensory inputs. A hierarchical soft actor-critic reinforcement learning framework with stage-wise training is proposed to control this high-dimensional system. The authors demonstrate three tasks: bipedal walking, vision-guided object manipulation, and bicycling, and argue that the simulations closely resemble natural human motor behaviours and reveal musculoskeletal dynamics that cannot be directly measured experimentally. The paper includes quantitative tracking errors and qualitative comparisons to human EMG and kinematics.

Significance. If the central claims were fully supported, this would be a valuable contribution: the model's anatomical detail is unprecedented, the integration of multisensory inputs into a closed-loop muscle-driven simulation addresses a long-standing challenge, and the hierarchical control scheme is a sensible approach to the redundancy problem. The demonstrations of walking, manipulation, and human-machine interaction indicate a genuinely functional whole-body platform. However, the evidence that the simulated behaviours resemble natural human movements is weakened by the fact that the walking and bicycling results are evaluated largely against the same reference trajectories used as training targets, and the EMG comparison is only qualitative. The claim to reveal unmeasurable deep-muscle dynamics rests on parameter estimates that are not validated. These gaps prevent the paper from currently meeting the bar for the strength of its stated conclusions, though the underlying model and control framework are plausible and potentially important.

major comments (5)

[Simulation of bipedal walking, Fig. 4b] The walking controller was trained by rewarding match to kinematic references from motion capture of an adult male, and the evaluation in Fig. 4b compares simulated joint angles to motion capture data from the same human subject. This is an in-sample fit: close tracking is expected because the reward directly penalizes deviation from these trajectories. The claim of 'close resemblance between natural and simulated human motor behaviours' requires held-out validation, for example forward prediction on a different subject's gait, or at minimum a clear statement that the result demonstrates tracking performance rather than independent evidence of naturalness.
[Fig. 4c, EMG comparison] The muscle-activation comparison is more independent than the kinematics, because EMG was not directly rewarded, but the manuscript provides only qualitative agreement, acknowledges 'time-locked phase shifts,' and reports no quantitative error metric. Eight muscles are shown, and the text says the patterns 'largely match' without specifying how many of the recorded muscles exhibit phase shifts. The authors should report quantitative metrics (e.g., Pearson correlation, root-mean-square difference, or phase offset in milliseconds) across all measured muscles, or explicitly moderate the claim of 'closely match' to 'qualitatively resemble in select muscles.'
[Construction of embodied human sensory-musculoskeletal model] For muscle-tendon units without published architectural parameters, the authors use a 'systematic approach for parameter estimation based on primary anatomical data,' but no validation of these estimates is provided. The highlighted deep-muscle activations (flexor digitorum longus, internal oblique) and the general claim of predicting unmeasurable dynamics depend entirely on these unvalidated parameters and on the simplified Hill-type muscle model. A sensitivity analysis of key muscle parameters, or a validation of the estimation method on a subset of muscles with known values, is needed to support the central claim that the model reveals dynamics that cannot be directly measured.
[Bicycle riding, Fig. 6] The bicycling task is trained to match 'synthetic 3D trajectories from pre-generated bicycling motion,' and the reported maximum spatial tracking error of 2.38 cm measures deviation from that synthetic reference. This demonstrates that the controller can reproduce a specified trajectory, but it does not validate resemblance to human cycling, since the reference is not experimental. The authors should either compare against human cycling motion-capture data (such as Ref. 51) or rephrase the claim from 'natural cycling posture' to 'task-specific trajectory tracking.'
[Vision-guided object manipulation, Fig. 5] The manipulation task's quantitative evaluation reports translational and rotational errors between object and target, which are the direct components of the reward function. The comparison to a human subject in Fig. 5b is qualitative, with no quantitative kinematics of eye, head, and hand movements reported for the human. To support the claim of human-like visuomotor coordination, the authors should provide quantitative model-to-human comparisons of the eye-head-hand trajectories, or at least report statistics on the human subject's fixation and error profiles.

minor comments (6)

[Fig. 1j,k and Deep reinforcement learning] The action space is described as 1266x1, but the hierarchical actor includes a group action network and a unit action network. Please clarify the dimensions of the group-level actions and the unit adjustment weights, and how they are combined to produce the final 1266-dimensional excitation vector.
[Abstract] The abstract states 'precise anatomical representations' and 'accurate spatial arrangement and validated functional parameters,' but the Discussion acknowledges that the Hill-type muscle model simplifies muscle-tendon dynamics and that some parameters are estimated. Consider softening the abstract's phrasing to avoid overstatement.
[Introduction] The sentence noting that 'precise counts of joints and muscles vary with the classification methods' is immediately followed by specific counts (206 bones, over 200 joints, more than 600 muscles). Please clarify how the model's exact numbers (278 joints, 1,266 muscle-tendon units) were derived from the varying anatomical classifications.
[Bicycle riding] The 'synthetic 3D trajectories from pre-generated bicycling motion' are not described; please state how these trajectories were generated, whether they are physically consistent with the bicycle model, and why this choice was made instead of using experimental cycling data.
[Fig. 4c caption] The caption reports 'n=7 for EMG,' but the main text does not specify the number of EMG trials or how the EMG was processed (e.g., filtering, normalization). Please provide these details.
[Supplementary materials] The text refers to Supplementary Videos 1-3 but does not describe their content; please add brief descriptions in the text or a supplementary note.

Circularity Check

3 steps flagged · score 6.0 of 10

Reported successes for bipedal walking, bicycling, and object manipulation are evaluated against the same reference trajectories and reward objectives used for training; the independent EMG evidence is qualitative and limited, so the headline claim of close resemblance and unmeasurable dynamics is not independently established.

fitted input called prediction [Results, Simulation of bipedal walking, first two paragraphs and Fig. 4b]
"During training, the network controller received kinematic references derived from motion capture data of body keypoint trajectories from an adult male performing bipedal walking... the DRL algorithm maximized a reward function that encouraged simulated movements to match the reference motion capture data... joint angle trajectories of eight representative joints closely tracked the motion capture data from the human subject throughout three gait cycles of bipedal walking."

The motion-capture trajectories are the training target, and the reward is defined to minimize tracking error against them. The reported 'close tracking' is therefore the optimized objective, not an independent check of naturalness. The same adult male is both the source of the reference and the comparison subject, so the evaluation is in-sample by construction. This is a fitted input presented as a quantitative evaluation of resemblance.
fitted input called prediction [Results, Bicycle riding, first two paragraphs and Fig. 6c]
"During training, the network controller was rewarded for maintaining stable upper body posture while holding the fixed handlebar via sensory feedback signals and for matching the model's body keypoint trajectories to synthetic 3D trajectories from pre-generated bicycling motion... Quantitative analysis reveals high precision in the control of simulated body movements, with a maximum spatial tracking error of 2.38 ±0.27 cm (mean ±SEM, n=10) for body keypoints across all three axes."

The synthetic 3D trajectories are the inputs used to define the bicycling reward; the reported maximum keypoint error is the same quantity the controller was trained to minimize. Consequently the error magnitude is an in-sample fit, not independent evidence about human cycling. The additional comparison to human cycling data is only qualitative ('similar to human cycling motion capture experiments'), with no quantitative error metric reported.

1 more flagged steps

fitted input called prediction [Results, Vision-guided object manipulation, third and fifth paragraphs and Fig. 5c]
"a specifically designed reward function was used to encourage the model to reduce both positional and orientational differences between the object and target during training... During the target tracking phase (after 0.5 s), the object stayed close to the target with an average translational difference of 0.15±0.25 cm and a rotational difference of 0.01±0.02 radians."

The manipulation reward directly penalizes positional and orientational differences between object and target; the reported small differences restate that optimized objective. Because no held-out human data or generalization measure is used, the 'success' of the task is the reward itself, so it cannot independently support the claim of human-like visuomotor control.

full rationale

The central claim that SMS-Human reproduces natural human motor behaviours and reveals unmeasurable musculoskeletal dynamics rests primarily on three demonstrated tasks. For bipedal walking, the controller is trained with a reward that matches motion-capture reference trajectories, and the reported joint-angle agreement in Fig. 4b is evaluated against the same reference data from the same adult male. This is an in-sample fit by construction, not an independent prediction. The bicycling task is trained to match synthetic 3D trajectories, so its keypoint-tracking error is likewise the training objective rather than evidence of human-likeness. The manipulation task defines success as minimizing object-to-target differences, and the reported small offsets are exactly that reward objective. These three items reduce to the training inputs, giving a partial circularity score of 6. The EMG comparisons in Fig. 4c are more independent because muscle activity is not directly rewarded, but the paper itself notes that 'the available human EMG data are limited' and reports only qualitative agreement with 'time-locked phase shifts' for eight muscles, with no quantitative error metric. The highlighted deep-muscle activations (flexor digitorum longus, internal oblique) are unvalidated model outputs generated by a highly redundant 1,266-actuator system, so they cannot independently confirm the claim of revealed dynamics. No load-bearing self-citation chain was found: the authors' prior ICRA paper is cited only as related work, not as the justification for the central result. Overall, the paper's main 'predictions' reduce to the fitted training objectives, while the more independent muscle-activation evidence is acknowledged by the authors to be limited.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central predictive claims depend on unvalidated muscle parameters, simplified muscle physiology, and in-sample training references. No code or model files are released, so the ledger contains several hand-estimated inputs.

free parameters (3)

Muscle-tendon architectural parameters for muscles absent from literature = Not reported; estimated from anatomical data
Section 'Construction' states a systematic estimation approach based on refs 34-42. These parameters (optimal fiber length, PCSA, pennation, tendon slack) determine force output and thus all simulated muscle activations.
Joint range-of-motion and axis parameters = Not reported
Joint configurations were determined from anatomical studies and previous models (refs 18-26); any inaccuracies propagate to all tasks.
Reward function weights and stage-wise training schedules = Not reported
Reward design and curriculum thresholds are hand-chosen and shape the learned behaviors; no sensitivity analysis is provided.

assumptions (4)

domain assumption Hill-type muscle model with normalized force-length-velocity relationships captures physiologically plausible muscle behaviour
Invoked via refs 43,44 in Section 'Construction'. If the simplified muscle model is inaccurate, simulated forces and activations, including deep-muscle predictions, are unreliable.
domain assumption MuJoCo physics engine accurately simulates rigid-body contacts and joint constraints for whole-body human movement
MuJoCo (ref 15) is the simulator for all tasks; contact and friction models are approximations.
domain assumption The motion capture and EMG data from one adult male are representative of natural human gait
Section 'Simulation of bipedal walking' uses a single reference subject for kinematic and EMG comparison; generalization to human population is assumed.
ad hoc to paper A systematic estimation of missing muscle parameters from primary anatomical data yields accurate values
Section 'Construction' uses this method for parameters not in existing literature; no validation against direct measurements is provided.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Human sensory-musculoskeletal modeling and control of whole-body movements." pith.science (2026). https://pith.science/paper/US3KF4A2

@misc{pith2026250600071,
  author       = {Pith},
  title        = {Pith review of: Human sensory-musculoskeletal modeling and control of whole-body movements},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/US3KF4A2}},
  note         = {Machine review of arXiv:2506.00071}
}

read the original abstract

Coordinated human movement depends on the integration of multisensory inputs, sensorimotor transformation, and motor execution, as well as sensory feedback resulting from body-environment interaction. Building dynamic models of the sensory-musculoskeletal system is essential for understanding movement control and investigating human behaviours. Here, we report a human sensory-musculoskeletal model, termed SMS-Human, that integrates precise anatomical representations of bones, joints, and muscle-tendon units with multimodal sensory inputs involving visual, vestibular, proprioceptive, and tactile components. A stage-wise hierarchical deep reinforcement learning framework was developed to address the inherent challenges of high-dimensional control in musculoskeletal systems with integrated multisensory information. Using this framework, we demonstrated the simulation of three representative movement tasks, including bipedal locomotion, vision-guided object manipulation, and human-machine interaction during bicycling. Our results showed a close resemblance between natural and simulated human motor behaviours. The simulation also revealed musculoskeletal dynamics that could not be directly measured. This work sheds deeper insights into the sensorimotor dynamics of human movements, facilitates quantitative understanding of human behaviours in interactive contexts, and informs the design of systems with embodied intelligence.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

51 extracted references · 51 canonical work pages

[1]

Gray’s Anatomy: The Anatomical Basis of Clinical Practice 41st edn

Standring, S. Gray’s Anatomy: The Anatomical Basis of Clinical Practice 41st edn. Gray’s Anatomy Series (Elsevier Limited, 2016)

work page 2016
[2]

Saul, K. R. et al. Benchmarking of dynamic simulation predictions in two software platforms using an upper limb musculoskeletal model. Computer methods in biomechanics and biomedical engineering 18, 1445 –1458 (2015)

work page 2015
[3]

McFarland, D. C. et al. A musculoskeletal model of the hand and wrist capable of simulating functional tasks. IEEE Transactions on Biomedical Engineering 70, 1424 –1435 (2022)

work page 2022
[4]

Delp, S. L. et al. An interactive graphics -based model of the lower extremity to study orthopaedic surgical procedures. IEEE Transactions on Biomedical engineering 37, 757 –767 (1990)

work page 1990
[5]

Rajagopal, A. et al. Full-body musculoskeletal model for muscle -driven simulation of human gait. IEEE Transactions on Biomedical Engineering 63, 2068 –2079 (2016)

work page 2016
[6]

K., Arnold, A

Lai, A. K., Arnold, A. S. & Wakeling, J. M. Why are antagonist muscles co -activated in my simulation? a musculoskeletal model for analysing human locomotor tasks. Annals of biomedical engineering 45, 2762 –2774 (2017)

work page 2017
[7]

H., Sifakis, E

Lee, S. H., Sifakis, E. & Terzopoulos, D. Comprehensive biomechanical modeling and simulation of the upper body. ACM Transactions on Graphics (TOG) 28, 1–17 (2009)

work page 2009
[8]

A., Allaire, B

Schmid, S., Burkhart, K. A., Allaire, B. T., Grindle, D. & Anderson, D. E. Musculoskeletal full - body models including a detailed thoracolumbar spine for children and adolescents aged 6 –18 years. Journal of biomechanics 102, 109305 (2020)

work page 2020

Show all 51 references

[9]

Kuo, A. D. & Donelan, J. M. Dynamic principles of gait and their clinical implications. Physical therapy 90, 157 –174 (2010)

2010
[10]

G., Yeadon, M

Lewis, M. G., Yeadon, M. R. & King, M. A. Are torque -driven simulation models of human movement limited by an assumption of monoarticularity? Applied Sciences 11, 3852 (2021)

2021
[11]

T., Surma, E

Damsgaard, M., Rasmussen, J., Christensen, S. T., Surma, E. & De Zee, M. Analysis of muscu - loskeletal systems in the anybody modeling system. Simulation Modelling Practice and Theory 14, 1100 –1111 (2006)

2006
[12]

Delp, S. L. et al. Opensim: open -source software to create and analyze dynamic simulations of movement. IEEE Transactions on Biomedical Engineering 54, 1940 –1950 (2007)

2007
[13]

& Sui, Y

Zuo, C., He, K., Shao, J. & Sui, Y. Self model for embodied intelligence: Modeling full- body human musculoskeletal system and locomotion control with hierarchical low -dimensional representation. 2024 IEEE International Conference on Robotics and Automation (ICRA) , 13062– 13...

2024
[14]

& Kumar, V

Caggiano, V., Wang, H., Durandau, G., Sartori, M. & Kumar, V. Myosuite: A contact -rich sim - ulation suite for musculoskeletal motor control. Learning for Dynamics and Control Conference , 492–507 (PMLR, 2022)

2022
[15]

& Tassa, Y

Todorov, E., Erez, T. & Tassa, Y. Mujoco: A physics engine for model -based control. 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems , 5026–5033 (IEEE, 2012)

2012
[16]

Mitsuhashi, N. et al. Bodyparts3d: 3d structure database for anatomical concepts. Nucleic acids research 37, D782 –D785 (2009)

2009
[17]

Maharaj, J. N. et al. Modelling the complexity of the foot and ankle during human locomotion: the development and validation of a multi -segment foot model using biplanar videoradiography. Computer Methods in Biomechanics and Biomedical Engineering 25, 554–565 (2022). 14

2022
[18]

Nester, C. et al. In vitro study of foot kinematics using a dynamic walking cadaver model. Journal of biomechanics 40, 1927–1937 (2007)

2007
[19]

& Ferrario, V

Mapelli, A., Galante, D., Lovecchio, N., Sforza, C. & Ferrario, V. F. Translation and rotation movements of the mandible during mouth opening and closing. Clinical Anatomy: The Offi- cial Journal of the American Association of Clinical Anatomists and the British Association of...

2009
[20]

Borkowski, S. L. et al. Challenging the conventional standard for thoracic spine range of motion: a systematic review. Jbjs Reviews 4, e5 (2016)

2016
[21]

J., Kim, J

Lee, W. J., Kim, J. H., Shin, Y. U., Hwang, S. & Lim, H. W. Differences in eye movement range based on age and gaze direction. Eye 33, 1145 –1151 (2019)

2019
[22]

P., Z˙ urawski, A

Kiebzak, W. P., Z˙ urawski, A. L- . & Kosztol -owicz, M. Alignment of the sternum and sacrum as a marker of sitting body posture in children. International Journal of Environmental Research and Public Health 19, 16287 (2022)

2022
[23]

& Hayashi, M

Negishi, K., Watanabe, K., Teramoto, A., Yamatsu, K. & Hayashi, M. Three -dimensional motion of the toes with simulated contraction of individual toe flexors and extensors: A cadaver study. The Foot 56, 102044 (2023)

2023
[24]

A., Lotz, J

Christophy, M., Faruk Senan, N. A., Lotz, J. C. & O’Reilly, O. M. A musculoskeletal model for the lumbar spine. Biomechanics and modeling in mechanobiology 11, 19–34 (2012)

2012
[25]

D., Jackson, R

Uhlrich, S. D., Jackson, R. W., Seth, A., Kolesar, J. A. & Delp, S. L. Muscle coordination retraining inspired by musculoskeletal simulations reduces knee contact force. Scientific reports 12, 9842 (2022)

2022
[26]

D., Vasavada, A

Mortensen, J. D., Vasavada, A. N. & Merryweather, A. S. The inclusion of hyoid muscles improve moment generating capacity and dynamic simulations in musculoskeletal models of the head and neck. PloS one 13, e0199912 (2018)

2018
[27]

L., Roy, R

Wickiewicz, T. L., Roy, R. R., Powell, P. L. & Edgerton, V. R. Muscle architecture of the human lower limb. Clinical Orthopaedics and Related Research 179, 275 –283 (1983)

1983
[28]

G., Zajac, F

Hoy, M. G., Zajac, F. E. & Gordon, M. E. A musculoskeletal model of the human lower extremity: the effect of muscle, tendon, and moment arm on the moment -angle relationship of musculotendon actuators at the hip, knee, and ankle. Journal of biomechanics 23, 157–169 (1990)

1990
[29]

Friederich, J. A. & Brand, R. A. Muscle fiber architecture in the human lower limb. Journal of biomechanics 23, 91–95 (1990)

1990
[30]

Anderson, F. C. & Pandy, M. G. Dynamic optimization of human walking. J. Biomech. Eng. 123, 381–390 (2001)

2001
[31]

& Triesch, J

Priamikov, A., Fronius, M., Shi, B. & Triesch, J. Openeyesim: A biomechanical model for simulation of closed -loop visual perception. Journal of vision 16, 25–25 (2016)

2016
[32]

D., van Leeuwen, T

Kerkhof, F. D., van Leeuwen, T. & Vereecke, E. E. The digital human forearm and hand. Journal of Anatomy 233, 557 –566 (2018)

2018
[33]

& Delp, S

Seth, A., Dong, M., Matias, R. & Delp, S. Muscle contributions to upper-extremity movement and work from a musculoskeletal model of the human shoulder. Frontiers in neurorobotics 13, 90 (2019)

2019
[34]

& Brugman, P

Van Eijden, T., Korfage, J. & Brugman, P. Architecture of the human jaw -closing and jaw - opening muscles. The Anatomical Record: An Official Publication of the American Association of Anatomists 248, 464–474 (1997). 15

1997
[35]

Breteler, M. D. K., Spoor, C. W. & Van der Helm, F. C. Measuring muscle and joint geometry parameters of a shoulder for modeling purposes. Journal of biomechanics 32, 1191 –1197 (1999)

1999
[36]

L., Suryanarayanan, S., Murray, W

Delp, S. L., Suryanarayanan, S., Murray, W. M., Uhlir, J. & Triolo, R. J. Architecture of the rectus abdominis, quadratus lumborum, and erector spinae. Journal of biomechanics 34, 371 –375 (2001)

2001
[37]

& Van Eijden, T

Koolstra, J. & Van Eijden, T. Combined finite -element and rigid -body analysis of human jaw joint dynamics. Journal of biomechanics 38, 2431 –2439 (2005)

2005
[38]

R., Ranes, A

Lachowitzer, M. R., Ranes, A. & Yamaguchi, G. T. Musculotendon parameters and mus - culoskeletal pathways within the human foot. Journal of applied biomechanics 23, 20–41 (2007)

2007
[39]

R., Eng, C

Ward, S. R., Eng, C. M., Smallwood, L. H. & Lieber, R. L. Are current measurements of lower extremity muscle architecture accurate? Clinical orthopaedics and related research 467, 1074–1082 (2009)

2009
[40]

A., Happee, R

Borst, J., Forbes, P. A., Happee, R. & Veeger, D. H. Muscle parameters for musculoskeletal modelling of the human neck. Clinical biomechanics 26, 343–351 (2011)

2011
[41]

G., Meyer, C

Handsfield, G. G., Meyer, C. H., Hart, J. M., Abel, M. F. & Blemker, S. S. Relationships of 35 lower limb muscles to height and body mass quantified using mri. Journal of biomechanics 47, 631–638 (2014)

2014
[42]

Mirakhorlo, M. et al. Anatomical parameters for musculoskeletal modeling of the hand and wrist. International Biomechanics 3, 40–49 (2016)

2016
[43]

& Delp, S

Millard, M., Uchida, T., Seth, A. & Delp, S. L. Flexing computational muscle: modeling and simulation of musculotendon dynamics. Journal of biomechanical engineering 135, 021005 (2013)

2013
[44]

Wolpert, D. M. & Ghahramani, Z. Computational principles of movement neuroscience. Nature neuroscience 3, 1212 –1217 (2000)

2000
[45]

Primal pictures anatomy teaching resources: 3d anatomy software and 3d real -time viewer

Wilkinson, T. Primal pictures anatomy teaching resources: 3d anatomy software and 3d real -time viewer. Journal of Anatomy 220, 118 (2012)

2012
[46]

The co -ordination and regulation of movements

Bernstein, N. The co -ordination and regulation of movements. The co -ordination and regulation of movements (1966)

1966
[47]

H., Chvatal, S

Ting, L. H., Chvatal, S. A., Safavynia, S. A. & Lucas McKay, J. Review and perspective: neurome - chanical considerations for predicting muscle activation patterns for movement. International journal for numerical methods in biomedical engineering 28, 1003 –1014 (2012)

2012
[48]

Neurobiological bases of rhythmic motor acts in vertebrates

Grillner, S. Neurobiological bases of rhythmic motor acts in vertebrates. Science 228, 143 –149 (1985)

1985
[49]

Dominici, N. et al. Locomotor primitives in newborn babies and their development. Science 334, 997–999 (2011)

2011
[50]

& Levine, S

Haarnoja, T., Zhou, A., Abbeel, P. & Levine, S. Soft actor -critic: Off -policy maximum entropy deep reinforcement learning with a stochastic actor. International Conference on Machine Learning , 1861 –1870 (PMLR, 2018)

2018
[51]

E., Gatti, A

Clancy, C. E., Gatti, A. A., Ong, C. F., Maly, M. R. & Delp, S. L. Muscle -driven simulations and experimental data of cycling. Scientific Reports 13, 21534 (2023)

2023

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Gray’s Anatomy: The Anatomical Basis of Clinical Practice 41st edn

Standring, S. Gray’s Anatomy: The Anatomical Basis of Clinical Practice 41st edn. Gray’s Anatomy Series (Elsevier Limited, 2016)

work page 2016

[2] [2]

Saul, K. R. et al. Benchmarking of dynamic simulation predictions in two software platforms using an upper limb musculoskeletal model. Computer methods in biomechanics and biomedical engineering 18, 1445 –1458 (2015)

work page 2015

[3] [3]

McFarland, D. C. et al. A musculoskeletal model of the hand and wrist capable of simulating functional tasks. IEEE Transactions on Biomedical Engineering 70, 1424 –1435 (2022)

work page 2022

[4] [4]

Delp, S. L. et al. An interactive graphics -based model of the lower extremity to study orthopaedic surgical procedures. IEEE Transactions on Biomedical engineering 37, 757 –767 (1990)

work page 1990

[5] [5]

Rajagopal, A. et al. Full-body musculoskeletal model for muscle -driven simulation of human gait. IEEE Transactions on Biomedical Engineering 63, 2068 –2079 (2016)

work page 2016

[6] [6]

K., Arnold, A

Lai, A. K., Arnold, A. S. & Wakeling, J. M. Why are antagonist muscles co -activated in my simulation? a musculoskeletal model for analysing human locomotor tasks. Annals of biomedical engineering 45, 2762 –2774 (2017)

work page 2017

[7] [7]

H., Sifakis, E

Lee, S. H., Sifakis, E. & Terzopoulos, D. Comprehensive biomechanical modeling and simulation of the upper body. ACM Transactions on Graphics (TOG) 28, 1–17 (2009)

work page 2009

[8] [8]

A., Allaire, B

Schmid, S., Burkhart, K. A., Allaire, B. T., Grindle, D. & Anderson, D. E. Musculoskeletal full - body models including a detailed thoracolumbar spine for children and adolescents aged 6 –18 years. Journal of biomechanics 102, 109305 (2020)

work page 2020