Reconstructing Unobservable Biomechanical States Through Inverse Simulation

22 concepts18 papers

When observing a complex biological movement—whether it is an athlete lunging for a shot, a stroke patient struggling to stand, or a speaker articulating a vowel—we only see the external effects of motion. We see limbs changing position, skin deforming, and feet striking the ground. However, the true drivers of these movements—the internal joint torques, deep muscle forces, and neural activation patterns—remain entirely hidden from view. Direct measurement of these internal variables is highly invasive or technologically impossible in living subjects.

To bridge this gap, researchers use inverse biomechanical simulation. The core intuition behind this approach is to treat the physical body as a known mathematical constraint. By replaying observed external movements through a high-fidelity physics-based digital twin, we can work backward from the observed effects to reconstruct the unobservable causes. As explained by Sibo Wang-Chen et al. in The embodied brain: Bridging the brain, body, and behavior with neuromechanical digital twins, this framework allows scientists to fill in the gaps of sparse experimental datasets by applying known biological constraints like skeletal topology and joint limits.

The Physics of Working Backward: Kinematics and Dynamics

The pipeline of inverse simulation operates in two primary stages: inverse kinematics and inverse dynamics.

First, raw experimental data (such as motion capture marker positions or medical imaging frames) must be translated into joint angles and structural coordinates. This is known as inverse kinematics. Once the motion is quantified, inverse dynamics applies classical Newtonian physics to calculate the net forces and torques required to generate that motion.

As demonstrated in High-density EMG, IMU, kinetic, and kinematic open-source data for comprehensive locomotion activities, this process translates external motion capture and ground reaction forces into internal, joint-level mechanical states. By applying Newton-Euler equations to a biomechanical model with known segment masses and inertial properties, Shuang Ren et al. showed in Identification of Kinetic Abnormalities in Male Patients after Anterior Cruciate Ligament Deficiency Combined with Meniscal Injury: A Musculoskeletal Model Study of Lower Limbs during Jogging how to isolate specific kinetic abnormalities, such as reduced knee joint forces during jogging.

This mathematical reverse-engineering is equally valuable for clinical and ergonomic interventions. For instance, Chung‐Ying Tsai et al. utilized inverse dynamics in The Relationship between Independent Transfer Skills and Upper Limb Kinetics in Wheelchair Users to determine the exact internal joint loads experienced by the shoulders and elbows during wheelchair transfers. Similarly, in robotic rehabilitation, Maël Gallois et al. applied these calculations in Human-Robot Co-Simulation method for upper limb assistive force calculation using polytopes to map out the distribution of effort between a patient's arm and an assistive exoskeleton.

Resolving the Redundancy Problem via Optimization

Calculating net joint torques is only half the battle. The human body is fundamentally over-actuated: multiple muscles cross a single joint, meaning an infinite number of distinct muscle activation combinations can produce the exact same net torque. This is known as the muscle redundancy problem.

To resolve this ambiguity, inverse simulations incorporate static optimization. The underlying assumption is that evolution has optimized the nervous system to minimize a biological cost function, such as metabolic energy expenditure or overall muscle activation. As reviewed by Tanja Lerchl et al. in Multibody Models of the Thoracolumbar Spine: A Review on Applications, Limitations, and Challenges, combining inverse dynamics with static optimization allows models to mathematically distribute joint torques among individual muscles of the spine.

This optimization framework makes it possible to observe deep, stabilizing muscles that surface sensors cannot reach. For example, Datao Xu et al. demonstrated in A foot joint and muscle force assessment of the running stance phase whilst wearing normal shoes and bionic shoes how custom footwear alters deep muscle recruitment patterns during running. In sports performance, Qincheng Ge et al. used the same combination of inverse dynamics and static optimization in Exploring Lower Limb Biomechanical Differences in Competitive Aerobics Athletes of Different Ability Levels During Rotational Jump Landings to expose how lower-level athletes exhibit poor joint stability and elevated injury risks due to suboptimal muscle recruitment strategies.

Integrating Electromyography and Hybrid Dynamics

While pure optimization provides a mathematically elegant solution, actual human movement is driven by neural commands that may not always be perfectly optimal. To ground these simulations in biological reality, researchers frequently integrate surface electromyography (EMG) data.

In Spatiotemporal Maps of Proprioceptive Inputs to the Cervical Spinal Cord During Three-Dimensional Reaching and Grasping, Pierre Kibleur et al. combined experimental kinematics with EMG data to drive computed muscle controls, mapping out the resulting muscle lengths, velocities, and forces during 3D arm movements. Similarly, Jun Wen et al. processed EMG data within a musculoskeletal simulation framework in Effects of foot rotation on ankle control and injury risk during badminton lunge landings to link surface muscle activity directly to internal joint reaction forces during complex maneuvers.

However, using EMG data presents a major challenge in clinical populations (such as stroke survivors) who cannot perform the standard maximum voluntary contraction tests required for calibration. To circumvent this, Qi An et al. developed a hybrid approach in Clarify Sit-to-Stand Muscle Synergy and Tension Changes in Subacute Stroke Rehabilitation by Musculoskeletal Modeling. Their method uses inverse dynamics to calculate joint torques from motion capture, and then feeds those torques into a forward dynamics simulation to scale and normalize muscle activity, successfully mapping out how patients regain coordinated control during rehabilitation.

Extending to Soft Tissues: Inverse Finite Element Modeling

The principles of rigid-body inverse dynamics break down when applied to complex, bone-free soft tissues like the human tongue or vocal tract. In these domains, researchers replace rigid skeletal links with Inverse Finite Element Modeling (FEM) and iterative optimization loops.

Instead of solving algebraic physics equations, the simulation treats the tissue as a continuous deformable mesh. As pioneered by Jianwu Dang 0001 et al. in A model based investigation of activation patterns of the tongue muscles for vowel production, a gradient descent algorithm iteratively adjusts internal muscle activation parameters until the simulated model's shape perfectly matches real-world MRI data.

This soft-tissue inverse simulation allows researchers to uncover personalized motor control strategies. In Speaker-Specific Biomechanical Model-Based Investigation of a Simple Speech Task Based on Tagged-MRI, Keyi Tang et al. demonstrated that by solving for the forces that match tracked tissue trajectories from tagged MRI, they could accurately infer the unique muscular strategies different speakers use to produce identical sounds.

To validate the mathematical reliability of this approach, Peter Anderson et al. showed in A comprehensive 3d biomechanically-driven vocal tract model including inverse dynamics for speech research that inverse dynamics simulations could faithfully recreate known muscle activations from forward speech simulations. This mathematical consistency enables non-invasive comparative studies, such as the work by Ursa Maity et al. in Motor Control Similarity Between Speakers Saying "A Souk" Using Inverse Atlas Tongue Modeling, which utilized an inverse atlas tongue model to uncover shared underlying motor control strategies across different individuals.

Simplifying the Control Space: Neural Primitives

If inverse simulation reveals that the body can be controlled in infinite ways, how does the brain avoid getting overwhelmed? Evidence suggests that the nervous system simplifies this high-dimensional problem by grouping muscles into low-dimensional functional units.

In A musculoskeletal model of human locomotion driven by a low dimensional set of impulsive excitation primitives, Dario Farina demonstrated that mapping muscle groups to a small set of fundamental, impulsive excitation primitives can accurately predict the activity of deep, unmeasurable muscles across diverse locomotor tasks. By constraining the inverse simulation's search space to these low-dimensional neural primitives, researchers can achieve highly accurate biomechanical reconstructions even when comprehensive experimental data is unavailable.

Key Foundational Work

A model based investigation of activation patterns of the tongue muscles for vowel production (2008) — contributed 1 concept elements to this direction.
Effective utilization of gravity during arm downswing in keystrokes by expert pianists (2009) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_inverse_biomechanical_simulation_b9c02426,
  title = {inverse biomechanical simulation},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/inverse-biomechanical-simulation/rdr_b9c02426941d48fdad3c802d6435afda},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-21},
}

This exact version (version 1, frozen text):

@misc{lacuna_inverse_biomechanical_simulation_b9c02426_v1,
  title = {inverse biomechanical simulation},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/inverse-biomechanical-simulation/rdr_b9c02426941d48fdad3c802d6435afda/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-21},
}

See something wrong? Report a mistake