• inverse dynamics
• MuJoCo
• muscle force estimation
• musculoskeletal models
• neuroprosthetics
• synthetic data

Electrical stimulation of peripheral nerves or spinal cord for movement restoration has been investigated in preclinical and clinical settings in cases of paralysis following trauma or illness. Nonetheless, the optimization and personalization of neuromodulation techniques and of the rehabilitation protocols that are often coupled with electrical stimulation requires in depth quantitative descriptions of the elicited movements. Moreover, the fast-paced development of new neuromodulation techniques and experimental protocols poses a fundamental challenge for rapid prototyping. Integration of modelling techniques in the continuum from fundamental research to clinical application is crucial to quickly bring innovative technologies into the patient's life.
Musculoskeletal models (MSMs) allow to investigate how the musculoskeletal system adapts to various movement tasks and pathological conditions. MSM are powerful tools for estimating variables like muscle-generated forces and joint torques, which are otherwise difficult to measure experimentally. Synthetic data generation offers potential advantages such as early problem identification and flexibility in experimental design.
Simulations of a biologically accurate lower limb MSM are conducted in MuJoCo, a platform known for its efficiency in MSMs simulations, facilitating rapid and precise computations with complex and anatomically faithful models. A method for stochastic generation of muscle controls is developed to obtain synthetic data for muscular activation. Leveraging on the mechanical transmission properties in the MSM, a surrogate correlation matrix is derived. This information makes it possible to introduce meaningful correlations in the stochastically generated data, thereby capturing relevant dynamics in completely synthetic MSMs simulations. This approach allows for testing and rapid prototyping of new experimental protocols on a MSM before acquiring real-world data. A simple and interpretable approach for muscle controls estimation is considered as a benchmark. The efficacy of synthetic control data in tuning and designing the estimator is analyzed and tested using synthetic data. Finally, the effectiveness of the estimator is evaluated on real data from a single healthy subject. Data include recordings from inertial mass units and differential electromyographic sensors to capture kinematics and obtain a ground-truth for muscular controls, represented by the linear envelope of the electromyographic signal. The MSM was used in two distinct conditions of sitting position and upright suspended position, reflecting the experimental design the characterized the real data. Experiments considered multiple repetitions of simple movements and are thought to ignore reaction forces. Hence, the testing and prototyping of the benchmark estimator can be carried out without the confounding factors related to typical gait analysis.
Synthetic data show promising results for the MSM characterization and have been fundamental to plug real data into the simulated MSM. The muscle control estimation algorithm is shown to yield meaningful results with data obtained from inertial and electromyographic sensors. Nevertheless, the stability inverse dynamics results with synthetic data is shown to remain a challenging problem and tuning based on synthetic data has been observed to bring no significant contribution. The need for real data to guide the creation of synthetic muscle controls is emphasized as the main future development required to refine the data generation method and effectively integrate it for rapid prototyping in MSMs. The synthetic data approach holds potential for broader clinical application of MSMs and better integration with neural modelling. MSMs are poised to become a fundamental tool in novel diagnostic and prognostic experimental protocols. Continued development in this direction promises valuable insights for future research aimed at translating MSM techniques into clinical practice.