Current rehabilitation techniques for treating subjects post-stroke are costly and have highly inconsistent outcomes. To improve the reliability of treatment, computational simulations of human locomotion using patient-specific neuromusculoskeletal models could be used be used to determine or design ideal treatment for individual subjects. However, current limitations in musculoskeletal model creation and in dynamic simulations prevent researchers from developing such tools.

For the first part of this dissertation, we developed a novel method for calibrating musculoskeletal models to match individual subjects. Constructing subject-specific musculoskeletal models is expensive and arduous. As a result, most musculoskeletal simulation studies utilize much less accurate scaled versions of generic musculoskeletal models to represent individuals. To improve the accuracy of these scaled models, we developed a novel technique for calibrating scaled musculoskeletal models to better match individual subjects. Muscle-tendon lengths, velocities, and moment arms, as well as other commonly calibrated model parameters were calibrated to match moment data from a hemi-paretic subject while keeping musculoskeletal geometry close to an initial scaled model. We found that our EMG-to-moment method incorporating automated calibration of musculoskeletal geometry predicted net joint moments during walking more accurately than the same method without geometric calibration. This result demonstrates that scaled musculoskeletal models can be calibrated to much more closely match individual subjects using only easily recorded data.

For the second part of this dissertation, we use our improved musculoskeletal models within a direct collocation optimal control framework in order to predict how two hemiparetic subjects walk faster. Most gait prediction studies utilize computationally inefficient algorithms, or use methods that produce physically inconsistent results, requiring fictitious loads to satisfy dynamics equations. In order to make dynamically consistent gait predictions with reasonable computation times, we interfaced our musculoskeletal models with a direct collocation optimal control toolbox to predict faster gait for subjects post-stroke. Using this new simulation framework, we were able to predict plausible fast gait motions. This optimal control gait prediction framework could be used to address a variety of gait related queries.