A reconstruction method of missing torque data using a simplified neuromusculoskeletal (NMS) model

Area of Science:

• Robotics
• Rehabilitation Engineering
• Biomechanics

Background:

• Exoskeleton rehabilitation requires accurate torque control for effectiveness and comfort.
• Data loss from sensor failures or transmission issues hampers predictive model accuracy.
• Existing data reconstruction methods are insufficient for complex rehabilitation data.

Purpose of the Study:

• To develop a novel torque data reconstruction method for exoskeleton rehabilitation.
• To improve the predictive accuracy of models trained on incomplete torque data.
• To enhance the overall effectiveness and comfort of exoskeleton-assisted rehabilitation.

Main Methods:

• Designed a torque data reconstruction method utilizing a simplified neuromusculoskeletal model.
• Employed an interpretable model to estimate and reconstruct missing torque data segments.
• Simulated the temporal characteristics of torque data for realistic reconstruction.

Main Results:

• Training neural network models with reconstructed data improved predicted torque accuracy by an average of 12.70%.
• The proposed method outperformed periodic shift interpolation, chained equations multiple imputation, and neural network imputation.
• Demonstrated superior performance in reconstructing missing time series data for rehabilitation applications.

Conclusions:

• The neuromusculoskeletal model-based reconstruction effectively addresses data loss in exoskeleton rehabilitation.
• This method enhances the accuracy of predictive models, leading to better rehabilitation outcomes.
• The approach offers a significant advancement over existing imputation techniques for time-series rehabilitation data.