Neuromusculoskeletal model-informed machine learning-based control of a knee exoskeleton with uncertainties...

Area of Science:

• Biomechanics
• Robotics
• Neuroscience

Background:

• Exoskeleton assistance strategies are increasingly incorporating user torque capacity.
• Predicting user torque capacity involves uncertainty, impacting exoskeleton-user interface safety.

Purpose of the Study:

• Propose an adaptive control framework for knee exoskeletons.
• Quantify uncertainty in predicted user torque using confidence bounds.
• Enhance the safety of the exoskeleton-user interface.

Main Methods:

• Utilized muscle electromyography (EMG) signals and joint kinematics.
• Developed a neuromusculoskeletal (NMS) solver-informed Bayesian Neural Network (NMS-BNN) to predict torque with confidence bounds.
• Compared NMS-BNN performance against a Gaussian process (NMS-GP) model.

Main Results:

• Both NMS-BNN and NMS-GP models accurately predicted knee joint torque with low error, outperforming traditional NMS models.
• High uncertainties were observed at movement initiation, terminal stance, and terminal swing during walking.
• The knee exoskeleton delivered desired assistive torque with low error, though variations occurred with walking speed.

Conclusions:

• The developed framework successfully predicted knee torque with quantifiable uncertainty.
• Adaptive assistive torque was provided, improving the safety of the exoskeleton-user interface.
• This approach has significant potential for developing personalized and safe exoskeleton assistance.