Predicting walking response to ankle exoskeletons using data-driven models.

Area of Science:

• Biomechanics
• Robotics
• Human-Computer Interaction

Background:

• Predicting subject-specific responses to exoskeletons is crucial for effective device design and control.
• Existing models often struggle to capture individual variations in gait dynamics.

Purpose of the Study:

• To evaluate the predictive accuracy of three subject-specific phase-varying (PV) models for exoskeleton impacts on gait.
• To assess the models' ability to predict kinematic and myoelectric responses without prior user-specific data.

Main Methods:

• Utilized Floquet theory to develop PV, linear PV (LPV), and nonlinear PV (NPV) models.
• Collected kinematic and myoelectric data from 12 adults walking with passive ankle exoskeletons under varying torque conditions.

Main Results:

• The LPV model outperformed the basic PV model for short-term predictions ( < 12.5% stride).
• LPV models explained 49-70% of variance in kinematic responses and matched NPV model accuracy.
• Myoelectric responses were difficult to predict, with models explaining < 10% of variance.

Conclusions:

• Data-driven PV models, particularly LPV, demonstrate potential for predicting subject-specific exoskeleton effects on gait kinematics.
• Further advancements are needed to accurately predict myoelectric responses for comprehensive exoskeleton control.