A personalized quasi-passive ankleexoskeleton using human-in-the loop optimization approaches

Sammanfattning: Passive ankle exoskeletons have been proven to successfully decrease muscle demand for walking. The movement of the user’s ankle drives a mechanical clutch and spring, which eventually stores and returns mechanical energy to the user. However, the applicability of the device is limited as the clutch engages at a fixed angle and the stiffness of the spring is fixed. These limitations restrict the usage of passive ankle exoskeletons for different users and for different walking speeds. To improve the passive ankle exoskeleton, a bilateral quasi-passive ankle exoskeleton has been developed during this thesis work. The new developed exoskeleton consists of multiple springs engaged by electromagnetic clutches for different assistance torque configurations, with the aim of adapting the assistance torque profiles for different users and walking speeds. Control software has been developed for exoskeleton control using human-in-the-loop optimization approaches. Muscle activity of the user is estimated during walking and the control software in the computer provides different control parameters configurations for evaluations. The evaluated result is sent from the controller to the computer, and the software eventually generates new configurations. Two different optimization algorithms are investigated in this project through experimentation, focusing on the magnitude and consistency of muscle activity reduction. The effects of assistance torque’s timing and magnitude on the muscle activity reduction at different walking speeds are also studied. The results showed that Bayesian Optimization has a better magnitude of muscle activity reduction in general, and the Covariance matrix adaptation evolution strategy has a more consistent reduction in muscle activity. The results also showed that changing the torque magnitudes has a larger effect on muscle activity decline than changing the duration of the torque profiles.