User preference optimization for control of ankle exoskeletons using sample efficient active learning

Area of Science:

• Robotics
• Human-Computer Interaction
• Biomechanics

Background:

• Widespread adoption of augmentative exoskeletons is hindered by challenges in controller tuning.
• Current methods often require extensive resources and focus on single objectives, neglecting multifaceted user experience factors like comfort and stability.

Purpose of the Study:

• To introduce a convenient approach for tuning exoskeleton controller parameters to maximize user preference.
• To leverage wearer feedback for balancing multiple experiential factors in exoskeleton assistance.

Main Methods:

• An evolutionary algorithm recommended controller parameters, ranked by a pre-trained neural network using user preference data.
• Real-time user feedback via forced-choice comparisons guided the tuning process for a partial-assist ankle exoskeleton.

Main Results:

• The approach achieved an average accuracy of 88% in converging on wearer-preferred controller parameters compared to random generation.
• User-preferred settings were typically stabilized within 43 ± 7 queries.

Conclusions:

• User preference can be effectively leveraged for real-time tuning of partial-assist ankle exoskeletons.
• This intuitive interface demonstrates potential for advancing lower-limb wearable technologies for daily use.