Balancing Biomechanics and Preference in Assistive Device Tuning via Metric-Regularized Optimization

Gait-assistive technology has the potential to benefit millions, but adoption is limited by challenges in tuning assistance to simultaneously provide biomechanical benefits and satisfy patient and clinician preference. In this study, we quantify the dissonance between these outcomes, inspect its sources, and propose methods to address it. Meth ods: Wecollected biomechanics and preference data from nine individuals post-stroke using a plantarflexion neuroprosthesis, and 96 corresponding preference datasets from 36 clinicians. We inspected the biomechanics and preference modeled out comes occurring when either outcome was optimized in isolation. Then, we used weighted sums of biomechanical principal components to identify determinants of preference for patients and clinicians, and inspected their anatomical locations. Finally, we extended this weighting method to biomechanical metrics, and developed a method of balancing preference with multiple metric outcomes.

We found that maximizing modeled preference or biomechanics produced poor modeled outcomes in the other domain. Patient and clinician preference could be strongly approximated with fewer than five extracted biomechanical determinants, though heterogeneity of determinants across individuals was high. Our metric-preference balanced method of tuning assistance significantly improved preference outcomes compared to metric-optimal assistance and prevented negative biomechanical outcomes for individualized sets of both one and ten metrics.

This work demonstrates the importance of both biomechanics and preference in gait-assistive device tuning, highlights the individualized nature of the biomechanical determinants of preference, and demonstrates, via offline modeling, that balancing biomechanics and preference is possible.

This work highlights the necessity and feasibility of balanced tuning in gait-assistive devices.