Preliminary Investigation of Symmetry Learning Control for Powered Ankle-Foot Prostheses

Abstract

This article proposes a human-in-the-loop optimization method, targeting gait symmetry, for powered ankle-foot prostheses (PAFPs). Individuals with unilateral below-knee amputations have distinctly asymmetrical gaits, which predisposes them to a host of secondary musculoskeletal impairments, including osteoarthritis of the intact limb joints. PAFPs can restore some ankle function, however current control methodologies rely on able-bodied gait data for trajectory synthesis, require expert tuning, and are limited in their ability to adapt. Human-in-the-loop methods, where the control signal is adjusted based on the achieved actions of the coupled human-robot system, would allow for automatic personalization and continuous adaptation. An adaptive gain iterative learning control algorithm adjusts the PAFPs torque to match the achieved intact ankle torque while maintaining boundedness of the control signal. The method is experimentally assessed during a pilot (N=1) study with a prototype PAFP. Results indicate a 25% reduction in the difference of mean peak ankle torques, and a reduction in ankle toque, ankle power and support moment asymmetry. This work demonstrates the practical implementation of a symmetry-based learning controller, which resulted in beneficial biomechanic adaptations, therefore providing motivation for future investigations of symmetry-based controllers for PAFPs.