Human-in-the-loop optimization of rocker shoes via different cost functions during walking

Abstract

Changing the apex position and angle of a rocker shoe can modify the gear ratio around the ankle [1], base of support and roll-over direction [2], and therefore affect different gait related objectives (e.g. metabolic cost, mechanical load or stability). Optimal apex parameters for these different objectives are dependent on individual musculoskeletal characteristics and the voluntary, yet unpredictable, gait adaptations of the user in response to changes in apex parameters [3]. A method to overcome these challenges is human-in-the-loop optimization [4], in which the human is included ‘in vivo’ in the control loop and apex parameters are systematically varied using an optimization algorithm in response to measured performances to optimize human performance. However, the outcome of this process might depend on the selected optimization objective, but knowledge about how different cost functions affect this outcome is lacking. The aim of the study is to investigate whether human-in-the-loop optimization via different cost functions, i.e. metabolic cost, external mechanical work, and gait stability, affects the optimal apex position and angle for individuals during walking. Seven healthy participants underwent three different optimization protocols while walking on a treadmill. With the different optimization protocols, we aimed to minimize (1) metabolic cost of walking, (2) negative collision work on the centre of mass, and (3) step distance (vector step length and step width) variability (as measure of gait stability) by optimizing the rocker profile of experimental shoes, with tuneable apex position and angle, using an evolutionary optimization algorithm [5]. Optimal shoe settings for the different cost functions and standard settings were compared. Optimized apex lines for the different cost functions are presented in Fig. 1. The optimized apex positions (percentage total shoe length) were located more distal compared to the standard position (64.0%) and significant difference between cost functions was approached (metabolic cost: 70.3±4.3%, collision work: 76.5±12.4%, step distance variability: 73.4±4.4%, p=0.05). The optimized apex angles tended to be larger compared to the standard angle (88.0˚), but were quite variable among participants (metabolic cost: 118.0±16.0˚, collision work: 93.2±33.5˚, and step distance variability: 103.0±27.7˚). Consequently, significant differences in apex angle between cost functions were not found.Download : Download high-res image (108KB)Download : Download full-size image Cost function tended to have an effect on optimal apex parameters. Optimizing for metabolic cost tended to result in a more proximal apex position compared to the other cost functions, while high variability in optimal angles between participants were found for most cost functions. The variety in optimal apex parameters between participants emphasizes the importance of an individualized approach. Our next step is to investigate how these optimized rocker profiles influence gait mechanics and energetics to further explore whether selecting the proper cost function should be taken into account while designing an individual shoe.