Human-in-the-loop Optimisation in Robot-assisted Gait Training

TL;DR

The paper addresses the challenge of personalising robot-assisted gait rehabilitation controllers amid high inter- and intra-subject gait variability. It applies human-in-the-loop optimisation using CMA-ES to tune a lower-limb exoskeleton's assist-as-needed impedance controller over multi-day sessions with healthy participants. While CMA-ES demonstrates continuous adaptation, especially increasing knee stiffness in several cases, no conclusive performance gains were observed in validation trials, likely due to pronounced human variability and co-adaptation. These findings highlight the limitations of current personalisation approaches and motivate development of more robust, hybrid, or clinically targeted HILO strategies to realize tangible rehabilitation benefits.

Abstract

Wearable robots offer a promising solution for quantitatively monitoring gait and providing systematic, adaptive assistance to promote patient independence and improve gait. However, due to significant interpersonal and intrapersonal variability in walking patterns, it is important to design robot controllers that can adapt to the unique characteristics of each individual. This paper investigates the potential of human-in-the-loop optimisation (HILO) to deliver personalised assistance in gait training. The Covariance Matrix Adaptation Evolution Strategy (CMA-ES) was employed to continuously optimise an assist-as-needed controller of a lower-limb exoskeleton. Six healthy individuals participated over a two-day experiment. Our results suggest that while the CMA-ES appears to converge to a unique set of stiffnesses for each individual, no measurable impact on the subjects' performance was observed during the validation trials. These findings highlight the impact of human-robot co-adaptation and human behaviour variability, whose effect may be greater than potential benefits of personalising rule-based assistive controllers. Our work contributes to understanding the limitations of current personalisation approaches in exoskeleton-assisted gait rehabilitation and identifies key challenges for effective implementation of human-in-the-loop optimisation in this domain.

Figures (6)

• Figure 1: HILO pipeline using the CMA-ES to personalise the open parameters of an impedance controller to provide assistance as needed.
• Figure 2: Illustration of the reference kinematic path, ${\bf Q}_{\text{ref}}$, surrounded by a dead band and the mapping of the kinematic configuration of the model, ${\bf q}_{\text{act}}$, to the reference point, ${\bf q}_{\text{ref}}$, on the reference path.
• Figure 3: (a) Healthy participant walking on a self-paced treadmill with real-time visual feedback and assistance from the exoskeleton, Exo-H3. (b) Experimental protocol for HILO following a continuous optimisation protocol over multiple days.
• Figure 4: Adaptation of covariance matrix and the CMA-ES generation mean, indicate from G1-G11, for all generations and all subjects. Yellow arrows show the CMA-ES step size and direction, and map out the progression of the generation mean.
• Figure 5: Results from validation trials with the baseline stiffness, ${\bf K}_{base}$, the optimised stiffness, ${\bf K}_{opt}$, and the last mean stiffness, ${\bf K}_{G11}$, for day 1 and day 2 separately. Error bars show the standard deviation of cost (*P$<$0.05, **P$<$0.01).