Assisting walking balance using a bio-inspired exoskeleton controller

Abstract

Background: Balance control is important for mobility, yet exoskeleton research has mainly focused on improving metabolic energy efficiency. Here we present a biomimetic exoskeleton controller that supports walking balance and reduces muscle activity.

Methods: Humans restore balance after a perturbation by adjusting activity of the muscles actuating the ankle in proportion to deviations from steady-state center of mass kinematics. We designed a controller that mimics the neural control of steady-state walking and the balance recovery responses to perturbations. This controller uses both feedback from ankle kinematics in accordance with an existing model and feedback from the center of mass velocity. Control parameters were estimated by fitting the experimental relation between kinematics and ankle moments observed in humans that were walking while being perturbed by push and pull perturbations. This identified model was implemented on a bilateral ankle exoskeleton.

Results: Across twelve subjects, exoskeleton support reduced calf muscle activity in steady-state walking by 19% with respect to a minimal impedance controller (p < 0.001). Proportional feedback of the center of mass velocity improved balance support after perturbation. Muscle activity is reduced in response to push and pull perturbations by 10% (p = 0.006) and 16% (p < 0.001) and center of mass deviations by 9% (p = 0.026) and 18% (p = 0.002) with respect to the same controller without center of mass feedback.

Conclusion: Our control approach implemented on bilateral ankle exoskeletons can thus effectively support steady-state walking and balance control and therefore has the potential to improve mobility in balance-impaired individuals.

Keywords: Assist balance control; Biomimetic control; Exoskeleton; Musculoskeletal modelling.

Fig. 1

Control parameter identification. We estimated the control parameters of a model without (blue) and with (green) additional feedback of whole body center of mass (COM) velocity by minimizing the difference between the experimental ankle moment and the ankle moment simulated with the neuromechanical model (representative example in A). Control parameters of each model were estimated by tracking eight steady-state and 16 perturbed gait cycles in one optimization problem. Perturbations were applied at toe-off of the contralateral leg while the subjects walked at 0.62 m/s. The root mean square error (RMSE) between experimental and simulated ankle moments was smaller in the model with additional COM feedback for (B) steady-state walking, (C) perturbed walking, and (D) a validation perturbation trial that was not used in the parameter estimation (the dots represent the RMSE in each of the five subjects and the bar represents the average across subjects). The lower RMSE for the model with COM feedback compared to the default model reflects the simulated change in ankle moment in response to pelvis push and pull perturbations (F) that is in agreement with experimental observations (G), whereas the default reflex model without COM feedback cannot capture the experimental data (E). (E–G contains data of one representative subject with the two-trial average response of each unique perturbation)

Fig. 2

Perturbed walking with ankle–foot exoskeleton. Twelve subjects walked with a bilateral ankle–foot exoskeleton in minimal-impedance mode and controlled with the neuromechanical model with and without additional COM velocity feedback. A External forces were applied at the pelvis after right heel strike in forward (push) or backward (pull) direction to perturb human walking. B, C The neuromuscular controller uses the encoder on the ankle joint, ground reaction forces and deviation in COM velocity from the reference trajectory as input and ankle joint moment as output. D The exoskeleton delivered 30% of the ankle joint moment computed with the neuromuscular controller. E Surface Electromyography was used to quantify muscle activity to evaluate controller performance. This figure contains data of one backward directed perturbation of a typical subject walking with the neuromuscular controller with additional feedback of COM velocity

Fig. 3

Effect of exoskeleton assistance on muscle activity in steady-state walking. We observed a gradual decrease in soleus muscle activity during the 20 min adaptation. Compared to minimal impedance mode (gray), average soleus activity was 19% lower at the end of the adaptation period (A). The soleus activity was decreased during the full duration of the stance phase (B), which is the result of the plantarflexion assistance provided by the exoskeleton (C). The desired exoskeleton moment was applied with a RMSE of 2.02 Nm with mainly larger differences between desired and actual moment during the first part of the stance phase. A paired t-test was used to compare muscle activity between controllers. A contains data of all subjects with the dots representing the median muscle activity during 3 min of walking and the bars the averaged data across subjects. The time series in B and C are based on data of one representative subject

Fig. 4

Exoskeleton assistance in perturbed walking. A The exoskeleton moment increased in response to pelvis push perturbations in the neuromuscular controller with COM feedback compared to the default neuromuscular controller and compared to steady-state walking assistance. B The type of controller had a significant influence on the average exoskeleton moment during the perturbed right stance phase after push perturbations [F(2,22) = 29.5001, p < 0.001]. C The assistive ankle moment decreased in response to pelvis pull perturbations in the neuromuscular controller with COM feedback compared to the default neuromuscular controller and compared to steady-state walking. D The type of controller had a significant influence on the average exoskeleton moment during the perturbed right stance phase after push perturbations [F(2,20) = 81.1381, p < 0.001]. A and C contain data of one representative subject. The bar plots in B and D contain data of all subjects with the dots representing the response of individual subjects and the bars the averages across subjects. A repeated measures anova with Tukey’s HSD post-hoc test was used to compare the exoskeleton moment between controllers

Fig. 5

Effect of perturbation force and exoskeleton controller on pelvis displacement. The controller type influenced the movement of the pelvis after push perturbations [F(2,22) = 4.1593, p = 0.04)]. Forward displacement of the pelvis during the perturbed stance phase was smaller in the controller with COM feedback compared to the controller without COM feedback (p = 0.026) (A, B). The controller type also influenced the backward pelvis displacement after pull perturbations [F(2,20) = 21.92, p < 0.001]. Backward pelvis displacement at the first heel strike after the perturbation was smaller in the controllers with COM feedback compared to the controller without COM feedback (p = 0.002) (C, D). A and C contain data of one representative subject. The bar plots in B and D contain data of all subjects with the dots representing the response of individual subjects and the bars the averages across subjects. A repeated measures anova with Tukey’s HSD post-hoc test was used to compare the pelvis displacement between controllers

Fig. 6

Effect of perturbation force and exoskeleton controller on muscle activity. The exoskeleton controller type influenced the increase in soleus activity after push perturbations [F(2,22) = 5.2763, p = 0.013]. Soleus activity in the first 500 ms after perturbation onset was smaller for the neuromuscular controller with COM feedback compared to the default neuromuscular controller (p = 0.006) and compared to the minimal impedance controller (although not significant, p = 0.057) (A, B). The exoskeleton controller type also influenced the increase in tibialis anterior activity after pull perturbations [F(2,18) = 15.1174, p = 0.0001]. Tibialis anterior activity was smaller in the neuromuscular controller with COM feedback compared to the default neuromuscular controller (p < 0.001) and compared to the minimal impedance controller (p = 0.036) (C, D). A and C contain data of one representative subject. The bar plots in B and D contain data of all subjects with the dots representing the response of individual subjects and the bars the averages across subjects. A repeated measures anova with Tukey’s HSD post-hoc test was used to compare the pelvis displacement between controllers