Simulation of human gait with body weight support: benchmarking models and unloading strategies

Gait training with partial body weight support (BWS) has become an established rehabilitation technique. Besides passive unloading mechanisms such as springs or counterweights, also active systems that allow rendering constant or modulated vertical forces have been proposed. However, only pilot studies have been conducted to compare different unloading or modulation strategies, and conducting experimental studies is costly and time-consuming. Simulation models that predict the influence of unloading force on human walking may help select the most promising candidates for further evaluation. However, the reliability of simulation results depends on the chosen gait model. The purpose of this paper is two-fold: First, using human experimental data, we evaluate the accuracy of some of the most prevalent walking models in replicating human walking under the influence of Constant-Force BWS: The Simplest Walking model (SW), the Spring-Loaded Inverted Pendulum model (SLIP) and the Muscle-Reflex (MR) gait model. Second, three realizations of BWS, based on Constant-Force (CF), Counterweight (CW) and Tuned-Spring (TS) approaches, are compared to each other in terms of their influence on gait parameters.

We conducted simulations in Matlab/Simulink to model the behaviour of each gait model under all three BWS conditions. Nine simulations were undertaken in total and gait parameter response was analysed in each case. Root mean square error (mrmse) w.r.t human data was used to compare the accuracy of gait models. The metrics of interest were spatiotemporal parameters and the vertical ground reaction forces. To scrutinize the BWS strategies, loss of dynamic similarity was calculated in terms of root mean square difference in gait dynamics (Δgd) with respect to the reference gait under zero unloading. The gait dynamics were characterized by a dimensionless number Modela-w.

SLIP model showed the lowest mrmse for 6 out of 8 gait parameters and for 1 other, the mrmse value were comparable to the MR model; SW model had the highest mrmse. Out of three BWS strategies, Tuned-Spring strategies led to the lowest Δgd values.

The results of this work demonstrate the usefulness of gait models for BWS simulation and suggest the SLIP model to be more suitable for BWS simulations than the Simplest Walker and the Muscle-reflex models. Further, the Tuned-Spring approach appears to cause less distortions to the gait pattern than the more established Counterweight and Constant-Force approaches and merits experimental verification.