Model of Gait Control in Parkinson's Disease and Prediction of Robotic Assistance

Gait variability of healthy adults exhibits Long-Range Autocorrelations (LRA), meaning that the stride interval at any time statistically depends on previous gait cycles; and this dependency spans over several hundreds of strides. Previous works have shown that this property is altered in patients with Parkinson's disease, such that their gait pattern corresponds to a more random process. Here, we adapted a model of gait control to interpret the reduction in LRA that characterized patients in a computational framework. Gait regulation was modeled as a Linear-Quadratic-Gaussian control problem where the objective was to maintain a fixed velocity through the coordinated regulation of stride duration and length. This objective offers a degree of redundancy in the way the controller can maintain a given velocity, resulting in the emergence of LRA. In this framework, the model suggested that patients exploited less the task redundancy, likely to compensate for an increased stride-to-stride variability. Furthermore, we used this model to predict the potential benefit of an active orthosis on the gait pattern of patients. The orthosis was embedded in the model as a low-pass filter on the series of stride parameters. We show in simulations that, with a suitable level of assistance, the orthosis could help patients recovering a gait pattern with LRA comparable to that of healthy controls. Assuming that the presence of LRA in a stride series is a marker of healthy gait control, our study provides a rationale for developing gait assistance technology to reduce the fall risk associated with Parkinson's disease.