Clustering of Human Motion Trajectory for Lower Limb Rehabilitation Robot Design Based on Machine Learning

Abstract

Lower limb rehabilitation robots, which usually produce repeated rehabilitative motion, not only simulate general human walking to help patients practice, but also do benefits to the remodel central nervous system to learn and store correct motion model. However, patients with different body parameters usually have different lower limb motion trajectories, and sometimes even the same person’s multiple motion trajectories could differ, thus the task of designing a specific lower limb rehabilitation mechanism for the realization of every motion trajectory is not practical. In this paper, we propose an approach to the clustering of motion trajectories of human lower limb to obtain a limited number of rehabilitation task motion types. Firstly, Gaussian distribution is adopted for the fitting of multiple trajectories of the same person. Through comparison of various clustering algorithms according to separation and compactness, Hierarchical clustering algorithm is chosen to obtain the partitions of the clusters. Finally, the Gaussian process regression (GPR) model of each cluster is established. Results show that clusters generated by this approach can reflect motion trajectory of the subjects and predict human lower limb motion pattern. With a limited number of lower-limb motion patterns, the design task of rehabilitation robots could be greatly simplified.