Framework for visual-feedback training based on a modified self-organizing map to imitate complex motion

Abstract

The goal of this research was to develop a visual-feedback system, based on motion sensing and computational technologies, to help athletes and patients imitate desired motor skills. To accomplish this objective, the authors used a self-organizing map to visualize high-dimensional, time-series motion data. The cyclic motion of one expert and five non-experts was captured as they pedaled a bicycle ergometer. A self-organizing map algorithm was used to display the corresponding circular motion trajectories on a two-dimensional motor skills map. The non-experts modified their motion to make their real-time motion trajectory approach that of the expert, thereby training themselves to imitate the expert motion. The root mean square error, which represents the difference between the non-expert motion and the expert motion, was significantly reduced upon using the proposed visual-feedback system. This indicates that the non-expert subjects successfully approximated the expert motion by repeated comparison of their trajectories on the motor skills map with that of the expert. The results demonstrate that the self-organizing map algorithm provides a unique way to visualize human movement and greatly facilitates the task of imitating a desired motion. By capturing the appropriate movements for display in the visual-feedback system, the proposed framework may be adopted for sports training or clinical practice.