A Multi-Kernel Low-Dimension Fuzzy Modeling Method for Personalized Gait Patterns

Abstract

Personalized gait pattern construction is challenging due to the inherent nonlinearity of gait patterns and limited available samples. To address these issues, we propose a multi-kernel low-dimensional fuzzy modeling method. Our approach first employs bootstrap resampling to expand the dataset into multiple subsets, each modeled using a low-dimensional fuzzy model to capture local nonlinear dynamics. However, these sub-models may be biased due to limited samples. To overcome this limitation, a global model is then formulated by integrating these sub-models with weights inversely related to their modeling errors. Finally, an objective function is formulated for the global model, and solving it yields a gait construction model. The proposed method achieves mean RMSEs of 1.86° (hip) and 2.43° (knee) on the KIST dataset, and 1.55° (left hip) and 1.18° (left knee) in practical experiments, with maximum errors capped at under 4° and 3.5° respectively, demonstrating outstanding precision.