Neural Network-Based Gait Stability Enhancement for Lower-Limb Robots in Rehabilitation

Abstract

Intelligent lower-limb exoskeletons are gradually supplanting traditional rehabilitation equipment and have become a prominent research focus. However, existing systems still face gait compliance challenges due to various disturbances during actual walking. To address this, we propose a novel RBF neural network-based control method. The network approximates the uncertainties induced by disturbances in the patient's lower-limb musculature, enabling adaptive parameter tuning, and we concurrently introduce a new dynamic stability assessment approach. By analysed the principles and inherent limitations of the conventional zero-moment point stability criterion during the late single-leg support phase, the proposed strategy enhances both the stability of gait motion and the smoothness of the trajectory, thereby satisfying the real-time demands of rehabilitation training. Theoretical analysis and comparative experiments confirm the stability and reliability of the exoskeleton's walking posture, demonstrating that the method meets the control requirements for rehabilitation robots while ensuring patient safety.