Real-time trajectory prediction of lower-limb exoskeleton based on motion pattern recognition and central pattern generator

Abstract

Current control strategies for rehabilitation lower limb exoskeleton robots face critical challenges in adaptive trajectory planning and reliable motion pattern recognition. To address these limitations, this study introduces a novel real-time trajectory prediction method based on multimodal motion recognition and central pattern generator (CPG) model architecture. This model retains learned motion pattern characteristics to predict joint trajectories when encountering similar scenarios, enabling autonomous trajectory adaptation across diverse working conditions. A multimodal recognition framework is developed by fusing inertial measurement unit signals with plantar pressure sensor data, achieving real-time motion pattern identification at gait cycle initiation. Validation experiments demonstrate two key outcomes. In simulations, the CPG model maintained trajectory learning errors within ±1°. During physical exoskeleton trials, stabilized joint angle prediction errors remained below ±6°. The recognition system attained 93% accuracy in motion pattern classification, effectively supporting real-time control requirements. The proposed scheme successfully balances pattern recognition accuracy with trajectory prediction precision, effectively enhancing exoskeleton robot functionality in practical rehabilitation applications.