This study aims to address the lag in real-time human motion tracking for weight-loading lower-limb exoskeletons by proposing a novel movement prediction method. The purpose is to enhance exoskeleton responsiveness through accurate prediction of lower-limb movement (LLM), enabling seamless human–robot interaction in industrial scenarios.
An adaptive temporal movement primitives (ATMPs)-based neuromorphic framework is developed, inspired by alpha motor neuron mechanisms. The method decomposes LLM into three primitive types (W-TMPs, S-TMPs and B-TMPs) and uses online adaptive algorithms (MDA-OGF) for real-time parameter tuning. A bilateral synchronization mechanism ensures robustness across locomotion modes.
Experimental validation demonstrated a prediction horizon of 148 ms with 4.25% root mean square error, outperforming the state-of-the-art methods. The algorithm showed robustness across seven locomotion modes and three transitional modes, with transient PRMSE <= 11.1% during mode switches.
This work introduces a neuroscience-inspired ATMPs framework that combines the advantages of different prediction methods, achieving a balance between prediction accuracy and prediction horizon. The method’s scalability to diverse wearable systems with high-frequency joint angle sensing.
