A brain–computer interface system for lower-limb exoskeletons based on motor imagery and stacked ensemble approach.

This article focuses on the development and evaluation of an online brain–computer interface (BCI) system for controlling a lower limb exoskeleton (LLE) based on motor imagery and a novel stacked ensemble classification algorithm called weighted random forests-support vector machines (WRF-SVM). The system integrates EEG signal acquisition, preprocessing, multi-domain feature extraction, and classification to decode users’ motion intentions, enabling closed-loop control with real-time visual and proprioceptive feedback during rehabilitation. Tested on eight healthy subjects, the WRF-SVM algorithm demonstrated superior classification accuracy compared to other methods in both offline and online experiments, achieving average accuracies above 77% offline and 80% online. The study validates the feasibility of the proposed BCI system for active rehabilitation with LLEs and suggests future integration of multimodal physiological signals to enhance control performance.