A Hierarchical Intelligent Rehabilitation Robotic System Based on MI-EEG

In 2022, global stroke cases reached 101 million, posing a significant health threat. Traditional rehabilitation methods are primarily passive, while active recovery therapy leveraging brain plasticity proves more effective. Brain-computer interfaces(BCIs) establish a pathway between individuals and external devices. To address this, we have developed a robotic system that combines a BCI system with active rehabilitation. This system consists of two components: signals acquisition and classification, and robot control. For electroencephalography (EEG) signals, we use the Filter Bank Common Spatial Pattern(FBCSP) algorithm to extract spatial-frequency features and employ machine learning methods for classification. Additionally, we enhance classification effectiveness using EEGNet. The robotic system, divided into navigation, vision, and robotic arm modules, simulates patient-robot interaction through motor imagery(MI) tasks. Tests on five participants demonstrate EEGNet’s superior classification accuracy (81.75% vs. 75.45%), with an overall task completion accuracy of 70.91%. The results show that the robotic system has a high accuracy in classifying MI-EEG, which is expected to create an intelligent rehabilitation system for future stroke patients.