Human-Robot Interaction Based on Biosignals

This paper introduces a novel manner for human-robot interaction (HRI) based on electroencephalogram (EEG) and surface electromyography (sEMG) signals. The P300-based BCI system aims to provide a target point for the mobile robot and indicates whether or not the robot will perform the grab task. Considering the relative position of the robot and the target cube will lead to the failure of the grasping task, we use different gestures which evoke corresponding sEMG signals to fine-tune the position of the robot. To get the target point closely related to the target character from the BCI system, we use linear discriminant analysis (LDA) to do the P300 detection task. And the average charater recognition for all subjects offline can reach 92%. Similarly, a simple convolutional neural network (CNN) is constructed for gestures classification and obtained 98% accuracy offline. In addition, the validation of the proposed method is verfied by the experimental results on mobile robot and Gaussian mechanical arm.