Authors: Ho Jin Choi, Satyajeet Das, Shaoting Peng, Ruzena Bajcsy and Nadia Figueroa
Abstract: This paper investigates the feasibility of using
EEG-based intention detection for real-time robot assistive
control, with a focus on motor intention prediction. The
proposed approach involves two pipelines: i) an offline pipeline
that collects and processes EEG data as well as motion data to
train a classifier for motion intention prediction and biological
interpretation, and ii) an online pipeline that uses the trained
classifier to predict a human’s motor intention and couples it
with a robot to perform assistive control. We adopt and modify
the state-of-the-art EEG sample covariance matrix feature
representation by using EEG signal derivatives and tangent
space projection as features for an SVM classifier that can
run in real-time. With this, Our system excels with the highest
accuracy of 86.88% in real-time settings, and it achieves an
impressive 70% accuracy in real robot experiments. We show
in a real-robot experiment that our online pipeline is able to
detect the onset of motion purely from EEG signals and trigger
a robot to perform an assistive task.
12 сен 2024
