A Lightweight Real-Time EEG-Based Brain–Computer Interface for Imagined Speech-Controlled Robotics via Machine Learning

This study aims to develop a lightweight, real-time brain–computer interface system for classifying imagined speech commands using electroencephalography (EEG) signals acquired from a 14-channel Emotiv headset. Four motor-related imagined speech commands—push, pull, left, and right—are recorded from 15 participants, with a neutral state serving as a baseline. Continuous wavelet transform (CWT) is applied to extract time–frequency features from the EEG signals, which are converted into scalograms representations and used as inputs for machine learning models. Among the evaluated classifiers, XGBoost achieves the highest classification accuracy of 98.10%. The system is deployed on a Raspberry Pi-controlled robotic platform, enabling real-time inference within 900 ms. The results demonstrate efficiency, low computational overhead, and highlight the potential of consumer-grade EEG devices for real-time assistive robotics applications.