BCI Course Module — MATLAB Implementation

Overview

This module is developed as part of an Advanced BCI Course at UMBC. Students learn to classify brain signals using EEG data and control a virtual robot using motor imagery — implemented entirely in MATLAB.

Tracks

• 400 Track — Foundations: Load, visualize, filter EEG, train SVM
• 600 Track — Advanced: CSP features, EEGNet, CNN classifiers

Dataset

PhysioNet EEG Motor Movement/Imagery Dataset

• 10 subjects (S001–S010)
• 3 runs each (R03, R07, R11)
• 64 channels, 160 Hz sampling rate
• Classes: T0 = Rest, T1 = Left fist, T2 = Right fist

https://physionet.org/content/eegmmidb/1.0.0/

Tools

• MATLAB R2023b
• EEGLAB v2026.0.0
• Statistics and Machine Learning Toolbox
• Signal Processing Toolbox

Live Scripts

• 01_load_and_visualize.mlx — Load and plot raw EEG signals
• 02_epoch_extraction.mlx — Extract and combine epochs across subjects
• 03_svm_classifier.mlx — Train SVM and generate robot commands

Results

• SVM Accuracy: 51.8%
• Confidence threshold command distribution:
  • LEFT: 32.3%
  • RIGHT: 29.3%
  • REST: 38.3%

Robot Control (Coming Soon)

• LEFT prediction → raise robot left arm
• RIGHT prediction → raise robot right arm
• REST prediction → robot returns to neutral

GitHub

• Python version: https://github.com/SatwikaKandula/BCI-Course-Module
• MATLAB version: https://github.com/SatwikaKandula/BCI-Course-Module-MATLAB

Progress

• Project setup
• Load and visualize EEG data
• Band-pass filtering (8-30 Hz)
• Epoch extraction (T0, T1, T2)
• Combine 10 subjects x 3 runs = 900 epochs
• SVM classifier — 51.8% accuracy
• Confidence threshold approach
• 2D MATLAB robot animation using classifier output
• VRsink virtual robot integration [If toolbox support is available]