Controlling a drone, or any object, with just your mind may seem to the subject of science fiction. In truth, researchers have spent decades investigating brain-computer interfaces (BCIs) that connect the human mind to external technology.
Dr. Uttam Ghosh, associate professor of cybersecurity, and colleagues have developed a one-of-a-kind prototype for smooth and reliable control of a UAV by the human brain. The device uses a motor imagery signal for semi-autonomous navigation and performs well in tests for accuracy, signal detection and response time.
Their work builds on previous research using the Common Spatial Pattern or CSP method.
“CSP is a common method to classify motor imagery data, and has been improved and improvised in such a way that it can work with a low quantity of training samples and noisy data, says Dr. Ghosh.”
He adds that researchers have proposed methods like CSP, Filter Bank CSP (FBCSP), and Common Spatio-Spectral Patterns (CSSP) to overcome challenges like extracting spatial spectrum filters and a shortage of training set data. However, they have not developed a single method to overcome these and other issues affecting model performance.
Dr. Ghosh and colleagues Ms. Sricheta Parui, Mr. Deborsi Basu, and Prof. Raja Datta from the Indian Institute of Technology in Kharagpur, India, have proposed a Brain to UAV Communication Model using the Stacked Ensemble CSP algorithm based on Motor Imagery EEG signal for accurate brain-to-UAV communication.
“Our prototype solves two problems by extracting features using several modified CSPs, followed by the use of Random Forest and SVM classifiers to categorize the data,” says Dr. Ghosh. “Then, we combined each prediction from the classifier after feeding a different set of features to it.
Dr. Ghosh and his team followed a unique strategy that combines several machine learning algorithms to help classify multiple classes at once. Their approach projects various commands for the UAV, which makes the system multifunctional.
We performed several experiments to test our method and we also used the bagging technique to enhance the classifier performance,” says Dr. Ghosh.
The authors intend to continue to improve their prototype.
“One issue we hope to address is that CSP only considers one frequency band linked to the motor imagery task. Other bands may include useful information for diverse subjects,” says Dr. Ghosh. “Further research might overcome these difficulties, and an optimal CSP-based algorithm as a bagging technique base learner could be constructed. We are working on it. “
The full paper, A Brain to UAV Communication Model using Stacked Ensemble CSP algorithm based on Motor Imagery EEG signal, is available online.