#Brain-Machine Interface Learns to Control Robot Arm Based on User Error Brain signals Brain-machine interfaces (BMIS) restore or replace motor or sensory function in individuals who are disabled by neuromuscular disorders, stroke, or spinal cord injury. Electrical signals, acquired through either invasive or noninvasive neural interfaces, are decoded to subsequently control external devices. However, patients must spend a significant amount of time training their brain to successfully control such neuroprosthetic devices. Jose Millán group at École Polytechnique Fédérale de Lausanne in Switzerland have published in Nature Scientific Reports a BMI paradigm that enables neuroprosthetic devices to learn on the fly thus decreasing patient training time. The technology is based on the error-related potential (Errp) measured non-invasively using electroencephalographic (EEG) electrode arrays. The Errp is elicited when an error is made, or when the action does not match the user expectation. The user error signal is integrated into the neuroprosthetic controller, enabling the neuroprosthetic device to learn incorrect movements and modify its behavior. Twelve subjects trained the Errp decoder to detect their individual Errp signals by observing 350 robot movements where 20%of the movements were incorrect (robot arm moved away from the target. The average time to train the Errp decoder was merely 25 minutes. Then, in the most complicated of three experiments, the subject controlled a real robot arm to move in 2d space (right, left, up and down) to reach a target location. At first, the robot moved randomly, but subsequently adapted its movements based on the detection of the subject Errp signals
Overtext Web Module V3.0 Alpha
Copyright Semantic-Knowledge, 1994-2011