BCIs have exhibited much success in recent years, translating the brain’s activity into robotic manoeuvres and showing potential to assist disabled people with everyday tasks. However, measuring brainwaves for precision control has, until now, only been possible by surgically placing implants in the brain, which brings issues of cost and patient risk.
BCIs that use non-invasive external sensing receive ‘dirtier’ signals, and as such can not be used to exert the same level of control as brain implants. The Carnegie Mellon team, working in collaboration with the University of Minnesota, used novel sensing techniques combined with machine learning to improve the neural decoding of EEG (electroencephalogram) signals. These improvements facilitated the real-time continuous control of a robotic arm in two dimensions, smoothly following a cursor around a screen. The research is published in Science Robotics.
“This work represents an important step in non-invasive brain-computer interfaces, a technology that someday may become a pervasive assistive technology aiding everyone, like smartphones,” said Bin He, head of Carnegie Mellon’s Biomedical Engineering Department.
“Despite technical challenges using non-invasive signals, we are fully committed to bringing this safe and economic technology to people who can benefit from it.”
Previous attempts at non-invasive control of robotics would result in jerky movements, as if the arm was trying to catch up with the brain’s activity. With the new system, the researchers found that continuous tracking of a cursor by a robotic arm was improved by 500 per cent. The team said the technology is directly applicable to patients and it’s hoped that non-invasive neurorobotics could be available in the near future, with clinical trials already being planned.
“There have been major advances in mind-controlled robotic devices using brain implants. It’s excellent science,” said Prof He. “But non-invasive is the ultimate goal. Advances in neural decoding and the practical utility of non-invasive robotic arm control will have major implications on the eventual development of non-invasive neurorobotics.”