Scientists Make Major Improvements in “Mind-Controlled” Robotics without Brain Implants

It has been said that the mind can be capable of anything. As new research now shows us, with the right tools, this just might be true. Using a noninvasive brain-computer interface (BCI), researchers at Carnegie Mellon, in collaboration with a team from the University of Minnesota, have recently developed the first mind-controlled robotic arm that has the ability to continuously track and follow a cursor without the need for brain implants.

A research subject tests out the noninvasive BCI framework. Carnegie Mellon

BCIs developed in the past have been able to control robotic devices in completing a number of tasks, but up until now this technology has only been possible through the use of invasive brain implants. Scientists and researchers in this field have been working toward developing noninvasive BCIs to help paralyzed patients control their environments and use their “thoughts” to control robotic limbs. Prior to this most recent advancement, BCIs using noninvasive external sensing rather than brain implants tended to receive lower-quality signals, leading to less overall control and precision.

Bin He, trustee professor and head of the Biomedical Engineering Department at Carnegie Mellon, and former faculty member in the Department of Biomedical Engineering at the University of Minnesota, explained to Twin Cities Geek: “The signals are ‘dirtier’ when detected from the outside of the scalp as compared to recordings made by probes placed within the brain because of the physical properties of our body tissue—brain and skull in this case. So any attempt to decode intention signals from a noninvasive recording using whatever approaches will face the challenge to handle the ‘dirtier’ signals.”

In an effort to combat this and allow for significantly greater control over the robotic arm, Dr. He and his team of researchers have pursued novel sensing and machine learning techniques to access signals within the brain. “We used several new techniques for sensing and machine learning to allow us to sense the extremely weak ‘intention’ signals,” he said. “For example, one technique is to extract electrical signals of activated neurons at region of interest from noninvasively recorded brainwaves.” This allows the team to overcome the noisy electroencephalogram (EEG) signals prevalent in previous attempts, allowing for much more perspicuous and direct neural decoding and enabling greater real-time continuous robotic control

Using these methods, Dr. He and his team have shown human subjects can “mind-control” a robotic arm to smoothly and continually follow a cursor on a computer screen, whereas previous attempts to do this have been jerky and sporadic. According to the press release from Carnegie Mellon, the team has “established a new framework that addresses and improves upon the ‘brain’ and ‘computer’ components of BCI by increasing user engagement and training, as well as spatial resolution of noninvasive neural data through EEG source imaging.” The releases notes the results of this work have “enhanced BCI learning by nearly 60% for traditional center-out tasks . . . [and] also enhanced continuous tracking of a computer cursor by over 500%.”

As of this July 2019, this has been successfully tested on 68 nondisabled participants. However, there is still work to be done. “A natural next step is to test in patients with motor dysfunctions or paralysis to control assistive robotic devices,” explained Dr. He. “A future goal would be to develop mind-controlled prosthetic limbs.”

The researchers’ paper, “Noninvasive neuroimaging enhances continuous neural tracking for robotic device control,” is published in Science Robotics.