Down the Line: Brain-Trained Robots
Some day soon, human brains and robots may be able to interact seamlessly. Traditionally, brain-machine interface (BMI) has required invasive surgical procedures, such as electrode array implants. But, scientists in Japan have discovered a more direct way for people to communicate with robots.
Researchers at the Advanced Telecommunications Research Institute International (Kyoto, Japan) and Honda Research Institute Japan Co. (Tokyo) have developed a way to manipulate robots by using brain activity signals. The new technology has enabled the decoding of natural brain activity, and the use of the extracted data for the near real-time operation of a robot, without an invasive incision of the head and brain.
Neural decoding can allow a robot hand to mimic a human's basic finger movements by tracking the hemodynamic responses in the brain. Although there is a 7-second time lag between the subject's movement and the robot's mimicking movement, the researchers succeeded in gaining a decoding accuracy of 85 percent.
"This technology is potentially applicable to other types of noninvasive brain measurements, such as the brain's electric and magnetic fields and brain waves," says Tomohiko Kawanabe, president of the Honda Research Institute, which developed the Asimo humanoid robot several years ago. "By utilizing such methods, it is expected that the same result could be achieved with less time lag and more compact BMI system devices."
During the experiment, a person enclosed in an MRI scanner made a finger gesture mimicking the popular "rock, paper, scissors" game. The changes in the individual's hemodynamic responses associated with brain activity were monitored every second. Specific signals generating rock-paper-scissors movements were extracted and decoded by a computer program. The decoded information was transferred to a hand-shaped robot to simulate the original movement performed by the subject.
While conventional machine-interfaces are operated using button switches controlled by human hands or feet, BMI uses brain activity measured by various devices and allows noncontact control of the terminal machines. Implanted electrode arrays and brain waves have been commonly used in the past.
According to Kawanabe, conventional noninvasive BMI requires humans to undergo intensive training in order to generate detectable brain activities. Because the brain activity associated with an intention is very hard to track, the subject is typically instructed to perform a mental task with easily detectable brain activity, such as calculation. The individual must also learn to control such brain activity to express an intention.
"The new BMI technology is different in that natural brain activity associated with specific movements can be decoded without using alternative brain activity," says Kawanabe. "The experiment revealed that rock-paper-scissors movements were decoded directly from an untrained subject's real-time brain activity."