Brain Waves Guide Walking Robot

Jan. 10, 2007 — Forget the remote control — scientists are learning how to let you control a robot with signals straight from your brain.

Eventually, the technique could lead to semi-autonomous robots able to assist disabled people or perform routine tasks in the home.

"We're using a well-known, well-characterized response that occurs in the brain to control a physical device in the world," said research leader Rajesh Rao, an associate professor of computer science and engineering at the University of Washington in Seattle.

That brain "response" is the same one triggered whenever you lose and then find your car keys, said Rao.

As you scan tables, desks and bureaus for lost keys, you're focused on a mental image of them. When you finally spot your keys, your brain's reaction is known to scientists as a P300 response ("P" is for positive and 300 for the number of milliseconds it takes your neurons to produce the reaction).

The P300 response is strong and distinctive, and therefore can be picked up by external sensors. And because it is such a well-characterized response, it can be used reliably again and again.

Rao and his team wrote software that allows a computer to recognize the P300 response, and use it to guide a robot.

The person controlling the robot is fitted with an electrode cap designed to pick up brain waves.

The person focuses on an object, for example a banana, displayed on a monitor. The computer flashes the boundaries of random objects one object at a time. When the banana flashes, the subject's P300 response triggers and the computer picks it up. After about 10 minutes of training with different objects, the computer becomes calibrated to the person's unique P300 signal.

After that initial callibration, the controller can interact with the robot.

To make the robot walk toward a table, for example, the wearer then focuses on an image of the table. Meanwhile, random images flash on the monitor. Each time the table flashes, the operator's P300 sparks. It takes 5 to 10 seconds for the computer to confirm that "table" is correct and then directs the robot to walk toward it.

The same process is used to get the robot to pick up an object and return to a specific location.

Currently, the thought commands are limited to a few basic instructions. But so far, the robot responds to those instructions with 94 percent accuracy.

But using the P300 response may not be the most efficient way of controlling a robot, said Paul Sajda, associate professor in the department of biomedical engineering at Columbia University in New York.

"The signals related to eye movements are 1,000 times stronger than scalp ECG," he said. So unless the controlling person is completely immobile, "you're better off using an eye tracker, which could be mounted on a pair of glasses."

However, Sajda does think the P300 response is an important avenue to explore, as it could lead to a better understanding of other signals, such as the "N170," triggered when a person recognizes a face.