The Brain as User Interface

By Samuel K. Moore, Associate Editor

Scientists hijack a rat’s brain to robotize the rodent and train a monkey’s brain to move a cursor

ROBOTICS • Recent experiments have shown that direct control of prosthetic limbs by the brain may be less difficult to achieve than was supposed. In early May, neuroscientist John Chapin and his colleagues at a Brooklyn, N.Y., medical center used a wireless receiver and electrodes implanted in a rat’s brain to steer the rodent anywhere they wanted it to go. About a month later, bioengineer Andrew Schwartz of Arizona State University (Tempe) announced a method to quickly train a small number of cells in a monkey’s brain to accurately control the 3-D movements of a dot on a monitor.

Scientists have known for decades that nerve cells, or neurons, in the area in the brain called the motor cortex, become active just before the limb they control moves. This observation sparked dreams of implanting electrodes in paralyzed people, extracting these signals, and funneling them into controls for artificial limbs or other devices.

Schwartz based his approach on a breakthrough made by Chapin, Miguel Nicolelis, and others. They had shown that a robotic arm could be steered with signals from electrodes implanted in the part of a monkey’s brain that controls its own arm [see "A mind-Internet-machine connection," IEEE Spectrum, January 2001, p. 33].

In Schwartz’s experiments, the signal controlled the 3-D motion of a dot on a display. In contrast to earlier tests, the monkey was permitted to see how well the brain signal controlled the dot, and could try to adjust for errors. The adaptations that then arose in the monkey’s brain controlled the dot better. At the same time, the algorithm used to transform the neural signals into what showed up on the display was also altered so as to match the monkey’s adapting signal more closely.

In this coadaptive system, as Schwartz calls it, a very small number of brain cells can control the movement. "A lot of the decoding computation is actually being done by the neurons," he says. "We’re only sampling 44 neurons–normally when you move, there are millions of neurons coding for it." And if fewer cells are involved in controlling a motion, it should be easier to use their signal to control future prosthetics. Schwartz’s work, which appeared in the 7 June issue of Science, is similar to a study published in March in Nature by a group led by John Donoghue, a professor at Brown University (Providence, R.I.).

All the same, even with better understanding of how to extract and interpret brain signals, a key component of brain-controlled prosthetics is still missing–conversion of the signals from a prosthetic’s pressure or position sensors into something the brain can understand. Without feedback, an artificial arm may reach out and grasp a glass of water, but the patient cannot feel the glass, Chapin told IEEE Spectrum in his office at the Downstate Medical Center (New York City), part of the State University of New York (SUNY). So there’s nothing to keep the artificial hand from dropping the glass or crushing it.

As a first step, Chapin and his colleagues wanted to learn how well a rat could understand and respond to a perception–touch in this case–produced electronically within its own brain.

Remotely regulated rats
The SUNY effort’s offspring has been popularly dubbed "roborat," and at least in some respects it beats its mechanical cousins. Suppose the need is to remotely guide a small robotic system through collapsed buildings to search for survivors. In pure robotics, this is a tough job, requiring a lightweight, low-power, agile, and controllable system that can circumnavigate obstacles with ease. But remotely guided rats carrying wireless video cameras fit the bill nicely.

The brain implant developed by Chapin and his colleagues lets them instruct a trained rat to turn right, left, or move forward according to keystrokes from a laptop as far as 500 meters away. Electrodes are implanted in three areas of the brain: one in the medial forebrain bundle (MFB), which is associated with feelings of pleasure, and one each in the left and right somatosensory cortex, part of the brain that handles the sense of touch. In particular, they implanted the electrodes in the parts of the cortex that sense the rat’s whiskers.

The rats were then trained to turn right when the brain cells representing their right whiskers were stimulated, left when the left ones were, and forward when the MFB, sometimes called the pleasure center, was electrically tickled. The training worked because the rats were rewarded with an additional stimulus to the pleasure center whenever they made a correct move. Sanjiv Talwar, a member of Chapin’s research team, calls such training "virtual learning," because "everything takes place inside the rat’s head."

Last fall, the instrumented rats were put through their paces at the Southwest Research Institute (San Antonio, Texas), where the project’s funder, the Defense Advanced Research Projects Agency (DARPA, Arlington, Va.), evaluates robots. There, under the electronic guidance of SUNY student Shaohua Xu, the intrepid animals scrambled over and through crumbled blocks of concrete, in addition to climbing a tree, walking along a railroad track, and doing other things lab rats just don’t do.

Also among the first tests was seeing if the rats could be distracted from their tasks by people, loud noises, and goodies like cheese. But with enough stimulation to the pleasure center the rats stayed on the job. "These guys are having too much fun to eat anything–not even chocolate, and rats are chocoholics," says Chapin.

Rodent-area networks?
Most systems for interfacing with brain cells include a constant current stimulator for each channel. Because the stimulator is about as big as a human head, it is clearly unsuitable for a radio-controlled rat. The constant current control overcomes what were thought to be inevitable impedance changes as the electrode gradually gums up.

But Emerson Hawley, an instrumentation expert and an IEEE member in Chapin’s lab, bet that such current control devices were overkill for what their lab was trying to accomplish. It turned out that simple transistor-to-transistor logic (TTL) could drive the right amount of current as long as the electrodes were of the right thickness and the length and duration of the electrical pulses used could be adjusted.

"The electronics were remarkably simple," noted Hawley. The main components are a battery-operated 433-MHz radio and a hobbiest’s microcontroller. The receiver picks up commands as a short string of ASCII text and translates that into which electrode of the three to activate, how long the bipolar pulse of current should last, and how much time should come between each pulse. The electronics package is finished off with a small video camera to provide the rat’s-eye-view of the world, and it all fits in a rodent-sized backpack.