View articles by date | ON SCIENCE Home Page | View articles by topic

Rats Control a Robot by Thought Alone

Among the most devastating of disabilities are spinal injuries that sever communication between the brain and the limbs. The nerve cells in the brain devoted to controlling limb movement are still there, but unable to communicate with the limb muscles. It takes a long time for the brain to adjust to this harsh reality, and in the meantime the injured person often experiences "phantom" limbs -- the cells in the brain have not yet ccome to terms with the fact that the muscles are not going to respond. Recent amputees suffer the same disconcerting period of adjustment.

Many researchers have tried to find a way to put these motor nerve cells to work, reestablishing communication between them and disconnected or artificial limbs. To do this, scientists have been trying to learn how to speak the language of the limbs -- how to unscramble and decode the electrical signals sent by the brain nerve cells to the muscles. If they knew what signal to send, then perhaps they could produce commands that a disconnected or prosthetic limb would understand and obey. This has proven very difficult to do, however. To intercept the signals that control limbs, researchers inserted a minute electrode into the brain, so that the electrode's tip touched a single cell in the motor cortex, the region of the brain that controls movement. Recording the output of single nerve cells in this way, they searched for a simple electrical signal that said "move," but could not find it.

Success, when it came, required a scientist able to step back from these largely unsuccessful experiments and ask the question a different way. John Chapin, of the MCP Hahnemann School of Medicine in Philadelphia, led a team of researchers that did just that. In work reported two weeks ago, Dr. Chapin learned how rats speak to their muscles so well that Chapin was able to train his experimental rats to move things around by the power of thought alone!

What Chapin did differently was to recognize that it is not necessary to understand what the cells are saying, so long as you can reproduce it. In the same way, a parrot can learn to copy what you say quite recognizably, without knowing what the words mean. Chapin set out to mimic, like a parrot, the message that a rat sends to its limbs when it moves.

When you look at the problem from this perspective, it quickly becomes obvious why earlier approaches hadn't worked. The earlier researchers recorded signals from single cells, while in a real rat many cells contribute to the complex signal the limb receives, both cells in the motor cortex and cells in the thalamus region which processes the signals. Recording the signal from only one cell was like a parrot mimicing only one vocal frequency. To record a more realistic signal, Chapin implanted an array of electrodes that recorded the signals from each of 32 nerve cells of the motor cortex and thalamus at the same time.

In order to have any hope of sorting out what was going on in the complex signal such a 32-electrode array generates, it was necessary to analyze the signal associated with a single, well-defined limb movement. Chapin put his rats in a cage where to get a drink of water a rat had to press a lever. When a rat did so, a robot arm would carry water from a dispenser too high for the rat to reach to a place where the rat could drink. Whenever the rats became thirsty, they learned to go push the lever to get a drink. All the while, Chapin was monitoring their brain waves with his 32-electrode array.

Looking at the complex data the rats were putting out, Chapin noticed a curious thing -- the nerve cells of the motor cortex and thalamus began to fire off signals long before the rat went over to push the lever. It was as if the rat was thinking about going to get a drink before actually doing it!

Focusing on that interval, Chapin set out to decode the signal being broadcast by the electrode array. Using a statistical trick called principle component analysis to separate signal from noise, and using an artificial neural network to model the effectiveness of candidate bits of the signal, Chapin finally arrived at a cleaned-up signal that seemed to reflect what was going on in the mind of the rat as it contemplated pushing the lever.

Now comes the fun part. Chapin rewired the robot arm so that it would respond directly to the cleaned-up electrical signal coming out of the 32-electrode array -- and disconnected the arm from the lever. The rats soon stopped pushing the lever. All they had to do when they were thirsty was think about pressing it, and the robot arm would obediently present water to them. In a very real sense, the rats were controlling the robot arm by thought alone.

Chapin's results hint strongly that in future work a way may be found to aid the human brain in its attempts to communicate with and move disconnected or artificial limbs. There is of course a long way to go, but for the first time the road to progress looks open.

©Txtwriter Inc.





Txtwriter Home Page | BACKGROUNDERS | ON SCIENCE | Text Books | Book Reviews
This website is the copyright ©1998-2013 of Txtwriter Inc. All rights reserved.
Google