From: | pgrosse@********.com (Paul Grosse) |
---|---|
Subject: | Cybertechnology: From the Times science section |
Date: | Fri, 16 May 2003 07:41:41 -0500 |
Wired to the Brain of a Rat, a Robot Takes On the World
By ANNE EISENBERG
THE nerve center of a conventional robot is a microprocessor of
silicon and metal. But for a robot under development at Georgia Tech,
commands are relayed by 2,000 or so cells from a rat's brain.
A group led by a university researcher has created a part mechanical,
part biological robot that operates on the basis of the neural
activity of rat brain cells grown in a dish. The neural signals are
analyzed by a computer that looks for patterns emitted by the brain
cells and then translates those patterns into robotic movement. If the
neurons fire a certain way, for example, the robot's right wheel
rotates once.
The leader of the group, Steve M. Potter, a professor in the
Laboratory for Neuroengineering at Georgia Tech, calls his creation a
Hybrot, short for hybrid robot.
"It's very much a symbiosis," he said, "a digital computer and a
living neural network working together."
Dr. Potter has been building the system of hardware, software,
incubators and rat neurons that constitute the Hybrot since 1993, when
he was a postdoctoral student at the California Institute of
Technology. He and his group have not only introduced the neurons to
the world outside their dish; the team has also closely monitored
minute changes that take place in the shape and connections of the
neurons as they are stimulated, using techniques like time-lapse
photography and laser imaging.
Dr. Potter hopes that close observation of how brain cells behave as
they are exposed to a world of sensation will help researchers
understand the way small groups of neurons go about learning. "If the
network begins to get better at a job," he said, "we will watch what
changed within the network to allow it to do that."
Dr. Jonathan Wolpaw, laboratory chief and professor of neuroscience at
the Wadsworth Center of the New York State Department of Health and
the State University of New York at Albany, said that Dr. Potter's
research could yield a simple system for exploring the capacity of
neurons and circuits to change based on incoming activity.
"These changes could be analogues of what happens in learning," Dr.
Wolpaw said. "You are dealing with neurons, the same tissue as in a
brain," although in a different setting and with different circuitry.
"Some things presumably are in common, for example, the neuron's
capacity for plasticity," he said.
In Dr. Potter's hybrid system, the layer of rat neurons is grown over
an array of electrodes that pick up the neurons' electrical activity.
A computer analyzes the activity of the several thousand brain cells
in real time to detect spikes produced by neurons firing near an
electrode.
A silver three-wheeled model of the robot is commercially available
through the Swiss robotics maker K-Team (www.k-team.com) for about
$3,000 and is about the size of a hockey puck. It trundles along at a
top speed of one meter per second.
"We assign a direction of movement, say, a step forward, that is
automatically triggered by a pattern of spikes," said Thomas DeMarse,
a former member of Dr. Potter's group who is an assistant professor in
the department of biomedical engineering at the University of Florida.
"Twenty of these patterns, for instance, means 20 rotations of the
wheel."
As the robot moves, it functions as a sensory system, delivering
feedback to the neurons through the electrodes. For example, Mr.
DeMarse said, the robot has sensors for light and feeds electrical
signals proportional to the light back to the electrodes. "We return
information to the dish on the intensity of light as the robot gets
closer and the light gets brighter."
The researchers monitor the activity of the neurons for new signals
and new connections. Dr. Potter said that the feedback mechanism was
crucial to the functioning of the neural network. In traditional,
isolated cultured networks, he said, in which neurons are not
connected to a body, the activity patterns of the neurons are largely
pathological. "They behave in an aberrant way," he said. "It's a
symptom of sensory deprivation, because the neurons are not receiving
the input they usually get."
He decided to provide a body for the neurons early in his research,
first in computer simulation and then in reality, so that neurons
would have feedback. In that way, if the cells learned, he and his
group might observe the changes that came about in the network.
"People say learning is a change in behavior that comes from
experience," he said. "For a cultured network to learn, it must first
be able to behave."
There is an analogy to the human nervous system in the feedback loop
developed by Dr. Potter, said Nicholas Hatsopoulos, an assistant
professor in the department of organismal biology and anatomy at the
University of Chicago.
Dr. Hatsopoulos also works on brain-machine interfaces, including ways
that brain signals may one day be used to move prosthetic devices.
"Potter's device has sensors that pick up information, and then the
signals go back to the dish and stimulate the cells," he said.
Similarly, he said, "signals out of the brain control the arm, but
there are also sensors in the muscles and skin that send information
back, too."
Such feedback loops are necessary to basic research in brain-machine
interactions, he said. Researchers need not only to record signals
that drive a device but also take signals from sensors and stimulate
the nervous system. "Closing the loop will be a key issue in moving
this field to the next level, for the feedback presumably helps
learning," he said.
Miguel A. L. Nicolelis, a neuroscientist at Duke University, has
identified signals generated by a monkey's brain as it gets ready to
move, and then used the signals to move a robotic arm. "We are
discovering that when animals learn to operate a robotic device, the
operation changes the sensory and motor maps of the animal," he said.
"Steve is looking for the same thing at the cellular level."
Dr. Potter has not yet demonstrated learning in his network but said
he might be able to do so within six months. In experiments, Dr.
Potter said he hoped to observe the Hybrot following an object at a
certain distance.
"The next step is to watch it to see if it becomes better at following
this object," he said. "That would become exciting."