Researchers from the Cybernetic Intelligence Research Group at the University of Reading have developed a robot whose movements are controlled by neurons growing in a culture dish.
The robot's "brain" consists of several hundred thousand neurons isolated from embryonic rat neocortex. The cortical tissue was first dissected out, then treated with enzymes which caused the cells to dissociate from one another. The resulting cell suspension was then added to a culture dish containing nutrients.
Rather than plating the cells onto a standard culture dish, the researchers instead grew them on one with a multi-electrode array embedded in its base. In the culture dish, the cells began to extend processes, then spontaneously formed synaptic connections and started to signal to one another. The dish contains 60 electrodes in its base and allows for bidirectional communication: the electrodes can not only detect the electrical signals produced by the neurons, but also generate their own signals and send them to the cells.
In this system, the neuronal culture "learns" how to control the robot. The spontaneous activity of the network directs the movements of the robot, which in turn feeds information about its environment back to the cells. This learning process was aided by the researchers, who added various chemicals to the cell culture to strengthen some connections and weaken others. By observing the behaviour of the cultured neurons, the researchers hope to gain insights into the cellular mechanisms of memory formation.
Multi-electrode arrays have been in use for several years now. In 2006, a team from the Neuroengineering Lab at Georgia Tech reported using electrode arrays to control the movements of virtual animals, and last year, Israeli researchers used a similar method to show that cultured neurons can store information. However, this is the first time that multi-electrode arrays have been used without a computer; in this case, the information is relayed between the neuronal culture and the robot by means of short-wave Bluetooth radio.
There's more information about the research in the film clip below; see also the Combining Cognits blog, by Paul Baxter, who is a member of the Cybernetics Intelligence Research Group.
That has got to be the best intersection of mad science and actual science I've seen in a while. Awesome!
When I hear the word 'culture' I reach for my vaccination gun....
I am disappointed to see Kevin Warwick again overstating things, but am especially bothered when it is about things we are also doing in my lab. He said there's no computer in the loop which is clearly not true, and if you listen to Ben Whalley at the end of the interview, he even says the neural recording "...goes through fairly complex processing steps..." before it controls the robot. It has to. Robots and neural cultures don't speak the same language and something has to do the translation. The difficulty with this type of work (and the fun, sometimes) is that we can only make educated guesses at the neural "language" since we don't understand it yet.
I see nothing new here beyond what we and others (e.g. Sergio Martinoia and Suguru Kudoh) have been doing for the past 5 years. Believe what you read in peer reviewed papers.
Steve Potter, PhD
Laboratory for Neuroengineering
Georgia Inst. of Technology
Can this 'brain' experience pain, distress, boredom... Then it's not ethical what you're doing.
"This learning process was aided by the researchers, who added various chemicals to the cell culture to strengthen some connections and weaken others."
So the neurons aren't learning on their own then?
@Clare D: Well, the chemicals added to the cultures were neurotransmitters, so they were topping up levels of substances which are already synthesized by the cells.
Thanks, Mo. When I read that, it sounded to me like they were being manipulated and not learning on their own - as if certain connections were selected to be strengthened. I guess that's not the case then.