|
|
Well you're here because you want the get the basic idea of what a nervous (Nv) net is. I can't pretend to be able to decribe it as well as Mark Tilden so I still suggest you read the academic version, as for simple language version here goes....
|
|
Nervous net technology is based on the idea that a brain is no good without a spinal cord. In fact a nervous net acts very much like a silicon spinal cord. Look around at the different types of life on this planet and you'll notice that about 80% of it dosn't actually have a brain.
So out of this observation comes Biomorphic Robobiology (biomorphic comes from the Latin words for "living" and "form"). Biomorphics is basically the science of building machines that are "Survivors" as opposed to "thinkers" or "workers". Once you've got a machine that can successfully fend for itself then you can consider giving it a task. Biomorphics and nervous net tech have many parallels to existing life so in a sense they are a form artificial life or Alife. Now most people confuse Alife with artificial intelligence, but we generally consider a geranium to be alive but we don't think of it as being terribly intelligent. In fact many of the BEAM devices that have been built are little more than robot geraniums. This doesn’t make them invalid, in fact they are an intrinsic, and necessary, part of a robot ecosystem.
The Nervous Net gives a machine the ability to react to its environment on a dynamic way. Instead of either calculating or bullying its way through its universe it will react to the outside environment in such a way as to anneal or soften towards the best fit solution for the terrain at hand. In a walking robot this is a big advantage.
Using simple TTL logic a Silicon spinal cord can be constructed that provides behaviors that are anything but simple. Several years of research have gone into 200 or so machines and will still haven't seen all the possible behaviors. The most fundamental nervous net is the MicroCore consisting of 6 Nv elements each equivalent to two transistors. This 12 transistor nervous net is sufficient to control and maintain a steady walking state in just about any walking machine.
In the MicroCore all effort is made to Include the load in the control system, instead of Isolating or buffering the control circuitry to avoid feedback as many processor designs do. In fact it is load feedback, or rather "implex", that gives the MicroCore its flexability. By driving a motor through some sort of analogue driver (such as a transistor H switch) the MicroCore can "sense" (I use the term loosely its actually much more subtle than this) what’s up with the motor and adjust it's performance accordingly. The "implex" effect seems to be inherent in the MicroCore and has not been successfully modeled or simulated on any computer. So as a result the only way to see if something is gonna work is to build the sucker (a methodology I strongly endorse).
One and 2 Nv devices are some of the most successful and can be parallel to the more simple single cell lifeon this planet such as bacteria and paramecium. By building lower order devices such as these and then "Evolving" them into higher order machines such as walkers and snakes we can provide a evolutionary trail to more complex and capable Nv designs.
all4nw
