🚀 go-pugleaf

In article <gate.wDJ8oB1w165w@toz.buffalo.ny.us>, 
  cyberman@toz.buffalo.ny.us (Cyberman) writes:
> 
>  Do> the next few weeks I am going to try and develop a model neuron in
>  Do> hardware. If I can do this, I will try and reproduce Beer's work...
 
> Hmmm a hardware version.  So you intend to make a cpu microcoded
> with "NEURAL NET" code?  This would be faster than making an
> interpreted one.  Or do you intend to use analog processing?
> Actually a lisp interpretor on a chip would make it easier.  This
> would provide some real dynamically alterable code.
> 

I intend to use analog processing, no cpu's, no code, just very simple
circuits that are based on Beer's model neurons that can be connected
together to control sensors and motors. This kind of architecture should
be quite scalable, perhaps more than Brooks' subsumption architecure.

You (Stephen) also mentioned porting the code to a smaller machine. I 
think you should be warned that the code is 15000 C lines long. Each model
neuron is implemented as a differential equation solved using Eulers' method.
Each one (I used 85) is updated 4 times before the physical body is
updated. Of course, alot of the program deals with the interface, graphics,
etc. but it is still rather computationally intensive.

is updated 4 times