🚀 go-pugleaf

Randall Beer has written a wonderful book entitled "Intelligence as
Adaptive Behavior: An Experiment in Computational Neuroethology".

In it he describes a simulated cockroach he developed. He makes the
point that most scientific and technological endeavors attempt to
study and understand simple systems before attempting more complex
ones. His goal is to develop an autonomous, reactive, adaptive agent
that is capable of surviving and coping with an open environment.

There is currently no reason to believe that a representation-based
system will provide the kind of general intelligence needed to accom-
plish this task. Dennet and Brooks have both argued for this approach.

Apparently, there is an enormous amount of biological data on the 
neural bases of cockroach locomotion (Pearson is sited). Beer uses 
model neurons to develop a nervous system that simulates locomotion
in his artificial insect, then performs lesion studies which reproduce
findings of lesion studies using biological insects. On top of his
locomotion controller, he adds neural controllers to enable wandering,
edge-following and feeding. The neuron specifications are at a lower 
level than connectionist systems (obviously) with parameters like the
membrane conductance and capacitance, firing frequency, and intrinsic
currents (in nanoamps) specified. The book is so well written that the
simulation was reproduced and is available on the net. The code consists
of 15000 lines of Turbo C 2.0 source code (Beer did his in Lisp). It is 
called Nerves.zip. I think there is probably a second version that fixes
some bugs (pardon the pun) - maybe Nerves2.zip? For the code contact
Pat and Greg Williams, Rt. 1, Box 302, Gravel Switch, KY 40328. To order
the book call Academic Press at 1-800-321-5068, ISBN# 0-12-084730-2. 

I have added neural plasticity to the model in order for the insect to
be able to learn. Of course, the more advanced the behavior, the more
assumptions must be made about the neural mechanisms involved. Over the
next few weeks I am going to try and develop a model neuron in hardware.
If I can do this, I will try and reproduce Beer's work in hardware - at
least the locomotion controller.


At the A-Life III conference in Santa Fe, I was incredibly impressed with
Mark Tilden's robots. I think this paradigm is the logical one to follow.

-Scott 

Scott Dodson
Memphis State University
Memphis Speech & Hearing Center
Memphis, TN  38105