|
Project: TED*
A TED (= Two-degrees-of-freedom Experimental Device)is a simple sensor-driven walking machine: its walking modes can be influenced by input to its left and right light sensors (e.g. flashlight).
The purpose of a TED is threefold: 1.) It serves as a device to study the effect
of different leg designs on the walking speed of this simple machine or its
ability to overcome obstacles in its way. It is intended to be used also to
study "hand-made" evolution of systems morphologies much in the same sense as
BEAM robotics.
2.) It demonstrates how the discrete-time dynamics of a two-neuron network**,
operating near a Sacker-Neimarck bifurcation, can serve as an effective
neurocontroller for the two motors of the TED***. It also demonstrates how
sensor signals (here coming from the left and right photoresistors) can change
the behavior of the machine.
|
The network with two parameters.
|
|
Output (motor) signals of the network
|
3.) It demonstrates that a (small) neural network can be easily downloaded on a
low cost processor chip, so that the machine can move and react autonomously (as
far as batteries are operating). It also has the educational aspect, that one
can learn the dynamical effects of binary implementation (here 16 bits) of a
differential function like the sigmoid used here.
|
|
This tiny chip is TED's brain. It is as small as a fingernail.
|
The neural network can easily be converted and downloaded to TED via a
simple tool.
|
Contact:
Frank Pasemann
and
Manfred Hild
Literature
[1] D. Hrynkiw, M.W. Tilden: Junkbots, Bugbots and Bots on Wheels, McGraw-Hill, New York, 2002.
[2] F. Pasemann, Complex dynamics and the structure of small neural networks, Network, Computat. Neural Syst., 13, (2002).
[4] F. Pasemann, M. Hild, K. Zahedi, SO(2)-networks as neural networks, Proceedings IWANN’03, LNCS 2686, Springer, 2003
* Work done at the former Intelligent Dynamics (INDY) group at the
Fraunhofer Institute AIS (now FhI-IAIS).
|
|