 |
Submitted by , posted on 05 September 2002
|
 |
Image Description, by
This little demo is inspired by the famous tutorial of Mat Buckland about neural networks and genetic algorithms.
I programmed my personal evolution of his 'neural ants' sample.
Each ant is composed of a neural network and two virtual motors (1 for left legs, and 1 for right legs).
Inputs reveived by the NN are :
current left motor power
current right motor power
current global power (speed up both left and right motors )
current relative position to reach (x,z)
outputs are :
current left motor power
current right motor power
current global power (speed up both left and right motors )
These ants are just learning to go toward something on the map :
For the moment, the switching between position of nearest food and nearest base is done
by manual programmation. I tried a version of NN that 'decide' to search food, and then to carry it to green base
but convergence toward an efficient behaviour was really too slow and hazardous for this demo.
Download of this demo is avalable at http://klaudius.free.fr
Excellent tutorial of mat is here : http://www.gameai.com/buckland.html
Graphics are created by Bozo : http://www.bozo.free.fr
Klaudius
|
|