In order to generate a program that controls the eyebot with a neural net (MLP)
the following steps are required:

 1. make send_train
 2. make receive_train
 3. make trainer
 4. dl sendtrain.elf (to a robot with radio)
 5. start program on robot
 6. start program receive_train on pc with radio and store into file e.g.
    >./receive_train -d traindata.dat
 7. push robot through maze to collect data
    while pushing press "SEN" to pause recording press "STP"
    if finished press "Q"
 8. train a neural net with collected data
    net params are specified in npara.h
    initialize net
    >./trainer -w weights.dat -d traindata.dat -c 1 -l 1.0 -i

    train net until error is low e.g.
    >./trainer -w weights.dat -d traindata.dat -c 600 -l 1.0
    >./trainer -w weights.dat -d traindata.dat -c 400 -l 0.5
    >./trainer -w weights.dat -d traindata.dat -c 500 -l 0.1
    >./trainer -w weights.dat -d traindata.dat -c 500 -l 0.01
 9. make mlp_drive
10. dl mlp_drive.elf to eyebot
11. start program

results may vary by trainingdata and training process
in order to make it work in a narrow maze the psd setup on the robot must be
changed like in picture mod_all.jpg. the HDT for this setup is hdt_mlp_drive.c 


------------------------
Norman Apel, UWA 2004
