Experiment 1: see if a Simple Recurrent Network can learn to steer the robot from one end to the other, alternating the direction of the turn at the end of the hallway. For example, if the robot is trained to always turn left at the "north" end of the hallway, and turn right at the south end, can the network learn to perform the task? Inputs to the network are just the front 8 sonar readings, and the ouput of the network is the rotation and translation actions for the robot.
#################### shorthall.world:
resolution 0.02
include "pioneer.inc"
#GUI settings
window
(
size [357.000 489.000]
center [1.000 2.000]
scale 0.011 # 1 pixel = 0.01 meters
)
model
(
size [2.0 4.0]
color "white"
pose [1.000 2.000 0.000]
)
pioneer2dx
(
color "red"
name "Pioneer"
port 6665
pose [1.000 2.000 90.000]
)
####################