We present a neural network that learns to control approach and avoidance behaviors in a mobile robot using the mechanisms of classical and operant conditioning. Learning, which requires no supervision, takes place as the ...