[sixo]

Ideally it would learn a relationship between the sensor input and the correct actions, even if the sensor input is not realistic for the GG in reality.