Connectionist Learning for Control