Towards Adjustable Autonomy for the Real World