Learning effects in variable autonomy human-robot systems: how much training is enough?