Privileged Information Dropout in Reinforcement Learning