Acquiring Diverse Robot Skills Via Maximum Entropy Deep Reinforcement Learning