Diversity-Driven Exploration Strategy for Deep Reinforcement Learning