Efficient Parallel Methods for Deep Reinforcement Learning