Overcoming Sampling and Exploration Challenges in Deep Reinforcement Learning