Striving for Simplicity in Off-policy Deep Reinforcement Learning