Deep Reinforcement Learning for a Four Degree of Freedom Robot Arm Accelerated by Human Demonstration