Improved Deep Reinforcement Learning for Robotics Through Distribution-based Experience Retention