Learning-Driven Exploration for Reinforcement Learning