Heuristically Accelerated Q-Learning: A New Approach to Speed Up Reinforcement Learning