Reinforcement Learning Acceleration through Autonomous Subgoal Discovery