Near-OptimalGoal-Oriented Reinforcement LearninginNon-StationaryEnvironments