Online inductive learning from answer sets for efficient reinforcement learning exploration