Priority-objective reinforcement learning