Quantum reinforcement learning: the maze problem