Q-learning is a reinforcement learning method to solve Markovian decision problems with incomplete information. The design of reward function is an important factor that affects the learning results of Q-learning. A method to design the reward functi