Abstract: Aiming at the problem of low efficiency and slow learning in discrete state of Q-Learning algorithm．The improved algorithm adds a learning process on the basis of the original algorithm，and makes deep learning of the environment．An improved Q-Learning algorithm is proposed to simulate in grid environment． It has been successfully applied to the path planning of a mobile robot in a multi barrier environment，and the results prove the feasibility of the algorithm． The improved Q-Learning algorithm can converge faster，reduce the number of learning，and increase the efficiency by 20%． The framework of the algorithm has strong generality for solving the same kind of problems．

Key words: path planning, improved Q-Learning algorithm, reinforcement learning, grid method, robot