Aiming at the problems of long path planning time and more turns in the planned path of automatic guided vehicle used in customized furniture production and packaging workshop, a path planning algorithm based on improved Q-learning algorithm is proposed. According to the environment of customized furniture production and packaging workshop, the grid method was used to model the workshop. The artificial scalar field of the workshop environment was established to make purpose of the early search of the learning layer Q-learning algorithm more clear, and the turning penalty was added to the reward function, so as to quickly plan a short path with less turns. Simulation results show that the iterations of the improved algorithm reduced by 70.46%, 64.40%, 67.75%, and 30.49%, respectively compared with standard Q-learning algorithm, learning layer Q-learning, artificial potential field Q-learning, and deep double Q network algorithm. Turning times reduced by 80%, 80.95%, 83.33%, 73.33% respectively. The improved Q-learning algorithm can effectively improve the smoothness of the path, reduce the time of path planning, and improve the efficiency of the automatic guided vehicle.