In this paper, an improved A* algorithm is proposed to solve a path planning problem on a multiple AGVs (automatic guided vehicles) system in unmanned warehouse with bottom passable shelves. The multiple AGV system is prone to conflict, collision and deadlock in the process of carrying. Therefore, the A* algorithm is improved by introducing multi-valued grid and traffic rules to solve the path planning problem for the multiple AGV system. To improve handling efficiency of the multiple AGV system, the planned path is optimized by adding shelf recall mechanism, turning cost and heat cost. A binary heap data structure is also used to improve calculation speed of the path planning. Finally, the improved A* algorithm and the path optimization strategy are simulated and verified by a visual quadrilateral grid unmanned warehouse model built by Python.