Autonomous path planning in multi-floor environments is one of the main challenges currently for indoor mobile robots. Traditional path planning method for ground mobile robots based on 2D grid map is difficult to adapt to the complex multi-floor spaces. To address this issue, this paper introduces an approach for multi-floor path planning relying on a grid-topological hybrid map and realizes the real-time construction and maintenance of a multi-floor grid-topological hybrid map in a dynamic environment. Firstly, by identifying carrier motion mode, the stair topological nodes are generated in real-time, and a multi-floor grid-topological map is constructed based on floor information and accessible area. Then, for multi-floor environments, the DFS-A* hybrid planning method is used for multi-floor path searching to achieve global path planning. Finally, the method is tested and verified in simulation and real environments, followed by result analysis. The results indicate that this method has a good applicability to unknown complex multi-floor spaces and meets the requirements of real-time construction and path searching.