Autonomous mobile robots need to use spatial information about the environment in order to effectively plan and execute navigation tasks. This information can be represented at different level of abstractions, ranging from detailed geometric maps to coarse topological maps. Each level is adequate for some sub-tasks, but not for others. In this paper, we propose to use hybrid maps, patchworks of local metric maps connected into a topological network. Our hybrid maps are peculiar in that they use fuzzy sets to represent the uncertainty that affects metric information. We show how a robot can build hybrid maps from sensor data, and how it can use them in autonomous indoor navigation. We also report experiments performed on a real mobile robot that demonstrate the robustness of our approach with respect to inaccuracies in the map and noise in the sensor data.