Abstract: A method includes generating a local map of a real environment, the local map being defined by first spatial relationships between first feature descriptors corresponding to a visible feature in the real environment captured by a device. The device receives a downloaded map defined by second spatial relationships between an anchor point and second feature descriptors corresponding to visible features captured by another device, wherein the anchor point corresponds to a location of a virtual object. The local map is updated by merging the downloaded map with the local map based on a comparison between the first feature descriptors and the second feature descriptors and a pose of the device is determined relative to a particular feature descriptor in the updated local map. Virtual content is rendered based on the pose and one or more spatial relationships linking the particular feature descriptor and the anchor in the updated local map.