Summary: Among the solutions to the simultaneous localization and mapping (SLAM) problem with probabilistic techniques, the extended Kalman filter (EKF) is a very common approach. There are several approaches to deal with its computational cost, usually based on an adequate selection of features to be updated in real time, while the whole map update is delayed or processed in a background task, allowing one to map larger environments or to carry out multirobot experiments. Although these solutions are theoretically sound, there is a great lack of real experiments in large indoor environments due to several previously unknown problems derived from the geometric model of the map features and the inconsistency of the SLAM-EKF algorithm. For the first time, these problems are described and solved, and the implementation of the algorithms and solutions presented in this paper achieve excellent results in experiments in different real large indoor environments.