Merging topological data into kalman based slam

This paper presents an application of a well-known SLAM algorithm, based on an augmented state Kalman estimator, to self-localise the robot and builds a fuzzy gridmap of the environment at the same time. In an office-like environment, a vision system is used to single-out on the ceiling some lamps, that are considered as natural landmarks and included in the state of the filter. Information provided at each step by ultrasonic range finders is used to build the gridmap. Sonar uncertainties are modeled using the theory of fuzzy measures for its ability to highlight contradiction arising from an imperfect localisation. A rather interesting point is the use of the acquired gridmap itself (beside the lamps) as an input for the SLAM algorithm, in particular for the robot orientation. Some simulations conclude the paper and show the effectiveness of the approach