Describing the environment using semantic labelled polylines from 2D laser scanned raw data: Application to autonomous navigation

This paper describes a real-time method that obtains a hybrid description of the environment (both metric and semantic) from raw data perceived by a 2D laser scanner. A set of linguistically labelled polylines allows to build a compact geometrical representation of the indoor location where a set of representative points (or features) are semantically described. These features are processed in order to find a list of traversable segments whose middle points are heuristically clustered. Finally, a set of safe paths are calculated from these clusters. Both the environment representation and the safe paths can be used by a controller to carry out navigation and exploration tasks. The method has been successfully tested in simulation and on a real robot.