Model structuring and concept recognition: two aspects of learning for a mobile robot

Abstract

We present here a method for providing a mobile robot with learning capabilities. The method is based on a model of the environment with several hierarchical levels organized by degree of abstraction. The mathematical structuring tool used is the decomposition of a graph into its k-connected components (k=2 and k=3). This structure allows the robot to improve navigation procedures and to recognize some concepts, such as a door, a room, or a corridor.