Abstract
In this paper, we discuss a method which reconstructs a global map of the indoor environment from omni-directional view obtained at several robot locations. The omni-directional view obtained by our method is a kind of environment representation, which has both precise angular information and range estimation from the observation point. The processes to reconstruct the global map are as follows: (1) Obtaining Robot-Centered Representations (omni-directional views) at several robot locations for finding free areas in which the robot can move. (2) Obtaining local maps, represanted by Path-Centered coordinates, from omni-directional views taken at two neighbor robot locations. The relations between two neighbor robot locations are determined by active observations using feedbacks of the omni-directional views while the robot moves. (3) Building a global map (Environtment-Centered Representation) by fusing those Path-Centered Representations.
