Fast 3D Perception for Collision Avoidance and SLAM in Domestic Environments

Abstract

With a focus on navigation in dynamic and cluttered domestic environments, we have presented a sensor setup for a 3D scanner that is especially appropriate for a fast 3D perception of those regions in the robot's vicinity that are relevant for collision avoidance. The 3D scanner is continuously pitched in a nodding-like fashion where the range of rotation angles (area of interest) is adapted in size just as the angular velocity with which the scanner is rotated. The acquired 3D data is projected into the xy-plane in which the robot is moving and used to construct and update egocentric 2.5D maps storing either the coordinates of closest obstacles or environmental structures. The representation of these maps is defined in a way that