Map segmentation based SLAM using embodied data

Autonomous mobile robots offer the prospect of extending our knowledge of remote places in the solar system or in the ocean. They also have the potential to improve everyday life with ever increasing adaptability to a large variety of environments. One of the key technological elements is the ability to navigate unknown and uncooperative environments. A range of solutions for the simultaneous localisation and mapping (SLAM) problem have emerged in the last decade. One factor which is often neglected is the fact that the robot has a body which interacts with the environment. In this paper a method is presented, which utilises this information and uses visual and non-visual correlations to generate accurate local map segments. Further, a method is presented to combine particle filter based local map segments and constraint graph based global pose optimization to a single coherent map representation. The method is evaluated on a Leg/Wheel hybrid mobile robot and the resulting maps compared against high resolution environment models generated with a commercial laser scanner.