Real-time 6DoF localization with relative poses to walls of buildings

Abstract

This paper presents a real-time 6DoF localization method which corrects accumulative error by estimating relative poses to building walls for mobile robots in urban areas. This method exploits a fact that most of all artificial walls are built vertically. It estimates poses by not only an inertial sensor but also real-time SLAM and observations of normals from point-cloud of artificial walls for estimating absolute poses in the gravity coordinate system. Those three types of poses which are estimated by each way are combined by extended Kalman filter. To evaluate the proposed method, outdoor experiments with an actual robot were performed. They show the method keeps correcting accumulative error while the robot moves. The other experiments show how the method suppresses influence of non-vertical planes.