Simultaneous localization and map-building in outdoors semi-structured environments

This paper addresses the issue of Simultaneous localization and map-building (SLAM) in large scale outdoor environments. A dead-reckoning scheme appropriate also for skid-steered mobile robots is used. A line feature-based map is constructed on-the-fly and assists in a better estimation for the robot pose. The dead-reckoning is based on internal sensors (inertial data and odometry) only. The information from an experimentally derived kinematic model and an onboard Inertial Navigation System (INS) is fused using a simple and fast maximum likelihood formulation to produce estimates for the linear and angular velocity of the robot. The map is constructed using two laser sensors located to cover a full 360⁰ sector around the robot. A second estimate for the robot pose is derived from a map matching algorithm. The two estimates are fused together using a covariance intersection filter. We verify our approach with large scale experiments in outdoors structured environments following paths with steep turns and variable velocity.