Introduction

Rescue robotic systems are designed to assist rescue workers in earthquake, fire, flooded, explosive and chemical disaster areas. Currently, many robots are manufactured, but most of them lack a reliable mapping method. Nevertheless, a fundamental task of rescue is to localize injured persons and to map the environment. To solve these tasks satisfactorily, the generated map of the disaster environment has to be three-dimensional. Solving the problem of simultaneous localization and mapping (SLAM) for 3D maps turns the localization into a problem with six degrees of freedom. The $x$, $y$ and $z$ positions and the roll, yaw and pitch orientations of the robot have to be considered. We are calling the resulting SLAM variant 6D SLAM [10].

This paper addresses the problem of creating a consistent 3D scene in a common coordinate system from multiple views. The proposed algorithms allow to digitize large environments fast and reliably without any intervention and solve the 6D SLAM problem. A 360$^\circ$ 3D laser scanner acquires data of the environment and interprets the 3D points online. A fast variant of the iterative closest points (ICP) algorithm [3] registers the 3D scans in a common coordinate system and relocalizes the robot. The registration uses a forest of approximate $k$d-trees. The resulting approach is highly reliable and fast, such that it can be applied online to exploration and mapping in RoboCup Rescue.

The paper is organized as follows: The remainder of this section describes the state of the art in automatic 3D mapping and presents the autonomous mobile robot and the used 3D scanner. Section describes briefly the online extraction of semantic knowledge of the environment, followed by a discussion of the scan matching using forests of trees (section ). Section presents experiments and results and concludes.