Introduction

The mobile robot Kurt3D. [r] Digital 3D models of the environment are needed in rescue, exploration and inspection robotics, industrial automation, facility management, agriculture and architecture. Many robotic tasks require highly precise environment maps as well. Building them manually is tedious: Thrun et al. report a time of about one week hard work for creating a 2D map of the museum in Bonn for RHINO [26]. It is even harder when 3D maps are needed, and is getting nearly impossible when mapping general outdoor environments. Therefore, automatic mapping is crucial in robotics. Autonomous mobile robots equipped with 3D laser scanners are well suited for the gaging task [18]. Contrary to indoor applications, robot poses in natural outdoor environments also involve pitch, roll and elevation, turning pose estimation into a problem in six mathematical dimensions. 6D SLAM [18] considers all of these six dimensions for the robot pose while generating 3D maps. In our experiments we use the mobile robot Kurt3D (Fig. 1, and Fig. 5, left). It is equipped with a 3D laser scanner, which is built on the basis of a SICK 2D range finder by extension with a mount and a small servo motor [17,23]).

Mapping autonomously outdoor terrain adds another complication that is not present in typical indoor SLAM: Finding of drivable surfaces ahead. Indoor SLAM can typically assume that the robot may get along and extend its map wherever no obstacle is visible. This strategy has always required some care in the vicinity of staircases, but is good enough in many buildings. In outdoor terrain, actively looking for navigable surface is mandatory, as ragged ground, potholes, or steps are to be expected at any place.

This paper adds to our previous work on 6D SLAM the functionality of detecting geometrically navigable surfaces in the 3D map that is being built. This information may then be used to determine the next pose or poses to steer to. To make the paper self-sufficient, we include a brief review of the complete 6D SLAM process in which the surface interpretation process is embedded; for details of 6D SLAM, please refer to previous publications, i.e., [18,17].