Automatic environment sensing and modeling is a fundamental scientific issue in robotics, since the presence of maps is essential for many robot tasks. Manual mapping of environments is a hard and tedious job: Thrun et al. report a time of about one week hard work for creating a map of the museum in Bonn for the robot RHINO [25]. Especially mobile systems with 3D laser scanners that automatically perform multiple steps such as scanning, gaging and autonomous driving have the potential to greatly improve mapping. Many application areas benefit from 3D maps, e.g., industrial automation, architecture, agriculture, the construction or maintenance of tunnels and mines and rescue robotic systems.
The robotic mapping problem is that of acquiring a spatial model of a robot's environment. If the robot poses were known, the local sensor inputs of the robot, i.e., local maps, could be registered into a common coordinate system to create a map. Unfortunately, any mobile robot's self localization suffers from imprecision and therefore the structure of the local maps, e.g., of single scans, needs to be used to create a precise global map. Finally, robot poses in natural outdoor environments involve yaw, pitch, roll angles and elevation, turning pose estimation as well as scan registration into a problem in six mathematical dimensions.
This paper proposes algorithms that allow to digitize large environments and solve the 6D SLAM problem. In previous works we already presented partially our 6D SLAM algorithm [19,23,24]. In [19] we use a global relaxation scan matching algorithm to create a model of an abandoned mine and in [24] we presented our first 3D model containing a closed loop. This paper's main contribution is an octree-based matching heuristic that allows us to match scans with rudimentary starting guesses and to detect closed loops.