The Autonomous Mobile Robot and the AIS 3D Laser Scanner

The ARIADNE robot (fig. 1) is an industrial robot and about 80 cm $\times$ 60 cm large and 90 cm high. The mobile platform can carry a payload of 200 kg at speeds of up to 0.8 m/s. The core of the robot is a Pentium-III-800 MHz with 384 MB RAM for controlling the AIS 3D laser range finder. An embedded PC-104 system is used to control the motor. The 3D laser scanner [7] is built on the basis of a 2D laser range finder by extension with a mount and a servomotor. The 2D laser range finder is attached to the mount for being rotated. The rotation axis is horizontal (pitch). A standard servo is connected on the left side (fig. 1) and is controlled by a laptop running RT-Linux [7].

Figure 1: Left: The Ariadne robot platform equipped with the 3D scanner. Right: The AIS 3D laser scanner.

The area of $180^{\circ}\mbox{(h)} \times 120^{\circ}\mbox{(v)}$ is scanned with different horizontal (181, 361, 721) and vertical (128, 256) resolutions. A plane with 181 data points is scanned in 13ms by the 3D laser range finder, that is a rotating mirror device. Planes with more data points, e.g. 361, 721 duplicate or quadruplicate this time. Thus a scan with 181 $\times$ 256 data points needs 3.4 sec. In addition to the distance measurement the 3D laser range finder is capable of quantifying the amount of light returning to the scanner. Fig. 2 (left) shows the hall of castle Birlinghoven wheras each voxel has an intensity value. This scene is used throughout the paper.

Figure 2: 3D range scan in the main hall of castle Birlinghoven. Left: The scene is shown (see fig. 5 (top row, third image) for the reflectance image). Middle: Line extraction and surface approximation. Right: Computed bounding boxes around objects.

The basis of the map building and planing module are algorithms for reducing points, line detection, surface extraction and object segmentation. Descriptions of these algorithms can be found in [7]. While scanning a scene, lines are detected in every scanned slice. These lines are merged into surfaces and are the basis of object segemntation, which marks occupied space. Fig. 2 shows the result of these algorithms.

Several 3D scans are necessary to digitalize environments without occlusions. To create a correct and consistent model, the scans have to merged in one coordinate system. This process is called registration. Variants of the iterative closest points algorithm [8] are used to calculate a rotation $\MR$ and translation $\Vt$, which aligns the 3D scans. This transformation also corrects the estimated robot pose [9]. The 3D digitalization and map building is a stop, scan, plan and go setting. The next section describes the next best view planning module.