Environment Reconstruction for Robot Navigation

The United State Department of Energy (DOE) is facing a large task in characterizing and remediating waste tanks and their contents. Because of the hazardous materials inside the waste tanks, all of the work must be done remotely. The purpose of this paper is to show how to reconstruct an enclosed environment from various scans of a Laser Range Finder (LRF). The reconstructed environment can then be used by a robot for path planning, and by an operator to monitor the progress of the waste remediation process. Environment reconstruction consists of two tasks: image processing and laser sculpting. The image processing task focuses first on reducing the quantity of low-confidence data and on smoothing random fluctuations in the data. Then the processed range data must be converted into an XYZ Cartesian coordinate space, a process for which we examined two methods. The first method is a geometrical transform of the LRF data. The second uses an artificial neural network to transform the data to XYZ coordinates. Once an XYZ data set is computed, laser sculpting can be performed. Laser sculpting employs a hierarchical tree structure formally called an octree. The octree structure allows efficient storage of volumetric data and the ability to fuse multiple data sets. Our research has allowed us to examine the difficulties of fusing multiple LRF scans into an octree and to develop algorithms for converting an octree structure into a representation of polygon surfaces.