Two-Dimensional Mapping of Expansive Unknown Areas

Abstract

In field robotics, there is a need for maps. In some workspaces, the appropriate maps exist or can be easily created by a person. However there are many workspaces for which maps do not exist and map creation is not humanly possible or practical. The problem addressed in this thesis is for a robot to model the bounds of its workspace with no external reference. The model must be complete, detailed, and concise to support navigation and planning. Previous systems have addressed a subset of these requirements, but none is capable of mapping an expansive unknown area. CLUE (Cartographer for large Unknown Environments) combines overlapping ladar scans from many different viewpoints into a usable representation of the robot's environment. The map is constructed incrementally by adding one ladar c=scan at a time, extending and refining the map to reflect the latesst information gathered. One of the strengths of CLUE is that it can refine an entire map until the total error in the map is minimized. CLUE handles dynamic environments by using newer ladar scans to eliminate older sections of the map that conflict with the new scan information. A progression of mapping technologies is presented, starting with pose estimation in a static surveyed area, and ending with mapping of a dynamic expansive unknown area. The underground room-and-pillar coal mining application is used as an example throughout the dissertation, though the technology developed is readily available to other tasks. The primary example of the thesis is a map of a large portion of a mine, covering about 5000 square meters and over one kilometer of map line segments, reducing 60,000 range measurements to about 3000 line segments.