Mobile Robot 3D Perception and Mapping without Odometry Using Multi-Resolution Occupancy Lists

摘要

Many real-world applications require mobile robots to have 3D perception and mapping capabilities. This paper proposes a novel mechanism for augmenting a traditional 2D laser range finder to produce almost omni-directional 3D scans at degree resolution in 3 seconds. Also introduced is a multi-resolution sampling based algorithm for aligning generated range data stored as occupancy lists. This algorithm can align a sequence of scans to produce maps and localize within a prior global map. A 200 square meter office environment is mapped at 0.02 m resolution, resulting in a 200,000 voxel occupancy list. Global localization within this entire map with no prior pose estimate is done in 5 seconds on a 2 GHz processor.