Correlation-based Map Merging for Multi-Robot SLAM

Abstract

Multi-robot simultaneous localization and mapping (SLAM) is an advanced technique used by multiple robots and autonomous vehicles to build up a collective map within an unknown environment, or to update a collective map within a known environment, while at the same time keeping track of their current location. The collective map is obtained by merging individual maps built by different multiple robots exploring the environment. When robots do not know their initial poses one another, the problem of map merging becomes challenging because the robots have different coordinate systems. If robot-to-robot measurements are not available, the problem of map merging becomes more challenging because the map transformation matrix (MTM) among robots cannot be computed directly. This dissertation presents novel map merging techniques based on the analysis of the correlation among the individual maps, which do not need the knowledge of the relative initial poses of robots and the robot-to-robot measurements. After the cross-correlation function among the spectrometric or tomographic information extracted from the individual maps is generated, the MTM is computed by taking the rotation angle and the translation amounts corresponding to the maximum cross-correlation values. The correlation-based map merging techniques with spectral information presented in this dissertation are the extensions of a conventional map merging technique. One extension is spectrum-based feature map merging (SFMM), which extracts the spectral information of feature maps from virtual supporting lines and computes the MTM by matching the extracted spectral information. The other extension is enhanced-spectrum-based map merging (ESMM), which enhances grid maps using the locations of visual objects and computes the MTM by matching the spectral information extracted from the enhanced grid maps. The two extensions overcome successfully the limitation of the conventional map merging technique. The correlation-based map merging technique with tomographic information is a new map merging technique, which is named tomographic map merging (TMM). Since tomographic analysis can provide more detailed information on grid maps according to rotation and translation than spectral analysis, the more accurate MTM can be computed by matching the tomographic information. The TMM was tested on various pairs of partial maps from real experiments in indoor and outdoor environments. The improved accuracy was verified by showing smaller map merging errors than the conventional map merging technique and several existing map merging techniques.