Evaluation of Point Cloud Registration Algorithm for Global Map Generation in Multi Robot Slam in Uneven Terrain

Abstract

With the increase in space and extraterrestrial exploration mission, one can tell that knowing the topography of the unknown environment becomes a crucial step and acts as a platform for other research and experiments. A manned mission to an unknown environment is hazardous. So the use of unmanned ground vehicles is a better alternative as it doesn't put human life in danger.During global map generation in multi robot slam, multiple local maps should be concatenated. Pointcloud registration algorithm plays a vital role in this scenario. However, the ICP algorithm is known to be sensitive to noise and outliers, and its performance can degrade significantly in uneven terrain. This paper presents a study of the ICP algorithm in uneven terrain. The performance of the ICP algorithms (both using kernel functions and without using kernel functions) are evaluated on synthetic data obtained from marsyard terrain. In this specific terrain use of GM loss and Cauchy kernel functions were promising.