Registration algorithms formatching laser scans in robotics application

Abstract: In this study, we compare different variations of the Iterative ClosestPoint (ICP) algorithm for the purpose of matching laser scans generatedby an indoor robot. The study is mainly focused on investigating maxi-mum difference in the viewpoint the algorithms can handle, and if it canbe used for robot-pose estimation by matching laser scan data generatedat different positions in a home. This study was carried out at Electroluxusing the robotic vacuum cleaner PUREi9 for gathering the dataset tobe used for the comparison.The ICP algorithm and its variations can achieve improved perfor-mance by fine-tuning heuristics and correspondences, which often re-quires substantial manual assistance and the tuning result often varyingcase-by-case. This study limits this fine tuning to standard parametersfor the purpose of comparing standard implementations, and focuses theresult more as a guideline toward what version and format is suitable forour use case.The result confirms the superiority of the Generalized ICP (GICP)version over the other versions compared in this report. The GICP ver-sion performed better for estimating the correct transform for both thetranslation distance and rotational distance between the point clouds.Two data formats were also compared. One with the aim to create adense point cloud and another data format with a more sparse pointcloud. Comparing the result of on these two data formats, we also testedthe implicit assumption of the ICP algorithm that the point cloud have tobe dense for the algorithm to perform well. From the result obtained, weconclude that this implicit assumption does not affect the performanceof the algorithms for our usage.Keywords:, Iterative Closest Point, ICP