Mobile robot pose estimation by qualitative scan matching with 2d range scans

Abstract

This paper presents a novel qualitative scan matching approach based on the point cluster representation of laser scans for the coordination of mobile robots in indoor environments. According to the geometric characteristics of the scanned environment, a hierarchical clustering method is proposed to split the points in a laser scan into stable clusters, with which the point distribution in the scan can be approximated by a collection of probability distributions instead of the discrete data points. Based on the proposed compact and continuous description of captured sensor data, the pairwise constraints between clusters are exploited as heuristic information for efficient data association between consecutive scans. In order to find the relative transformation between two frames, a qualitative estimate of the rotation with the maximum support and a closed-form solution for the translation are derived. Experiments in a variety of scenarios demonstrate the effectiveness of the proposed approach for robot pose estimation in indoor environments.