Polygon-Based 3d Scan Registration With Dual-Robots In Structured Indoor Environments

This paper addresses the problems involved in 3D mapping in structured indoor environments. We explore a scenario where a dual-robot system jointly constructs the complete 3D map of an indoor environment. We assume that the environment is partially planar; lacking visual features and the robot odometry is unreliable. We introduce a framework to obtain the 6-degree of freedom relative pose between two partially overlapping range images acquired by the dual robots in stop-n-go fashion. Finally, the relative pose of the dual robots and their respective range images are transformed into a global frame to obtain a complete 3D map. The relative pose estimate between the two partially overlapping range maps acquired by the dual robots at any given instance is obtained in two steps, viz, a camera pose estimation algorithm that estimates the initial pose between the two range images and a 3D scan registration algorithm that further refines the pose estimate between the partially overlapping range images. The novel scan registration algorithm presented in this paper is initialized by a good pose estimate and registers two polygon sets extracted from the range images acquired by the dual robots. The experiments are intended to evaluate quantitatively the pose parameters for both the camera pose estimation algorithm and the scan registration algorithm. We experimentally validate our method with loop closure resulting in an accurate polygon map of an indoor environment.