Using EM to Learn 3D Environment Models with Mobile Robots

Abstract

This paper describes an algorithm for generating compact 3D models of indoor environments with mobile robots. Our algorithm employs the expectation maximization algorithm to fit a lowcomplexity planar model to 3D data collected by range finders and a panoramic camera. The complexity of the model is determined during model fitting, by incrementally adding and removing surfaces. In a final post-processing step, measurements are converted into polygons and projected onto the surface model where possible. Empirical results obtained with a mobile robot illustrate that high-resolutionmodels can be acquired in reasonable time. 1. Introduction This paper addresses the problem of acquiring volumetric 3D models of indoor environments with mobile robots. A large number of indoor mobile robots rely on environment maps for navigation (Kortenkamp et al., 1998). Almost all existing algorithms for acquiring such maps operate in 2D---despite the fact that robot environments are threedime...