Place Learning in Dynamic Real-World Environments

Abstract

In this paper, we present an approach for mobile robot localization designed for use in dynamic environments. Our approach integrates evidence grids within a topological /metric network that can be used for navigation. Place learning consists of associating evidence grids with places in the topological network. Place recognition consists of building an evidence grid at the current location and using a registration procedure based on hill climbing to find the best match between the current grid and the grids associated with places in the network. This approach has been implemented on a real mobile robot and has been tested in a real-world office environment containing multiple forms of dynamic change. In these experiments, this approach demonstrated robust localization in the presence of transient changes (such as moving people) and lasting changes (such as rearranged furniture) in the environment. 1. Introduction A central issue in mobile robotics is localization: how can a robot use...