Robust and Scalable Indoor Robot Localization Based on Fusion of Infrastructure Camera Feeds and On-Board Sensors

Abstract

In this paper, we propose a method to quantify the spatially-varying uncertainty associated with external-camera-based pose estimates of autonomous mobile robots in an indoor setting. We build an observation model for the camera based on an estimate of an upper bound on the uncertainty and demonstrate through experiments how it can be used to fuse information from an on-board lidar to arrive at accurate localization of the robot in feature-poor or changing environments.