Efficient Navigation Based on the Landmark-Tree Map and the Zinf Algorithm Using an Omnidirectional Camera

Abstract

Map based navigation is a crucial task for any mobile robot. On many platforms this problem is addressed by applying Simultaneous Localization and Mapping (SLAM) based on metric grid-maps. Such solutions work well on robots with adequate resources and limited workspaces. Platforms with limited payload which operate in unbounded workspaces, do often have insufficient resources to keep a metric world representation. Nevertheless, many applications demand that the robot can autonomously navigate between different operation areas. In this work the Landmark-Tree map (LT-map), a resource efficient topological map concept, is for the first time applied to a mobile robotic platform equipped with an omnidirectional camera. It enables the robot to efficiently adapt the acquired map online to the available memory. During map acquisition and navigation the motion is estimated by the Zinf-algorithm. Both methods are based on similar concepts, which results in a mutual benefit. An efficient navigation strategy based on the LT-map allows the robot to reliably follow previously recorded paths. The presented approach is evaluated on a mobile robot in indoor and outdoor scenarios. The experiments prove its feasibility and show that pruning the map just smooths the trajectories, which is the expected and desired behaviour.