An active visual loop closure detection and validation system for topological SLAM

Abstract

Loop closing is a vital component for mobile robot navigation without a priori information of the environment, since the mobile robot has to explore, build and at the same time maintain a globally consistent map. Moreover, it allows the mobile robot to recover from positional drifts due to errors associated with sensor measurements once loop closing is performed. In this paper, an active loop closing detection and validation system for a fully autonomous vision-based mobile robot performing topological SLAM is presented. It integrates an appearance-based mapping and localization system with the path planner to achieve an active loop closing system. This system does not only detect loop closures but actively validates them by taking advantage of the tight integration with the path planner. Experimental results conducted in a semi-outdoor environment validate the proposed system.