Towards Removing Artificial Landmarks for Autonomous Exploration in Structured Environments

This research takes one step towards using naturally-occurring and automatically detected features present in a structured environment for map building. We present a simple real-time algorithm for solving the vision task of extracting interesting cues from the environment, to be used as landmarks. The landmarks are used in a simultaneous localization and mapping (SLAM) algorithm based on an extended Kalman filter. The natural feature based map is embedded in a topological framework to map a large office environment on the order of 4000 m . This map is later used for global localization of the robot. Experimental results on a mobile robot are included. We outline a philosophy of good landmark while deriving an analytical expression of sparseness between two landmarks for good map building.