Appearance-based approach to hybrid metric-topological simultaneous localisation and mapping

In this study a unified framework to carry out the simultaneous localisation and mapping of a mobile robot combining metric and topological techniques is presented. The robot moves in a real indoor environment and the algorithm makes use of the information provided by an omnidirectional camera mounted on the robot and its internal odometry. The hybrid approach consists in constructing simultaneously two maps of the environment, one metric and other topological with relationships between them which are updated in each step. The robot goes through the environment to build up a map while continuously captures images. To build the topological map the most relevant information from the scenes is extracted using a global appearance descriptor. A new node is added to the map when the appearance between two images is sufficiently different. Also, the authors check if there is a loop closure with a previous node. At the same time, a metrical map of the environment is computed. With this aim, the authors estimate the position of the robot when it captures a new image using a Monte-Carlo algorithm. The authors show how it is possible to obtain a reasonable performance both in time and accuracy in an indoor environment, when the involved parameters are properly tuned.