A Separate Data Structure for Online Multi-hypothesis Topological Mapping

This paper proposes an algorithm for topological simultaneous localization and mapping (SLAM) using multi-hypothesis method. This algorithm focuses on improving on-board computational efficiency and capability of finding out the correct hypothesis as early as possible. In the algorithm, an innovative data structure is applied, in which the edges and vertexes of the topological graph are stored separately. So that detailed information of the vertexes has only one copy in the storage, which also benefits saving communication bandwidth. Then, lots of repetitive loop-closing tests in similar hypothesizes are simplified to one single test that only uses vertexes storage. Lastly, incorporating with the data structure, loop closure situations can be evaluated as soon as it happens. In a word, the algorithm is highly efficient to cope with the hyper-exponential growth disaster caused by perceptual aliasing. The work is evaluated by simulations and demonstrated on a maze-like scenario with a Micro-Aerial Vehicle (MAV) equipped a computational resources restricted computer.