Simultaneous Localization and Mapping


One of my secret weapons is possessing expertise in both 3D modeling in the field of computer vision and the simultaneous localization and mapping (SLAM) problem in robotics, two problems that share a similar mathematical formulation. I exploited this in my work on linear SLAM and Intrinsic Localization and Mapping (ILM), both advancing the state of the art in SLAM using computer vision style algorithms. My work on MCMC sampling over large discrete spaces also led to the development of a wholly new concept in SLAM: probabilistic topological maps. This very recent work enables one to build a probability distribution over topological maps rather than detailed metric maps as have been more popular. By sampling over topological maps to represent the uncertainty over them we combine the advantages of both metric maps (a sound probabilistic basis) and topological maps (scalability to large environments) in one representation. Although the space of topological maps is combinatorially large, MCMC sampling can still enable one to perform inference in these large spaces.


Relevant Publications

Related Links

Frank Dellaert Home
Computational Perception Lab

Frank Dellaert's Homepage