Traditionally the problem of finding the pose of a sensor in an unknown environment (a.k.a. SLAM: Simultaneous Localisation and Mapping) is done with very low-level representations such as keypoints. We are exploring new ways to bring domain knowledge into SLAM using higher-level priors such as objects, their properties and relationships to allow more meaningful predictions while saving computing resources.

Publications

elastic

ElasticFusion: Dense SLAM Without A Pose Graph, Thomas Whelan, Stefan Leutenegger, Renato F. Salas-Moreno, Ben Glocker, and Andrew J. Davison. Robotics: Science and Systems (RSS), July 2015. PDF File
thesis

Dense Semantic SLAM, Renato F. Salas-Moreno, PhD Thesis. Imperial College London, October 2014. PDF File
dps

Dense Planar SLAM, Renato F. Salas-Moreno, Ben Glocker, Paul H. J. Kelly and Andrew J. Davison. International Symposium on Mixed and Augmented Reality (ISMAR), IEEE, September 2014. Project Page, PDF File
slam++ SLAM++: Simultaneous Localisation and Mapping at the Level of Objects, Renato F. Salas-Moreno, Richard A. Newcombe, Hauke Strasdat, Paul H. J. Kelly and Andrew J. Davison. Computer Vision and Pattern Recognition (CVPR), IEEE, June 2013. Project Page, PDF File