Large-scale SLAM using the Atlas framework

This video shows the operation of the Atlas framework for real-time, large-scale mapping using the MIT Killian Court data set. Atlas employed graphs of coordinate frames. Each vertex in the graph represents a local coordinate frame, and each edge represents the transformation between adjacent local coordinate frames. In each local coordinate frame, extended Kalman filter SLAM (Chap. 46.3.1, Springer Handbook of Robotics, 2nd edn 2016) is performed to make a map of the local environment and to estimate the current robot pose, along with the uncertainties of each. Each map's uncertainties were modelled with respect to its own local frame. Probabilities of entities in relation to arbitrary map-frames were generated by following a path formed by the edges between adjacent map-frames, using Dijkstra's shortest path algorithm. Loop-closing was achieved via an efficient map matching algorithm. Reference: M. Bosse, P. M. Newman, J. Leonard, S. Teller: Simultaneous localization and map building in large-scale cyclic environments using the Atlas framework, Int. J. Robot. Res. 23(12), 1113-1139 (2004).