Diffusion Mapping

Frank Dellaert

NEW: see our paper:  Intrinsic Localization and Mapping with 2 Applications:Diffusion Mapping and Marco Polo Localization, Frank Dellaert, Fernando Alegre, and Eric Beowulf Martinson, submitted to ICRA 2003:

Diffusion Mapping is an approach where a highly redundant team of simple robots is used to map out a previously unknown environment, simply by virtue of recording the localization and line-of-sight traces, which provide a detailed picture of the navigable space. The pictures below show a simulated example where 15 robots are released on the left and execute a pure random walk control strategy in a large environment, except that they re- flect off walls. Shown are the traced trajectories at regular time intervals between 0 and 1000 steps, which collectively constitute a map of the empty space, and hence of the navigable environment. Gray lines indicate recorded lines of sight, which complement the trajectory information:

Frank Dellaert Home
Computational Perception Lab

Frank Dellaert's Homepage