Recovering Particle Diversity

Abstract

We presented in Chapter 7 a technique that allows an exploring robot to detect loops and to carry out place revisiting actions. We showed that the quality of a map constructed by a mobile robot depends on its trajectory during data acquisition. This is due to the fact that the vehicle needs to relocalize itself during exploration in order to build an accurate model of the environment. Our loop-closing technique uses a heuristic stopping criterion in order to continue with the new terrain acquisition and to reduce the risk that the particle depletion problem affects the filter. We showed that such an approach works well in practice, however, the general problem of particle depletion in the context of loop-closing still exists. To overcome this limitation, we present in this chapter a technique that is able to approximatively recover lost hypotheses. It enables a mobile robot to store the particle diversity of the filter before entering a loop. When leaving the loop, the robot is then able to recover that diversity and continue the exploration process. This technique allows a mobile robot to stay - at least theoretically - arbitrarily long in a loop without depleting hypotheses needed to close an additional, outer loop.