Grid map building based on prediction of local features

Abstract

In order to build more precise and stable map with sonar sensors, a method based on prediction of local geometrical features is proposed. The position and orientation of features in the robot's local space are predicted from reliable sonar readings, and confidence measures are assigned to each sonar reading to indicate the probability that it is reliable. The confidence values are used to weaken the effect of uncertain sonar readings on the global map, and to discard the readings completely if the confidence value is sufficiently low. Experiments carried out on Pioneer2 robot prove that the method performs well.