Robust mobile robot localisation from sparse and noisy proximity readings using Hough transform and probability grids

We present a robust position tracking method for a mobile robot with seven sonar sensors. The method is based on Hough transform and probability grids. The focus of the paper is on the problem how to handle sparse sensors and noisy data in order to develop a low-cost navigation system for real-world applications. The proposed method consists of three steps. It computes a two-dimensional feature space by applying a straight-line Hough transform to the sonar readings. The detected features are then matched with the world map as reference pattern. The correlation counts obtained in the previous step are used for updating the position probability grid. We demonstrate that this method, on the one hand, avoids the common problems of feature detection in sonar data such as erroneous lines through separate clusters, corner inference, and line artifacts through reflection. On the other hand, it achieves a robustness that dense sensor matching techniques, such as Markov localisation, can only deliver if they use a complex sensor model which takes into account the angle to the object reflecting the sonar beam.