Gaussian beam processes: A nonparametric Bayesian measurement model for range finders

Abstract

In probabilistic mobile robotics, the development ofmeasurement models plays a crucial role as it directly influencesthe efficiency and the robustness of the robot's performancein a great variety of tasks including localization, tracking, andmap building. In this paper, we present a novel probabilisticmeasurement model for range finders, called Gaussian beamprocesses, which treats the measurement modeling task as anonparametric Bayesian regression problem and solves it usingGaussian processes. The major benefit of our approach is itsability to generalize over entire range scans directly. This way,we can learn the distributions of range measurements for wholeregions of the robot's configuration space from only few recordedor simulated range scans. Especially in approximative approachesto state estimation like particle filtering or histogram filtering,this leads to a better approximation of the true likelihoodfunction. Experiments on real world and synthetic data showthat Gaussian beam processes combine the advantages of twopopular measurement models.