Using Gaussian Process Regression for the interpolation of missing 2.5D environment modelling data

Abstract

Due to kinematic, physical, or operational constraints in terrain types, it is necessary for a mobile robot to be able to quantify the traversability characteristics of its environment to ensure safe and efficient navigation. The Gaussian Process Regression approach is a supervised method which promises to add completeness to one of the aspects of traversability analyses, namely environment modelling. This paper presents experimental results which demonstrate the effectiveness of Gaussian Process Regression in predicting the values of missing data for artificial environment features as well as actual collected point cloud data. The study concludes that when there are sufficient points the regression fits more closely to the features in the data set, with less error. Also, the prediction model produced by the Gaussian Process Regression method can be useful during robot operation to improve the terrain modelling.