Recursive Bayesian updates for occupancy mapping and surface reconstruction

This paper proposes a new method to build two kinds of map representations, occupancy maps and surface meshes, in a single framework of Gaussian processes and update recursively using Bayesian Committee Machines. Previously, Gaussian processes were applied to robotic mapping as a batch process considering all the observations at once. However, that approach not only increases the number of training data, which is critical to the time complexity of Gaussian processes, but also is not able to update the final map with new observations. Therefore, we propose to recursively update Gaussian process maps using Bayesian Committee Machines based on the static world assumption. We demonstrate our method with a real dataset and compare the accuracy and run time with OctoMaps. Experimental results confirm that our method successfully works with a sequence of observations. Our method is slower than OctoMaps but generates more accurate occupancy maps as well as surface meshes without additional cost of computation.