Applied Partitioned Ordinary Kriging for Online Updates for Autonomous Vehicles

Abstract

Autonomous vehicles for exploration purposes are often limited by energy and computation capacity. Usually they are tasked with the goal of efficiently and optimally exploring a given region of space. Tasks involving path planning and spatial estimation can require computation time with exponential growth versus the number of measurements taken. This creates a problem if the number of measurements is large. This paper outlines an experiment to compare a spatial estimation method, ordinary kriging with a proposed method, partitioned ordinary kriging (POK) using real environmental data measured by a remote-operated autonomous surface vehicle (ASV). The ASV collected depth measurements of a small body of water, mapped to its GPS location while under remote-control. The mean absolute error (MAE) and computation time were compared as the number of measurements increased. The POK method demonstrated favorable error and computation time compared to ordinary kriging.