Implementation of a Navigation Solution for an Experimental Rover Platform, Extendable to Multiple Agents

Abstract

Navigation of rovers in planetary exploration is challenging due to the characteristics of the terrain. For localization, rovers cannot rely on external services but are limited to the use of on-board sensors. The navigation precision of rovers can be enhanced by collaboration of multiple units. We present a single-rover navigation solution, an application specifically designed for an experimental rover platform and an Intel D400 series RGB-D sensor for performing visual SLAM using ORB-SLAM2. Deployed on the rover platform, we perform an evaluation of the application in a planetary exploration laboratory and compare the achieved localization accuracy of our trajectory estimation against benchmarks of state-of-the-art SLAM solutions and experiments performed on a planetary surface analogue environment on Mt. Etna. While we cannot achieve benchmarking accuracies, our application can attain comparable results to the experiments on Mt. Etna. For multi-agent collaboration, we discuss a map data compression method that uses octrees for representing point cloud data and introduce a compact binary representation. Further, we propose a message protocol for exchanging map data among agents. Our approach does not exchange entire maps but only requested subsets of data using a grid-based map representation. Finally, we introduce a system architecture to deploy on each agent for implementing this message protocol.