Mars Rover Exploration Combining Remote and In Situ Measurements for Wide-Area Mapping

Abstract

We present an approach to autonomous rover exploration that enables higher science productivity. We first describe a machine learning model for wide-area mineral mapping that extrapolates signatures from just a few rover measurements. We use spectroscopic data because it is diagnostic of mineral composition. Exploration productivity is improved by incorporating notions from information theory and non-myopic path planning. We recently demonstrated the success of this approach in a field experiment in which our autonomous rover Zoe mapped a well-studied region of geological interest in Nevada. In this work, we apply and extend our methodology to actual Mars data and show performance in a Mars simulation study.