Autonomous Rock Instance Segmentation for Extra-Terrestrial Robotic Missions

Kurzfassung

The collection and analysis of extra-terrestrial matter are two of the main motivations for space exploration missions. Due to the inherent risks for participating astronauts during space missions, autonomous robotic systems are often consid- ered as a promising alternative. In recent years, many (in- ter)national space missions containing rovers to explore celestial bodies have been launched. Hereby, the communication delay as well as limited bandwidth creates a need for highly self-governed agents that require only infrequent interaction with scientists at a ground station. Such a setting is explored in the ARCHES mis- sion, which seeks to investigate different means of collaboration between scientists and autonomous robots in extra-terrestrial environments. The analog mission focuses a team of hetero- geneous agents (two Lightweight Rover Units and ARDEA, a drone), which together perform various complex tasks under strict communication constraints. In this paper, we highlight three of these tasks that were successfully demonstrated during a one-month test mission on Mt. Etna in Sicily, Italy, which was chosen due to its similarity to the Moon in terms of geological structure. All three tasks have in common, that they leverage an instance segmentation approach deployed on the rovers to detect rocks within camera imagery. The first application is a map- ping scheme that incorporates semantically detected rocks into its environment model to safely navigate to points of interest. Secondly, we present a method for the collection and extraction of in-situ samples with a rover, which uses rock detection to localize relevant candidates to grasp. For the third task, we show the usefulness of stone segmentation to autonomously conduct a spectrometer measurement experiment. We perform a throughout analysis of the presented methods and evaluate our experimental results. The demonstrations on Mt. Etna show that our approaches are well suited for navigation, geological analysis, and sample extraction tasks within autonomous robotic extra-terrestrial missions.