P2022XF72W - Combining mAchine Learning and optImization for Planetary remote Sensing missiOns (CALIPSO)

Robotic planetary missions constitute the only way for humanity to explore solar system bodies beyond the orbit of the Moon. Hindered by long time delays between transmission of commands, actual execution, and reporting of their outcome, their operations are planned sometimes months in advance, through a series of refinements and trade-offs to achieve mission goals. Given the diversity of requirements and capabilities of different sensors, no established procedures or common software exist to effect the selection and planning of observation opportunities. Starting from these considerations and the specific case study of the space experiment MARSIS on board ESA's Mars Express spacecraft, the goals of Combining mAchine Learning and optImization for Planetary remote Sensing missiOns (CALIPSO) project are: (i) to analyze the entire dataset of MARSIS, using data analysis and machine learning (ML) techniques; (ii) to derive a predictive model of MARSIS performance; (iii) to develop scheduling algorithms to effectively plan its future observations; (iv) to devise an integrated suite of tools that can be used in the planning and management of future planetary remote sensing missions. The initial analysis will explore in detail a large dataset, which has so far remained under-exploited due to lack of manpower, and will provide insights in the physical and instrumental parameters affecting performance, potentially detecting also interesting connections amongst observed variables. This would make a groundbreaking contribution of the CALIPSO project to the field of astrophysics. Besides the main contribution to Universe Sciences, the project will also significantly affect the research field of ML and Operation Research (OR). Specifically, ML algorithms will be used to predict the quality of MARSIS observations from variables available at the time of planning the operations. This predictive capability will be key in planning future observations of the experiment, and to achieve its overarching goal to map the distribution of liquid water beneath the Martian South polar cap through OR techniques. Indeed, the project exploits OR methodologies to provide substantial support to the generation of effective plans of the observations performed by MARSIS, by developing flexible and comprehensive optimization algorithms. As for ML, we also aim to exploit interpretable AI techniques to understand which are the factors that affect the most the quality of a MARSIS observation and, therefore, which make such observation interesting. All these methodologies and techniques will be integrated into a suite of tools, to be used by researchers working on the planning of MARSIS experiment and on planetary remote sensing missions in general. Indeed, the CALIPSO project plans to test the developed tools also on future missions’ scenarios and provide a flexible approach to effectively support the space exploration industry for data analysis and mission management.