Mattia Pugliatti

Small bodies such as asteroids and comets display great variability in terms of surface morphological features. These are often unknown beforehand but can be employed for hazard avoidance during landing, autonomous planning of scientific observations, and navigation purposes. Algorithms performing these tasks are often data driven, which means they require realistic, sizeable, and annotated datasets, which in turn may rely heavily on human intervention. This work develops a methodology to generate synthetic, automatically labeled datasets that are used in conjunction with real, manually labeled ones to train deep-learning architectures in the task of semantic segmentation. This functionality is achieved by designing U-shaped network architectures trained with different strategies. These show good generalization capabilities, implement uncertainty quantification estimates, and can be hybridized to exploit qualities from multiple networks.

In this work data-driven image processing options for a CubeSat mission around a binary asteroid system are investigated. The methods considered belong to two main branches of image processing methods: centroid and artificial intelligence. The former is represented by three variations of centroiding methods, and the latter by three neural networks and one convolutional neural network. The first contribution of this work is an enhanced center of brightness method with a data-driven scattering law. This method is demonstrated to share similarities with neural networks in terms of both design and performance, with the advantage of relying on a traditional, robust, and fully explainable algorithm. The second contribution is given by the performance assessment between the different families of image processing methods. For this purpose, the Milani mission is considered as a case study: a 6U CubeSat that will visit the Didymos system as part of the Hera mission. From this analysis, it emerges that convolutional networks perform better than other methods across all metrics considered. This hints to the importance of filtering techniques to extract spatial information from images, which is a unique feature of the convolutional approach over the other image processing methods considered.

CubeSats offer a flexible and low-cost option to increase the scientific and technological return of small-body exploration missions. ESA’s Hera mission, the European component of the Asteroid Impact and Deflection Assessment (AIDA) international collaboration, plans on deploying two CubeSats in the proximity of binary system 65803 Didymos, after arrival in 2027. In this work, we discuss the feasibility and preliminary mission profile of Hera’s Milani CubeSat. The CubeSat mission is designed to achieve both scientific and technological objectives. We identify the design challenges and discuss design criteria to find suitable solutions in terms of mission analysis, operational trajectories, and Guidance, Navigation, & Control (GNC) design. We present initial trajectories and GNC baseline, as a result of trade-off analyses. We assess the feasibility of the Milani CubeSat mission and provide a preliminary solution to cover the operational mission profile of Milani in the close-proximity of Didymos system.