The image processing of Milani: challenges after DART impact

Abstract

Milani is a 6U CubeSat that is designed to visit the Didymos binary system in early 2027 as part of the Hera mission. Milani is expected to provide new insights into the Didymos system as well as demonstrate the capabilities of CubeSat technology for interplanetary missions. Its objectives are both scientific and technological: to study and characterize the asteroid environment, estimate its gravity field, characterize the dust environment, demonstrate inter-satellite communication capabilities with Hera, and demonstrate the use of CubeSat technologies in deep space. The image processing pipeline of Milani is a crucial part of the mission as it allows to extract valuable information from the images of the system, in particular about the primary body called Didymos, that is used as the main target for navigation purposes while the secondary body, Dimorphos, acts as disturbing effect on the images. The image processing algorithm is supported in its most advanced configuration by a global data-driven method that updates itself through new datasets of images. The image processing pipeline uses coefficients that need tuning through data obtained at different phases but also from different missions. The pipeline is designed to extract low-level and high-level observables such as the center of figure, range, and phase angle, which will be used by the on-board guidance, navigation, and control subsystem. This design has been chosen to allow the algorithm to accommodate variations throughout the mission duration without the need for algorithm updates. This particular choice has been made since the Didymos system will be visited twice before Milani's release, a unique opportunity that leaves some space for non-invasive improvements of the image processing algorithms before deployment. First, the system has been preliminarily investigated by the DART spacecraft in late 2022, which impacted Dimorphos, successfully altering its orbital motion around Didymos. Second, the system will be characterized by Hera itself, the mothercraft carrying Milani and Juventas CubeSat, which will study the asteroids for some months before Milani's release. Both data and images from these missions as well as from Milani itself can be used to generate meaningful datasets. The former can be used to define an artificial environment and generate synthetic images, the latter could be used directly, making sure proper image transformations are applied to fit with Milani camera's properties. In this work, we analyze the challenges of the image processing algorithm of Milani after the impact of the DART spacecraft on Dimorphos. In particular, the focus is set on the differences between the expected shape models of the asteroids used during the design phase of the algorithm and the observed ones and how these will impact the performance of the image processing. Indeed, it has been observed that both bodies are more flattened than previously thought and this seems to put an important stress on the current design. One of the preliminary results indicates that the oblateness of Dimorphos exceeds the range previously considered during the design, invalidating some of the assumptions of the fitting function used to estimate the phase angle with the eccentricity of the blobs of pixel of Didymos. Such an issue would require a new fitting function or a change of features considered for the evaluation. Both tasks however are not trivial and different strategies exist that can be used to perform the change. Among the strategy, the possibility to use feature-based neural networks and symbolic neural networks is explored to find meaningful relationships that can estimate the phase angle with the updated shape of Didymos.