Small bodies non-uniform gravity field on-board learning through Hopfield Neural Networks

Pasquale, Andrea; Silvestrini, Stefano; Capannolo, Andrea; Lunghi, Paolo; Lavagna, Michèle

doi:10.1016/j.pss.2022.105425

Cited by 9 publications

(10 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Comparison with existing methods. The use of machine learning to represent the gravity field of small bodies has been the subject of two recent works 38,39 . The work of ref.…”

Section: Resultsmentioning

confidence: 99%

See 1 more Smart Citation

Geodesy of irregular small bodies via neural density fields

Izzo

Gómez

2022

Commun Eng

View full text Add to dashboard Cite

Asteroids’ and comets’ geodesy is a challenging yet important task for planetary science and spacecraft operations, such as ESA’s Hera mission tasked to look at the aftermath of the recent NASA DART spacecraft’s impact on Dimorphos. Here we present a machine learning approach based on so-called geodesyNets which learns accurate density models of irregular bodies using minimal prior information. geodesyNets are a three-dimensional, differentiable function representing the density of a target irregular body. We investigate six different bodies, including the asteroids Bennu, Eros, and Itokawa and the comet Churyumov-Gerasimenko, and validate on heterogeneous and homogeneous ground-truth density distributions. Induced gravitational accelerations and inferred body shape are accurate, resulting in a relative acceleration error of less than 1%, also close to the surface. With a shape model, geodesyNets can even learn heterogeneous density fields and thus provide insight into the body’s internal structure. This adds a powerful tool to consolidated approaches like spherical harmonics, mascon models, and polyhedral gravity.

show abstract

“…Comparison with existing methods. The use of machine learning to represent the gravity field of small bodies has been the subject of two recent works 38,39 . The work of ref.…”

Section: Resultsmentioning

confidence: 99%

“…The work of ref . 39 proposes the use of a Hopfield network to represent and learn on-board the spherical harmonic coefficients. Unlike GeodesyNets, such a representation, useful for the use of preliminary navigational models, is subject to the same convergence concerns as any model based on spherical harmonics.…”

Section: Resultsmentioning

confidence: 99%

Geodesy of irregular small bodies via neural density fields

Izzo

Gómez

2022

Commun Eng

View full text Add to dashboard Cite

show abstract

“…The formulation of the network was due to Hopfield [42], but the formulation by Abe [43] is reportedly the most suited for combinatorial optimization problems [44], which are of great interest in the space domain. For this reason, here the most recent architecture is reported.…”

Section: Hopfield Neural Networkmentioning

confidence: 99%

“…where the right-hand term is expressed in a compact form, with s, the vector of s neuron states, and b, the bias vector. A remarkable property of the network is that the trajectories always remain within the hypercube [−c i , c i ] as long as the initial values belong to the hypercube too [44,45]. For implementation purposes, the discrete version of the HNN is employed, as was done in [44,46].…”

Section: Hopfield Neural Networkmentioning

confidence: 99%

Deep Learning and Artificial Neural Networks for Spacecraft Dynamics, Navigation and Control

Silvestrini

Lavagna

2022

Drones

View full text Add to dashboard Cite

The growing interest in Artificial Intelligence is pervading several domains of technology and robotics research. Only recently has the space community started to investigate deep learning methods and artificial neural networks for space systems. This paper aims at introducing the most relevant characteristics of these topics for spacecraft dynamics control, guidance and navigation. The most common artificial neural network architectures and the associated training methods are examined, trying to highlight the advantages and disadvantages of their employment for specific problems. In particular, the applications of artificial neural networks to system identification, control synthesis and optical navigation are reviewed and compared using quantitative and qualitative metrics. This overview presents the end-to-end deep learning frameworks for spacecraft guidance, navigation and control together with the hybrid methods in which the neural techniques are coupled with traditional algorithms to enhance their performance levels.

show abstract

“…On the other hand, recent advancement in research demonstrate the use of Artificial Intelligence for different tasks in the space domain, from navigation [5,6] to formation flying guidance and control [7,8,9]. In this work, the development of a vision-based navigation system using AI to solve the task of pinpoint landing on the Moon is presented.…”

Section: Introductionmentioning

confidence: 99%