[Paper Review] Study of the asteroid Bennu using geodesyANNs and Osiris-Rex data

Moritz von Looz, Pablo Gómez+1 more|arXiv (Cornell University)|Sep 29, 2021

Astro and Planetary Science17 references4 citations

TL;DR

This paper proposes geodesyNets, a neural network approach to infer the mass density distribution of asteroid Bennu directly from trajectory data of pebble-sized particles orbiting it, using real Osiris-Rex mission data. The method achieves gravity field fidelity comparable to a uniform-density mascon model without requiring a shape model, demonstrating feasibility for on-board autonomous gravity field inversion despite noise from non-gravitational forces.

ABSTRACT

Asteroids and other small bodies in the solar system tend to have irregular shapes, owing to their low gravity. This irregularity does not only apply to the topology, but also to the underlying geology, potentially containing regions of different densities and materials. The topology can be derived from optical observations, while the mass density distribution of an object is only observable, to some extent, in its gravitational field. In a companion paper, we presented geodesyNets, a neural network approach to infer the mass density distribution of an object from measurements of its gravitational field. In the present work, we apply this approach to the asteroid Bennu using real data from the Osiris Rex mission. The mission measured the trajectories of not only the Osiris Rex spacecraft itself, but also of numerous pebble-sized rock particles which temporarily orbited Bennu. From these trajectory data, we obtain a representation of Bennu's mass density and validate it by propagating, in the resulting gravity field, multiple pebbles not used in the training process. The performance is comparable to that of a polyhedral gravity model of uniform density, but does not require a shape model. As little additional information is needed, we see this as a step towards autonomous on-board inversion of gravitational fields.

Motivation & Objective

• To develop a method for inferring the internal mass density distribution of small bodies like asteroids from gravitational trajectory data.
• To validate a neural network-based approach (geodesyNets) on real mission data from the Osiris-Rex mission to asteroid Bennu.
• To assess whether the method can achieve high-fidelity gravity field modeling without requiring a pre-existing shape model.
• To investigate the impact of non-gravitational forces (e.g., solar radiation pressure) on the accuracy of gravity field inversion from sparse, noisy trajectory data.
• To explore the potential for on-board, autonomous inversion of gravitational fields using neural networks in future planetary science missions.

Proposed method

• Train a neural network to predict the 3D mass density function f(x,y,z) within Bennu from observed accelerations derived via numerical differentiation of pebble trajectories.
• Use the trained network to compute the gravitational acceleration at external points by integrating the density function over the volume, using a modified Newtonian gravity formulation.
• Compare predicted accelerations with observed accelerations from pebble trajectories to compute a normalized loss function for training.
• Validate the model by propagating the orbits of pebbles not used in training, comparing predicted trajectories with reported trajectories from Scheeres et al. [4].
• Implement two scenarios: non-differential (no shape model) and differential (using a shape model to infer deviations from uniform density).
• Account for non-gravitational forces such as solar radiation pressure during orbit propagation, though not explicitly modeled in the network training.

Experimental results

Research questions

• RQ1 Can a neural network infer the mass density distribution of an asteroid from sparse, noisy trajectory data of orbiting particles without prior knowledge of the shape?
• RQ2 How does the performance of the geodesyNets approach compare to traditional gravity models like mascon or spherical harmonics when applied to real mission data?
• RQ3 To what extent do non-gravitational forces such as solar radiation pressure degrade the accuracy of gravity field inversion using neural networks?
• RQ4 Does the differential approach—modeling density variations relative to a uniform background—improve fidelity over the non-differential method when applied to real Bennu data?
• RQ5 Can this method be adapted for on-board autonomous gravity field inversion in future planetary exploration missions?

Key findings

• The non-differential geodesyNets approach achieved a trajectory propagation error of 44.47 meters after 3 hours, comparable to a uniform-density mascon model (43.26 meters), despite not requiring a shape model.
• The performance was significantly better than a point-mass model, which had a 73.47-meter error after the same time, confirming the network learned meaningful gravity field structure.
• The differential approach failed to capture meaningful internal heterogeneity, likely due to high uncertainty in pebble size and mass, which amplified noise relative to actual density variations.
• Synthetic experiments showed that the method can successfully recover artificial non-uniform mass distributions when non-gravitational forces are properly modeled, indicating the method is capable in principle.
• The neural network was able to infer the overall shape of Bennu from gravitational accelerations alone, even in the absence of a shape model, suggesting robustness to missing geometric priors.
• The results suggest that extending the model to include pebble size and mass estimates could significantly improve fidelity, especially for detecting small-scale internal heterogeneity.